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_call">'<tt>call</tt>' Instruction</a></li>
105 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
110 <li><a href="#intrinsics">Intrinsic Functions</a>
112 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
114 <li><a href="#i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
115 <li><a href="#i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
116 <li><a href="#i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
119 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
121 <li><a href="#i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
122 <li><a href="#i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
123 <li><a href="#i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
126 <li><a href="#int_codegen">Code Generator Intrinsics</a>
128 <li><a href="#i_bswap_i16">'<tt>llvm.bswap.i16</tt>' Intrinsic</a></li>
129 <li><a href="#i_bswap_i32">'<tt>llvm.bswap.i32</tt>' Intrinsic</a></li>
130 <li><a href="#i_bswap_i64">'<tt>llvm.bswap.i64</tt>' Intrinsic</a></li>
131 <li><a href="#i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
132 <li><a href="#i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
133 <li><a href="#i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
134 <li><a href="#i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
135 <li><a href="#i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
136 <li><a href="#i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
137 <li><a href="#i_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
140 <li><a href="#int_os">Operating System Intrinsics</a>
142 <li><a href="#i_readport">'<tt>llvm.readport</tt>' Intrinsic</a></li>
143 <li><a href="#i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a></li>
144 <li><a href="#i_readio">'<tt>llvm.readio</tt>' Intrinsic</a></li>
145 <li><a href="#i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a></li>
147 <li><a href="#int_libc">Standard C Library Intrinsics</a>
149 <li><a href="#i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a></li>
150 <li><a href="#i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a></li>
151 <li><a href="#i_memset">'<tt>llvm.memset</tt>' Intrinsic</a></li>
152 <li><a href="#i_isunordered">'<tt>llvm.isunordered</tt>' Intrinsic</a></li>
153 <li><a href="#i_sqrt">'<tt>llvm.sqrt</tt>' Intrinsic</a></li>
157 <li><a href="#int_count">Bit counting Intrinsics</a>
159 <li><a href="#int_ctpop">'<tt>llvm.ctpop</tt>' Intrinsic </a></li>
160 <li><a href="#int_ctlz">'<tt>llvm.ctlz</tt>' Intrinsic </a></li>
161 <li><a href="#int_cttz">'<tt>llvm.cttz</tt>' Intrinsic </a></li>
164 <li><a href="#int_debugger">Debugger intrinsics</a></li>
169 <div class="doc_author">
170 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
171 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
174 <!-- *********************************************************************** -->
175 <div class="doc_section"> <a name="abstract">Abstract </a></div>
176 <!-- *********************************************************************** -->
178 <div class="doc_text">
179 <p>This document is a reference manual for the LLVM assembly language.
180 LLVM is an SSA based representation that provides type safety,
181 low-level operations, flexibility, and the capability of representing
182 'all' high-level languages cleanly. It is the common code
183 representation used throughout all phases of the LLVM compilation
187 <!-- *********************************************************************** -->
188 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
189 <!-- *********************************************************************** -->
191 <div class="doc_text">
193 <p>The LLVM code representation is designed to be used in three
194 different forms: as an in-memory compiler IR, as an on-disk bytecode
195 representation (suitable for fast loading by a Just-In-Time compiler),
196 and as a human readable assembly language representation. This allows
197 LLVM to provide a powerful intermediate representation for efficient
198 compiler transformations and analysis, while providing a natural means
199 to debug and visualize the transformations. The three different forms
200 of LLVM are all equivalent. This document describes the human readable
201 representation and notation.</p>
203 <p>The LLVM representation aims to be light-weight and low-level
204 while being expressive, typed, and extensible at the same time. It
205 aims to be a "universal IR" of sorts, by being at a low enough level
206 that high-level ideas may be cleanly mapped to it (similar to how
207 microprocessors are "universal IR's", allowing many source languages to
208 be mapped to them). By providing type information, LLVM can be used as
209 the target of optimizations: for example, through pointer analysis, it
210 can be proven that a C automatic variable is never accessed outside of
211 the current function... allowing it to be promoted to a simple SSA
212 value instead of a memory location.</p>
216 <!-- _______________________________________________________________________ -->
217 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
219 <div class="doc_text">
221 <p>It is important to note that this document describes 'well formed'
222 LLVM assembly language. There is a difference between what the parser
223 accepts and what is considered 'well formed'. For example, the
224 following instruction is syntactically okay, but not well formed:</p>
227 %x = <a href="#i_add">add</a> int 1, %x
230 <p>...because the definition of <tt>%x</tt> does not dominate all of
231 its uses. The LLVM infrastructure provides a verification pass that may
232 be used to verify that an LLVM module is well formed. This pass is
233 automatically run by the parser after parsing input assembly and by
234 the optimizer before it outputs bytecode. The violations pointed out
235 by the verifier pass indicate bugs in transformation passes or input to
238 <!-- Describe the typesetting conventions here. --> </div>
240 <!-- *********************************************************************** -->
241 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
242 <!-- *********************************************************************** -->
244 <div class="doc_text">
246 <p>LLVM uses three different forms of identifiers, for different
250 <li>Named values are represented as a string of characters with a '%' prefix.
251 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
252 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
253 Identifiers which require other characters in their names can be surrounded
254 with quotes. In this way, anything except a <tt>"</tt> character can be used
257 <li>Unnamed values are represented as an unsigned numeric value with a '%'
258 prefix. For example, %12, %2, %44.</li>
260 <li>Constants, which are described in a <a href="#constants">section about
261 constants</a>, below.</li>
264 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
265 don't need to worry about name clashes with reserved words, and the set of
266 reserved words may be expanded in the future without penalty. Additionally,
267 unnamed identifiers allow a compiler to quickly come up with a temporary
268 variable without having to avoid symbol table conflicts.</p>
270 <p>Reserved words in LLVM are very similar to reserved words in other
271 languages. There are keywords for different opcodes ('<tt><a
272 href="#i_add">add</a></tt>', '<tt><a href="#i_cast">cast</a></tt>', '<tt><a
273 href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
274 href="#t_void">void</a></tt>', '<tt><a href="#t_uint">uint</a></tt>', etc...),
275 and others. These reserved words cannot conflict with variable names, because
276 none of them start with a '%' character.</p>
278 <p>Here is an example of LLVM code to multiply the integer variable
279 '<tt>%X</tt>' by 8:</p>
284 %result = <a href="#i_mul">mul</a> uint %X, 8
287 <p>After strength reduction:</p>
290 %result = <a href="#i_shl">shl</a> uint %X, ubyte 3
293 <p>And the hard way:</p>
296 <a href="#i_add">add</a> uint %X, %X <i>; yields {uint}:%0</i>
297 <a href="#i_add">add</a> uint %0, %0 <i>; yields {uint}:%1</i>
298 %result = <a href="#i_add">add</a> uint %1, %1
301 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
302 important lexical features of LLVM:</p>
306 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
309 <li>Unnamed temporaries are created when the result of a computation is not
310 assigned to a named value.</li>
312 <li>Unnamed temporaries are numbered sequentially</li>
316 <p>...and it also shows a convention that we follow in this document. When
317 demonstrating instructions, we will follow an instruction with a comment that
318 defines the type and name of value produced. Comments are shown in italic
323 <!-- *********************************************************************** -->
324 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
325 <!-- *********************************************************************** -->
327 <!-- ======================================================================= -->
328 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
331 <div class="doc_text">
333 <p>LLVM programs are composed of "Module"s, each of which is a
334 translation unit of the input programs. Each module consists of
335 functions, global variables, and symbol table entries. Modules may be
336 combined together with the LLVM linker, which merges function (and
337 global variable) definitions, resolves forward declarations, and merges
338 symbol table entries. Here is an example of the "hello world" module:</p>
340 <pre><i>; Declare the string constant as a global constant...</i>
341 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
342 href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
344 <i>; External declaration of the puts function</i>
345 <a href="#functionstructure">declare</a> int %puts(sbyte*) <i>; int(sbyte*)* </i>
347 <i>; Definition of main function</i>
348 int %main() { <i>; int()* </i>
349 <i>; Convert [13x sbyte]* to sbyte *...</i>
351 href="#i_getelementptr">getelementptr</a> [13 x sbyte]* %.LC0, long 0, long 0 <i>; sbyte*</i>
353 <i>; Call puts function to write out the string to stdout...</i>
355 href="#i_call">call</a> int %puts(sbyte* %cast210) <i>; int</i>
357 href="#i_ret">ret</a> int 0<br>}<br></pre>
359 <p>This example is made up of a <a href="#globalvars">global variable</a>
360 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
361 function, and a <a href="#functionstructure">function definition</a>
362 for "<tt>main</tt>".</p>
364 <p>In general, a module is made up of a list of global values,
365 where both functions and global variables are global values. Global values are
366 represented by a pointer to a memory location (in this case, a pointer to an
367 array of char, and a pointer to a function), and have one of the following <a
368 href="#linkage">linkage types</a>.</p>
372 <!-- ======================================================================= -->
373 <div class="doc_subsection">
374 <a name="linkage">Linkage Types</a>
377 <div class="doc_text">
380 All Global Variables and Functions have one of the following types of linkage:
385 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
387 <dd>Global values with internal linkage are only directly accessible by
388 objects in the current module. In particular, linking code into a module with
389 an internal global value may cause the internal to be renamed as necessary to
390 avoid collisions. Because the symbol is internal to the module, all
391 references can be updated. This corresponds to the notion of the
392 '<tt>static</tt>' keyword in C, or the idea of "anonymous namespaces" in C++.
395 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
397 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt> linkage, with
398 the twist that linking together two modules defining the same
399 <tt>linkonce</tt> globals will cause one of the globals to be discarded. This
400 is typically used to implement inline functions. Unreferenced
401 <tt>linkonce</tt> globals are allowed to be discarded.
404 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
406 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
407 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
408 used to implement constructs in C such as "<tt>int X;</tt>" at global scope.
411 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
413 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
414 pointer to array type. When two global variables with appending linkage are
415 linked together, the two global arrays are appended together. This is the
416 LLVM, typesafe, equivalent of having the system linker append together
417 "sections" with identical names when .o files are linked.
420 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
422 <dd>If none of the above identifiers are used, the global is externally
423 visible, meaning that it participates in linkage and can be used to resolve
424 external symbol references.
428 <p><a name="linkage_external">For example, since the "<tt>.LC0</tt>"
429 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
430 variable and was linked with this one, one of the two would be renamed,
431 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
432 external (i.e., lacking any linkage declarations), they are accessible
433 outside of the current module. It is illegal for a function <i>declaration</i>
434 to have any linkage type other than "externally visible".</a></p>
438 <!-- ======================================================================= -->
439 <div class="doc_subsection">
440 <a name="callingconv">Calling Conventions</a>
443 <div class="doc_text">
445 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
446 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
447 specified for the call. The calling convention of any pair of dynamic
448 caller/callee must match, or the behavior of the program is undefined. The
449 following calling conventions are supported by LLVM, and more may be added in
453 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
455 <dd>This calling convention (the default if no other calling convention is
456 specified) matches the target C calling conventions. This calling convention
457 supports varargs function calls and tolerates some mismatch in the declared
458 prototype and implemented declaration of the function (as does normal C).
461 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
463 <dd>This calling convention attempts to make calls as fast as possible
464 (e.g. by passing things in registers). This calling convention allows the
465 target to use whatever tricks it wants to produce fast code for the target,
466 without having to conform to an externally specified ABI. Implementations of
467 this convention should allow arbitrary tail call optimization to be supported.
468 This calling convention does not support varargs and requires the prototype of
469 all callees to exactly match the prototype of the function definition.
472 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
474 <dd>This calling convention attempts to make code in the caller as efficient
475 as possible under the assumption that the call is not commonly executed. As
476 such, these calls often preserve all registers so that the call does not break
477 any live ranges in the caller side. This calling convention does not support
478 varargs and requires the prototype of all callees to exactly match the
479 prototype of the function definition.
482 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
484 <dd>Any calling convention may be specified by number, allowing
485 target-specific calling conventions to be used. Target specific calling
486 conventions start at 64.
490 <p>More calling conventions can be added/defined on an as-needed basis, to
491 support pascal conventions or any other well-known target-independent
496 <!-- ======================================================================= -->
497 <div class="doc_subsection">
498 <a name="globalvars">Global Variables</a>
501 <div class="doc_text">
503 <p>Global variables define regions of memory allocated at compilation time
504 instead of run-time. Global variables may optionally be initialized, may have
505 an explicit section to be placed in, and may
506 have an optional explicit alignment specified. A
507 variable may be defined as a global "constant," which indicates that the
508 contents of the variable will <b>never</b> be modified (enabling better
509 optimization, allowing the global data to be placed in the read-only section of
510 an executable, etc). Note that variables that need runtime initialization
511 cannot be marked "constant" as there is a store to the variable.</p>
514 LLVM explicitly allows <em>declarations</em> of global variables to be marked
515 constant, even if the final definition of the global is not. This capability
516 can be used to enable slightly better optimization of the program, but requires
517 the language definition to guarantee that optimizations based on the
518 'constantness' are valid for the translation units that do not include the
522 <p>As SSA values, global variables define pointer values that are in
523 scope (i.e. they dominate) all basic blocks in the program. Global
524 variables always define a pointer to their "content" type because they
525 describe a region of memory, and all memory objects in LLVM are
526 accessed through pointers.</p>
528 <p>LLVM allows an explicit section to be specified for globals. If the target
529 supports it, it will emit globals to the section specified.</p>
531 <p>An explicit alignment may be specified for a global. If not present, or if
532 the alignment is set to zero, the alignment of the global is set by the target
533 to whatever it feels convenient. If an explicit alignment is specified, the
534 global is forced to have at least that much alignment. All alignments must be
540 <!-- ======================================================================= -->
541 <div class="doc_subsection">
542 <a name="functionstructure">Functions</a>
545 <div class="doc_text">
547 <p>LLVM function definitions consist of an optional <a href="#linkage">linkage
548 type</a>, an optional <a href="#callingconv">calling convention</a>, a return
549 type, a function name, a (possibly empty) argument list, an optional section,
550 an optional alignment, an opening curly brace,
551 a list of basic blocks, and a closing curly brace. LLVM function declarations
552 are defined with the "<tt>declare</tt>" keyword, an optional <a
553 href="#callingconv">calling convention</a>, a return type, a function name,
554 a possibly empty list of arguments, and an optional alignment.</p>
556 <p>A function definition contains a list of basic blocks, forming the CFG for
557 the function. Each basic block may optionally start with a label (giving the
558 basic block a symbol table entry), contains a list of instructions, and ends
559 with a <a href="#terminators">terminator</a> instruction (such as a branch or
560 function return).</p>
562 <p>The first basic block in a program is special in two ways: it is immediately
563 executed on entrance to the function, and it is not allowed to have predecessor
564 basic blocks (i.e. there can not be any branches to the entry block of a
565 function). Because the block can have no predecessors, it also cannot have any
566 <a href="#i_phi">PHI nodes</a>.</p>
568 <p>LLVM functions are identified by their name and type signature. Hence, two
569 functions with the same name but different parameter lists or return values are
570 considered different functions, and LLVM will resolve references to each
573 <p>LLVM allows an explicit section to be specified for functions. If the target
574 supports it, it will emit functions to the section specified.</p>
576 <p>An explicit alignment may be specified for a function. If not present, or if
577 the alignment is set to zero, the alignment of the function is set by the target
578 to whatever it feels convenient. If an explicit alignment is specified, the
579 function is forced to have at least that much alignment. All alignments must be
586 <!-- *********************************************************************** -->
587 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
588 <!-- *********************************************************************** -->
590 <div class="doc_text">
592 <p>The LLVM type system is one of the most important features of the
593 intermediate representation. Being typed enables a number of
594 optimizations to be performed on the IR directly, without having to do
595 extra analyses on the side before the transformation. A strong type
596 system makes it easier to read the generated code and enables novel
597 analyses and transformations that are not feasible to perform on normal
598 three address code representations.</p>
602 <!-- ======================================================================= -->
603 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
604 <div class="doc_text">
605 <p>The primitive types are the fundamental building blocks of the LLVM
606 system. The current set of primitive types is as follows:</p>
608 <table class="layout">
613 <tr><th>Type</th><th>Description</th></tr>
614 <tr><td><tt>void</tt></td><td>No value</td></tr>
615 <tr><td><tt>ubyte</tt></td><td>Unsigned 8-bit value</td></tr>
616 <tr><td><tt>ushort</tt></td><td>Unsigned 16-bit value</td></tr>
617 <tr><td><tt>uint</tt></td><td>Unsigned 32-bit value</td></tr>
618 <tr><td><tt>ulong</tt></td><td>Unsigned 64-bit value</td></tr>
619 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
620 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
627 <tr><th>Type</th><th>Description</th></tr>
628 <tr><td><tt>bool</tt></td><td>True or False value</td></tr>
629 <tr><td><tt>sbyte</tt></td><td>Signed 8-bit value</td></tr>
630 <tr><td><tt>short</tt></td><td>Signed 16-bit value</td></tr>
631 <tr><td><tt>int</tt></td><td>Signed 32-bit value</td></tr>
632 <tr><td><tt>long</tt></td><td>Signed 64-bit value</td></tr>
633 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
641 <!-- _______________________________________________________________________ -->
642 <div class="doc_subsubsection"> <a name="t_classifications">Type
643 Classifications</a> </div>
644 <div class="doc_text">
645 <p>These different primitive types fall into a few useful
648 <table border="1" cellspacing="0" cellpadding="4">
650 <tr><th>Classification</th><th>Types</th></tr>
652 <td><a name="t_signed">signed</a></td>
653 <td><tt>sbyte, short, int, long, float, double</tt></td>
656 <td><a name="t_unsigned">unsigned</a></td>
657 <td><tt>ubyte, ushort, uint, ulong</tt></td>
660 <td><a name="t_integer">integer</a></td>
661 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
664 <td><a name="t_integral">integral</a></td>
665 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt>
669 <td><a name="t_floating">floating point</a></td>
670 <td><tt>float, double</tt></td>
673 <td><a name="t_firstclass">first class</a></td>
674 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long,<br>
675 float, double, <a href="#t_pointer">pointer</a>,
676 <a href="#t_packed">packed</a></tt></td>
681 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
682 most important. Values of these types are the only ones which can be
683 produced by instructions, passed as arguments, or used as operands to
684 instructions. This means that all structures and arrays must be
685 manipulated either by pointer or by component.</p>
688 <!-- ======================================================================= -->
689 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
691 <div class="doc_text">
693 <p>The real power in LLVM comes from the derived types in the system.
694 This is what allows a programmer to represent arrays, functions,
695 pointers, and other useful types. Note that these derived types may be
696 recursive: For example, it is possible to have a two dimensional array.</p>
700 <!-- _______________________________________________________________________ -->
701 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
703 <div class="doc_text">
707 <p>The array type is a very simple derived type that arranges elements
708 sequentially in memory. The array type requires a size (number of
709 elements) and an underlying data type.</p>
714 [<# elements> x <elementtype>]
717 <p>The number of elements is a constant integer value; elementtype may
718 be any type with a size.</p>
721 <table class="layout">
724 <tt>[40 x int ]</tt><br/>
725 <tt>[41 x int ]</tt><br/>
726 <tt>[40 x uint]</tt><br/>
729 Array of 40 integer values.<br/>
730 Array of 41 integer values.<br/>
731 Array of 40 unsigned integer values.<br/>
735 <p>Here are some examples of multidimensional arrays:</p>
736 <table class="layout">
739 <tt>[3 x [4 x int]]</tt><br/>
740 <tt>[12 x [10 x float]]</tt><br/>
741 <tt>[2 x [3 x [4 x uint]]]</tt><br/>
744 3x4 array of integer values.<br/>
745 12x10 array of single precision floating point values.<br/>
746 2x3x4 array of unsigned integer values.<br/>
751 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
752 length array. Normally, accesses past the end of an array are undefined in
753 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
754 As a special case, however, zero length arrays are recognized to be variable
755 length. This allows implementation of 'pascal style arrays' with the LLVM
756 type "{ int, [0 x float]}", for example.</p>
760 <!-- _______________________________________________________________________ -->
761 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
762 <div class="doc_text">
764 <p>The function type can be thought of as a function signature. It
765 consists of a return type and a list of formal parameter types.
766 Function types are usually used to build virtual function tables
767 (which are structures of pointers to functions), for indirect function
768 calls, and when defining a function.</p>
770 The return type of a function type cannot be an aggregate type.
773 <pre> <returntype> (<parameter list>)<br></pre>
774 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
775 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
776 which indicates that the function takes a variable number of arguments.
777 Variable argument functions can access their arguments with the <a
778 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
780 <table class="layout">
783 <tt>int (int)</tt> <br/>
784 <tt>float (int, int *) *</tt><br/>
785 <tt>int (sbyte *, ...)</tt><br/>
788 function taking an <tt>int</tt>, returning an <tt>int</tt><br/>
789 <a href="#t_pointer">Pointer</a> to a function that takes an
790 <tt>int</tt> and a <a href="#t_pointer">pointer</a> to <tt>int</tt>,
791 returning <tt>float</tt>.<br/>
792 A vararg function that takes at least one <a href="#t_pointer">pointer</a>
793 to <tt>sbyte</tt> (signed char in C), which returns an integer. This is
794 the signature for <tt>printf</tt> in LLVM.<br/>
800 <!-- _______________________________________________________________________ -->
801 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
802 <div class="doc_text">
804 <p>The structure type is used to represent a collection of data members
805 together in memory. The packing of the field types is defined to match
806 the ABI of the underlying processor. The elements of a structure may
807 be any type that has a size.</p>
808 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
809 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
810 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
813 <pre> { <type list> }<br></pre>
815 <table class="layout">
818 <tt>{ int, int, int }</tt><br/>
819 <tt>{ float, int (int) * }</tt><br/>
822 a triple of three <tt>int</tt> values<br/>
823 A pair, where the first element is a <tt>float</tt> and the second element
824 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
825 that takes an <tt>int</tt>, returning an <tt>int</tt>.<br/>
831 <!-- _______________________________________________________________________ -->
832 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
833 <div class="doc_text">
835 <p>As in many languages, the pointer type represents a pointer or
836 reference to another object, which must live in memory.</p>
838 <pre> <type> *<br></pre>
840 <table class="layout">
843 <tt>[4x int]*</tt><br/>
844 <tt>int (int *) *</tt><br/>
847 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
848 four <tt>int</tt> values<br/>
849 A <a href="#t_pointer">pointer</a> to a <a
850 href="#t_function">function</a> that takes an <tt>int*</tt>, returning an
857 <!-- _______________________________________________________________________ -->
858 <div class="doc_subsubsection"> <a name="t_packed">Packed Type</a> </div>
859 <div class="doc_text">
863 <p>A packed type is a simple derived type that represents a vector
864 of elements. Packed types are used when multiple primitive data
865 are operated in parallel using a single instruction (SIMD).
866 A packed type requires a size (number of
867 elements) and an underlying primitive data type. Vectors must have a power
868 of two length (1, 2, 4, 8, 16 ...). Packed types are
869 considered <a href="#t_firstclass">first class</a>.</p>
874 < <# elements> x <elementtype> >
877 <p>The number of elements is a constant integer value; elementtype may
878 be any integral or floating point type.</p>
882 <table class="layout">
885 <tt><4 x int></tt><br/>
886 <tt><8 x float></tt><br/>
887 <tt><2 x uint></tt><br/>
890 Packed vector of 4 integer values.<br/>
891 Packed vector of 8 floating-point values.<br/>
892 Packed vector of 2 unsigned integer values.<br/>
898 <!-- _______________________________________________________________________ -->
899 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
900 <div class="doc_text">
904 <p>Opaque types are used to represent unknown types in the system. This
905 corresponds (for example) to the C notion of a foward declared structure type.
906 In LLVM, opaque types can eventually be resolved to any type (not just a
917 <table class="layout">
930 <!-- *********************************************************************** -->
931 <div class="doc_section"> <a name="constants">Constants</a> </div>
932 <!-- *********************************************************************** -->
934 <div class="doc_text">
936 <p>LLVM has several different basic types of constants. This section describes
937 them all and their syntax.</p>
941 <!-- ======================================================================= -->
942 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
944 <div class="doc_text">
947 <dt><b>Boolean constants</b></dt>
949 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
950 constants of the <tt><a href="#t_primitive">bool</a></tt> type.
953 <dt><b>Integer constants</b></dt>
955 <dd>Standard integers (such as '4') are constants of the <a
956 href="#t_integer">integer</a> type. Negative numbers may be used with signed
960 <dt><b>Floating point constants</b></dt>
962 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
963 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
964 notation (see below). Floating point constants must have a <a
965 href="#t_floating">floating point</a> type. </dd>
967 <dt><b>Null pointer constants</b></dt>
969 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
970 and must be of <a href="#t_pointer">pointer type</a>.</dd>
974 <p>The one non-intuitive notation for constants is the optional hexadecimal form
975 of floating point constants. For example, the form '<tt>double
976 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
977 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
978 (and the only time that they are generated by the disassembler) is when a
979 floating point constant must be emitted but it cannot be represented as a
980 decimal floating point number. For example, NaN's, infinities, and other
981 special values are represented in their IEEE hexadecimal format so that
982 assembly and disassembly do not cause any bits to change in the constants.</p>
986 <!-- ======================================================================= -->
987 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
990 <div class="doc_text">
991 <p>Aggregate constants arise from aggregation of simple constants
992 and smaller aggregate constants.</p>
995 <dt><b>Structure constants</b></dt>
997 <dd>Structure constants are represented with notation similar to structure
998 type definitions (a comma separated list of elements, surrounded by braces
999 (<tt>{}</tt>)). For example: "<tt>{ int 4, float 17.0, int* %G }</tt>",
1000 where "<tt>%G</tt>" is declared as "<tt>%G = external global int</tt>". Structure constants
1001 must have <a href="#t_struct">structure type</a>, and the number and
1002 types of elements must match those specified by the type.
1005 <dt><b>Array constants</b></dt>
1007 <dd>Array constants are represented with notation similar to array type
1008 definitions (a comma separated list of elements, surrounded by square brackets
1009 (<tt>[]</tt>)). For example: "<tt>[ int 42, int 11, int 74 ]</tt>". Array
1010 constants must have <a href="#t_array">array type</a>, and the number and
1011 types of elements must match those specified by the type.
1014 <dt><b>Packed constants</b></dt>
1016 <dd>Packed constants are represented with notation similar to packed type
1017 definitions (a comma separated list of elements, surrounded by
1018 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< int 42,
1019 int 11, int 74, int 100 ></tt>". Packed constants must have <a
1020 href="#t_packed">packed type</a>, and the number and types of elements must
1021 match those specified by the type.
1024 <dt><b>Zero initialization</b></dt>
1026 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1027 value to zero of <em>any</em> type, including scalar and aggregate types.
1028 This is often used to avoid having to print large zero initializers (e.g. for
1029 large arrays) and is always exactly equivalent to using explicit zero
1036 <!-- ======================================================================= -->
1037 <div class="doc_subsection">
1038 <a name="globalconstants">Global Variable and Function Addresses</a>
1041 <div class="doc_text">
1043 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1044 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1045 constants. These constants are explicitly referenced when the <a
1046 href="#identifiers">identifier for the global</a> is used and always have <a
1047 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1053 %Z = global [2 x int*] [ int* %X, int* %Y ]
1058 <!-- ======================================================================= -->
1059 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1060 <div class="doc_text">
1061 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1062 no specific value. Undefined values may be of any type and be used anywhere
1063 a constant is permitted.</p>
1065 <p>Undefined values indicate to the compiler that the program is well defined
1066 no matter what value is used, giving the compiler more freedom to optimize.
1070 <!-- ======================================================================= -->
1071 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1074 <div class="doc_text">
1076 <p>Constant expressions are used to allow expressions involving other constants
1077 to be used as constants. Constant expressions may be of any <a
1078 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1079 that does not have side effects (e.g. load and call are not supported). The
1080 following is the syntax for constant expressions:</p>
1083 <dt><b><tt>cast ( CST to TYPE )</tt></b></dt>
1085 <dd>Cast a constant to another type.</dd>
1087 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1089 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1090 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1091 instruction, the index list may have zero or more indexes, which are required
1092 to make sense for the type of "CSTPTR".</dd>
1094 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1096 <dd>Perform the <a href="#i_select">select operation</a> on
1099 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1101 <dd>Perform the <a href="#i_extractelement">extractelement
1102 operation</a> on constants.
1104 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1106 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1107 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1108 binary</a> operations. The constraints on operands are the same as those for
1109 the corresponding instruction (e.g. no bitwise operations on floating point
1110 values are allowed).</dd>
1114 <!-- *********************************************************************** -->
1115 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1116 <!-- *********************************************************************** -->
1118 <div class="doc_text">
1120 <p>The LLVM instruction set consists of several different
1121 classifications of instructions: <a href="#terminators">terminator
1122 instructions</a>, <a href="#binaryops">binary instructions</a>,
1123 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1124 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1125 instructions</a>.</p>
1129 <!-- ======================================================================= -->
1130 <div class="doc_subsection"> <a name="terminators">Terminator
1131 Instructions</a> </div>
1133 <div class="doc_text">
1135 <p>As mentioned <a href="#functionstructure">previously</a>, every
1136 basic block in a program ends with a "Terminator" instruction, which
1137 indicates which block should be executed after the current block is
1138 finished. These terminator instructions typically yield a '<tt>void</tt>'
1139 value: they produce control flow, not values (the one exception being
1140 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1141 <p>There are six different terminator instructions: the '<a
1142 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1143 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1144 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1145 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1146 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1150 <!-- _______________________________________________________________________ -->
1151 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1152 Instruction</a> </div>
1153 <div class="doc_text">
1155 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1156 ret void <i>; Return from void function</i>
1159 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1160 value) from a function back to the caller.</p>
1161 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1162 returns a value and then causes control flow, and one that just causes
1163 control flow to occur.</p>
1165 <p>The '<tt>ret</tt>' instruction may return any '<a
1166 href="#t_firstclass">first class</a>' type. Notice that a function is
1167 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1168 instruction inside of the function that returns a value that does not
1169 match the return type of the function.</p>
1171 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1172 returns back to the calling function's context. If the caller is a "<a
1173 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1174 the instruction after the call. If the caller was an "<a
1175 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1176 at the beginning of the "normal" destination block. If the instruction
1177 returns a value, that value shall set the call or invoke instruction's
1180 <pre> ret int 5 <i>; Return an integer value of 5</i>
1181 ret void <i>; Return from a void function</i>
1184 <!-- _______________________________________________________________________ -->
1185 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1186 <div class="doc_text">
1188 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1191 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1192 transfer to a different basic block in the current function. There are
1193 two forms of this instruction, corresponding to a conditional branch
1194 and an unconditional branch.</p>
1196 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1197 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
1198 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
1199 value as a target.</p>
1201 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
1202 argument is evaluated. If the value is <tt>true</tt>, control flows
1203 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1204 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1206 <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
1207 href="#i_ret">ret</a> int 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> int 0<br></pre>
1209 <!-- _______________________________________________________________________ -->
1210 <div class="doc_subsubsection">
1211 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1214 <div class="doc_text">
1218 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1223 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1224 several different places. It is a generalization of the '<tt>br</tt>'
1225 instruction, allowing a branch to occur to one of many possible
1231 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1232 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1233 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1234 table is not allowed to contain duplicate constant entries.</p>
1238 <p>The <tt>switch</tt> instruction specifies a table of values and
1239 destinations. When the '<tt>switch</tt>' instruction is executed, this
1240 table is searched for the given value. If the value is found, control flow is
1241 transfered to the corresponding destination; otherwise, control flow is
1242 transfered to the default destination.</p>
1244 <h5>Implementation:</h5>
1246 <p>Depending on properties of the target machine and the particular
1247 <tt>switch</tt> instruction, this instruction may be code generated in different
1248 ways. For example, it could be generated as a series of chained conditional
1249 branches or with a lookup table.</p>
1254 <i>; Emulate a conditional br instruction</i>
1255 %Val = <a href="#i_cast">cast</a> bool %value to int
1256 switch int %Val, label %truedest [int 0, label %falsedest ]
1258 <i>; Emulate an unconditional br instruction</i>
1259 switch uint 0, label %dest [ ]
1261 <i>; Implement a jump table:</i>
1262 switch uint %val, label %otherwise [ uint 0, label %onzero
1263 uint 1, label %onone
1264 uint 2, label %ontwo ]
1268 <!-- _______________________________________________________________________ -->
1269 <div class="doc_subsubsection">
1270 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1273 <div class="doc_text">
1278 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1279 to label <normal label> except label <exception label>
1284 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1285 function, with the possibility of control flow transfer to either the
1286 '<tt>normal</tt>' label or the
1287 '<tt>exception</tt>' label. If the callee function returns with the
1288 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1289 "normal" label. If the callee (or any indirect callees) returns with the "<a
1290 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1291 continued at the dynamically nearest "exception" label.</p>
1295 <p>This instruction requires several arguments:</p>
1299 The optional "cconv" marker indicates which <a href="callingconv">calling
1300 convention</a> the call should use. If none is specified, the call defaults
1301 to using C calling conventions.
1303 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1304 function value being invoked. In most cases, this is a direct function
1305 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1306 an arbitrary pointer to function value.
1309 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1310 function to be invoked. </li>
1312 <li>'<tt>function args</tt>': argument list whose types match the function
1313 signature argument types. If the function signature indicates the function
1314 accepts a variable number of arguments, the extra arguments can be
1317 <li>'<tt>normal label</tt>': the label reached when the called function
1318 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1320 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1321 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1327 <p>This instruction is designed to operate as a standard '<tt><a
1328 href="#i_call">call</a></tt>' instruction in most regards. The primary
1329 difference is that it establishes an association with a label, which is used by
1330 the runtime library to unwind the stack.</p>
1332 <p>This instruction is used in languages with destructors to ensure that proper
1333 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1334 exception. Additionally, this is important for implementation of
1335 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1339 %retval = invoke int %Test(int 15) to label %Continue
1340 except label %TestCleanup <i>; {int}:retval set</i>
1341 %retval = invoke <a href="#callingconv">coldcc</a> int %Test(int 15) to label %Continue
1342 except label %TestCleanup <i>; {int}:retval set</i>
1347 <!-- _______________________________________________________________________ -->
1349 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1350 Instruction</a> </div>
1352 <div class="doc_text">
1361 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1362 at the first callee in the dynamic call stack which used an <a
1363 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1364 primarily used to implement exception handling.</p>
1368 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1369 immediately halt. The dynamic call stack is then searched for the first <a
1370 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1371 execution continues at the "exceptional" destination block specified by the
1372 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1373 dynamic call chain, undefined behavior results.</p>
1376 <!-- _______________________________________________________________________ -->
1378 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1379 Instruction</a> </div>
1381 <div class="doc_text">
1390 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1391 instruction is used to inform the optimizer that a particular portion of the
1392 code is not reachable. This can be used to indicate that the code after a
1393 no-return function cannot be reached, and other facts.</p>
1397 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1402 <!-- ======================================================================= -->
1403 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1404 <div class="doc_text">
1405 <p>Binary operators are used to do most of the computation in a
1406 program. They require two operands, execute an operation on them, and
1407 produce a single value. The operands might represent
1408 multiple data, as is the case with the <a href="#t_packed">packed</a> data type.
1409 The result value of a binary operator is not
1410 necessarily the same type as its operands.</p>
1411 <p>There are several different binary operators:</p>
1413 <!-- _______________________________________________________________________ -->
1414 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1415 Instruction</a> </div>
1416 <div class="doc_text">
1418 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1421 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1423 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1424 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1425 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1426 Both arguments must have identical types.</p>
1428 <p>The value produced is the integer or floating point sum of the two
1431 <pre> <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
1434 <!-- _______________________________________________________________________ -->
1435 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1436 Instruction</a> </div>
1437 <div class="doc_text">
1439 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1442 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1444 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1445 instruction present in most other intermediate representations.</p>
1447 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1448 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1450 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1451 Both arguments must have identical types.</p>
1453 <p>The value produced is the integer or floating point difference of
1454 the two operands.</p>
1456 <pre> <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
1457 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
1460 <!-- _______________________________________________________________________ -->
1461 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1462 Instruction</a> </div>
1463 <div class="doc_text">
1465 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1468 <p>The '<tt>mul</tt>' instruction returns the product of its two
1471 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1472 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1474 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1475 Both arguments must have identical types.</p>
1477 <p>The value produced is the integer or floating point product of the
1479 <p>There is no signed vs unsigned multiplication. The appropriate
1480 action is taken based on the type of the operand.</p>
1482 <pre> <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1485 <!-- _______________________________________________________________________ -->
1486 <div class="doc_subsubsection"> <a name="i_div">'<tt>div</tt>'
1487 Instruction</a> </div>
1488 <div class="doc_text">
1490 <pre> <result> = div <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1493 <p>The '<tt>div</tt>' instruction returns the quotient of its two
1496 <p>The two arguments to the '<tt>div</tt>' instruction must be either <a
1497 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1499 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1500 Both arguments must have identical types.</p>
1502 <p>The value produced is the integer or floating point quotient of the
1505 <pre> <result> = div int 4, %var <i>; yields {int}:result = 4 / %var</i>
1508 <!-- _______________________________________________________________________ -->
1509 <div class="doc_subsubsection"> <a name="i_rem">'<tt>rem</tt>'
1510 Instruction</a> </div>
1511 <div class="doc_text">
1513 <pre> <result> = rem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1516 <p>The '<tt>rem</tt>' instruction returns the remainder from the
1517 division of its two operands.</p>
1519 <p>The two arguments to the '<tt>rem</tt>' instruction must be either <a
1520 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1522 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1523 Both arguments must have identical types.</p>
1525 <p>This returns the <i>remainder</i> of a division (where the result
1526 has the same sign as the divisor), not the <i>modulus</i> (where the
1527 result has the same sign as the dividend) of a value. For more
1528 information about the difference, see <a
1529 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
1532 <pre> <result> = rem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1535 <!-- _______________________________________________________________________ -->
1536 <div class="doc_subsubsection"> <a name="i_setcc">'<tt>set<i>cc</i></tt>'
1537 Instructions</a> </div>
1538 <div class="doc_text">
1540 <pre> <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1541 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1542 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1543 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1544 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1545 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1548 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean
1549 value based on a comparison of their two operands.</p>
1551 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must
1552 be of <a href="#t_firstclass">first class</a> type (it is not possible
1553 to compare '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>'
1554 or '<tt>void</tt>' values, etc...). Both arguments must have identical
1557 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1558 value if both operands are equal.<br>
1559 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1560 value if both operands are unequal.<br>
1561 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1562 value if the first operand is less than the second operand.<br>
1563 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1564 value if the first operand is greater than the second operand.<br>
1565 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1566 value if the first operand is less than or equal to the second operand.<br>
1567 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1568 value if the first operand is greater than or equal to the second
1571 <pre> <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1572 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1573 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1574 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1575 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1576 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1579 <!-- ======================================================================= -->
1580 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
1581 Operations</a> </div>
1582 <div class="doc_text">
1583 <p>Bitwise binary operators are used to do various forms of
1584 bit-twiddling in a program. They are generally very efficient
1585 instructions and can commonly be strength reduced from other
1586 instructions. They require two operands, execute an operation on them,
1587 and produce a single value. The resulting value of the bitwise binary
1588 operators is always the same type as its first operand.</p>
1590 <!-- _______________________________________________________________________ -->
1591 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
1592 Instruction</a> </div>
1593 <div class="doc_text">
1595 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1598 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1599 its two operands.</p>
1601 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1602 href="#t_integral">integral</a> values. Both arguments must have
1603 identical types.</p>
1605 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1607 <div style="align: center">
1608 <table border="1" cellspacing="0" cellpadding="4">
1639 <pre> <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1640 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1641 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1644 <!-- _______________________________________________________________________ -->
1645 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
1646 <div class="doc_text">
1648 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1651 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1652 or of its two operands.</p>
1654 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1655 href="#t_integral">integral</a> values. Both arguments must have
1656 identical types.</p>
1658 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1660 <div style="align: center">
1661 <table border="1" cellspacing="0" cellpadding="4">
1692 <pre> <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1693 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1694 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1697 <!-- _______________________________________________________________________ -->
1698 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
1699 Instruction</a> </div>
1700 <div class="doc_text">
1702 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1705 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
1706 or of its two operands. The <tt>xor</tt> is used to implement the
1707 "one's complement" operation, which is the "~" operator in C.</p>
1709 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
1710 href="#t_integral">integral</a> values. Both arguments must have
1711 identical types.</p>
1713 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
1715 <div style="align: center">
1716 <table border="1" cellspacing="0" cellpadding="4">
1748 <pre> <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
1749 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
1750 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
1751 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
1754 <!-- _______________________________________________________________________ -->
1755 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
1756 Instruction</a> </div>
1757 <div class="doc_text">
1759 <pre> <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1762 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
1763 the left a specified number of bits.</p>
1765 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
1766 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1769 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
1771 <pre> <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
1772 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
1773 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
1776 <!-- _______________________________________________________________________ -->
1777 <div class="doc_subsubsection"> <a name="i_shr">'<tt>shr</tt>'
1778 Instruction</a> </div>
1779 <div class="doc_text">
1781 <pre> <result> = shr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1784 <p>The '<tt>shr</tt>' instruction returns the first operand shifted to
1785 the right a specified number of bits.</p>
1787 <p>The first argument to the '<tt>shr</tt>' instruction must be an <a
1788 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1791 <p>If the first argument is a <a href="#t_signed">signed</a> type, the
1792 most significant bit is duplicated in the newly free'd bit positions.
1793 If the first argument is unsigned, zero bits shall fill the empty
1796 <pre> <result> = shr int 4, ubyte %var <i>; yields {int}:result = 4 >> %var</i>
1797 <result> = shr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
1798 <result> = shr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
1799 <result> = shr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
1800 <result> = shr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
1804 <!-- ======================================================================= -->
1805 <div class="doc_subsection">
1806 <a name="memoryops">Memory Access Operations</a>
1809 <div class="doc_text">
1811 <p>A key design point of an SSA-based representation is how it
1812 represents memory. In LLVM, no memory locations are in SSA form, which
1813 makes things very simple. This section describes how to read, write,
1814 allocate, and free memory in LLVM.</p>
1818 <!-- _______________________________________________________________________ -->
1819 <div class="doc_subsubsection">
1820 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
1823 <div class="doc_text">
1828 <result> = malloc <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
1833 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
1834 heap and returns a pointer to it.</p>
1838 <p>The '<tt>malloc</tt>' instruction allocates
1839 <tt>sizeof(<type>)*NumElements</tt>
1840 bytes of memory from the operating system and returns a pointer of the
1841 appropriate type to the program. If "NumElements" is specified, it is the
1842 number of elements allocated. If an alignment is specified, the value result
1843 of the allocation is guaranteed to be aligned to at least that boundary. If
1844 not specified, or if zero, the target can choose to align the allocation on any
1845 convenient boundary.</p>
1847 <p>'<tt>type</tt>' must be a sized type.</p>
1851 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
1852 a pointer is returned.</p>
1857 %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
1859 %size = <a href="#i_add">add</a> uint 2, 2 <i>; yields {uint}:size = uint 4</i>
1860 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
1861 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
1862 %array3 = malloc int, uint 4, align 1024 <i>; yields {int*}:array3</i>
1863 %array4 = malloc int, align 1024 <i>; yields {int*}:array4</i>
1867 <!-- _______________________________________________________________________ -->
1868 <div class="doc_subsubsection">
1869 <a name="i_free">'<tt>free</tt>' Instruction</a>
1872 <div class="doc_text">
1877 free <type> <value> <i>; yields {void}</i>
1882 <p>The '<tt>free</tt>' instruction returns memory back to the unused
1883 memory heap to be reallocated in the future.</p>
1887 <p>'<tt>value</tt>' shall be a pointer value that points to a value
1888 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
1893 <p>Access to the memory pointed to by the pointer is no longer defined
1894 after this instruction executes.</p>
1899 %array = <a href="#i_malloc">malloc</a> [4 x ubyte] <i>; yields {[4 x ubyte]*}:array</i>
1900 free [4 x ubyte]* %array
1904 <!-- _______________________________________________________________________ -->
1905 <div class="doc_subsubsection">
1906 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
1909 <div class="doc_text">
1914 <result> = alloca <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
1919 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
1920 stack frame of the procedure that is live until the current function
1921 returns to its caller.</p>
1925 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
1926 bytes of memory on the runtime stack, returning a pointer of the
1927 appropriate type to the program. If "NumElements" is specified, it is the
1928 number of elements allocated. If an alignment is specified, the value result
1929 of the allocation is guaranteed to be aligned to at least that boundary. If
1930 not specified, or if zero, the target can choose to align the allocation on any
1931 convenient boundary.</p>
1933 <p>'<tt>type</tt>' may be any sized type.</p>
1937 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
1938 memory is automatically released when the function returns. The '<tt>alloca</tt>'
1939 instruction is commonly used to represent automatic variables that must
1940 have an address available. When the function returns (either with the <tt><a
1941 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
1942 instructions), the memory is reclaimed.</p>
1947 %ptr = alloca int <i>; yields {int*}:ptr</i>
1948 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
1949 %ptr = alloca int, uint 4, align 1024 <i>; yields {int*}:ptr</i>
1950 %ptr = alloca int, align 1024 <i>; yields {int*}:ptr</i>
1954 <!-- _______________________________________________________________________ -->
1955 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
1956 Instruction</a> </div>
1957 <div class="doc_text">
1959 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
1961 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
1963 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
1964 address from which to load. The pointer must point to a <a
1965 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
1966 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
1967 the number or order of execution of this <tt>load</tt> with other
1968 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
1971 <p>The location of memory pointed to is loaded.</p>
1973 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
1975 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
1976 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
1979 <!-- _______________________________________________________________________ -->
1980 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
1981 Instruction</a> </div>
1983 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
1984 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
1987 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
1989 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
1990 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
1991 operand must be a pointer to the type of the '<tt><value></tt>'
1992 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
1993 optimizer is not allowed to modify the number or order of execution of
1994 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
1995 href="#i_store">store</a></tt> instructions.</p>
1997 <p>The contents of memory are updated to contain '<tt><value></tt>'
1998 at the location specified by the '<tt><pointer></tt>' operand.</p>
2000 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2002 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2003 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2005 <!-- _______________________________________________________________________ -->
2006 <div class="doc_subsubsection">
2007 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2010 <div class="doc_text">
2013 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2019 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2020 subelement of an aggregate data structure.</p>
2024 <p>This instruction takes a list of integer constants that indicate what
2025 elements of the aggregate object to index to. The actual types of the arguments
2026 provided depend on the type of the first pointer argument. The
2027 '<tt>getelementptr</tt>' instruction is used to index down through the type
2028 levels of a structure or to a specific index in an array. When indexing into a
2029 structure, only <tt>uint</tt>
2030 integer constants are allowed. When indexing into an array or pointer,
2031 <tt>int</tt> and <tt>long</tt> indexes are allowed of any sign.</p>
2033 <p>For example, let's consider a C code fragment and how it gets
2034 compiled to LLVM:</p>
2048 int *foo(struct ST *s) {
2049 return &s[1].Z.B[5][13];
2053 <p>The LLVM code generated by the GCC frontend is:</p>
2056 %RT = type { sbyte, [10 x [20 x int]], sbyte }
2057 %ST = type { int, double, %RT }
2061 int* %foo(%ST* %s) {
2063 %reg = getelementptr %ST* %s, int 1, uint 2, uint 1, int 5, int 13
2070 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2071 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2072 and <a href="#t_array">array</a> types require <tt>uint</tt>, <tt>int</tt>,
2073 <tt>ulong</tt>, or <tt>long</tt> values, and <a href="#t_struct">structure</a>
2074 types require <tt>uint</tt> <b>constants</b>.</p>
2076 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2077 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
2078 }</tt>' type, a structure. The second index indexes into the third element of
2079 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
2080 sbyte }</tt>' type, another structure. The third index indexes into the second
2081 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
2082 array. The two dimensions of the array are subscripted into, yielding an
2083 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2084 to this element, thus computing a value of '<tt>int*</tt>' type.</p>
2086 <p>Note that it is perfectly legal to index partially through a
2087 structure, returning a pointer to an inner element. Because of this,
2088 the LLVM code for the given testcase is equivalent to:</p>
2091 int* %foo(%ST* %s) {
2092 %t1 = getelementptr %ST* %s, int 1 <i>; yields %ST*:%t1</i>
2093 %t2 = getelementptr %ST* %t1, int 0, uint 2 <i>; yields %RT*:%t2</i>
2094 %t3 = getelementptr %RT* %t2, int 0, uint 1 <i>; yields [10 x [20 x int]]*:%t3</i>
2095 %t4 = getelementptr [10 x [20 x int]]* %t3, int 0, int 5 <i>; yields [20 x int]*:%t4</i>
2096 %t5 = getelementptr [20 x int]* %t4, int 0, int 13 <i>; yields int*:%t5</i>
2101 <p>Note that it is undefined to access an array out of bounds: array and
2102 pointer indexes must always be within the defined bounds of the array type.
2103 The one exception for this rules is zero length arrays. These arrays are
2104 defined to be accessible as variable length arrays, which requires access
2105 beyond the zero'th element.</p>
2110 <i>; yields [12 x ubyte]*:aptr</i>
2111 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, uint 1
2115 <!-- ======================================================================= -->
2116 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
2117 <div class="doc_text">
2118 <p>The instructions in this category are the "miscellaneous"
2119 instructions, which defy better classification.</p>
2121 <!-- _______________________________________________________________________ -->
2122 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
2123 Instruction</a> </div>
2124 <div class="doc_text">
2126 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
2128 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
2129 the SSA graph representing the function.</p>
2131 <p>The type of the incoming values are specified with the first type
2132 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
2133 as arguments, with one pair for each predecessor basic block of the
2134 current block. Only values of <a href="#t_firstclass">first class</a>
2135 type may be used as the value arguments to the PHI node. Only labels
2136 may be used as the label arguments.</p>
2137 <p>There must be no non-phi instructions between the start of a basic
2138 block and the PHI instructions: i.e. PHI instructions must be first in
2141 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
2142 value specified by the parameter, depending on which basic block we
2143 came from in the last <a href="#terminators">terminator</a> instruction.</p>
2145 <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>
2148 <!-- _______________________________________________________________________ -->
2149 <div class="doc_subsubsection">
2150 <a name="i_cast">'<tt>cast .. to</tt>' Instruction</a>
2153 <div class="doc_text">
2158 <result> = cast <ty> <value> to <ty2> <i>; yields ty2</i>
2164 The '<tt>cast</tt>' instruction is used as the primitive means to convert
2165 integers to floating point, change data type sizes, and break type safety (by
2173 The '<tt>cast</tt>' instruction takes a value to cast, which must be a first
2174 class value, and a type to cast it to, which must also be a <a
2175 href="#t_firstclass">first class</a> type.
2181 This instruction follows the C rules for explicit casts when determining how the
2182 data being cast must change to fit in its new container.
2186 When casting to bool, any value that would be considered true in the context of
2187 a C '<tt>if</tt>' condition is converted to the boolean '<tt>true</tt>' values,
2188 all else are '<tt>false</tt>'.
2192 When extending an integral value from a type of one signness to another (for
2193 example '<tt>sbyte</tt>' to '<tt>ulong</tt>'), the value is sign-extended if the
2194 <b>source</b> value is signed, and zero-extended if the source value is
2195 unsigned. <tt>bool</tt> values are always zero extended into either zero or
2202 %X = cast int 257 to ubyte <i>; yields ubyte:1</i>
2203 %Y = cast int 123 to bool <i>; yields bool:true</i>
2207 <!-- _______________________________________________________________________ -->
2208 <div class="doc_subsubsection">
2209 <a name="i_select">'<tt>select</tt>' Instruction</a>
2212 <div class="doc_text">
2217 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
2223 The '<tt>select</tt>' instruction is used to choose one value based on a
2224 condition, without branching.
2231 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.
2237 If the boolean condition evaluates to true, the instruction returns the first
2238 value argument; otherwise, it returns the second value argument.
2244 %X = select bool true, ubyte 17, ubyte 42 <i>; yields ubyte:17</i>
2249 <!-- _______________________________________________________________________ -->
2250 <div class="doc_subsubsection">
2251 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2254 <div class="doc_text">
2259 <result> = extractelement <n x <ty>> <val>, uint <idx> <i>; yields <ty></i>
2265 The '<tt>extractelement</tt>' instruction extracts a single scalar
2266 element from a vector at a specified index.
2273 The first operand of an '<tt>extractelement</tt>' instruction is a
2274 value of <a href="#t_packed">packed</a> type. The second operand is
2275 an index indicating the position from which to extract the element.
2276 The index may be a variable.</p>
2281 The result is a scalar of the same type as the element type of
2282 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2283 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2284 results are undefined.
2290 %result = extractelement <4 x int> %vec, uint 0 <i>; yields int</i>
2295 <!-- _______________________________________________________________________ -->
2296 <div class="doc_subsubsection">
2297 <a name="i_call">'<tt>call</tt>' Instruction</a>
2300 <div class="doc_text">
2304 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
2309 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
2313 <p>This instruction requires several arguments:</p>
2317 <p>The optional "tail" marker indicates whether the callee function accesses
2318 any allocas or varargs in the caller. If the "tail" marker is present, the
2319 function call is eligible for tail call optimization. Note that calls may
2320 be marked "tail" even if they do not occur before a <a
2321 href="#i_ret"><tt>ret</tt></a> instruction.
2324 <p>The optional "cconv" marker indicates which <a href="callingconv">calling
2325 convention</a> the call should use. If none is specified, the call defaults
2326 to using C calling conventions.
2329 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
2330 being invoked. The argument types must match the types implied by this
2331 signature. This type can be omitted if the function is not varargs and
2332 if the function type does not return a pointer to a function.</p>
2335 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
2336 be invoked. In most cases, this is a direct function invocation, but
2337 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
2338 to function value.</p>
2341 <p>'<tt>function args</tt>': argument list whose types match the
2342 function signature argument types. All arguments must be of
2343 <a href="#t_firstclass">first class</a> type. If the function signature
2344 indicates the function accepts a variable number of arguments, the extra
2345 arguments can be specified.</p>
2351 <p>The '<tt>call</tt>' instruction is used to cause control flow to
2352 transfer to a specified function, with its incoming arguments bound to
2353 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
2354 instruction in the called function, control flow continues with the
2355 instruction after the function call, and the return value of the
2356 function is bound to the result argument. This is a simpler case of
2357 the <a href="#i_invoke">invoke</a> instruction.</p>
2362 %retval = call int %test(int %argc)
2363 call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);
2364 %X = tail call int %foo()
2365 %Y = tail call <a href="#callingconv">fastcc</a> int %foo()
2370 <!-- _______________________________________________________________________ -->
2371 <div class="doc_subsubsection">
2372 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
2375 <div class="doc_text">
2380 <resultval> = va_arg <va_list*> <arglist>, <argty>
2385 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
2386 the "variable argument" area of a function call. It is used to implement the
2387 <tt>va_arg</tt> macro in C.</p>
2391 <p>This instruction takes a <tt>va_list*</tt> value and the type of
2392 the argument. It returns a value of the specified argument type and
2393 increments the <tt>va_list</tt> to point to the next argument. Again, the
2394 actual type of <tt>va_list</tt> is target specific.</p>
2398 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
2399 type from the specified <tt>va_list</tt> and causes the
2400 <tt>va_list</tt> to point to the next argument. For more information,
2401 see the variable argument handling <a href="#int_varargs">Intrinsic
2404 <p>It is legal for this instruction to be called in a function which does not
2405 take a variable number of arguments, for example, the <tt>vfprintf</tt>
2408 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
2409 href="#intrinsics">intrinsic function</a> because it takes a type as an
2414 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
2418 <!-- *********************************************************************** -->
2419 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
2420 <!-- *********************************************************************** -->
2422 <div class="doc_text">
2424 <p>LLVM supports the notion of an "intrinsic function". These functions have
2425 well known names and semantics and are required to follow certain
2426 restrictions. Overall, these instructions represent an extension mechanism for
2427 the LLVM language that does not require changing all of the transformations in
2428 LLVM to add to the language (or the bytecode reader/writer, the parser,
2431 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
2432 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
2433 this. Intrinsic functions must always be external functions: you cannot define
2434 the body of intrinsic functions. Intrinsic functions may only be used in call
2435 or invoke instructions: it is illegal to take the address of an intrinsic
2436 function. Additionally, because intrinsic functions are part of the LLVM
2437 language, it is required that they all be documented here if any are added.</p>
2440 <p>To learn how to add an intrinsic function, please see the <a
2441 href="ExtendingLLVM.html">Extending LLVM Guide</a>.
2446 <!-- ======================================================================= -->
2447 <div class="doc_subsection">
2448 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
2451 <div class="doc_text">
2453 <p>Variable argument support is defined in LLVM with the <a
2454 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
2455 intrinsic functions. These functions are related to the similarly
2456 named macros defined in the <tt><stdarg.h></tt> header file.</p>
2458 <p>All of these functions operate on arguments that use a
2459 target-specific value type "<tt>va_list</tt>". The LLVM assembly
2460 language reference manual does not define what this type is, so all
2461 transformations should be prepared to handle intrinsics with any type
2464 <p>This example shows how the <a href="#i_vanext"><tt>vanext</tt></a>
2465 instruction and the variable argument handling intrinsic functions are
2469 int %test(int %X, ...) {
2470 ; Initialize variable argument processing
2472 call void %<a href="#i_va_start">llvm.va_start</a>(sbyte** %ap)
2474 ; Read a single integer argument
2475 %tmp = va_arg sbyte** %ap, int
2477 ; Demonstrate usage of llvm.va_copy and llvm.va_end
2479 call void %<a href="#i_va_copy">llvm.va_copy</a>(sbyte** %aq, sbyte** %ap)
2480 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %aq)
2482 ; Stop processing of arguments.
2483 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %ap)
2489 <!-- _______________________________________________________________________ -->
2490 <div class="doc_subsubsection">
2491 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
2495 <div class="doc_text">
2497 <pre> declare void %llvm.va_start(<va_list>* <arglist>)<br></pre>
2499 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
2500 <tt>*<arglist></tt> for subsequent use by <tt><a
2501 href="#i_va_arg">va_arg</a></tt>.</p>
2505 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
2509 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
2510 macro available in C. In a target-dependent way, it initializes the
2511 <tt>va_list</tt> element the argument points to, so that the next call to
2512 <tt>va_arg</tt> will produce the first variable argument passed to the function.
2513 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
2514 last argument of the function, the compiler can figure that out.</p>
2518 <!-- _______________________________________________________________________ -->
2519 <div class="doc_subsubsection">
2520 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
2523 <div class="doc_text">
2525 <pre> declare void %llvm.va_end(<va_list*> <arglist>)<br></pre>
2527 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
2528 which has been initialized previously with <tt><a href="#i_va_start">llvm.va_start</a></tt>
2529 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
2531 <p>The argument is a <tt>va_list</tt> to destroy.</p>
2533 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
2534 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
2535 Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and <a
2536 href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
2537 with calls to <tt>llvm.va_end</tt>.</p>
2540 <!-- _______________________________________________________________________ -->
2541 <div class="doc_subsubsection">
2542 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
2545 <div class="doc_text">
2550 declare void %llvm.va_copy(<va_list>* <destarglist>,
2551 <va_list>* <srcarglist>)
2556 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
2557 the source argument list to the destination argument list.</p>
2561 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
2562 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
2567 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
2568 available in C. In a target-dependent way, it copies the source
2569 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
2570 because the <tt><a href="i_va_begin">llvm.va_begin</a></tt> intrinsic may be
2571 arbitrarily complex and require memory allocation, for example.</p>
2575 <!-- ======================================================================= -->
2576 <div class="doc_subsection">
2577 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
2580 <div class="doc_text">
2583 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
2584 Collection</a> requires the implementation and generation of these intrinsics.
2585 These intrinsics allow identification of <a href="#i_gcroot">GC roots on the
2586 stack</a>, as well as garbage collector implementations that require <a
2587 href="#i_gcread">read</a> and <a href="#i_gcwrite">write</a> barriers.
2588 Front-ends for type-safe garbage collected languages should generate these
2589 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
2590 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
2594 <!-- _______________________________________________________________________ -->
2595 <div class="doc_subsubsection">
2596 <a name="i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
2599 <div class="doc_text">
2604 declare void %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
2609 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
2610 the code generator, and allows some metadata to be associated with it.</p>
2614 <p>The first argument specifies the address of a stack object that contains the
2615 root pointer. The second pointer (which must be either a constant or a global
2616 value address) contains the meta-data to be associated with the root.</p>
2620 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
2621 location. At compile-time, the code generator generates information to allow
2622 the runtime to find the pointer at GC safe points.
2628 <!-- _______________________________________________________________________ -->
2629 <div class="doc_subsubsection">
2630 <a name="i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
2633 <div class="doc_text">
2638 declare sbyte* %llvm.gcread(sbyte** %Ptr)
2643 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
2644 locations, allowing garbage collector implementations that require read
2649 <p>The argument is the address to read from, which should be an address
2650 allocated from the garbage collector.</p>
2654 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
2655 instruction, but may be replaced with substantially more complex code by the
2656 garbage collector runtime, as needed.</p>
2661 <!-- _______________________________________________________________________ -->
2662 <div class="doc_subsubsection">
2663 <a name="i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
2666 <div class="doc_text">
2671 declare void %llvm.gcwrite(sbyte* %P1, sbyte** %P2)
2676 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
2677 locations, allowing garbage collector implementations that require write
2678 barriers (such as generational or reference counting collectors).</p>
2682 <p>The first argument is the reference to store, and the second is the heap
2683 location to store to.</p>
2687 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
2688 instruction, but may be replaced with substantially more complex code by the
2689 garbage collector runtime, as needed.</p>
2695 <!-- ======================================================================= -->
2696 <div class="doc_subsection">
2697 <a name="int_codegen">Code Generator Intrinsics</a>
2700 <div class="doc_text">
2702 These intrinsics are provided by LLVM to expose special features that may only
2703 be implemented with code generator support.
2708 <!-- _______________________________________________________________________ -->
2709 <div class="doc_subsubsection">
2710 <a name="i_bswap_i16">'<tt>llvm.bswap.i16</tt>' Intrinsic</a>
2713 <div class="doc_text">
2717 declare ushort %llvm.bswap.i16( ushort <id> )
2723 The '<tt>llvm.bwsap.i16</tt>' intrinsic is used to byteswap a 16 bit quantity.
2724 This is useful for performing operations on data that is not in the target's
2731 This intrinsic returns a ushort value that has the two bytes of the input ushort
2737 <!-- _______________________________________________________________________ -->
2738 <div class="doc_subsubsection">
2739 <a name="i_bswap_i32">'<tt>llvm.bswap.i32</tt>' Intrinsic</a>
2742 <div class="doc_text">
2746 declare uint %llvm.bswap.i32( uint <id> )
2752 The '<tt>llvm.bwsap.i32</tt>' intrinsic is used to byteswap a 32 bit quantity.
2753 This is useful for performing operations on data that is not in the target's
2760 This intrinsic returns a uint value that has the four bytes of the input uint
2761 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
2762 uint will have its bytes in 3, 2, 1, 0 order.
2767 <!-- _______________________________________________________________________ -->
2768 <div class="doc_subsubsection">
2769 <a name="i_bswap_i64">'<tt>llvm.bswap.i64</tt>' Intrinsic</a>
2772 <div class="doc_text">
2776 declare ulong %llvm.bswap.i64( ulong <id> )
2782 The '<tt>llvm.bwsap.i64</tt>' intrinsic is used to byteswap a 64 bit quantity.
2783 This is useful for performing operations on data that is not in the target's
2790 See the description for <a href="#i_bswap_i32"><tt>llvm.bswap.i32</tt></a>.
2795 <!-- _______________________________________________________________________ -->
2796 <div class="doc_subsubsection">
2797 <a name="i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
2800 <div class="doc_text">
2804 declare sbyte *%llvm.returnaddress(uint <level>)
2810 The '<tt>llvm.returnaddress</tt>' intrinsic returns a target-specific value
2811 indicating the return address of the current function or one of its callers.
2817 The argument to this intrinsic indicates which function to return the address
2818 for. Zero indicates the calling function, one indicates its caller, etc. The
2819 argument is <b>required</b> to be a constant integer value.
2825 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
2826 the return address of the specified call frame, or zero if it cannot be
2827 identified. The value returned by this intrinsic is likely to be incorrect or 0
2828 for arguments other than zero, so it should only be used for debugging purposes.
2832 Note that calling this intrinsic does not prevent function inlining or other
2833 aggressive transformations, so the value returned may not be that of the obvious
2834 source-language caller.
2839 <!-- _______________________________________________________________________ -->
2840 <div class="doc_subsubsection">
2841 <a name="i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
2844 <div class="doc_text">
2848 declare sbyte *%llvm.frameaddress(uint <level>)
2854 The '<tt>llvm.frameaddress</tt>' intrinsic returns the target-specific frame
2855 pointer value for the specified stack frame.
2861 The argument to this intrinsic indicates which function to return the frame
2862 pointer for. Zero indicates the calling function, one indicates its caller,
2863 etc. The argument is <b>required</b> to be a constant integer value.
2869 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
2870 the frame address of the specified call frame, or zero if it cannot be
2871 identified. The value returned by this intrinsic is likely to be incorrect or 0
2872 for arguments other than zero, so it should only be used for debugging purposes.
2876 Note that calling this intrinsic does not prevent function inlining or other
2877 aggressive transformations, so the value returned may not be that of the obvious
2878 source-language caller.
2882 <!-- _______________________________________________________________________ -->
2883 <div class="doc_subsubsection">
2884 <a name="i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
2887 <div class="doc_text">
2891 declare sbyte *%llvm.stacksave()
2897 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
2898 the function stack, for use with <a href="#i_stackrestore">
2899 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
2900 features like scoped automatic variable sized arrays in C99.
2906 This intrinsic returns a opaque pointer value that can be passed to <a
2907 href="#i_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
2908 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
2909 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
2910 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
2911 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
2912 that were allocated after the <tt>llvm.stacksave</tt> was executed.
2917 <!-- _______________________________________________________________________ -->
2918 <div class="doc_subsubsection">
2919 <a name="i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
2922 <div class="doc_text">
2926 declare void %llvm.stackrestore(sbyte* %ptr)
2932 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
2933 the function stack to the state it was in when the corresponding <a
2934 href="#llvm.stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
2935 useful for implementing language features like scoped automatic variable sized
2942 See the description for <a href="#i_stacksave"><tt>llvm.stacksave</tt></a>.
2948 <!-- _______________________________________________________________________ -->
2949 <div class="doc_subsubsection">
2950 <a name="i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
2953 <div class="doc_text">
2957 declare void %llvm.prefetch(sbyte * <address>,
2958 uint <rw>, uint <locality>)
2965 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
2966 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
2968 effect on the behavior of the program but can change its performance
2975 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
2976 determining if the fetch should be for a read (0) or write (1), and
2977 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
2978 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
2979 <tt>locality</tt> arguments must be constant integers.
2985 This intrinsic does not modify the behavior of the program. In particular,
2986 prefetches cannot trap and do not produce a value. On targets that support this
2987 intrinsic, the prefetch can provide hints to the processor cache for better
2993 <!-- _______________________________________________________________________ -->
2994 <div class="doc_subsubsection">
2995 <a name="i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
2998 <div class="doc_text">
3002 declare void %llvm.pcmarker( uint <id> )
3009 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
3011 code to simulators and other tools. The method is target specific, but it is
3012 expected that the marker will use exported symbols to transmit the PC of the marker.
3013 The marker makes no guarantees that it will remain with any specific instruction
3014 after optimizations. It is possible that the presence of a marker will inhibit
3015 optimizations. The intended use is to be inserted after optmizations to allow
3016 correlations of simulation runs.
3022 <tt>id</tt> is a numerical id identifying the marker.
3028 This intrinsic does not modify the behavior of the program. Backends that do not
3029 support this intrinisic may ignore it.
3034 <!-- _______________________________________________________________________ -->
3035 <div class="doc_subsubsection">
3036 <a name="i_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
3039 <div class="doc_text">
3043 declare ulong %llvm.readcyclecounter( )
3050 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
3051 counter register (or similar low latency, high accuracy clocks) on those targets
3052 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
3053 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
3054 should only be used for small timings.
3060 When directly supported, reading the cycle counter should not modify any memory.
3061 Implementations are allowed to either return a application specific value or a
3062 system wide value. On backends without support, this is lowered to a constant 0.
3068 <!-- ======================================================================= -->
3069 <div class="doc_subsection">
3070 <a name="int_os">Operating System Intrinsics</a>
3073 <div class="doc_text">
3075 These intrinsics are provided by LLVM to support the implementation of
3076 operating system level code.
3081 <!-- _______________________________________________________________________ -->
3082 <div class="doc_subsubsection">
3083 <a name="i_readport">'<tt>llvm.readport</tt>' Intrinsic</a>
3086 <div class="doc_text">
3090 declare <integer type> %llvm.readport (<integer type> <address>)
3096 The '<tt>llvm.readport</tt>' intrinsic reads data from the specified hardware
3103 The argument to this intrinsic indicates the hardware I/O address from which
3104 to read the data. The address is in the hardware I/O address namespace (as
3105 opposed to being a memory location for memory mapped I/O).
3111 The '<tt>llvm.readport</tt>' intrinsic reads data from the hardware I/O port
3112 specified by <i>address</i> and returns the value. The address and return
3113 value must be integers, but the size is dependent upon the platform upon which
3114 the program is code generated. For example, on x86, the address must be an
3115 unsigned 16-bit value, and the return value must be 8, 16, or 32 bits.
3120 <!-- _______________________________________________________________________ -->
3121 <div class="doc_subsubsection">
3122 <a name="i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a>
3125 <div class="doc_text">
3129 call void (<integer type>, <integer type>)*
3130 %llvm.writeport (<integer type> <value>,
3131 <integer type> <address>)
3137 The '<tt>llvm.writeport</tt>' intrinsic writes data to the specified hardware
3144 The first argument is the value to write to the I/O port.
3148 The second argument indicates the hardware I/O address to which data should be
3149 written. The address is in the hardware I/O address namespace (as opposed to
3150 being a memory location for memory mapped I/O).
3156 The '<tt>llvm.writeport</tt>' intrinsic writes <i>value</i> to the I/O port
3157 specified by <i>address</i>. The address and value must be integers, but the
3158 size is dependent upon the platform upon which the program is code generated.
3159 For example, on x86, the address must be an unsigned 16-bit value, and the
3160 value written must be 8, 16, or 32 bits in length.
3165 <!-- _______________________________________________________________________ -->
3166 <div class="doc_subsubsection">
3167 <a name="i_readio">'<tt>llvm.readio</tt>' Intrinsic</a>
3170 <div class="doc_text">
3174 declare <result> %llvm.readio (<ty> * <pointer>)
3180 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
3187 The argument to this intrinsic is a pointer indicating the memory address from
3188 which to read the data. The data must be a
3189 <a href="#t_firstclass">first class</a> type.
3195 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
3196 location specified by <i>pointer</i> and returns the value. The argument must
3197 be a pointer, and the return value must be a
3198 <a href="#t_firstclass">first class</a> type. However, certain architectures
3199 may not support I/O on all first class types. For example, 32-bit processors
3200 may only support I/O on data types that are 32 bits or less.
3204 This intrinsic enforces an in-order memory model for llvm.readio and
3205 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
3206 scheduled processors may execute loads and stores out of order, re-ordering at
3207 run time accesses to memory mapped I/O registers. Using these intrinsics
3208 ensures that accesses to memory mapped I/O registers occur in program order.
3213 <!-- _______________________________________________________________________ -->
3214 <div class="doc_subsubsection">
3215 <a name="i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a>
3218 <div class="doc_text">
3222 declare void %llvm.writeio (<ty1> <value>, <ty2> * <pointer>)
3228 The '<tt>llvm.writeio</tt>' intrinsic writes data to the specified memory
3235 The first argument is the value to write to the memory mapped I/O location.
3236 The second argument is a pointer indicating the memory address to which the
3237 data should be written.
3243 The '<tt>llvm.writeio</tt>' intrinsic writes <i>value</i> to the memory mapped
3244 I/O address specified by <i>pointer</i>. The value must be a
3245 <a href="#t_firstclass">first class</a> type. However, certain architectures
3246 may not support I/O on all first class types. For example, 32-bit processors
3247 may only support I/O on data types that are 32 bits or less.
3251 This intrinsic enforces an in-order memory model for llvm.readio and
3252 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
3253 scheduled processors may execute loads and stores out of order, re-ordering at
3254 run time accesses to memory mapped I/O registers. Using these intrinsics
3255 ensures that accesses to memory mapped I/O registers occur in program order.
3260 <!-- ======================================================================= -->
3261 <div class="doc_subsection">
3262 <a name="int_libc">Standard C Library Intrinsics</a>
3265 <div class="doc_text">
3267 LLVM provides intrinsics for a few important standard C library functions.
3268 These intrinsics allow source-language front-ends to pass information about the
3269 alignment of the pointer arguments to the code generator, providing opportunity
3270 for more efficient code generation.
3275 <!-- _______________________________________________________________________ -->
3276 <div class="doc_subsubsection">
3277 <a name="i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
3280 <div class="doc_text">
3284 declare void %llvm.memcpy(sbyte* <dest>, sbyte* <src>,
3285 uint <len>, uint <align>)
3291 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
3292 location to the destination location.
3296 Note that, unlike the standard libc function, the <tt>llvm.memcpy</tt> intrinsic
3297 does not return a value, and takes an extra alignment argument.
3303 The first argument is a pointer to the destination, the second is a pointer to
3304 the source. The third argument is an (arbitrarily sized) integer argument
3305 specifying the number of bytes to copy, and the fourth argument is the alignment
3306 of the source and destination locations.
3310 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3311 the caller guarantees that the size of the copy is a multiple of the alignment
3312 and that both the source and destination pointers are aligned to that boundary.
3318 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
3319 location to the destination location, which are not allowed to overlap. It
3320 copies "len" bytes of memory over. If the argument is known to be aligned to
3321 some boundary, this can be specified as the fourth argument, otherwise it should
3327 <!-- _______________________________________________________________________ -->
3328 <div class="doc_subsubsection">
3329 <a name="i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
3332 <div class="doc_text">
3336 declare void %llvm.memmove(sbyte* <dest>, sbyte* <src>,
3337 uint <len>, uint <align>)
3343 The '<tt>llvm.memmove</tt>' intrinsic moves a block of memory from the source
3344 location to the destination location. It is similar to the '<tt>llvm.memcpy</tt>'
3345 intrinsic but allows the two memory locations to overlap.
3349 Note that, unlike the standard libc function, the <tt>llvm.memmove</tt> intrinsic
3350 does not return a value, and takes an extra alignment argument.
3356 The first argument is a pointer to the destination, the second is a pointer to
3357 the source. The third argument is an (arbitrarily sized) integer argument
3358 specifying the number of bytes to copy, and the fourth argument is the alignment
3359 of the source and destination locations.
3363 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3364 the caller guarantees that the size of the copy is a multiple of the alignment
3365 and that both the source and destination pointers are aligned to that boundary.
3371 The '<tt>llvm.memmove</tt>' intrinsic copies a block of memory from the source
3372 location to the destination location, which may overlap. It
3373 copies "len" bytes of memory over. If the argument is known to be aligned to
3374 some boundary, this can be specified as the fourth argument, otherwise it should
3380 <!-- _______________________________________________________________________ -->
3381 <div class="doc_subsubsection">
3382 <a name="i_memset">'<tt>llvm.memset</tt>' Intrinsic</a>
3385 <div class="doc_text">
3389 declare void %llvm.memset(sbyte* <dest>, ubyte <val>,
3390 uint <len>, uint <align>)
3396 The '<tt>llvm.memset</tt>' intrinsic fills a block of memory with a particular
3401 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
3402 does not return a value, and takes an extra alignment argument.
3408 The first argument is a pointer to the destination to fill, the second is the
3409 byte value to fill it with, the third argument is an (arbitrarily sized) integer
3410 argument specifying the number of bytes to fill, and the fourth argument is the
3411 known alignment of destination location.
3415 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3416 the caller guarantees that the size of the copy is a multiple of the alignment
3417 and that the destination pointer is aligned to that boundary.
3423 The '<tt>llvm.memset</tt>' intrinsic fills "len" bytes of memory starting at the
3424 destination location. If the argument is known to be aligned to some boundary,
3425 this can be specified as the fourth argument, otherwise it should be set to 0 or
3431 <!-- _______________________________________________________________________ -->
3432 <div class="doc_subsubsection">
3433 <a name="i_isunordered">'<tt>llvm.isunordered</tt>' Intrinsic</a>
3436 <div class="doc_text">
3440 declare bool %llvm.isunordered(<float or double> Val1, <float or double> Val2)
3446 The '<tt>llvm.isunordered</tt>' intrinsic returns true if either or both of the
3447 specified floating point values is a NAN.
3453 The arguments are floating point numbers of the same type.
3459 If either or both of the arguments is a SNAN or QNAN, it returns true, otherwise
3465 <!-- _______________________________________________________________________ -->
3466 <div class="doc_subsubsection">
3467 <a name="i_sqrt">'<tt>llvm.sqrt</tt>' Intrinsic</a>
3470 <div class="doc_text">
3474 declare <float or double> %llvm.sqrt(<float or double> Val)
3480 The '<tt>llvm.sqrt</tt>' intrinsic returns the sqrt of the specified operand,
3481 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
3482 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
3483 negative numbers (which allows for better optimization).
3489 The argument and return value are floating point numbers of the same type.
3495 This function returns the sqrt of the specified operand if it is a positive
3496 floating point number.
3500 <!-- ======================================================================= -->
3501 <div class="doc_subsection">
3502 <a name="int_count">Bit Counting Intrinsics</a>
3505 <div class="doc_text">
3507 LLVM provides intrinsics for a few important bit counting operations.
3508 These allow efficient code generation for some algorithms.
3513 <!-- _______________________________________________________________________ -->
3514 <div class="doc_subsubsection">
3515 <a name="int_ctpop">'<tt>llvm.ctpop</tt>' Intrinsic</a>
3518 <div class="doc_text">
3522 declare int %llvm.ctpop(int <src>)
3528 The '<tt>llvm.ctpop</tt>' intrinsic counts the number of ones in a variable.
3534 The only argument is the value to be counted. The argument may be of any
3535 integer type. The return type must match the argument type.
3541 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
3545 <!-- _______________________________________________________________________ -->
3546 <div class="doc_subsubsection">
3547 <a name="int_ctlz">'<tt>llvm.ctlz</tt>' Intrinsic</a>
3550 <div class="doc_text">
3554 declare int %llvm.ctlz(int <src>)
3561 The '<tt>llvm.ctlz</tt>' intrinsic counts the number of leading zeros in a
3568 The only argument is the value to be counted. The argument may be of any
3569 integer type. The return type must match the argument type.
3575 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
3576 in a variable. If the src == 0 then the result is the size in bits of the type
3577 of src. For example, <tt>llvm.cttz(int 2) = 30</tt>.
3583 <!-- _______________________________________________________________________ -->
3584 <div class="doc_subsubsection">
3585 <a name="int_cttz">'<tt>llvm.cttz</tt>' Intrinsic</a>
3588 <div class="doc_text">
3592 declare int %llvm.cttz(int <src>)
3598 The '<tt>llvm.cttz</tt>' intrinsic counts the number of trailing zeros.
3604 The only argument is the value to be counted. The argument may be of any
3605 integer type. The return type must match the argument type.
3611 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
3612 in a variable. If the src == 0 then the result is the size in bits of the type
3613 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
3617 <!-- ======================================================================= -->
3618 <div class="doc_subsection">
3619 <a name="int_debugger">Debugger Intrinsics</a>
3622 <div class="doc_text">
3624 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
3625 are described in the <a
3626 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
3627 Debugging</a> document.
3632 <!-- *********************************************************************** -->
3635 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3636 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
3637 <a href="http://validator.w3.org/check/referer"><img
3638 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
3640 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3641 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br>
3642 Last modified: $Date$