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">Function Structure</a></li>
29 <li><a href="#typesystem">Type System</a>
31 <li><a href="#t_primitive">Primitive Types</a>
33 <li><a href="#t_classifications">Type Classifications</a></li>
36 <li><a href="#t_derived">Derived Types</a>
38 <li><a href="#t_array">Array Type</a></li>
39 <li><a href="#t_function">Function Type</a></li>
40 <li><a href="#t_pointer">Pointer Type</a></li>
41 <li><a href="#t_struct">Structure Type</a></li>
42 <li><a href="#t_packed">Packed Type</a></li>
43 <li><a href="#t_opaque">Opaque Type</a></li>
48 <li><a href="#constants">Constants</a>
50 <li><a href="#simpleconstants">Simple Constants</a>
51 <li><a href="#aggregateconstants">Aggregate Constants</a>
52 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
53 <li><a href="#undefvalues">Undefined Values</a>
54 <li><a href="#constantexprs">Constant Expressions</a>
57 <li><a href="#instref">Instruction Reference</a>
59 <li><a href="#terminators">Terminator Instructions</a>
61 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
62 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
63 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
64 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
65 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
66 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
69 <li><a href="#binaryops">Binary Operations</a>
71 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
72 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
73 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
74 <li><a href="#i_div">'<tt>div</tt>' Instruction</a></li>
75 <li><a href="#i_rem">'<tt>rem</tt>' Instruction</a></li>
76 <li><a href="#i_setcc">'<tt>set<i>cc</i></tt>' Instructions</a></li>
79 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
81 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
82 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
83 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
84 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
85 <li><a href="#i_shr">'<tt>shr</tt>' Instruction</a></li>
88 <li><a href="#memoryops">Memory Access Operations</a>
90 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
91 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
92 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
93 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
94 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
95 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
98 <li><a href="#otherops">Other Operations</a>
100 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
101 <li><a href="#i_cast">'<tt>cast .. to</tt>' Instruction</a></li>
102 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
103 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
104 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
105 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
106 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
111 <li><a href="#intrinsics">Intrinsic Functions</a>
113 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
115 <li><a href="#i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
116 <li><a href="#i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
117 <li><a href="#i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
120 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
122 <li><a href="#i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
123 <li><a href="#i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
124 <li><a href="#i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
127 <li><a href="#int_codegen">Code Generator Intrinsics</a>
129 <li><a href="#i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
130 <li><a href="#i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
131 <li><a href="#i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
132 <li><a href="#i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
133 <li><a href="#i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
134 <li><a href="#i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
135 <li><a href="#i_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
138 <li><a href="#int_os">Operating System Intrinsics</a>
140 <li><a href="#i_readport">'<tt>llvm.readport</tt>' Intrinsic</a></li>
141 <li><a href="#i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a></li>
142 <li><a href="#i_readio">'<tt>llvm.readio</tt>' Intrinsic</a></li>
143 <li><a href="#i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a></li>
145 <li><a href="#int_libc">Standard C Library Intrinsics</a>
147 <li><a href="#i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a></li>
148 <li><a href="#i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a></li>
149 <li><a href="#i_memset">'<tt>llvm.memset</tt>' Intrinsic</a></li>
150 <li><a href="#i_isunordered">'<tt>llvm.isunordered</tt>' Intrinsic</a></li>
151 <li><a href="#i_sqrt">'<tt>llvm.sqrt</tt>' Intrinsic</a></li>
155 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
157 <li><a href="#i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
158 <li><a href="#int_ctpop">'<tt>llvm.ctpop</tt>' Intrinsic </a></li>
159 <li><a href="#int_ctlz">'<tt>llvm.ctlz</tt>' Intrinsic </a></li>
160 <li><a href="#int_cttz">'<tt>llvm.cttz</tt>' Intrinsic </a></li>
163 <li><a href="#int_debugger">Debugger intrinsics</a></li>
168 <div class="doc_author">
169 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
170 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
173 <!-- *********************************************************************** -->
174 <div class="doc_section"> <a name="abstract">Abstract </a></div>
175 <!-- *********************************************************************** -->
177 <div class="doc_text">
178 <p>This document is a reference manual for the LLVM assembly language.
179 LLVM is an SSA based representation that provides type safety,
180 low-level operations, flexibility, and the capability of representing
181 'all' high-level languages cleanly. It is the common code
182 representation used throughout all phases of the LLVM compilation
186 <!-- *********************************************************************** -->
187 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
188 <!-- *********************************************************************** -->
190 <div class="doc_text">
192 <p>The LLVM code representation is designed to be used in three
193 different forms: as an in-memory compiler IR, as an on-disk bytecode
194 representation (suitable for fast loading by a Just-In-Time compiler),
195 and as a human readable assembly language representation. This allows
196 LLVM to provide a powerful intermediate representation for efficient
197 compiler transformations and analysis, while providing a natural means
198 to debug and visualize the transformations. The three different forms
199 of LLVM are all equivalent. This document describes the human readable
200 representation and notation.</p>
202 <p>The LLVM representation aims to be light-weight and low-level
203 while being expressive, typed, and extensible at the same time. It
204 aims to be a "universal IR" of sorts, by being at a low enough level
205 that high-level ideas may be cleanly mapped to it (similar to how
206 microprocessors are "universal IR's", allowing many source languages to
207 be mapped to them). By providing type information, LLVM can be used as
208 the target of optimizations: for example, through pointer analysis, it
209 can be proven that a C automatic variable is never accessed outside of
210 the current function... allowing it to be promoted to a simple SSA
211 value instead of a memory location.</p>
215 <!-- _______________________________________________________________________ -->
216 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
218 <div class="doc_text">
220 <p>It is important to note that this document describes 'well formed'
221 LLVM assembly language. There is a difference between what the parser
222 accepts and what is considered 'well formed'. For example, the
223 following instruction is syntactically okay, but not well formed:</p>
226 %x = <a href="#i_add">add</a> int 1, %x
229 <p>...because the definition of <tt>%x</tt> does not dominate all of
230 its uses. The LLVM infrastructure provides a verification pass that may
231 be used to verify that an LLVM module is well formed. This pass is
232 automatically run by the parser after parsing input assembly and by
233 the optimizer before it outputs bytecode. The violations pointed out
234 by the verifier pass indicate bugs in transformation passes or input to
237 <!-- Describe the typesetting conventions here. --> </div>
239 <!-- *********************************************************************** -->
240 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
241 <!-- *********************************************************************** -->
243 <div class="doc_text">
245 <p>LLVM uses three different forms of identifiers, for different
249 <li>Named values are represented as a string of characters with a '%' prefix.
250 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
251 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
252 Identifiers which require other characters in their names can be surrounded
253 with quotes. In this way, anything except a <tt>"</tt> character can be used
256 <li>Unnamed values are represented as an unsigned numeric value with a '%'
257 prefix. For example, %12, %2, %44.</li>
259 <li>Constants, which are described in a <a href="#constants">section about
260 constants</a>, below.</li>
263 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
264 don't need to worry about name clashes with reserved words, and the set of
265 reserved words may be expanded in the future without penalty. Additionally,
266 unnamed identifiers allow a compiler to quickly come up with a temporary
267 variable without having to avoid symbol table conflicts.</p>
269 <p>Reserved words in LLVM are very similar to reserved words in other
270 languages. There are keywords for different opcodes ('<tt><a
271 href="#i_add">add</a></tt>', '<tt><a href="#i_cast">cast</a></tt>', '<tt><a
272 href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
273 href="#t_void">void</a></tt>', '<tt><a href="#t_uint">uint</a></tt>', etc...),
274 and others. These reserved words cannot conflict with variable names, because
275 none of them start with a '%' character.</p>
277 <p>Here is an example of LLVM code to multiply the integer variable
278 '<tt>%X</tt>' by 8:</p>
283 %result = <a href="#i_mul">mul</a> uint %X, 8
286 <p>After strength reduction:</p>
289 %result = <a href="#i_shl">shl</a> uint %X, ubyte 3
292 <p>And the hard way:</p>
295 <a href="#i_add">add</a> uint %X, %X <i>; yields {uint}:%0</i>
296 <a href="#i_add">add</a> uint %0, %0 <i>; yields {uint}:%1</i>
297 %result = <a href="#i_add">add</a> uint %1, %1
300 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
301 important lexical features of LLVM:</p>
305 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
308 <li>Unnamed temporaries are created when the result of a computation is not
309 assigned to a named value.</li>
311 <li>Unnamed temporaries are numbered sequentially</li>
315 <p>...and it also shows a convention that we follow in this document. When
316 demonstrating instructions, we will follow an instruction with a comment that
317 defines the type and name of value produced. Comments are shown in italic
322 <!-- *********************************************************************** -->
323 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
324 <!-- *********************************************************************** -->
326 <!-- ======================================================================= -->
327 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
330 <div class="doc_text">
332 <p>LLVM programs are composed of "Module"s, each of which is a
333 translation unit of the input programs. Each module consists of
334 functions, global variables, and symbol table entries. Modules may be
335 combined together with the LLVM linker, which merges function (and
336 global variable) definitions, resolves forward declarations, and merges
337 symbol table entries. Here is an example of the "hello world" module:</p>
339 <pre><i>; Declare the string constant as a global constant...</i>
340 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
341 href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
343 <i>; External declaration of the puts function</i>
344 <a href="#functionstructure">declare</a> int %puts(sbyte*) <i>; int(sbyte*)* </i>
346 <i>; Definition of main function</i>
347 int %main() { <i>; int()* </i>
348 <i>; Convert [13x sbyte]* to sbyte *...</i>
350 href="#i_getelementptr">getelementptr</a> [13 x sbyte]* %.LC0, long 0, long 0 <i>; sbyte*</i>
352 <i>; Call puts function to write out the string to stdout...</i>
354 href="#i_call">call</a> int %puts(sbyte* %cast210) <i>; int</i>
356 href="#i_ret">ret</a> int 0<br>}<br></pre>
358 <p>This example is made up of a <a href="#globalvars">global variable</a>
359 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
360 function, and a <a href="#functionstructure">function definition</a>
361 for "<tt>main</tt>".</p>
363 <p>In general, a module is made up of a list of global values,
364 where both functions and global variables are global values. Global values are
365 represented by a pointer to a memory location (in this case, a pointer to an
366 array of char, and a pointer to a function), and have one of the following <a
367 href="#linkage">linkage types</a>.</p>
371 <!-- ======================================================================= -->
372 <div class="doc_subsection">
373 <a name="linkage">Linkage Types</a>
376 <div class="doc_text">
379 All Global Variables and Functions have one of the following types of linkage:
384 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
386 <dd>Global values with internal linkage are only directly accessible by
387 objects in the current module. In particular, linking code into a module with
388 an internal global value may cause the internal to be renamed as necessary to
389 avoid collisions. Because the symbol is internal to the module, all
390 references can be updated. This corresponds to the notion of the
391 '<tt>static</tt>' keyword in C, or the idea of "anonymous namespaces" in C++.
394 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
396 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt> linkage, with
397 the twist that linking together two modules defining the same
398 <tt>linkonce</tt> globals will cause one of the globals to be discarded. This
399 is typically used to implement inline functions. Unreferenced
400 <tt>linkonce</tt> globals are allowed to be discarded.
403 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
405 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
406 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
407 used to implement constructs in C such as "<tt>int X;</tt>" at global scope.
410 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
412 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
413 pointer to array type. When two global variables with appending linkage are
414 linked together, the two global arrays are appended together. This is the
415 LLVM, typesafe, equivalent of having the system linker append together
416 "sections" with identical names when .o files are linked.
419 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
421 <dd>If none of the above identifiers are used, the global is externally
422 visible, meaning that it participates in linkage and can be used to resolve
423 external symbol references.
427 <p><a name="linkage_external">For example, since the "<tt>.LC0</tt>"
428 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
429 variable and was linked with this one, one of the two would be renamed,
430 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
431 external (i.e., lacking any linkage declarations), they are accessible
432 outside of the current module. It is illegal for a function <i>declaration</i>
433 to have any linkage type other than "externally visible".</a></p>
437 <!-- ======================================================================= -->
438 <div class="doc_subsection">
439 <a name="callingconv">Calling Conventions</a>
442 <div class="doc_text">
444 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
445 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
446 specified for the call. The calling convention of any pair of dynamic
447 caller/callee must match, or the behavior of the program is undefined. The
448 following calling conventions are supported by LLVM, and more may be added in
452 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
454 <dd>This calling convention (the default if no other calling convention is
455 specified) matches the target C calling conventions. This calling convention
456 supports varargs function calls and tolerates some mismatch in the declared
457 prototype and implemented declaration of the function (as does normal C).
460 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
462 <dd>This calling convention attempts to make calls as fast as possible
463 (e.g. by passing things in registers). This calling convention allows the
464 target to use whatever tricks it wants to produce fast code for the target,
465 without having to conform to an externally specified ABI. Implementations of
466 this convention should allow arbitrary tail call optimization to be supported.
467 This calling convention does not support varargs and requires the prototype of
468 all callees to exactly match the prototype of the function definition.
471 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
473 <dd>This calling convention attempts to make code in the caller as efficient
474 as possible under the assumption that the call is not commonly executed. As
475 such, these calls often preserve all registers so that the call does not break
476 any live ranges in the caller side. This calling convention does not support
477 varargs and requires the prototype of all callees to exactly match the
478 prototype of the function definition.
481 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
483 <dd>Any calling convention may be specified by number, allowing
484 target-specific calling conventions to be used. Target specific calling
485 conventions start at 64.
489 <p>More calling conventions can be added/defined on an as-needed basis, to
490 support pascal conventions or any other well-known target-independent
495 <!-- ======================================================================= -->
496 <div class="doc_subsection">
497 <a name="globalvars">Global Variables</a>
500 <div class="doc_text">
502 <p>Global variables define regions of memory allocated at compilation time
503 instead of run-time. Global variables may optionally be initialized, may have
504 an explicit section to be placed in, and may
505 have an optional explicit alignment specified. A
506 variable may be defined as a global "constant," which indicates that the
507 contents of the variable will <b>never</b> be modified (enabling better
508 optimization, allowing the global data to be placed in the read-only section of
509 an executable, etc). Note that variables that need runtime initialization
510 cannot be marked "constant" as there is a store to the variable.</p>
513 LLVM explicitly allows <em>declarations</em> of global variables to be marked
514 constant, even if the final definition of the global is not. This capability
515 can be used to enable slightly better optimization of the program, but requires
516 the language definition to guarantee that optimizations based on the
517 'constantness' are valid for the translation units that do not include the
521 <p>As SSA values, global variables define pointer values that are in
522 scope (i.e. they dominate) all basic blocks in the program. Global
523 variables always define a pointer to their "content" type because they
524 describe a region of memory, and all memory objects in LLVM are
525 accessed through pointers.</p>
527 <p>LLVM allows an explicit section to be specified for globals. If the target
528 supports it, it will emit globals to the section specified.</p>
530 <p>An explicit alignment may be specified for a global. If not present, or if
531 the alignment is set to zero, the alignment of the global is set by the target
532 to whatever it feels convenient. If an explicit alignment is specified, the
533 global is forced to have at least that much alignment. All alignments must be
539 <!-- ======================================================================= -->
540 <div class="doc_subsection">
541 <a name="functionstructure">Functions</a>
544 <div class="doc_text">
546 <p>LLVM function definitions consist of an optional <a href="#linkage">linkage
547 type</a>, an optional <a href="#callingconv">calling convention</a>, a return
548 type, a function name, a (possibly empty) argument list, an optional section,
549 an optional alignment, an opening curly brace,
550 a list of basic blocks, and a closing curly brace. LLVM function declarations
551 are defined with the "<tt>declare</tt>" keyword, an optional <a
552 href="#callingconv">calling convention</a>, a return type, a function name,
553 a possibly empty list of arguments, and an optional alignment.</p>
555 <p>A function definition contains a list of basic blocks, forming the CFG for
556 the function. Each basic block may optionally start with a label (giving the
557 basic block a symbol table entry), contains a list of instructions, and ends
558 with a <a href="#terminators">terminator</a> instruction (such as a branch or
559 function return).</p>
561 <p>The first basic block in a program is special in two ways: it is immediately
562 executed on entrance to the function, and it is not allowed to have predecessor
563 basic blocks (i.e. there can not be any branches to the entry block of a
564 function). Because the block can have no predecessors, it also cannot have any
565 <a href="#i_phi">PHI nodes</a>.</p>
567 <p>LLVM functions are identified by their name and type signature. Hence, two
568 functions with the same name but different parameter lists or return values are
569 considered different functions, and LLVM will resolve references to each
572 <p>LLVM allows an explicit section to be specified for functions. If the target
573 supports it, it will emit functions to the section specified.</p>
575 <p>An explicit alignment may be specified for a function. If not present, or if
576 the alignment is set to zero, the alignment of the function is set by the target
577 to whatever it feels convenient. If an explicit alignment is specified, the
578 function is forced to have at least that much alignment. All alignments must be
585 <!-- *********************************************************************** -->
586 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
587 <!-- *********************************************************************** -->
589 <div class="doc_text">
591 <p>The LLVM type system is one of the most important features of the
592 intermediate representation. Being typed enables a number of
593 optimizations to be performed on the IR directly, without having to do
594 extra analyses on the side before the transformation. A strong type
595 system makes it easier to read the generated code and enables novel
596 analyses and transformations that are not feasible to perform on normal
597 three address code representations.</p>
601 <!-- ======================================================================= -->
602 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
603 <div class="doc_text">
604 <p>The primitive types are the fundamental building blocks of the LLVM
605 system. The current set of primitive types is as follows:</p>
607 <table class="layout">
612 <tr><th>Type</th><th>Description</th></tr>
613 <tr><td><tt>void</tt></td><td>No value</td></tr>
614 <tr><td><tt>ubyte</tt></td><td>Unsigned 8-bit value</td></tr>
615 <tr><td><tt>ushort</tt></td><td>Unsigned 16-bit value</td></tr>
616 <tr><td><tt>uint</tt></td><td>Unsigned 32-bit value</td></tr>
617 <tr><td><tt>ulong</tt></td><td>Unsigned 64-bit value</td></tr>
618 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
619 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
626 <tr><th>Type</th><th>Description</th></tr>
627 <tr><td><tt>bool</tt></td><td>True or False value</td></tr>
628 <tr><td><tt>sbyte</tt></td><td>Signed 8-bit value</td></tr>
629 <tr><td><tt>short</tt></td><td>Signed 16-bit value</td></tr>
630 <tr><td><tt>int</tt></td><td>Signed 32-bit value</td></tr>
631 <tr><td><tt>long</tt></td><td>Signed 64-bit value</td></tr>
632 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
640 <!-- _______________________________________________________________________ -->
641 <div class="doc_subsubsection"> <a name="t_classifications">Type
642 Classifications</a> </div>
643 <div class="doc_text">
644 <p>These different primitive types fall into a few useful
647 <table border="1" cellspacing="0" cellpadding="4">
649 <tr><th>Classification</th><th>Types</th></tr>
651 <td><a name="t_signed">signed</a></td>
652 <td><tt>sbyte, short, int, long, float, double</tt></td>
655 <td><a name="t_unsigned">unsigned</a></td>
656 <td><tt>ubyte, ushort, uint, ulong</tt></td>
659 <td><a name="t_integer">integer</a></td>
660 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
663 <td><a name="t_integral">integral</a></td>
664 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt>
668 <td><a name="t_floating">floating point</a></td>
669 <td><tt>float, double</tt></td>
672 <td><a name="t_firstclass">first class</a></td>
673 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long,<br>
674 float, double, <a href="#t_pointer">pointer</a>,
675 <a href="#t_packed">packed</a></tt></td>
680 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
681 most important. Values of these types are the only ones which can be
682 produced by instructions, passed as arguments, or used as operands to
683 instructions. This means that all structures and arrays must be
684 manipulated either by pointer or by component.</p>
687 <!-- ======================================================================= -->
688 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
690 <div class="doc_text">
692 <p>The real power in LLVM comes from the derived types in the system.
693 This is what allows a programmer to represent arrays, functions,
694 pointers, and other useful types. Note that these derived types may be
695 recursive: For example, it is possible to have a two dimensional array.</p>
699 <!-- _______________________________________________________________________ -->
700 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
702 <div class="doc_text">
706 <p>The array type is a very simple derived type that arranges elements
707 sequentially in memory. The array type requires a size (number of
708 elements) and an underlying data type.</p>
713 [<# elements> x <elementtype>]
716 <p>The number of elements is a constant integer value; elementtype may
717 be any type with a size.</p>
720 <table class="layout">
723 <tt>[40 x int ]</tt><br/>
724 <tt>[41 x int ]</tt><br/>
725 <tt>[40 x uint]</tt><br/>
728 Array of 40 integer values.<br/>
729 Array of 41 integer values.<br/>
730 Array of 40 unsigned integer values.<br/>
734 <p>Here are some examples of multidimensional arrays:</p>
735 <table class="layout">
738 <tt>[3 x [4 x int]]</tt><br/>
739 <tt>[12 x [10 x float]]</tt><br/>
740 <tt>[2 x [3 x [4 x uint]]]</tt><br/>
743 3x4 array of integer values.<br/>
744 12x10 array of single precision floating point values.<br/>
745 2x3x4 array of unsigned integer values.<br/>
750 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
751 length array. Normally, accesses past the end of an array are undefined in
752 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
753 As a special case, however, zero length arrays are recognized to be variable
754 length. This allows implementation of 'pascal style arrays' with the LLVM
755 type "{ int, [0 x float]}", for example.</p>
759 <!-- _______________________________________________________________________ -->
760 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
761 <div class="doc_text">
763 <p>The function type can be thought of as a function signature. It
764 consists of a return type and a list of formal parameter types.
765 Function types are usually used to build virtual function tables
766 (which are structures of pointers to functions), for indirect function
767 calls, and when defining a function.</p>
769 The return type of a function type cannot be an aggregate type.
772 <pre> <returntype> (<parameter list>)<br></pre>
773 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
774 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
775 which indicates that the function takes a variable number of arguments.
776 Variable argument functions can access their arguments with the <a
777 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
779 <table class="layout">
782 <tt>int (int)</tt> <br/>
783 <tt>float (int, int *) *</tt><br/>
784 <tt>int (sbyte *, ...)</tt><br/>
787 function taking an <tt>int</tt>, returning an <tt>int</tt><br/>
788 <a href="#t_pointer">Pointer</a> to a function that takes an
789 <tt>int</tt> and a <a href="#t_pointer">pointer</a> to <tt>int</tt>,
790 returning <tt>float</tt>.<br/>
791 A vararg function that takes at least one <a href="#t_pointer">pointer</a>
792 to <tt>sbyte</tt> (signed char in C), which returns an integer. This is
793 the signature for <tt>printf</tt> in LLVM.<br/>
799 <!-- _______________________________________________________________________ -->
800 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
801 <div class="doc_text">
803 <p>The structure type is used to represent a collection of data members
804 together in memory. The packing of the field types is defined to match
805 the ABI of the underlying processor. The elements of a structure may
806 be any type that has a size.</p>
807 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
808 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
809 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
812 <pre> { <type list> }<br></pre>
814 <table class="layout">
817 <tt>{ int, int, int }</tt><br/>
818 <tt>{ float, int (int) * }</tt><br/>
821 a triple of three <tt>int</tt> values<br/>
822 A pair, where the first element is a <tt>float</tt> and the second element
823 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
824 that takes an <tt>int</tt>, returning an <tt>int</tt>.<br/>
830 <!-- _______________________________________________________________________ -->
831 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
832 <div class="doc_text">
834 <p>As in many languages, the pointer type represents a pointer or
835 reference to another object, which must live in memory.</p>
837 <pre> <type> *<br></pre>
839 <table class="layout">
842 <tt>[4x int]*</tt><br/>
843 <tt>int (int *) *</tt><br/>
846 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
847 four <tt>int</tt> values<br/>
848 A <a href="#t_pointer">pointer</a> to a <a
849 href="#t_function">function</a> that takes an <tt>int*</tt>, returning an
856 <!-- _______________________________________________________________________ -->
857 <div class="doc_subsubsection"> <a name="t_packed">Packed Type</a> </div>
858 <div class="doc_text">
862 <p>A packed type is a simple derived type that represents a vector
863 of elements. Packed types are used when multiple primitive data
864 are operated in parallel using a single instruction (SIMD).
865 A packed type requires a size (number of
866 elements) and an underlying primitive data type. Vectors must have a power
867 of two length (1, 2, 4, 8, 16 ...). Packed types are
868 considered <a href="#t_firstclass">first class</a>.</p>
873 < <# elements> x <elementtype> >
876 <p>The number of elements is a constant integer value; elementtype may
877 be any integral or floating point type.</p>
881 <table class="layout">
884 <tt><4 x int></tt><br/>
885 <tt><8 x float></tt><br/>
886 <tt><2 x uint></tt><br/>
889 Packed vector of 4 integer values.<br/>
890 Packed vector of 8 floating-point values.<br/>
891 Packed vector of 2 unsigned integer values.<br/>
897 <!-- _______________________________________________________________________ -->
898 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
899 <div class="doc_text">
903 <p>Opaque types are used to represent unknown types in the system. This
904 corresponds (for example) to the C notion of a foward declared structure type.
905 In LLVM, opaque types can eventually be resolved to any type (not just a
916 <table class="layout">
929 <!-- *********************************************************************** -->
930 <div class="doc_section"> <a name="constants">Constants</a> </div>
931 <!-- *********************************************************************** -->
933 <div class="doc_text">
935 <p>LLVM has several different basic types of constants. This section describes
936 them all and their syntax.</p>
940 <!-- ======================================================================= -->
941 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
943 <div class="doc_text">
946 <dt><b>Boolean constants</b></dt>
948 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
949 constants of the <tt><a href="#t_primitive">bool</a></tt> type.
952 <dt><b>Integer constants</b></dt>
954 <dd>Standard integers (such as '4') are constants of the <a
955 href="#t_integer">integer</a> type. Negative numbers may be used with signed
959 <dt><b>Floating point constants</b></dt>
961 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
962 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
963 notation (see below). Floating point constants must have a <a
964 href="#t_floating">floating point</a> type. </dd>
966 <dt><b>Null pointer constants</b></dt>
968 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
969 and must be of <a href="#t_pointer">pointer type</a>.</dd>
973 <p>The one non-intuitive notation for constants is the optional hexadecimal form
974 of floating point constants. For example, the form '<tt>double
975 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
976 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
977 (and the only time that they are generated by the disassembler) is when a
978 floating point constant must be emitted but it cannot be represented as a
979 decimal floating point number. For example, NaN's, infinities, and other
980 special values are represented in their IEEE hexadecimal format so that
981 assembly and disassembly do not cause any bits to change in the constants.</p>
985 <!-- ======================================================================= -->
986 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
989 <div class="doc_text">
990 <p>Aggregate constants arise from aggregation of simple constants
991 and smaller aggregate constants.</p>
994 <dt><b>Structure constants</b></dt>
996 <dd>Structure constants are represented with notation similar to structure
997 type definitions (a comma separated list of elements, surrounded by braces
998 (<tt>{}</tt>)). For example: "<tt>{ int 4, float 17.0, int* %G }</tt>",
999 where "<tt>%G</tt>" is declared as "<tt>%G = external global int</tt>". Structure constants
1000 must have <a href="#t_struct">structure type</a>, and the number and
1001 types of elements must match those specified by the type.
1004 <dt><b>Array constants</b></dt>
1006 <dd>Array constants are represented with notation similar to array type
1007 definitions (a comma separated list of elements, surrounded by square brackets
1008 (<tt>[]</tt>)). For example: "<tt>[ int 42, int 11, int 74 ]</tt>". Array
1009 constants must have <a href="#t_array">array type</a>, and the number and
1010 types of elements must match those specified by the type.
1013 <dt><b>Packed constants</b></dt>
1015 <dd>Packed constants are represented with notation similar to packed type
1016 definitions (a comma separated list of elements, surrounded by
1017 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< int 42,
1018 int 11, int 74, int 100 ></tt>". Packed constants must have <a
1019 href="#t_packed">packed type</a>, and the number and types of elements must
1020 match those specified by the type.
1023 <dt><b>Zero initialization</b></dt>
1025 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1026 value to zero of <em>any</em> type, including scalar and aggregate types.
1027 This is often used to avoid having to print large zero initializers (e.g. for
1028 large arrays) and is always exactly equivalent to using explicit zero
1035 <!-- ======================================================================= -->
1036 <div class="doc_subsection">
1037 <a name="globalconstants">Global Variable and Function Addresses</a>
1040 <div class="doc_text">
1042 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1043 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1044 constants. These constants are explicitly referenced when the <a
1045 href="#identifiers">identifier for the global</a> is used and always have <a
1046 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1052 %Z = global [2 x int*] [ int* %X, int* %Y ]
1057 <!-- ======================================================================= -->
1058 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1059 <div class="doc_text">
1060 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1061 no specific value. Undefined values may be of any type and be used anywhere
1062 a constant is permitted.</p>
1064 <p>Undefined values indicate to the compiler that the program is well defined
1065 no matter what value is used, giving the compiler more freedom to optimize.
1069 <!-- ======================================================================= -->
1070 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1073 <div class="doc_text">
1075 <p>Constant expressions are used to allow expressions involving other constants
1076 to be used as constants. Constant expressions may be of any <a
1077 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1078 that does not have side effects (e.g. load and call are not supported). The
1079 following is the syntax for constant expressions:</p>
1082 <dt><b><tt>cast ( CST to TYPE )</tt></b></dt>
1084 <dd>Cast a constant to another type.</dd>
1086 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1088 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1089 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1090 instruction, the index list may have zero or more indexes, which are required
1091 to make sense for the type of "CSTPTR".</dd>
1093 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1095 <dd>Perform the <a href="#i_select">select operation</a> on
1098 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1100 <dd>Perform the <a href="#i_extractelement">extractelement
1101 operation</a> on constants.
1103 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1105 <dd>Perform the <a href="#i_insertelement">insertelement
1106 operation</a> on constants.
1108 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1110 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1111 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1112 binary</a> operations. The constraints on operands are the same as those for
1113 the corresponding instruction (e.g. no bitwise operations on floating point
1114 values are allowed).</dd>
1118 <!-- *********************************************************************** -->
1119 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1120 <!-- *********************************************************************** -->
1122 <div class="doc_text">
1124 <p>The LLVM instruction set consists of several different
1125 classifications of instructions: <a href="#terminators">terminator
1126 instructions</a>, <a href="#binaryops">binary instructions</a>,
1127 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1128 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1129 instructions</a>.</p>
1133 <!-- ======================================================================= -->
1134 <div class="doc_subsection"> <a name="terminators">Terminator
1135 Instructions</a> </div>
1137 <div class="doc_text">
1139 <p>As mentioned <a href="#functionstructure">previously</a>, every
1140 basic block in a program ends with a "Terminator" instruction, which
1141 indicates which block should be executed after the current block is
1142 finished. These terminator instructions typically yield a '<tt>void</tt>'
1143 value: they produce control flow, not values (the one exception being
1144 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1145 <p>There are six different terminator instructions: the '<a
1146 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1147 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1148 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1149 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1150 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1154 <!-- _______________________________________________________________________ -->
1155 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1156 Instruction</a> </div>
1157 <div class="doc_text">
1159 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1160 ret void <i>; Return from void function</i>
1163 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1164 value) from a function back to the caller.</p>
1165 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1166 returns a value and then causes control flow, and one that just causes
1167 control flow to occur.</p>
1169 <p>The '<tt>ret</tt>' instruction may return any '<a
1170 href="#t_firstclass">first class</a>' type. Notice that a function is
1171 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1172 instruction inside of the function that returns a value that does not
1173 match the return type of the function.</p>
1175 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1176 returns back to the calling function's context. If the caller is a "<a
1177 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1178 the instruction after the call. If the caller was an "<a
1179 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1180 at the beginning of the "normal" destination block. If the instruction
1181 returns a value, that value shall set the call or invoke instruction's
1184 <pre> ret int 5 <i>; Return an integer value of 5</i>
1185 ret void <i>; Return from a void function</i>
1188 <!-- _______________________________________________________________________ -->
1189 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1190 <div class="doc_text">
1192 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1195 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1196 transfer to a different basic block in the current function. There are
1197 two forms of this instruction, corresponding to a conditional branch
1198 and an unconditional branch.</p>
1200 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1201 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
1202 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
1203 value as a target.</p>
1205 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
1206 argument is evaluated. If the value is <tt>true</tt>, control flows
1207 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1208 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1210 <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
1211 href="#i_ret">ret</a> int 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> int 0<br></pre>
1213 <!-- _______________________________________________________________________ -->
1214 <div class="doc_subsubsection">
1215 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1218 <div class="doc_text">
1222 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1227 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1228 several different places. It is a generalization of the '<tt>br</tt>'
1229 instruction, allowing a branch to occur to one of many possible
1235 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1236 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1237 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1238 table is not allowed to contain duplicate constant entries.</p>
1242 <p>The <tt>switch</tt> instruction specifies a table of values and
1243 destinations. When the '<tt>switch</tt>' instruction is executed, this
1244 table is searched for the given value. If the value is found, control flow is
1245 transfered to the corresponding destination; otherwise, control flow is
1246 transfered to the default destination.</p>
1248 <h5>Implementation:</h5>
1250 <p>Depending on properties of the target machine and the particular
1251 <tt>switch</tt> instruction, this instruction may be code generated in different
1252 ways. For example, it could be generated as a series of chained conditional
1253 branches or with a lookup table.</p>
1258 <i>; Emulate a conditional br instruction</i>
1259 %Val = <a href="#i_cast">cast</a> bool %value to int
1260 switch int %Val, label %truedest [int 0, label %falsedest ]
1262 <i>; Emulate an unconditional br instruction</i>
1263 switch uint 0, label %dest [ ]
1265 <i>; Implement a jump table:</i>
1266 switch uint %val, label %otherwise [ uint 0, label %onzero
1267 uint 1, label %onone
1268 uint 2, label %ontwo ]
1272 <!-- _______________________________________________________________________ -->
1273 <div class="doc_subsubsection">
1274 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1277 <div class="doc_text">
1282 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1283 to label <normal label> except label <exception label>
1288 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1289 function, with the possibility of control flow transfer to either the
1290 '<tt>normal</tt>' label or the
1291 '<tt>exception</tt>' label. If the callee function returns with the
1292 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1293 "normal" label. If the callee (or any indirect callees) returns with the "<a
1294 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1295 continued at the dynamically nearest "exception" label.</p>
1299 <p>This instruction requires several arguments:</p>
1303 The optional "cconv" marker indicates which <a href="callingconv">calling
1304 convention</a> the call should use. If none is specified, the call defaults
1305 to using C calling conventions.
1307 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1308 function value being invoked. In most cases, this is a direct function
1309 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1310 an arbitrary pointer to function value.
1313 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1314 function to be invoked. </li>
1316 <li>'<tt>function args</tt>': argument list whose types match the function
1317 signature argument types. If the function signature indicates the function
1318 accepts a variable number of arguments, the extra arguments can be
1321 <li>'<tt>normal label</tt>': the label reached when the called function
1322 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1324 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1325 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1331 <p>This instruction is designed to operate as a standard '<tt><a
1332 href="#i_call">call</a></tt>' instruction in most regards. The primary
1333 difference is that it establishes an association with a label, which is used by
1334 the runtime library to unwind the stack.</p>
1336 <p>This instruction is used in languages with destructors to ensure that proper
1337 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1338 exception. Additionally, this is important for implementation of
1339 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1343 %retval = invoke int %Test(int 15) to label %Continue
1344 except label %TestCleanup <i>; {int}:retval set</i>
1345 %retval = invoke <a href="#callingconv">coldcc</a> int %Test(int 15) to label %Continue
1346 except label %TestCleanup <i>; {int}:retval set</i>
1351 <!-- _______________________________________________________________________ -->
1353 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1354 Instruction</a> </div>
1356 <div class="doc_text">
1365 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1366 at the first callee in the dynamic call stack which used an <a
1367 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1368 primarily used to implement exception handling.</p>
1372 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1373 immediately halt. The dynamic call stack is then searched for the first <a
1374 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1375 execution continues at the "exceptional" destination block specified by the
1376 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1377 dynamic call chain, undefined behavior results.</p>
1380 <!-- _______________________________________________________________________ -->
1382 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1383 Instruction</a> </div>
1385 <div class="doc_text">
1394 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1395 instruction is used to inform the optimizer that a particular portion of the
1396 code is not reachable. This can be used to indicate that the code after a
1397 no-return function cannot be reached, and other facts.</p>
1401 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1406 <!-- ======================================================================= -->
1407 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1408 <div class="doc_text">
1409 <p>Binary operators are used to do most of the computation in a
1410 program. They require two operands, execute an operation on them, and
1411 produce a single value. The operands might represent
1412 multiple data, as is the case with the <a href="#t_packed">packed</a> data type.
1413 The result value of a binary operator is not
1414 necessarily the same type as its operands.</p>
1415 <p>There are several different binary operators:</p>
1417 <!-- _______________________________________________________________________ -->
1418 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1419 Instruction</a> </div>
1420 <div class="doc_text">
1422 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1425 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1427 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1428 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1429 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1430 Both arguments must have identical types.</p>
1432 <p>The value produced is the integer or floating point sum of the two
1435 <pre> <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
1438 <!-- _______________________________________________________________________ -->
1439 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1440 Instruction</a> </div>
1441 <div class="doc_text">
1443 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1446 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1448 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1449 instruction present in most other intermediate representations.</p>
1451 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1452 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1454 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1455 Both arguments must have identical types.</p>
1457 <p>The value produced is the integer or floating point difference of
1458 the two operands.</p>
1460 <pre> <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
1461 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
1464 <!-- _______________________________________________________________________ -->
1465 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1466 Instruction</a> </div>
1467 <div class="doc_text">
1469 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1472 <p>The '<tt>mul</tt>' instruction returns the product of its two
1475 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1476 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1478 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1479 Both arguments must have identical types.</p>
1481 <p>The value produced is the integer or floating point product of the
1483 <p>There is no signed vs unsigned multiplication. The appropriate
1484 action is taken based on the type of the operand.</p>
1486 <pre> <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1489 <!-- _______________________________________________________________________ -->
1490 <div class="doc_subsubsection"> <a name="i_div">'<tt>div</tt>'
1491 Instruction</a> </div>
1492 <div class="doc_text">
1494 <pre> <result> = div <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1497 <p>The '<tt>div</tt>' instruction returns the quotient of its two
1500 <p>The two arguments to the '<tt>div</tt>' instruction must be either <a
1501 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1503 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1504 Both arguments must have identical types.</p>
1506 <p>The value produced is the integer or floating point quotient of the
1509 <pre> <result> = div int 4, %var <i>; yields {int}:result = 4 / %var</i>
1512 <!-- _______________________________________________________________________ -->
1513 <div class="doc_subsubsection"> <a name="i_rem">'<tt>rem</tt>'
1514 Instruction</a> </div>
1515 <div class="doc_text">
1517 <pre> <result> = rem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1520 <p>The '<tt>rem</tt>' instruction returns the remainder from the
1521 division of its two operands.</p>
1523 <p>The two arguments to the '<tt>rem</tt>' instruction must be either <a
1524 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1526 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1527 Both arguments must have identical types.</p>
1529 <p>This returns the <i>remainder</i> of a division (where the result
1530 has the same sign as the divisor), not the <i>modulus</i> (where the
1531 result has the same sign as the dividend) of a value. For more
1532 information about the difference, see <a
1533 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
1536 <pre> <result> = rem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1539 <!-- _______________________________________________________________________ -->
1540 <div class="doc_subsubsection"> <a name="i_setcc">'<tt>set<i>cc</i></tt>'
1541 Instructions</a> </div>
1542 <div class="doc_text">
1544 <pre> <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1545 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1546 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1547 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1548 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1549 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1552 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean
1553 value based on a comparison of their two operands.</p>
1555 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must
1556 be of <a href="#t_firstclass">first class</a> type (it is not possible
1557 to compare '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>'
1558 or '<tt>void</tt>' values, etc...). Both arguments must have identical
1561 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1562 value if both operands are equal.<br>
1563 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1564 value if both operands are unequal.<br>
1565 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1566 value if the first operand is less than the second operand.<br>
1567 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1568 value if the first operand is greater than the second operand.<br>
1569 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1570 value if the first operand is less than or equal to the second operand.<br>
1571 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1572 value if the first operand is greater than or equal to the second
1575 <pre> <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1576 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1577 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1578 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1579 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1580 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1583 <!-- ======================================================================= -->
1584 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
1585 Operations</a> </div>
1586 <div class="doc_text">
1587 <p>Bitwise binary operators are used to do various forms of
1588 bit-twiddling in a program. They are generally very efficient
1589 instructions and can commonly be strength reduced from other
1590 instructions. They require two operands, execute an operation on them,
1591 and produce a single value. The resulting value of the bitwise binary
1592 operators is always the same type as its first operand.</p>
1594 <!-- _______________________________________________________________________ -->
1595 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
1596 Instruction</a> </div>
1597 <div class="doc_text">
1599 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1602 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1603 its two operands.</p>
1605 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1606 href="#t_integral">integral</a> values. Both arguments must have
1607 identical types.</p>
1609 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1611 <div style="align: center">
1612 <table border="1" cellspacing="0" cellpadding="4">
1643 <pre> <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1644 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1645 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1648 <!-- _______________________________________________________________________ -->
1649 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
1650 <div class="doc_text">
1652 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1655 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1656 or of its two operands.</p>
1658 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1659 href="#t_integral">integral</a> values. Both arguments must have
1660 identical types.</p>
1662 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1664 <div style="align: center">
1665 <table border="1" cellspacing="0" cellpadding="4">
1696 <pre> <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1697 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1698 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1701 <!-- _______________________________________________________________________ -->
1702 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
1703 Instruction</a> </div>
1704 <div class="doc_text">
1706 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1709 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
1710 or of its two operands. The <tt>xor</tt> is used to implement the
1711 "one's complement" operation, which is the "~" operator in C.</p>
1713 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
1714 href="#t_integral">integral</a> values. Both arguments must have
1715 identical types.</p>
1717 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
1719 <div style="align: center">
1720 <table border="1" cellspacing="0" cellpadding="4">
1752 <pre> <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
1753 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
1754 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
1755 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
1758 <!-- _______________________________________________________________________ -->
1759 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
1760 Instruction</a> </div>
1761 <div class="doc_text">
1763 <pre> <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1766 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
1767 the left a specified number of bits.</p>
1769 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
1770 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1773 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
1775 <pre> <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
1776 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
1777 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
1780 <!-- _______________________________________________________________________ -->
1781 <div class="doc_subsubsection"> <a name="i_shr">'<tt>shr</tt>'
1782 Instruction</a> </div>
1783 <div class="doc_text">
1785 <pre> <result> = shr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1788 <p>The '<tt>shr</tt>' instruction returns the first operand shifted to
1789 the right a specified number of bits.</p>
1791 <p>The first argument to the '<tt>shr</tt>' instruction must be an <a
1792 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1795 <p>If the first argument is a <a href="#t_signed">signed</a> type, the
1796 most significant bit is duplicated in the newly free'd bit positions.
1797 If the first argument is unsigned, zero bits shall fill the empty
1800 <pre> <result> = shr int 4, ubyte %var <i>; yields {int}:result = 4 >> %var</i>
1801 <result> = shr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
1802 <result> = shr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
1803 <result> = shr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
1804 <result> = shr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
1808 <!-- ======================================================================= -->
1809 <div class="doc_subsection">
1810 <a name="memoryops">Memory Access Operations</a>
1813 <div class="doc_text">
1815 <p>A key design point of an SSA-based representation is how it
1816 represents memory. In LLVM, no memory locations are in SSA form, which
1817 makes things very simple. This section describes how to read, write,
1818 allocate, and free memory in LLVM.</p>
1822 <!-- _______________________________________________________________________ -->
1823 <div class="doc_subsubsection">
1824 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
1827 <div class="doc_text">
1832 <result> = malloc <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
1837 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
1838 heap and returns a pointer to it.</p>
1842 <p>The '<tt>malloc</tt>' instruction allocates
1843 <tt>sizeof(<type>)*NumElements</tt>
1844 bytes of memory from the operating system and returns a pointer of the
1845 appropriate type to the program. If "NumElements" is specified, it is the
1846 number of elements allocated. If an alignment is specified, the value result
1847 of the allocation is guaranteed to be aligned to at least that boundary. If
1848 not specified, or if zero, the target can choose to align the allocation on any
1849 convenient boundary.</p>
1851 <p>'<tt>type</tt>' must be a sized type.</p>
1855 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
1856 a pointer is returned.</p>
1861 %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
1863 %size = <a href="#i_add">add</a> uint 2, 2 <i>; yields {uint}:size = uint 4</i>
1864 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
1865 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
1866 %array3 = malloc int, uint 4, align 1024 <i>; yields {int*}:array3</i>
1867 %array4 = malloc int, align 1024 <i>; yields {int*}:array4</i>
1871 <!-- _______________________________________________________________________ -->
1872 <div class="doc_subsubsection">
1873 <a name="i_free">'<tt>free</tt>' Instruction</a>
1876 <div class="doc_text">
1881 free <type> <value> <i>; yields {void}</i>
1886 <p>The '<tt>free</tt>' instruction returns memory back to the unused
1887 memory heap to be reallocated in the future.</p>
1891 <p>'<tt>value</tt>' shall be a pointer value that points to a value
1892 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
1897 <p>Access to the memory pointed to by the pointer is no longer defined
1898 after this instruction executes.</p>
1903 %array = <a href="#i_malloc">malloc</a> [4 x ubyte] <i>; yields {[4 x ubyte]*}:array</i>
1904 free [4 x ubyte]* %array
1908 <!-- _______________________________________________________________________ -->
1909 <div class="doc_subsubsection">
1910 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
1913 <div class="doc_text">
1918 <result> = alloca <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
1923 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
1924 stack frame of the procedure that is live until the current function
1925 returns to its caller.</p>
1929 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
1930 bytes of memory on the runtime stack, returning a pointer of the
1931 appropriate type to the program. If "NumElements" is specified, it is the
1932 number of elements allocated. If an alignment is specified, the value result
1933 of the allocation is guaranteed to be aligned to at least that boundary. If
1934 not specified, or if zero, the target can choose to align the allocation on any
1935 convenient boundary.</p>
1937 <p>'<tt>type</tt>' may be any sized type.</p>
1941 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
1942 memory is automatically released when the function returns. The '<tt>alloca</tt>'
1943 instruction is commonly used to represent automatic variables that must
1944 have an address available. When the function returns (either with the <tt><a
1945 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
1946 instructions), the memory is reclaimed.</p>
1951 %ptr = alloca int <i>; yields {int*}:ptr</i>
1952 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
1953 %ptr = alloca int, uint 4, align 1024 <i>; yields {int*}:ptr</i>
1954 %ptr = alloca int, align 1024 <i>; yields {int*}:ptr</i>
1958 <!-- _______________________________________________________________________ -->
1959 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
1960 Instruction</a> </div>
1961 <div class="doc_text">
1963 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
1965 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
1967 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
1968 address from which to load. The pointer must point to a <a
1969 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
1970 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
1971 the number or order of execution of this <tt>load</tt> with other
1972 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
1975 <p>The location of memory pointed to is loaded.</p>
1977 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
1979 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
1980 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
1983 <!-- _______________________________________________________________________ -->
1984 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
1985 Instruction</a> </div>
1987 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
1988 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
1991 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
1993 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
1994 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
1995 operand must be a pointer to the type of the '<tt><value></tt>'
1996 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
1997 optimizer is not allowed to modify the number or order of execution of
1998 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
1999 href="#i_store">store</a></tt> instructions.</p>
2001 <p>The contents of memory are updated to contain '<tt><value></tt>'
2002 at the location specified by the '<tt><pointer></tt>' operand.</p>
2004 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2006 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2007 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2009 <!-- _______________________________________________________________________ -->
2010 <div class="doc_subsubsection">
2011 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2014 <div class="doc_text">
2017 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2023 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2024 subelement of an aggregate data structure.</p>
2028 <p>This instruction takes a list of integer constants that indicate what
2029 elements of the aggregate object to index to. The actual types of the arguments
2030 provided depend on the type of the first pointer argument. The
2031 '<tt>getelementptr</tt>' instruction is used to index down through the type
2032 levels of a structure or to a specific index in an array. When indexing into a
2033 structure, only <tt>uint</tt>
2034 integer constants are allowed. When indexing into an array or pointer,
2035 <tt>int</tt> and <tt>long</tt> indexes are allowed of any sign.</p>
2037 <p>For example, let's consider a C code fragment and how it gets
2038 compiled to LLVM:</p>
2052 int *foo(struct ST *s) {
2053 return &s[1].Z.B[5][13];
2057 <p>The LLVM code generated by the GCC frontend is:</p>
2060 %RT = type { sbyte, [10 x [20 x int]], sbyte }
2061 %ST = type { int, double, %RT }
2065 int* %foo(%ST* %s) {
2067 %reg = getelementptr %ST* %s, int 1, uint 2, uint 1, int 5, int 13
2074 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2075 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2076 and <a href="#t_array">array</a> types require <tt>uint</tt>, <tt>int</tt>,
2077 <tt>ulong</tt>, or <tt>long</tt> values, and <a href="#t_struct">structure</a>
2078 types require <tt>uint</tt> <b>constants</b>.</p>
2080 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2081 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
2082 }</tt>' type, a structure. The second index indexes into the third element of
2083 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
2084 sbyte }</tt>' type, another structure. The third index indexes into the second
2085 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
2086 array. The two dimensions of the array are subscripted into, yielding an
2087 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2088 to this element, thus computing a value of '<tt>int*</tt>' type.</p>
2090 <p>Note that it is perfectly legal to index partially through a
2091 structure, returning a pointer to an inner element. Because of this,
2092 the LLVM code for the given testcase is equivalent to:</p>
2095 int* %foo(%ST* %s) {
2096 %t1 = getelementptr %ST* %s, int 1 <i>; yields %ST*:%t1</i>
2097 %t2 = getelementptr %ST* %t1, int 0, uint 2 <i>; yields %RT*:%t2</i>
2098 %t3 = getelementptr %RT* %t2, int 0, uint 1 <i>; yields [10 x [20 x int]]*:%t3</i>
2099 %t4 = getelementptr [10 x [20 x int]]* %t3, int 0, int 5 <i>; yields [20 x int]*:%t4</i>
2100 %t5 = getelementptr [20 x int]* %t4, int 0, int 13 <i>; yields int*:%t5</i>
2105 <p>Note that it is undefined to access an array out of bounds: array and
2106 pointer indexes must always be within the defined bounds of the array type.
2107 The one exception for this rules is zero length arrays. These arrays are
2108 defined to be accessible as variable length arrays, which requires access
2109 beyond the zero'th element.</p>
2114 <i>; yields [12 x ubyte]*:aptr</i>
2115 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, uint 1
2119 <!-- ======================================================================= -->
2120 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
2121 <div class="doc_text">
2122 <p>The instructions in this category are the "miscellaneous"
2123 instructions, which defy better classification.</p>
2125 <!-- _______________________________________________________________________ -->
2126 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
2127 Instruction</a> </div>
2128 <div class="doc_text">
2130 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
2132 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
2133 the SSA graph representing the function.</p>
2135 <p>The type of the incoming values are specified with the first type
2136 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
2137 as arguments, with one pair for each predecessor basic block of the
2138 current block. Only values of <a href="#t_firstclass">first class</a>
2139 type may be used as the value arguments to the PHI node. Only labels
2140 may be used as the label arguments.</p>
2141 <p>There must be no non-phi instructions between the start of a basic
2142 block and the PHI instructions: i.e. PHI instructions must be first in
2145 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
2146 value specified by the parameter, depending on which basic block we
2147 came from in the last <a href="#terminators">terminator</a> instruction.</p>
2149 <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>
2152 <!-- _______________________________________________________________________ -->
2153 <div class="doc_subsubsection">
2154 <a name="i_cast">'<tt>cast .. to</tt>' Instruction</a>
2157 <div class="doc_text">
2162 <result> = cast <ty> <value> to <ty2> <i>; yields ty2</i>
2168 The '<tt>cast</tt>' instruction is used as the primitive means to convert
2169 integers to floating point, change data type sizes, and break type safety (by
2177 The '<tt>cast</tt>' instruction takes a value to cast, which must be a first
2178 class value, and a type to cast it to, which must also be a <a
2179 href="#t_firstclass">first class</a> type.
2185 This instruction follows the C rules for explicit casts when determining how the
2186 data being cast must change to fit in its new container.
2190 When casting to bool, any value that would be considered true in the context of
2191 a C '<tt>if</tt>' condition is converted to the boolean '<tt>true</tt>' values,
2192 all else are '<tt>false</tt>'.
2196 When extending an integral value from a type of one signness to another (for
2197 example '<tt>sbyte</tt>' to '<tt>ulong</tt>'), the value is sign-extended if the
2198 <b>source</b> value is signed, and zero-extended if the source value is
2199 unsigned. <tt>bool</tt> values are always zero extended into either zero or
2206 %X = cast int 257 to ubyte <i>; yields ubyte:1</i>
2207 %Y = cast int 123 to bool <i>; yields bool:true</i>
2211 <!-- _______________________________________________________________________ -->
2212 <div class="doc_subsubsection">
2213 <a name="i_select">'<tt>select</tt>' Instruction</a>
2216 <div class="doc_text">
2221 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
2227 The '<tt>select</tt>' instruction is used to choose one value based on a
2228 condition, without branching.
2235 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.
2241 If the boolean condition evaluates to true, the instruction returns the first
2242 value argument; otherwise, it returns the second value argument.
2248 %X = select bool true, ubyte 17, ubyte 42 <i>; yields ubyte:17</i>
2253 <!-- _______________________________________________________________________ -->
2254 <div class="doc_subsubsection">
2255 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2258 <div class="doc_text">
2263 <result> = extractelement <n x <ty>> <val>, uint <idx> <i>; yields <ty></i>
2269 The '<tt>extractelement</tt>' instruction extracts a single scalar
2270 element from a packed vector at a specified index.
2277 The first operand of an '<tt>extractelement</tt>' instruction is a
2278 value of <a href="#t_packed">packed</a> type. The second operand is
2279 an index indicating the position from which to extract the element.
2280 The index may be a variable.</p>
2285 The result is a scalar of the same type as the element type of
2286 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2287 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2288 results are undefined.
2294 %result = extractelement <4 x int> %vec, uint 0 <i>; yields int</i>
2299 <!-- _______________________________________________________________________ -->
2300 <div class="doc_subsubsection">
2301 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2304 <div class="doc_text">
2309 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, uint <idx> <i>; yields <n x <ty>></i>
2315 The '<tt>insertelement</tt>' instruction inserts a scalar
2316 element into a packed vector at a specified index.
2323 The first operand of an '<tt>insertelement</tt>' instruction is a
2324 value of <a href="#t_packed">packed</a> type. The second operand is a
2325 scalar value whose type must equal the element type of the first
2326 operand. The third operand is an index indicating the position at
2327 which to insert the value. The index may be a variable.</p>
2332 The result is a packed vector of the same type as <tt>val</tt>. Its
2333 element values are those of <tt>val</tt> except at position
2334 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2335 exceeds the length of <tt>val</tt>, the results are undefined.
2341 %result = insertelement <4 x int> %vec, int 1, uint 0 <i>; yields <4 x int></i>
2346 <!-- _______________________________________________________________________ -->
2347 <div class="doc_subsubsection">
2348 <a name="i_call">'<tt>call</tt>' Instruction</a>
2351 <div class="doc_text">
2355 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
2360 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
2364 <p>This instruction requires several arguments:</p>
2368 <p>The optional "tail" marker indicates whether the callee function accesses
2369 any allocas or varargs in the caller. If the "tail" marker is present, the
2370 function call is eligible for tail call optimization. Note that calls may
2371 be marked "tail" even if they do not occur before a <a
2372 href="#i_ret"><tt>ret</tt></a> instruction.
2375 <p>The optional "cconv" marker indicates which <a href="callingconv">calling
2376 convention</a> the call should use. If none is specified, the call defaults
2377 to using C calling conventions.
2380 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
2381 being invoked. The argument types must match the types implied by this
2382 signature. This type can be omitted if the function is not varargs and
2383 if the function type does not return a pointer to a function.</p>
2386 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
2387 be invoked. In most cases, this is a direct function invocation, but
2388 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
2389 to function value.</p>
2392 <p>'<tt>function args</tt>': argument list whose types match the
2393 function signature argument types. All arguments must be of
2394 <a href="#t_firstclass">first class</a> type. If the function signature
2395 indicates the function accepts a variable number of arguments, the extra
2396 arguments can be specified.</p>
2402 <p>The '<tt>call</tt>' instruction is used to cause control flow to
2403 transfer to a specified function, with its incoming arguments bound to
2404 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
2405 instruction in the called function, control flow continues with the
2406 instruction after the function call, and the return value of the
2407 function is bound to the result argument. This is a simpler case of
2408 the <a href="#i_invoke">invoke</a> instruction.</p>
2413 %retval = call int %test(int %argc)
2414 call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);
2415 %X = tail call int %foo()
2416 %Y = tail call <a href="#callingconv">fastcc</a> int %foo()
2421 <!-- _______________________________________________________________________ -->
2422 <div class="doc_subsubsection">
2423 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
2426 <div class="doc_text">
2431 <resultval> = va_arg <va_list*> <arglist>, <argty>
2436 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
2437 the "variable argument" area of a function call. It is used to implement the
2438 <tt>va_arg</tt> macro in C.</p>
2442 <p>This instruction takes a <tt>va_list*</tt> value and the type of
2443 the argument. It returns a value of the specified argument type and
2444 increments the <tt>va_list</tt> to point to the next argument. Again, the
2445 actual type of <tt>va_list</tt> is target specific.</p>
2449 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
2450 type from the specified <tt>va_list</tt> and causes the
2451 <tt>va_list</tt> to point to the next argument. For more information,
2452 see the variable argument handling <a href="#int_varargs">Intrinsic
2455 <p>It is legal for this instruction to be called in a function which does not
2456 take a variable number of arguments, for example, the <tt>vfprintf</tt>
2459 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
2460 href="#intrinsics">intrinsic function</a> because it takes a type as an
2465 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
2469 <!-- *********************************************************************** -->
2470 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
2471 <!-- *********************************************************************** -->
2473 <div class="doc_text">
2475 <p>LLVM supports the notion of an "intrinsic function". These functions have
2476 well known names and semantics and are required to follow certain
2477 restrictions. Overall, these instructions represent an extension mechanism for
2478 the LLVM language that does not require changing all of the transformations in
2479 LLVM to add to the language (or the bytecode reader/writer, the parser,
2482 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
2483 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
2484 this. Intrinsic functions must always be external functions: you cannot define
2485 the body of intrinsic functions. Intrinsic functions may only be used in call
2486 or invoke instructions: it is illegal to take the address of an intrinsic
2487 function. Additionally, because intrinsic functions are part of the LLVM
2488 language, it is required that they all be documented here if any are added.</p>
2491 <p>To learn how to add an intrinsic function, please see the <a
2492 href="ExtendingLLVM.html">Extending LLVM Guide</a>.
2497 <!-- ======================================================================= -->
2498 <div class="doc_subsection">
2499 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
2502 <div class="doc_text">
2504 <p>Variable argument support is defined in LLVM with the <a
2505 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
2506 intrinsic functions. These functions are related to the similarly
2507 named macros defined in the <tt><stdarg.h></tt> header file.</p>
2509 <p>All of these functions operate on arguments that use a
2510 target-specific value type "<tt>va_list</tt>". The LLVM assembly
2511 language reference manual does not define what this type is, so all
2512 transformations should be prepared to handle intrinsics with any type
2515 <p>This example shows how the <a href="#i_vanext"><tt>vanext</tt></a>
2516 instruction and the variable argument handling intrinsic functions are
2520 int %test(int %X, ...) {
2521 ; Initialize variable argument processing
2523 call void %<a href="#i_va_start">llvm.va_start</a>(sbyte** %ap)
2525 ; Read a single integer argument
2526 %tmp = va_arg sbyte** %ap, int
2528 ; Demonstrate usage of llvm.va_copy and llvm.va_end
2530 call void %<a href="#i_va_copy">llvm.va_copy</a>(sbyte** %aq, sbyte** %ap)
2531 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %aq)
2533 ; Stop processing of arguments.
2534 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %ap)
2540 <!-- _______________________________________________________________________ -->
2541 <div class="doc_subsubsection">
2542 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
2546 <div class="doc_text">
2548 <pre> declare void %llvm.va_start(<va_list>* <arglist>)<br></pre>
2550 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
2551 <tt>*<arglist></tt> for subsequent use by <tt><a
2552 href="#i_va_arg">va_arg</a></tt>.</p>
2556 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
2560 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
2561 macro available in C. In a target-dependent way, it initializes the
2562 <tt>va_list</tt> element the argument points to, so that the next call to
2563 <tt>va_arg</tt> will produce the first variable argument passed to the function.
2564 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
2565 last argument of the function, the compiler can figure that out.</p>
2569 <!-- _______________________________________________________________________ -->
2570 <div class="doc_subsubsection">
2571 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
2574 <div class="doc_text">
2576 <pre> declare void %llvm.va_end(<va_list*> <arglist>)<br></pre>
2578 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
2579 which has been initialized previously with <tt><a href="#i_va_start">llvm.va_start</a></tt>
2580 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
2582 <p>The argument is a <tt>va_list</tt> to destroy.</p>
2584 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
2585 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
2586 Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and <a
2587 href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
2588 with calls to <tt>llvm.va_end</tt>.</p>
2591 <!-- _______________________________________________________________________ -->
2592 <div class="doc_subsubsection">
2593 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
2596 <div class="doc_text">
2601 declare void %llvm.va_copy(<va_list>* <destarglist>,
2602 <va_list>* <srcarglist>)
2607 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
2608 the source argument list to the destination argument list.</p>
2612 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
2613 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
2618 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
2619 available in C. In a target-dependent way, it copies the source
2620 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
2621 because the <tt><a href="i_va_begin">llvm.va_begin</a></tt> intrinsic may be
2622 arbitrarily complex and require memory allocation, for example.</p>
2626 <!-- ======================================================================= -->
2627 <div class="doc_subsection">
2628 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
2631 <div class="doc_text">
2634 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
2635 Collection</a> requires the implementation and generation of these intrinsics.
2636 These intrinsics allow identification of <a href="#i_gcroot">GC roots on the
2637 stack</a>, as well as garbage collector implementations that require <a
2638 href="#i_gcread">read</a> and <a href="#i_gcwrite">write</a> barriers.
2639 Front-ends for type-safe garbage collected languages should generate these
2640 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
2641 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
2645 <!-- _______________________________________________________________________ -->
2646 <div class="doc_subsubsection">
2647 <a name="i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
2650 <div class="doc_text">
2655 declare void %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
2660 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
2661 the code generator, and allows some metadata to be associated with it.</p>
2665 <p>The first argument specifies the address of a stack object that contains the
2666 root pointer. The second pointer (which must be either a constant or a global
2667 value address) contains the meta-data to be associated with the root.</p>
2671 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
2672 location. At compile-time, the code generator generates information to allow
2673 the runtime to find the pointer at GC safe points.
2679 <!-- _______________________________________________________________________ -->
2680 <div class="doc_subsubsection">
2681 <a name="i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
2684 <div class="doc_text">
2689 declare sbyte* %llvm.gcread(sbyte** %Ptr)
2694 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
2695 locations, allowing garbage collector implementations that require read
2700 <p>The argument is the address to read from, which should be an address
2701 allocated from the garbage collector.</p>
2705 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
2706 instruction, but may be replaced with substantially more complex code by the
2707 garbage collector runtime, as needed.</p>
2712 <!-- _______________________________________________________________________ -->
2713 <div class="doc_subsubsection">
2714 <a name="i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
2717 <div class="doc_text">
2722 declare void %llvm.gcwrite(sbyte* %P1, sbyte** %P2)
2727 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
2728 locations, allowing garbage collector implementations that require write
2729 barriers (such as generational or reference counting collectors).</p>
2733 <p>The first argument is the reference to store, and the second is the heap
2734 location to store to.</p>
2738 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
2739 instruction, but may be replaced with substantially more complex code by the
2740 garbage collector runtime, as needed.</p>
2746 <!-- ======================================================================= -->
2747 <div class="doc_subsection">
2748 <a name="int_codegen">Code Generator Intrinsics</a>
2751 <div class="doc_text">
2753 These intrinsics are provided by LLVM to expose special features that may only
2754 be implemented with code generator support.
2759 <!-- _______________________________________________________________________ -->
2760 <div class="doc_subsubsection">
2761 <a name="i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
2764 <div class="doc_text">
2768 declare sbyte *%llvm.returnaddress(uint <level>)
2774 The '<tt>llvm.returnaddress</tt>' intrinsic returns a target-specific value
2775 indicating the return address of the current function or one of its callers.
2781 The argument to this intrinsic indicates which function to return the address
2782 for. Zero indicates the calling function, one indicates its caller, etc. The
2783 argument is <b>required</b> to be a constant integer value.
2789 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
2790 the return address of the specified call frame, or zero if it cannot be
2791 identified. The value returned by this intrinsic is likely to be incorrect or 0
2792 for arguments other than zero, so it should only be used for debugging purposes.
2796 Note that calling this intrinsic does not prevent function inlining or other
2797 aggressive transformations, so the value returned may not be that of the obvious
2798 source-language caller.
2803 <!-- _______________________________________________________________________ -->
2804 <div class="doc_subsubsection">
2805 <a name="i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
2808 <div class="doc_text">
2812 declare sbyte *%llvm.frameaddress(uint <level>)
2818 The '<tt>llvm.frameaddress</tt>' intrinsic returns the target-specific frame
2819 pointer value for the specified stack frame.
2825 The argument to this intrinsic indicates which function to return the frame
2826 pointer for. Zero indicates the calling function, one indicates its caller,
2827 etc. The argument is <b>required</b> to be a constant integer value.
2833 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
2834 the frame address of the specified call frame, or zero if it cannot be
2835 identified. The value returned by this intrinsic is likely to be incorrect or 0
2836 for arguments other than zero, so it should only be used for debugging purposes.
2840 Note that calling this intrinsic does not prevent function inlining or other
2841 aggressive transformations, so the value returned may not be that of the obvious
2842 source-language caller.
2846 <!-- _______________________________________________________________________ -->
2847 <div class="doc_subsubsection">
2848 <a name="i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
2851 <div class="doc_text">
2855 declare sbyte *%llvm.stacksave()
2861 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
2862 the function stack, for use with <a href="#i_stackrestore">
2863 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
2864 features like scoped automatic variable sized arrays in C99.
2870 This intrinsic returns a opaque pointer value that can be passed to <a
2871 href="#i_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
2872 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
2873 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
2874 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
2875 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
2876 that were allocated after the <tt>llvm.stacksave</tt> was executed.
2881 <!-- _______________________________________________________________________ -->
2882 <div class="doc_subsubsection">
2883 <a name="i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
2886 <div class="doc_text">
2890 declare void %llvm.stackrestore(sbyte* %ptr)
2896 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
2897 the function stack to the state it was in when the corresponding <a
2898 href="#llvm.stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
2899 useful for implementing language features like scoped automatic variable sized
2906 See the description for <a href="#i_stacksave"><tt>llvm.stacksave</tt></a>.
2912 <!-- _______________________________________________________________________ -->
2913 <div class="doc_subsubsection">
2914 <a name="i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
2917 <div class="doc_text">
2921 declare void %llvm.prefetch(sbyte * <address>,
2922 uint <rw>, uint <locality>)
2929 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
2930 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
2932 effect on the behavior of the program but can change its performance
2939 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
2940 determining if the fetch should be for a read (0) or write (1), and
2941 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
2942 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
2943 <tt>locality</tt> arguments must be constant integers.
2949 This intrinsic does not modify the behavior of the program. In particular,
2950 prefetches cannot trap and do not produce a value. On targets that support this
2951 intrinsic, the prefetch can provide hints to the processor cache for better
2957 <!-- _______________________________________________________________________ -->
2958 <div class="doc_subsubsection">
2959 <a name="i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
2962 <div class="doc_text">
2966 declare void %llvm.pcmarker( uint <id> )
2973 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
2975 code to simulators and other tools. The method is target specific, but it is
2976 expected that the marker will use exported symbols to transmit the PC of the marker.
2977 The marker makes no guarantees that it will remain with any specific instruction
2978 after optimizations. It is possible that the presence of a marker will inhibit
2979 optimizations. The intended use is to be inserted after optmizations to allow
2980 correlations of simulation runs.
2986 <tt>id</tt> is a numerical id identifying the marker.
2992 This intrinsic does not modify the behavior of the program. Backends that do not
2993 support this intrinisic may ignore it.
2998 <!-- _______________________________________________________________________ -->
2999 <div class="doc_subsubsection">
3000 <a name="i_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
3003 <div class="doc_text">
3007 declare ulong %llvm.readcyclecounter( )
3014 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
3015 counter register (or similar low latency, high accuracy clocks) on those targets
3016 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
3017 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
3018 should only be used for small timings.
3024 When directly supported, reading the cycle counter should not modify any memory.
3025 Implementations are allowed to either return a application specific value or a
3026 system wide value. On backends without support, this is lowered to a constant 0.
3032 <!-- ======================================================================= -->
3033 <div class="doc_subsection">
3034 <a name="int_os">Operating System Intrinsics</a>
3037 <div class="doc_text">
3039 These intrinsics are provided by LLVM to support the implementation of
3040 operating system level code.
3045 <!-- _______________________________________________________________________ -->
3046 <div class="doc_subsubsection">
3047 <a name="i_readport">'<tt>llvm.readport</tt>' Intrinsic</a>
3050 <div class="doc_text">
3054 declare <integer type> %llvm.readport (<integer type> <address>)
3060 The '<tt>llvm.readport</tt>' intrinsic reads data from the specified hardware
3067 The argument to this intrinsic indicates the hardware I/O address from which
3068 to read the data. The address is in the hardware I/O address namespace (as
3069 opposed to being a memory location for memory mapped I/O).
3075 The '<tt>llvm.readport</tt>' intrinsic reads data from the hardware I/O port
3076 specified by <i>address</i> and returns the value. The address and return
3077 value must be integers, but the size is dependent upon the platform upon which
3078 the program is code generated. For example, on x86, the address must be an
3079 unsigned 16-bit value, and the return value must be 8, 16, or 32 bits.
3084 <!-- _______________________________________________________________________ -->
3085 <div class="doc_subsubsection">
3086 <a name="i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a>
3089 <div class="doc_text">
3093 call void (<integer type>, <integer type>)*
3094 %llvm.writeport (<integer type> <value>,
3095 <integer type> <address>)
3101 The '<tt>llvm.writeport</tt>' intrinsic writes data to the specified hardware
3108 The first argument is the value to write to the I/O port.
3112 The second argument indicates the hardware I/O address to which data should be
3113 written. The address is in the hardware I/O address namespace (as opposed to
3114 being a memory location for memory mapped I/O).
3120 The '<tt>llvm.writeport</tt>' intrinsic writes <i>value</i> to the I/O port
3121 specified by <i>address</i>. The address and value must be integers, but the
3122 size is dependent upon the platform upon which the program is code generated.
3123 For example, on x86, the address must be an unsigned 16-bit value, and the
3124 value written must be 8, 16, or 32 bits in length.
3129 <!-- _______________________________________________________________________ -->
3130 <div class="doc_subsubsection">
3131 <a name="i_readio">'<tt>llvm.readio</tt>' Intrinsic</a>
3134 <div class="doc_text">
3138 declare <result> %llvm.readio (<ty> * <pointer>)
3144 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
3151 The argument to this intrinsic is a pointer indicating the memory address from
3152 which to read the data. The data must be a
3153 <a href="#t_firstclass">first class</a> type.
3159 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
3160 location specified by <i>pointer</i> and returns the value. The argument must
3161 be a pointer, and the return value must be a
3162 <a href="#t_firstclass">first class</a> type. However, certain architectures
3163 may not support I/O on all first class types. For example, 32-bit processors
3164 may only support I/O on data types that are 32 bits or less.
3168 This intrinsic enforces an in-order memory model for llvm.readio and
3169 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
3170 scheduled processors may execute loads and stores out of order, re-ordering at
3171 run time accesses to memory mapped I/O registers. Using these intrinsics
3172 ensures that accesses to memory mapped I/O registers occur in program order.
3177 <!-- _______________________________________________________________________ -->
3178 <div class="doc_subsubsection">
3179 <a name="i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a>
3182 <div class="doc_text">
3186 declare void %llvm.writeio (<ty1> <value>, <ty2> * <pointer>)
3192 The '<tt>llvm.writeio</tt>' intrinsic writes data to the specified memory
3199 The first argument is the value to write to the memory mapped I/O location.
3200 The second argument is a pointer indicating the memory address to which the
3201 data should be written.
3207 The '<tt>llvm.writeio</tt>' intrinsic writes <i>value</i> to the memory mapped
3208 I/O address specified by <i>pointer</i>. The value must be a
3209 <a href="#t_firstclass">first class</a> type. However, certain architectures
3210 may not support I/O on all first class types. For example, 32-bit processors
3211 may only support I/O on data types that are 32 bits or less.
3215 This intrinsic enforces an in-order memory model for llvm.readio and
3216 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
3217 scheduled processors may execute loads and stores out of order, re-ordering at
3218 run time accesses to memory mapped I/O registers. Using these intrinsics
3219 ensures that accesses to memory mapped I/O registers occur in program order.
3224 <!-- ======================================================================= -->
3225 <div class="doc_subsection">
3226 <a name="int_libc">Standard C Library Intrinsics</a>
3229 <div class="doc_text">
3231 LLVM provides intrinsics for a few important standard C library functions.
3232 These intrinsics allow source-language front-ends to pass information about the
3233 alignment of the pointer arguments to the code generator, providing opportunity
3234 for more efficient code generation.
3239 <!-- _______________________________________________________________________ -->
3240 <div class="doc_subsubsection">
3241 <a name="i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
3244 <div class="doc_text">
3248 declare void %llvm.memcpy(sbyte* <dest>, sbyte* <src>,
3249 uint <len>, uint <align>)
3255 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
3256 location to the destination location.
3260 Note that, unlike the standard libc function, the <tt>llvm.memcpy</tt> intrinsic
3261 does not return a value, and takes an extra alignment argument.
3267 The first argument is a pointer to the destination, the second is a pointer to
3268 the source. The third argument is an (arbitrarily sized) integer argument
3269 specifying the number of bytes to copy, and the fourth argument is the alignment
3270 of the source and destination locations.
3274 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3275 the caller guarantees that the size of the copy is a multiple of the alignment
3276 and that both the source and destination pointers are aligned to that boundary.
3282 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
3283 location to the destination location, which are not allowed to overlap. It
3284 copies "len" bytes of memory over. If the argument is known to be aligned to
3285 some boundary, this can be specified as the fourth argument, otherwise it should
3291 <!-- _______________________________________________________________________ -->
3292 <div class="doc_subsubsection">
3293 <a name="i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
3296 <div class="doc_text">
3300 declare void %llvm.memmove(sbyte* <dest>, sbyte* <src>,
3301 uint <len>, uint <align>)
3307 The '<tt>llvm.memmove</tt>' intrinsic moves a block of memory from the source
3308 location to the destination location. It is similar to the '<tt>llvm.memcpy</tt>'
3309 intrinsic but allows the two memory locations to overlap.
3313 Note that, unlike the standard libc function, the <tt>llvm.memmove</tt> intrinsic
3314 does not return a value, and takes an extra alignment argument.
3320 The first argument is a pointer to the destination, the second is a pointer to
3321 the source. The third argument is an (arbitrarily sized) integer argument
3322 specifying the number of bytes to copy, and the fourth argument is the alignment
3323 of the source and destination locations.
3327 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3328 the caller guarantees that the size of the copy is a multiple of the alignment
3329 and that both the source and destination pointers are aligned to that boundary.
3335 The '<tt>llvm.memmove</tt>' intrinsic copies a block of memory from the source
3336 location to the destination location, which may overlap. It
3337 copies "len" bytes of memory over. If the argument is known to be aligned to
3338 some boundary, this can be specified as the fourth argument, otherwise it should
3344 <!-- _______________________________________________________________________ -->
3345 <div class="doc_subsubsection">
3346 <a name="i_memset">'<tt>llvm.memset</tt>' Intrinsic</a>
3349 <div class="doc_text">
3353 declare void %llvm.memset(sbyte* <dest>, ubyte <val>,
3354 uint <len>, uint <align>)
3360 The '<tt>llvm.memset</tt>' intrinsic fills a block of memory with a particular
3365 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
3366 does not return a value, and takes an extra alignment argument.
3372 The first argument is a pointer to the destination to fill, the second is the
3373 byte value to fill it with, the third argument is an (arbitrarily sized) integer
3374 argument specifying the number of bytes to fill, and the fourth argument is the
3375 known alignment of destination location.
3379 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3380 the caller guarantees that the size of the copy is a multiple of the alignment
3381 and that the destination pointer is aligned to that boundary.
3387 The '<tt>llvm.memset</tt>' intrinsic fills "len" bytes of memory starting at the
3388 destination location. If the argument is known to be aligned to some boundary,
3389 this can be specified as the fourth argument, otherwise it should be set to 0 or
3395 <!-- _______________________________________________________________________ -->
3396 <div class="doc_subsubsection">
3397 <a name="i_isunordered">'<tt>llvm.isunordered</tt>' Intrinsic</a>
3400 <div class="doc_text">
3404 declare bool %llvm.isunordered(<float or double> Val1, <float or double> Val2)
3410 The '<tt>llvm.isunordered</tt>' intrinsic returns true if either or both of the
3411 specified floating point values is a NAN.
3417 The arguments are floating point numbers of the same type.
3423 If either or both of the arguments is a SNAN or QNAN, it returns true, otherwise
3429 <!-- _______________________________________________________________________ -->
3430 <div class="doc_subsubsection">
3431 <a name="i_sqrt">'<tt>llvm.sqrt</tt>' Intrinsic</a>
3434 <div class="doc_text">
3438 declare <float or double> %llvm.sqrt(<float or double> Val)
3444 The '<tt>llvm.sqrt</tt>' intrinsic returns the sqrt of the specified operand,
3445 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
3446 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
3447 negative numbers (which allows for better optimization).
3453 The argument and return value are floating point numbers of the same type.
3459 This function returns the sqrt of the specified operand if it is a positive
3460 floating point number.
3464 <!-- ======================================================================= -->
3465 <div class="doc_subsection">
3466 <a name="int_manip">Bit Manipulation Intrinsics</a>
3469 <div class="doc_text">
3471 LLVM provides intrinsics for a few important bit manipulation operations.
3472 These allow efficient code generation for some algorithms.
3477 <!-- _______________________________________________________________________ -->
3478 <div class="doc_subsubsection">
3479 <a name="i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
3482 <div class="doc_text">
3486 declare ushort %llvm.bswap.i16( ushort <id> )
3487 declare uint %llvm.bswap.i32( uint <id> )
3488 declare ulong %llvm.bswap.i64( ulong <id> )
3494 The '<tt>llvm.bwsap</tt>' family of intrinsics is used to byteswap a 16, 32 or
3495 64 bit quantity. These are useful for performing operations on data that is not
3496 in the target's native byte order.
3502 The llvm.bswap.16 intrinsic returns a ushort value that has the high and low
3503 byte of the input ushort swapped. Similarly, the llvm.bswap.i32 intrinsic
3504 returns a uint value that has the four bytes of the input uint swapped, so that
3505 if the input bytes are numbered 0, 1, 2, 3 then the returned uint will have its
3506 bytes in 3, 2, 1, 0 order. The llvm.bswap.i64 intrinsic extends this concept
3512 <!-- _______________________________________________________________________ -->
3513 <div class="doc_subsubsection">
3514 <a name="int_ctpop">'<tt>llvm.ctpop</tt>' Intrinsic</a>
3517 <div class="doc_text">
3521 declare int %llvm.ctpop(int <src>)
3527 The '<tt>llvm.ctpop</tt>' intrinsic counts the number of ones in a variable.
3533 The only argument is the value to be counted. The argument may be of any
3534 integer type. The return type must match the argument type.
3540 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
3544 <!-- _______________________________________________________________________ -->
3545 <div class="doc_subsubsection">
3546 <a name="int_ctlz">'<tt>llvm.ctlz</tt>' Intrinsic</a>
3549 <div class="doc_text">
3553 declare int %llvm.ctlz(int <src>)
3560 The '<tt>llvm.ctlz</tt>' intrinsic counts the number of leading zeros in a
3567 The only argument is the value to be counted. The argument may be of any
3568 integer type. The return type must match the argument type.
3574 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
3575 in a variable. If the src == 0 then the result is the size in bits of the type
3576 of src. For example, <tt>llvm.cttz(int 2) = 30</tt>.
3582 <!-- _______________________________________________________________________ -->
3583 <div class="doc_subsubsection">
3584 <a name="int_cttz">'<tt>llvm.cttz</tt>' Intrinsic</a>
3587 <div class="doc_text">
3591 declare int %llvm.cttz(int <src>)
3597 The '<tt>llvm.cttz</tt>' intrinsic counts the number of trailing zeros.
3603 The only argument is the value to be counted. The argument may be of any
3604 integer type. The return type must match the argument type.
3610 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
3611 in a variable. If the src == 0 then the result is the size in bits of the type
3612 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
3616 <!-- ======================================================================= -->
3617 <div class="doc_subsection">
3618 <a name="int_debugger">Debugger Intrinsics</a>
3621 <div class="doc_text">
3623 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
3624 are described in the <a
3625 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
3626 Debugging</a> document.
3631 <!-- *********************************************************************** -->
3634 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3635 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
3636 <a href="http://validator.w3.org/check/referer"><img
3637 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
3639 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3640 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br>
3641 Last modified: $Date$