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_vaarg">'<tt>vaarg</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_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
129 <li><a href="#i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
130 <li><a href="#i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
131 <li><a href="#i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
132 <li><a href="#i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
133 <li><a href="#i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
134 <li><a href="#i_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
137 <li><a href="#int_os">Operating System Intrinsics</a>
139 <li><a href="#i_readport">'<tt>llvm.readport</tt>' Intrinsic</a></li>
140 <li><a href="#i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a></li>
141 <li><a href="#i_readio">'<tt>llvm.readio</tt>' Intrinsic</a></li>
142 <li><a href="#i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a></li>
144 <li><a href="#int_libc">Standard C Library Intrinsics</a>
146 <li><a href="#i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a></li>
147 <li><a href="#i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a></li>
148 <li><a href="#i_memset">'<tt>llvm.memset</tt>' Intrinsic</a></li>
149 <li><a href="#i_isunordered">'<tt>llvm.isunordered</tt>' Intrinsic</a></li>
150 <li><a href="#i_sqrt">'<tt>llvm.sqrt</tt>' Intrinsic</a></li>
154 <li><a href="#int_count">Bit counting Intrinsics</a>
156 <li><a href="#int_ctpop">'<tt>llvm.ctpop</tt>' Intrinsic </a></li>
157 <li><a href="#int_ctlz">'<tt>llvm.ctlz</tt>' Intrinsic </a></li>
158 <li><a href="#int_cttz">'<tt>llvm.cttz</tt>' Intrinsic </a></li>
161 <li><a href="#int_debugger">Debugger intrinsics</a></li>
166 <div class="doc_author">
167 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
168 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
171 <!-- *********************************************************************** -->
172 <div class="doc_section"> <a name="abstract">Abstract </a></div>
173 <!-- *********************************************************************** -->
175 <div class="doc_text">
176 <p>This document is a reference manual for the LLVM assembly language.
177 LLVM is an SSA based representation that provides type safety,
178 low-level operations, flexibility, and the capability of representing
179 'all' high-level languages cleanly. It is the common code
180 representation used throughout all phases of the LLVM compilation
184 <!-- *********************************************************************** -->
185 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
186 <!-- *********************************************************************** -->
188 <div class="doc_text">
190 <p>The LLVM code representation is designed to be used in three
191 different forms: as an in-memory compiler IR, as an on-disk bytecode
192 representation (suitable for fast loading by a Just-In-Time compiler),
193 and as a human readable assembly language representation. This allows
194 LLVM to provide a powerful intermediate representation for efficient
195 compiler transformations and analysis, while providing a natural means
196 to debug and visualize the transformations. The three different forms
197 of LLVM are all equivalent. This document describes the human readable
198 representation and notation.</p>
200 <p>The LLVM representation aims to be light-weight and low-level
201 while being expressive, typed, and extensible at the same time. It
202 aims to be a "universal IR" of sorts, by being at a low enough level
203 that high-level ideas may be cleanly mapped to it (similar to how
204 microprocessors are "universal IR's", allowing many source languages to
205 be mapped to them). By providing type information, LLVM can be used as
206 the target of optimizations: for example, through pointer analysis, it
207 can be proven that a C automatic variable is never accessed outside of
208 the current function... allowing it to be promoted to a simple SSA
209 value instead of a memory location.</p>
213 <!-- _______________________________________________________________________ -->
214 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
216 <div class="doc_text">
218 <p>It is important to note that this document describes 'well formed'
219 LLVM assembly language. There is a difference between what the parser
220 accepts and what is considered 'well formed'. For example, the
221 following instruction is syntactically okay, but not well formed:</p>
224 %x = <a href="#i_add">add</a> int 1, %x
227 <p>...because the definition of <tt>%x</tt> does not dominate all of
228 its uses. The LLVM infrastructure provides a verification pass that may
229 be used to verify that an LLVM module is well formed. This pass is
230 automatically run by the parser after parsing input assembly and by
231 the optimizer before it outputs bytecode. The violations pointed out
232 by the verifier pass indicate bugs in transformation passes or input to
235 <!-- Describe the typesetting conventions here. --> </div>
237 <!-- *********************************************************************** -->
238 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
239 <!-- *********************************************************************** -->
241 <div class="doc_text">
243 <p>LLVM uses three different forms of identifiers, for different
247 <li>Named values are represented as a string of characters with a '%' prefix.
248 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
249 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
250 Identifiers which require other characters in their names can be surrounded
251 with quotes. In this way, anything except a <tt>"</tt> character can be used
254 <li>Unnamed values are represented as an unsigned numeric value with a '%'
255 prefix. For example, %12, %2, %44.</li>
257 <li>Constants, which are described in a <a href="#constants">section about
258 constants</a>, below.</li>
261 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
262 don't need to worry about name clashes with reserved words, and the set of
263 reserved words may be expanded in the future without penalty. Additionally,
264 unnamed identifiers allow a compiler to quickly come up with a temporary
265 variable without having to avoid symbol table conflicts.</p>
267 <p>Reserved words in LLVM are very similar to reserved words in other
268 languages. There are keywords for different opcodes ('<tt><a
269 href="#i_add">add</a></tt>', '<tt><a href="#i_cast">cast</a></tt>', '<tt><a
270 href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
271 href="#t_void">void</a></tt>', '<tt><a href="#t_uint">uint</a></tt>', etc...),
272 and others. These reserved words cannot conflict with variable names, because
273 none of them start with a '%' character.</p>
275 <p>Here is an example of LLVM code to multiply the integer variable
276 '<tt>%X</tt>' by 8:</p>
281 %result = <a href="#i_mul">mul</a> uint %X, 8
284 <p>After strength reduction:</p>
287 %result = <a href="#i_shl">shl</a> uint %X, ubyte 3
290 <p>And the hard way:</p>
293 <a href="#i_add">add</a> uint %X, %X <i>; yields {uint}:%0</i>
294 <a href="#i_add">add</a> uint %0, %0 <i>; yields {uint}:%1</i>
295 %result = <a href="#i_add">add</a> uint %1, %1
298 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
299 important lexical features of LLVM:</p>
303 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
306 <li>Unnamed temporaries are created when the result of a computation is not
307 assigned to a named value.</li>
309 <li>Unnamed temporaries are numbered sequentially</li>
313 <p>...and it also shows a convention that we follow in this document. When
314 demonstrating instructions, we will follow an instruction with a comment that
315 defines the type and name of value produced. Comments are shown in italic
320 <!-- *********************************************************************** -->
321 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
322 <!-- *********************************************************************** -->
324 <!-- ======================================================================= -->
325 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
328 <div class="doc_text">
330 <p>LLVM programs are composed of "Module"s, each of which is a
331 translation unit of the input programs. Each module consists of
332 functions, global variables, and symbol table entries. Modules may be
333 combined together with the LLVM linker, which merges function (and
334 global variable) definitions, resolves forward declarations, and merges
335 symbol table entries. Here is an example of the "hello world" module:</p>
337 <pre><i>; Declare the string constant as a global constant...</i>
338 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
339 href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
341 <i>; External declaration of the puts function</i>
342 <a href="#functionstructure">declare</a> int %puts(sbyte*) <i>; int(sbyte*)* </i>
344 <i>; Definition of main function</i>
345 int %main() { <i>; int()* </i>
346 <i>; Convert [13x sbyte]* to sbyte *...</i>
348 href="#i_getelementptr">getelementptr</a> [13 x sbyte]* %.LC0, long 0, long 0 <i>; sbyte*</i>
350 <i>; Call puts function to write out the string to stdout...</i>
352 href="#i_call">call</a> int %puts(sbyte* %cast210) <i>; int</i>
354 href="#i_ret">ret</a> int 0<br>}<br></pre>
356 <p>This example is made up of a <a href="#globalvars">global variable</a>
357 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
358 function, and a <a href="#functionstructure">function definition</a>
359 for "<tt>main</tt>".</p>
361 <p>In general, a module is made up of a list of global values,
362 where both functions and global variables are global values. Global values are
363 represented by a pointer to a memory location (in this case, a pointer to an
364 array of char, and a pointer to a function), and have one of the following <a
365 href="#linkage">linkage types</a>.</p>
369 <!-- ======================================================================= -->
370 <div class="doc_subsection">
371 <a name="linkage">Linkage Types</a>
374 <div class="doc_text">
377 All Global Variables and Functions have one of the following types of linkage:
382 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
384 <dd>Global values with internal linkage are only directly accessible by
385 objects in the current module. In particular, linking code into a module with
386 an internal global value may cause the internal to be renamed as necessary to
387 avoid collisions. Because the symbol is internal to the module, all
388 references can be updated. This corresponds to the notion of the
389 '<tt>static</tt>' keyword in C, or the idea of "anonymous namespaces" in C++.
392 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
394 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt> linkage, with
395 the twist that linking together two modules defining the same
396 <tt>linkonce</tt> globals will cause one of the globals to be discarded. This
397 is typically used to implement inline functions. Unreferenced
398 <tt>linkonce</tt> globals are allowed to be discarded.
401 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
403 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
404 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
405 used to implement constructs in C such as "<tt>int X;</tt>" at global scope.
408 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
410 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
411 pointer to array type. When two global variables with appending linkage are
412 linked together, the two global arrays are appended together. This is the
413 LLVM, typesafe, equivalent of having the system linker append together
414 "sections" with identical names when .o files are linked.
417 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
419 <dd>If none of the above identifiers are used, the global is externally
420 visible, meaning that it participates in linkage and can be used to resolve
421 external symbol references.
425 <p><a name="linkage_external">For example, since the "<tt>.LC0</tt>"
426 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
427 variable and was linked with this one, one of the two would be renamed,
428 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
429 external (i.e., lacking any linkage declarations), they are accessible
430 outside of the current module. It is illegal for a function <i>declaration</i>
431 to have any linkage type other than "externally visible".</a></p>
435 <!-- ======================================================================= -->
436 <div class="doc_subsection">
437 <a name="callingconv">Calling Conventions</a>
440 <div class="doc_text">
442 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
443 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
444 specified for the call. The calling convention of any pair of dynamic
445 caller/callee must match, or the behavior of the program is undefined. The
446 following calling conventions are supported by LLVM, and more may be added in
450 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
452 <dd>This calling convention (the default if no other calling convention is
453 specified) matches the target C calling conventions. This calling convention
454 supports varargs function calls and tolerates some mismatch in the declared
455 prototype and implemented declaration of the function (as does normal C).
458 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
460 <dd>This calling convention attempts to make calls as fast as possible
461 (e.g. by passing things in registers). This calling convention allows the
462 target to use whatever tricks it wants to produce fast code for the target,
463 without having to conform to an externally specified ABI. Implementations of
464 this convention should allow arbitrary tail call optimization to be supported.
465 This calling convention does not support varargs and requires the prototype of
466 all callees to exactly match the prototype of the function definition.
469 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
471 <dd>This calling convention attempts to make code in the caller as efficient
472 as possible under the assumption that the call is not commonly executed. As
473 such, these calls often preserve all registers so that the call does not break
474 any live ranges in the caller side. This calling convention does not support
475 varargs and requires the prototype of all callees to exactly match the
476 prototype of the function definition.
479 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
481 <dd>Any calling convention may be specified by number, allowing
482 target-specific calling conventions to be used. Target specific calling
483 conventions start at 64.
487 <p>More calling conventions can be added/defined on an as-needed basis, to
488 support pascal conventions or any other well-known target-independent
493 <!-- ======================================================================= -->
494 <div class="doc_subsection">
495 <a name="globalvars">Global Variables</a>
498 <div class="doc_text">
500 <p>Global variables define regions of memory allocated at compilation time
501 instead of run-time. Global variables may optionally be initialized, may have
502 an explicit section to be placed in, and may
503 have an optional explicit alignment specified. A
504 variable may be defined as a global "constant," which indicates that the
505 contents of the variable will <b>never</b> be modified (enabling better
506 optimization, allowing the global data to be placed in the read-only section of
507 an executable, etc). Note that variables that need runtime initialization
508 cannot be marked "constant" as there is a store to the variable.</p>
511 LLVM explicitly allows <em>declarations</em> of global variables to be marked
512 constant, even if the final definition of the global is not. This capability
513 can be used to enable slightly better optimization of the program, but requires
514 the language definition to guarantee that optimizations based on the
515 'constantness' are valid for the translation units that do not include the
519 <p>As SSA values, global variables define pointer values that are in
520 scope (i.e. they dominate) all basic blocks in the program. Global
521 variables always define a pointer to their "content" type because they
522 describe a region of memory, and all memory objects in LLVM are
523 accessed through pointers.</p>
525 <p>LLVM allows an explicit section to be specified for globals. If the target
526 supports it, it will emit globals to the section specified.</p>
528 <p>An explicit alignment may be specified for a global. If not present, or if
529 the alignment is set to zero, the alignment of the global is set by the target
530 to whatever it feels convenient. If an explicit alignment is specified, the
531 global is forced to have at least that much alignment. All alignments must be
537 <!-- ======================================================================= -->
538 <div class="doc_subsection">
539 <a name="functionstructure">Functions</a>
542 <div class="doc_text">
544 <p>LLVM function definitions consist of an optional <a href="#linkage">linkage
545 type</a>, an optional <a href="#callingconv">calling convention</a>, a return
546 type, a function name, a (possibly empty) argument list, an optional section,
547 an optional alignment, an opening curly brace,
548 a list of basic blocks, and a closing curly brace. LLVM function declarations
549 are defined with the "<tt>declare</tt>" keyword, an optional <a
550 href="#callingconv">calling convention</a>, a return type, a function name,
551 a possibly empty list of arguments, and an optional alignment.</p>
553 <p>A function definition contains a list of basic blocks, forming the CFG for
554 the function. Each basic block may optionally start with a label (giving the
555 basic block a symbol table entry), contains a list of instructions, and ends
556 with a <a href="#terminators">terminator</a> instruction (such as a branch or
557 function return).</p>
559 <p>The first basic block in a program is special in two ways: it is immediately
560 executed on entrance to the function, and it is not allowed to have predecessor
561 basic blocks (i.e. there can not be any branches to the entry block of a
562 function). Because the block can have no predecessors, it also cannot have any
563 <a href="#i_phi">PHI nodes</a>.</p>
565 <p>LLVM functions are identified by their name and type signature. Hence, two
566 functions with the same name but different parameter lists or return values are
567 considered different functions, and LLVM will resolve references to each
570 <p>LLVM allows an explicit section to be specified for functions. If the target
571 supports it, it will emit functions to the section specified.</p>
573 <p>An explicit alignment may be specified for a function. If not present, or if
574 the alignment is set to zero, the alignment of the function is set by the target
575 to whatever it feels convenient. If an explicit alignment is specified, the
576 function is forced to have at least that much alignment. All alignments must be
583 <!-- *********************************************************************** -->
584 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
585 <!-- *********************************************************************** -->
587 <div class="doc_text">
589 <p>The LLVM type system is one of the most important features of the
590 intermediate representation. Being typed enables a number of
591 optimizations to be performed on the IR directly, without having to do
592 extra analyses on the side before the transformation. A strong type
593 system makes it easier to read the generated code and enables novel
594 analyses and transformations that are not feasible to perform on normal
595 three address code representations.</p>
599 <!-- ======================================================================= -->
600 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
601 <div class="doc_text">
602 <p>The primitive types are the fundamental building blocks of the LLVM
603 system. The current set of primitive types is as follows:</p>
605 <table class="layout">
610 <tr><th>Type</th><th>Description</th></tr>
611 <tr><td><tt>void</tt></td><td>No value</td></tr>
612 <tr><td><tt>ubyte</tt></td><td>Unsigned 8-bit value</td></tr>
613 <tr><td><tt>ushort</tt></td><td>Unsigned 16-bit value</td></tr>
614 <tr><td><tt>uint</tt></td><td>Unsigned 32-bit value</td></tr>
615 <tr><td><tt>ulong</tt></td><td>Unsigned 64-bit value</td></tr>
616 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
617 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
624 <tr><th>Type</th><th>Description</th></tr>
625 <tr><td><tt>bool</tt></td><td>True or False value</td></tr>
626 <tr><td><tt>sbyte</tt></td><td>Signed 8-bit value</td></tr>
627 <tr><td><tt>short</tt></td><td>Signed 16-bit value</td></tr>
628 <tr><td><tt>int</tt></td><td>Signed 32-bit value</td></tr>
629 <tr><td><tt>long</tt></td><td>Signed 64-bit value</td></tr>
630 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
638 <!-- _______________________________________________________________________ -->
639 <div class="doc_subsubsection"> <a name="t_classifications">Type
640 Classifications</a> </div>
641 <div class="doc_text">
642 <p>These different primitive types fall into a few useful
645 <table border="1" cellspacing="0" cellpadding="4">
647 <tr><th>Classification</th><th>Types</th></tr>
649 <td><a name="t_signed">signed</a></td>
650 <td><tt>sbyte, short, int, long, float, double</tt></td>
653 <td><a name="t_unsigned">unsigned</a></td>
654 <td><tt>ubyte, ushort, uint, ulong</tt></td>
657 <td><a name="t_integer">integer</a></td>
658 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
661 <td><a name="t_integral">integral</a></td>
662 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt>
666 <td><a name="t_floating">floating point</a></td>
667 <td><tt>float, double</tt></td>
670 <td><a name="t_firstclass">first class</a></td>
671 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long,<br>
672 float, double, <a href="#t_pointer">pointer</a>,
673 <a href="#t_packed">packed</a></tt></td>
678 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
679 most important. Values of these types are the only ones which can be
680 produced by instructions, passed as arguments, or used as operands to
681 instructions. This means that all structures and arrays must be
682 manipulated either by pointer or by component.</p>
685 <!-- ======================================================================= -->
686 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
688 <div class="doc_text">
690 <p>The real power in LLVM comes from the derived types in the system.
691 This is what allows a programmer to represent arrays, functions,
692 pointers, and other useful types. Note that these derived types may be
693 recursive: For example, it is possible to have a two dimensional array.</p>
697 <!-- _______________________________________________________________________ -->
698 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
700 <div class="doc_text">
704 <p>The array type is a very simple derived type that arranges elements
705 sequentially in memory. The array type requires a size (number of
706 elements) and an underlying data type.</p>
711 [<# elements> x <elementtype>]
714 <p>The number of elements is a constant integer value; elementtype may
715 be any type with a size.</p>
718 <table class="layout">
721 <tt>[40 x int ]</tt><br/>
722 <tt>[41 x int ]</tt><br/>
723 <tt>[40 x uint]</tt><br/>
726 Array of 40 integer values.<br/>
727 Array of 41 integer values.<br/>
728 Array of 40 unsigned integer values.<br/>
732 <p>Here are some examples of multidimensional arrays:</p>
733 <table class="layout">
736 <tt>[3 x [4 x int]]</tt><br/>
737 <tt>[12 x [10 x float]]</tt><br/>
738 <tt>[2 x [3 x [4 x uint]]]</tt><br/>
741 3x4 array of integer values.<br/>
742 12x10 array of single precision floating point values.<br/>
743 2x3x4 array of unsigned integer values.<br/>
748 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
749 length array. Normally, accesses past the end of an array are undefined in
750 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
751 As a special case, however, zero length arrays are recognized to be variable
752 length. This allows implementation of 'pascal style arrays' with the LLVM
753 type "{ int, [0 x float]}", for example.</p>
757 <!-- _______________________________________________________________________ -->
758 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
759 <div class="doc_text">
761 <p>The function type can be thought of as a function signature. It
762 consists of a return type and a list of formal parameter types.
763 Function types are usually used to build virtual function tables
764 (which are structures of pointers to functions), for indirect function
765 calls, and when defining a function.</p>
767 The return type of a function type cannot be an aggregate type.
770 <pre> <returntype> (<parameter list>)<br></pre>
771 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
772 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
773 which indicates that the function takes a variable number of arguments.
774 Variable argument functions can access their arguments with the <a
775 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
777 <table class="layout">
780 <tt>int (int)</tt> <br/>
781 <tt>float (int, int *) *</tt><br/>
782 <tt>int (sbyte *, ...)</tt><br/>
785 function taking an <tt>int</tt>, returning an <tt>int</tt><br/>
786 <a href="#t_pointer">Pointer</a> to a function that takes an
787 <tt>int</tt> and a <a href="#t_pointer">pointer</a> to <tt>int</tt>,
788 returning <tt>float</tt>.<br/>
789 A vararg function that takes at least one <a href="#t_pointer">pointer</a>
790 to <tt>sbyte</tt> (signed char in C), which returns an integer. This is
791 the signature for <tt>printf</tt> in LLVM.<br/>
797 <!-- _______________________________________________________________________ -->
798 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
799 <div class="doc_text">
801 <p>The structure type is used to represent a collection of data members
802 together in memory. The packing of the field types is defined to match
803 the ABI of the underlying processor. The elements of a structure may
804 be any type that has a size.</p>
805 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
806 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
807 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
810 <pre> { <type list> }<br></pre>
812 <table class="layout">
815 <tt>{ int, int, int }</tt><br/>
816 <tt>{ float, int (int) * }</tt><br/>
819 a triple of three <tt>int</tt> values<br/>
820 A pair, where the first element is a <tt>float</tt> and the second element
821 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
822 that takes an <tt>int</tt>, returning an <tt>int</tt>.<br/>
828 <!-- _______________________________________________________________________ -->
829 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
830 <div class="doc_text">
832 <p>As in many languages, the pointer type represents a pointer or
833 reference to another object, which must live in memory.</p>
835 <pre> <type> *<br></pre>
837 <table class="layout">
840 <tt>[4x int]*</tt><br/>
841 <tt>int (int *) *</tt><br/>
844 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
845 four <tt>int</tt> values<br/>
846 A <a href="#t_pointer">pointer</a> to a <a
847 href="#t_function">function</a> that takes an <tt>int*</tt>, returning an
854 <!-- _______________________________________________________________________ -->
855 <div class="doc_subsubsection"> <a name="t_packed">Packed Type</a> </div>
856 <div class="doc_text">
860 <p>A packed type is a simple derived type that represents a vector
861 of elements. Packed types are used when multiple primitive data
862 are operated in parallel using a single instruction (SIMD).
863 A packed type requires a size (number of
864 elements) and an underlying primitive data type. Vectors must have a power
865 of two length (1, 2, 4, 8, 16 ...). Packed types are
866 considered <a href="#t_firstclass">first class</a>.</p>
871 < <# elements> x <elementtype> >
874 <p>The number of elements is a constant integer value; elementtype may
875 be any integral or floating point type.</p>
879 <table class="layout">
882 <tt><4 x int></tt><br/>
883 <tt><8 x float></tt><br/>
884 <tt><2 x uint></tt><br/>
887 Packed vector of 4 integer values.<br/>
888 Packed vector of 8 floating-point values.<br/>
889 Packed vector of 2 unsigned integer values.<br/>
895 <!-- _______________________________________________________________________ -->
896 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
897 <div class="doc_text">
901 <p>Opaque types are used to represent unknown types in the system. This
902 corresponds (for example) to the C notion of a foward declared structure type.
903 In LLVM, opaque types can eventually be resolved to any type (not just a
914 <table class="layout">
927 <!-- *********************************************************************** -->
928 <div class="doc_section"> <a name="constants">Constants</a> </div>
929 <!-- *********************************************************************** -->
931 <div class="doc_text">
933 <p>LLVM has several different basic types of constants. This section describes
934 them all and their syntax.</p>
938 <!-- ======================================================================= -->
939 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
941 <div class="doc_text">
944 <dt><b>Boolean constants</b></dt>
946 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
947 constants of the <tt><a href="#t_primitive">bool</a></tt> type.
950 <dt><b>Integer constants</b></dt>
952 <dd>Standard integers (such as '4') are constants of the <a
953 href="#t_integer">integer</a> type. Negative numbers may be used with signed
957 <dt><b>Floating point constants</b></dt>
959 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
960 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
961 notation (see below). Floating point constants must have a <a
962 href="#t_floating">floating point</a> type. </dd>
964 <dt><b>Null pointer constants</b></dt>
966 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
967 and must be of <a href="#t_pointer">pointer type</a>.</dd>
971 <p>The one non-intuitive notation for constants is the optional hexadecimal form
972 of floating point constants. For example, the form '<tt>double
973 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
974 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
975 (and the only time that they are generated by the disassembler) is when a
976 floating point constant must be emitted but it cannot be represented as a
977 decimal floating point number. For example, NaN's, infinities, and other
978 special values are represented in their IEEE hexadecimal format so that
979 assembly and disassembly do not cause any bits to change in the constants.</p>
983 <!-- ======================================================================= -->
984 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
987 <div class="doc_text">
988 <p>Aggregate constants arise from aggregation of simple constants
989 and smaller aggregate constants.</p>
992 <dt><b>Structure constants</b></dt>
994 <dd>Structure constants are represented with notation similar to structure
995 type definitions (a comma separated list of elements, surrounded by braces
996 (<tt>{}</tt>)). For example: "<tt>{ int 4, float 17.0, int* %G }</tt>",
997 where "<tt>%G</tt>" is declared as "<tt>%G = external global int</tt>". Structure constants
998 must have <a href="#t_struct">structure type</a>, and the number and
999 types of elements must match those specified by the type.
1002 <dt><b>Array constants</b></dt>
1004 <dd>Array constants are represented with notation similar to array type
1005 definitions (a comma separated list of elements, surrounded by square brackets
1006 (<tt>[]</tt>)). For example: "<tt>[ int 42, int 11, int 74 ]</tt>". Array
1007 constants must have <a href="#t_array">array type</a>, and the number and
1008 types of elements must match those specified by the type.
1011 <dt><b>Packed constants</b></dt>
1013 <dd>Packed constants are represented with notation similar to packed type
1014 definitions (a comma separated list of elements, surrounded by
1015 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< int 42,
1016 int 11, int 74, int 100 ></tt>". Packed constants must have <a
1017 href="#t_packed">packed type</a>, and the number and types of elements must
1018 match those specified by the type.
1021 <dt><b>Zero initialization</b></dt>
1023 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1024 value to zero of <em>any</em> type, including scalar and aggregate types.
1025 This is often used to avoid having to print large zero initializers (e.g. for
1026 large arrays) and is always exactly equivalent to using explicit zero
1033 <!-- ======================================================================= -->
1034 <div class="doc_subsection">
1035 <a name="globalconstants">Global Variable and Function Addresses</a>
1038 <div class="doc_text">
1040 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1041 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1042 constants. These constants are explicitly referenced when the <a
1043 href="#identifiers">identifier for the global</a> is used and always have <a
1044 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1050 %Z = global [2 x int*] [ int* %X, int* %Y ]
1055 <!-- ======================================================================= -->
1056 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1057 <div class="doc_text">
1058 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1059 no specific value. Undefined values may be of any type and be used anywhere
1060 a constant is permitted.</p>
1062 <p>Undefined values indicate to the compiler that the program is well defined
1063 no matter what value is used, giving the compiler more freedom to optimize.
1067 <!-- ======================================================================= -->
1068 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1071 <div class="doc_text">
1073 <p>Constant expressions are used to allow expressions involving other constants
1074 to be used as constants. Constant expressions may be of any <a
1075 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1076 that does not have side effects (e.g. load and call are not supported). The
1077 following is the syntax for constant expressions:</p>
1080 <dt><b><tt>cast ( CST to TYPE )</tt></b></dt>
1082 <dd>Cast a constant to another type.</dd>
1084 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1086 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1087 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1088 instruction, the index list may have zero or more indexes, which are required
1089 to make sense for the type of "CSTPTR".</dd>
1091 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1093 <dd>Perform the <a href="#i_select">select operation</a> on
1096 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1098 <dd>Perform the <a href="#i_extractelement">extractelement
1099 operation</a> on constants.
1101 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1103 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1104 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1105 binary</a> operations. The constraints on operands are the same as those for
1106 the corresponding instruction (e.g. no bitwise operations on floating point
1107 values are allowed).</dd>
1111 <!-- *********************************************************************** -->
1112 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1113 <!-- *********************************************************************** -->
1115 <div class="doc_text">
1117 <p>The LLVM instruction set consists of several different
1118 classifications of instructions: <a href="#terminators">terminator
1119 instructions</a>, <a href="#binaryops">binary instructions</a>,
1120 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1121 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1122 instructions</a>.</p>
1126 <!-- ======================================================================= -->
1127 <div class="doc_subsection"> <a name="terminators">Terminator
1128 Instructions</a> </div>
1130 <div class="doc_text">
1132 <p>As mentioned <a href="#functionstructure">previously</a>, every
1133 basic block in a program ends with a "Terminator" instruction, which
1134 indicates which block should be executed after the current block is
1135 finished. These terminator instructions typically yield a '<tt>void</tt>'
1136 value: they produce control flow, not values (the one exception being
1137 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1138 <p>There are six different terminator instructions: the '<a
1139 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1140 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1141 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1142 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1143 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1147 <!-- _______________________________________________________________________ -->
1148 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1149 Instruction</a> </div>
1150 <div class="doc_text">
1152 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1153 ret void <i>; Return from void function</i>
1156 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1157 value) from a function back to the caller.</p>
1158 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1159 returns a value and then causes control flow, and one that just causes
1160 control flow to occur.</p>
1162 <p>The '<tt>ret</tt>' instruction may return any '<a
1163 href="#t_firstclass">first class</a>' type. Notice that a function is
1164 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1165 instruction inside of the function that returns a value that does not
1166 match the return type of the function.</p>
1168 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1169 returns back to the calling function's context. If the caller is a "<a
1170 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1171 the instruction after the call. If the caller was an "<a
1172 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1173 at the beginning of the "normal" destination block. If the instruction
1174 returns a value, that value shall set the call or invoke instruction's
1177 <pre> ret int 5 <i>; Return an integer value of 5</i>
1178 ret void <i>; Return from a void function</i>
1181 <!-- _______________________________________________________________________ -->
1182 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1183 <div class="doc_text">
1185 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1188 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1189 transfer to a different basic block in the current function. There are
1190 two forms of this instruction, corresponding to a conditional branch
1191 and an unconditional branch.</p>
1193 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1194 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
1195 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
1196 value as a target.</p>
1198 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
1199 argument is evaluated. If the value is <tt>true</tt>, control flows
1200 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1201 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1203 <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
1204 href="#i_ret">ret</a> int 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> int 0<br></pre>
1206 <!-- _______________________________________________________________________ -->
1207 <div class="doc_subsubsection">
1208 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1211 <div class="doc_text">
1215 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1220 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1221 several different places. It is a generalization of the '<tt>br</tt>'
1222 instruction, allowing a branch to occur to one of many possible
1228 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1229 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1230 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1231 table is not allowed to contain duplicate constant entries.</p>
1235 <p>The <tt>switch</tt> instruction specifies a table of values and
1236 destinations. When the '<tt>switch</tt>' instruction is executed, this
1237 table is searched for the given value. If the value is found, control flow is
1238 transfered to the corresponding destination; otherwise, control flow is
1239 transfered to the default destination.</p>
1241 <h5>Implementation:</h5>
1243 <p>Depending on properties of the target machine and the particular
1244 <tt>switch</tt> instruction, this instruction may be code generated in different
1245 ways. For example, it could be generated as a series of chained conditional
1246 branches or with a lookup table.</p>
1251 <i>; Emulate a conditional br instruction</i>
1252 %Val = <a href="#i_cast">cast</a> bool %value to int
1253 switch int %Val, label %truedest [int 0, label %falsedest ]
1255 <i>; Emulate an unconditional br instruction</i>
1256 switch uint 0, label %dest [ ]
1258 <i>; Implement a jump table:</i>
1259 switch uint %val, label %otherwise [ uint 0, label %onzero
1260 uint 1, label %onone
1261 uint 2, label %ontwo ]
1265 <!-- _______________________________________________________________________ -->
1266 <div class="doc_subsubsection">
1267 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1270 <div class="doc_text">
1275 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1276 to label <normal label> except label <exception label>
1281 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1282 function, with the possibility of control flow transfer to either the
1283 '<tt>normal</tt>' label or the
1284 '<tt>exception</tt>' label. If the callee function returns with the
1285 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1286 "normal" label. If the callee (or any indirect callees) returns with the "<a
1287 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1288 continued at the dynamically nearest "exception" label.</p>
1292 <p>This instruction requires several arguments:</p>
1296 The optional "cconv" marker indicates which <a href="callingconv">calling
1297 convention</a> the call should use. If none is specified, the call defaults
1298 to using C calling conventions.
1300 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1301 function value being invoked. In most cases, this is a direct function
1302 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1303 an arbitrary pointer to function value.
1306 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1307 function to be invoked. </li>
1309 <li>'<tt>function args</tt>': argument list whose types match the function
1310 signature argument types. If the function signature indicates the function
1311 accepts a variable number of arguments, the extra arguments can be
1314 <li>'<tt>normal label</tt>': the label reached when the called function
1315 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1317 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1318 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1324 <p>This instruction is designed to operate as a standard '<tt><a
1325 href="#i_call">call</a></tt>' instruction in most regards. The primary
1326 difference is that it establishes an association with a label, which is used by
1327 the runtime library to unwind the stack.</p>
1329 <p>This instruction is used in languages with destructors to ensure that proper
1330 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1331 exception. Additionally, this is important for implementation of
1332 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1336 %retval = invoke int %Test(int 15) to label %Continue
1337 except label %TestCleanup <i>; {int}:retval set</i>
1338 %retval = invoke <a href="#callingconv">coldcc</a> int %Test(int 15) to label %Continue
1339 except label %TestCleanup <i>; {int}:retval set</i>
1344 <!-- _______________________________________________________________________ -->
1346 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1347 Instruction</a> </div>
1349 <div class="doc_text">
1358 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1359 at the first callee in the dynamic call stack which used an <a
1360 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1361 primarily used to implement exception handling.</p>
1365 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1366 immediately halt. The dynamic call stack is then searched for the first <a
1367 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1368 execution continues at the "exceptional" destination block specified by the
1369 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1370 dynamic call chain, undefined behavior results.</p>
1373 <!-- _______________________________________________________________________ -->
1375 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1376 Instruction</a> </div>
1378 <div class="doc_text">
1387 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1388 instruction is used to inform the optimizer that a particular portion of the
1389 code is not reachable. This can be used to indicate that the code after a
1390 no-return function cannot be reached, and other facts.</p>
1394 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1399 <!-- ======================================================================= -->
1400 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1401 <div class="doc_text">
1402 <p>Binary operators are used to do most of the computation in a
1403 program. They require two operands, execute an operation on them, and
1404 produce a single value. The operands might represent
1405 multiple data, as is the case with the <a href="#t_packed">packed</a> data type.
1406 The result value of a binary operator is not
1407 necessarily the same type as its operands.</p>
1408 <p>There are several different binary operators:</p>
1410 <!-- _______________________________________________________________________ -->
1411 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1412 Instruction</a> </div>
1413 <div class="doc_text">
1415 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1418 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1420 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1421 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1422 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1423 Both arguments must have identical types.</p>
1425 <p>The value produced is the integer or floating point sum of the two
1428 <pre> <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
1431 <!-- _______________________________________________________________________ -->
1432 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1433 Instruction</a> </div>
1434 <div class="doc_text">
1436 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1439 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1441 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1442 instruction present in most other intermediate representations.</p>
1444 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1445 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1447 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1448 Both arguments must have identical types.</p>
1450 <p>The value produced is the integer or floating point difference of
1451 the two operands.</p>
1453 <pre> <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
1454 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
1457 <!-- _______________________________________________________________________ -->
1458 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1459 Instruction</a> </div>
1460 <div class="doc_text">
1462 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1465 <p>The '<tt>mul</tt>' instruction returns the product of its two
1468 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1469 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1471 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1472 Both arguments must have identical types.</p>
1474 <p>The value produced is the integer or floating point product of the
1476 <p>There is no signed vs unsigned multiplication. The appropriate
1477 action is taken based on the type of the operand.</p>
1479 <pre> <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1482 <!-- _______________________________________________________________________ -->
1483 <div class="doc_subsubsection"> <a name="i_div">'<tt>div</tt>'
1484 Instruction</a> </div>
1485 <div class="doc_text">
1487 <pre> <result> = div <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1490 <p>The '<tt>div</tt>' instruction returns the quotient of its two
1493 <p>The two arguments to the '<tt>div</tt>' instruction must be either <a
1494 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1496 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1497 Both arguments must have identical types.</p>
1499 <p>The value produced is the integer or floating point quotient of the
1502 <pre> <result> = div int 4, %var <i>; yields {int}:result = 4 / %var</i>
1505 <!-- _______________________________________________________________________ -->
1506 <div class="doc_subsubsection"> <a name="i_rem">'<tt>rem</tt>'
1507 Instruction</a> </div>
1508 <div class="doc_text">
1510 <pre> <result> = rem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1513 <p>The '<tt>rem</tt>' instruction returns the remainder from the
1514 division of its two operands.</p>
1516 <p>The two arguments to the '<tt>rem</tt>' instruction must be either <a
1517 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1519 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1520 Both arguments must have identical types.</p>
1522 <p>This returns the <i>remainder</i> of a division (where the result
1523 has the same sign as the divisor), not the <i>modulus</i> (where the
1524 result has the same sign as the dividend) of a value. For more
1525 information about the difference, see <a
1526 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
1529 <pre> <result> = rem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1532 <!-- _______________________________________________________________________ -->
1533 <div class="doc_subsubsection"> <a name="i_setcc">'<tt>set<i>cc</i></tt>'
1534 Instructions</a> </div>
1535 <div class="doc_text">
1537 <pre> <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1538 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1539 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1540 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1541 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1542 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1545 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean
1546 value based on a comparison of their two operands.</p>
1548 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must
1549 be of <a href="#t_firstclass">first class</a> type (it is not possible
1550 to compare '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>'
1551 or '<tt>void</tt>' values, etc...). Both arguments must have identical
1554 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1555 value if both operands are equal.<br>
1556 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1557 value if both operands are unequal.<br>
1558 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1559 value if the first operand is less than the second operand.<br>
1560 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1561 value if the first operand is greater than the second operand.<br>
1562 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1563 value if the first operand is less than or equal to the second operand.<br>
1564 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1565 value if the first operand is greater than or equal to the second
1568 <pre> <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1569 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1570 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1571 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1572 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1573 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1576 <!-- ======================================================================= -->
1577 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
1578 Operations</a> </div>
1579 <div class="doc_text">
1580 <p>Bitwise binary operators are used to do various forms of
1581 bit-twiddling in a program. They are generally very efficient
1582 instructions and can commonly be strength reduced from other
1583 instructions. They require two operands, execute an operation on them,
1584 and produce a single value. The resulting value of the bitwise binary
1585 operators is always the same type as its first operand.</p>
1587 <!-- _______________________________________________________________________ -->
1588 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
1589 Instruction</a> </div>
1590 <div class="doc_text">
1592 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1595 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1596 its two operands.</p>
1598 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1599 href="#t_integral">integral</a> values. Both arguments must have
1600 identical types.</p>
1602 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1604 <div style="align: center">
1605 <table border="1" cellspacing="0" cellpadding="4">
1636 <pre> <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1637 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1638 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1641 <!-- _______________________________________________________________________ -->
1642 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
1643 <div class="doc_text">
1645 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1648 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1649 or of its two operands.</p>
1651 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1652 href="#t_integral">integral</a> values. Both arguments must have
1653 identical types.</p>
1655 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1657 <div style="align: center">
1658 <table border="1" cellspacing="0" cellpadding="4">
1689 <pre> <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1690 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1691 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1694 <!-- _______________________________________________________________________ -->
1695 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
1696 Instruction</a> </div>
1697 <div class="doc_text">
1699 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1702 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
1703 or of its two operands. The <tt>xor</tt> is used to implement the
1704 "one's complement" operation, which is the "~" operator in C.</p>
1706 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
1707 href="#t_integral">integral</a> values. Both arguments must have
1708 identical types.</p>
1710 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
1712 <div style="align: center">
1713 <table border="1" cellspacing="0" cellpadding="4">
1745 <pre> <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
1746 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
1747 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
1748 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
1751 <!-- _______________________________________________________________________ -->
1752 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
1753 Instruction</a> </div>
1754 <div class="doc_text">
1756 <pre> <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1759 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
1760 the left a specified number of bits.</p>
1762 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
1763 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1766 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
1768 <pre> <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
1769 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
1770 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
1773 <!-- _______________________________________________________________________ -->
1774 <div class="doc_subsubsection"> <a name="i_shr">'<tt>shr</tt>'
1775 Instruction</a> </div>
1776 <div class="doc_text">
1778 <pre> <result> = shr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1781 <p>The '<tt>shr</tt>' instruction returns the first operand shifted to
1782 the right a specified number of bits.</p>
1784 <p>The first argument to the '<tt>shr</tt>' instruction must be an <a
1785 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1788 <p>If the first argument is a <a href="#t_signed">signed</a> type, the
1789 most significant bit is duplicated in the newly free'd bit positions.
1790 If the first argument is unsigned, zero bits shall fill the empty
1793 <pre> <result> = shr int 4, ubyte %var <i>; yields {int}:result = 4 >> %var</i>
1794 <result> = shr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
1795 <result> = shr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
1796 <result> = shr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
1797 <result> = shr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
1801 <!-- ======================================================================= -->
1802 <div class="doc_subsection">
1803 <a name="memoryops">Memory Access Operations</a>
1806 <div class="doc_text">
1808 <p>A key design point of an SSA-based representation is how it
1809 represents memory. In LLVM, no memory locations are in SSA form, which
1810 makes things very simple. This section describes how to read, write,
1811 allocate, and free memory in LLVM.</p>
1815 <!-- _______________________________________________________________________ -->
1816 <div class="doc_subsubsection">
1817 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
1820 <div class="doc_text">
1825 <result> = malloc <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
1830 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
1831 heap and returns a pointer to it.</p>
1835 <p>The '<tt>malloc</tt>' instruction allocates
1836 <tt>sizeof(<type>)*NumElements</tt>
1837 bytes of memory from the operating system and returns a pointer of the
1838 appropriate type to the program. If "NumElements" is specified, it is the
1839 number of elements allocated. If an alignment is specified, the value result
1840 of the allocation is guaranteed to be aligned to at least that boundary. If
1841 not specified, or if zero, the target can choose to align the allocation on any
1842 convenient boundary.</p>
1844 <p>'<tt>type</tt>' must be a sized type.</p>
1848 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
1849 a pointer is returned.</p>
1854 %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
1856 %size = <a href="#i_add">add</a> uint 2, 2 <i>; yields {uint}:size = uint 4</i>
1857 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
1858 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
1859 %array3 = malloc int, uint 4, align 1024 <i>; yields {int*}:array3</i>
1860 %array4 = malloc int, align 1024 <i>; yields {int*}:array4</i>
1864 <!-- _______________________________________________________________________ -->
1865 <div class="doc_subsubsection">
1866 <a name="i_free">'<tt>free</tt>' Instruction</a>
1869 <div class="doc_text">
1874 free <type> <value> <i>; yields {void}</i>
1879 <p>The '<tt>free</tt>' instruction returns memory back to the unused
1880 memory heap to be reallocated in the future.</p>
1884 <p>'<tt>value</tt>' shall be a pointer value that points to a value
1885 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
1890 <p>Access to the memory pointed to by the pointer is no longer defined
1891 after this instruction executes.</p>
1896 %array = <a href="#i_malloc">malloc</a> [4 x ubyte] <i>; yields {[4 x ubyte]*}:array</i>
1897 free [4 x ubyte]* %array
1901 <!-- _______________________________________________________________________ -->
1902 <div class="doc_subsubsection">
1903 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
1906 <div class="doc_text">
1911 <result> = alloca <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
1916 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
1917 stack frame of the procedure that is live until the current function
1918 returns to its caller.</p>
1922 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
1923 bytes of memory on the runtime stack, returning a pointer of the
1924 appropriate type to the program. If "NumElements" is specified, it is the
1925 number of elements allocated. If an alignment is specified, the value result
1926 of the allocation is guaranteed to be aligned to at least that boundary. If
1927 not specified, or if zero, the target can choose to align the allocation on any
1928 convenient boundary.</p>
1930 <p>'<tt>type</tt>' may be any sized type.</p>
1934 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
1935 memory is automatically released when the function returns. The '<tt>alloca</tt>'
1936 instruction is commonly used to represent automatic variables that must
1937 have an address available. When the function returns (either with the <tt><a
1938 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
1939 instructions), the memory is reclaimed.</p>
1944 %ptr = alloca int <i>; yields {int*}:ptr</i>
1945 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
1946 %ptr = alloca int, uint 4, align 1024 <i>; yields {int*}:ptr</i>
1947 %ptr = alloca int, align 1024 <i>; yields {int*}:ptr</i>
1951 <!-- _______________________________________________________________________ -->
1952 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
1953 Instruction</a> </div>
1954 <div class="doc_text">
1956 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
1958 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
1960 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
1961 address from which to load. The pointer must point to a <a
1962 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
1963 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
1964 the number or order of execution of this <tt>load</tt> with other
1965 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
1968 <p>The location of memory pointed to is loaded.</p>
1970 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
1972 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
1973 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
1976 <!-- _______________________________________________________________________ -->
1977 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
1978 Instruction</a> </div>
1980 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
1981 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
1984 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
1986 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
1987 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
1988 operand must be a pointer to the type of the '<tt><value></tt>'
1989 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
1990 optimizer is not allowed to modify the number or order of execution of
1991 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
1992 href="#i_store">store</a></tt> instructions.</p>
1994 <p>The contents of memory are updated to contain '<tt><value></tt>'
1995 at the location specified by the '<tt><pointer></tt>' operand.</p>
1997 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
1999 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2000 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2002 <!-- _______________________________________________________________________ -->
2003 <div class="doc_subsubsection">
2004 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2007 <div class="doc_text">
2010 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2016 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2017 subelement of an aggregate data structure.</p>
2021 <p>This instruction takes a list of integer constants that indicate what
2022 elements of the aggregate object to index to. The actual types of the arguments
2023 provided depend on the type of the first pointer argument. The
2024 '<tt>getelementptr</tt>' instruction is used to index down through the type
2025 levels of a structure or to a specific index in an array. When indexing into a
2026 structure, only <tt>uint</tt>
2027 integer constants are allowed. When indexing into an array or pointer,
2028 <tt>int</tt> and <tt>long</tt> indexes are allowed of any sign.</p>
2030 <p>For example, let's consider a C code fragment and how it gets
2031 compiled to LLVM:</p>
2045 int *foo(struct ST *s) {
2046 return &s[1].Z.B[5][13];
2050 <p>The LLVM code generated by the GCC frontend is:</p>
2053 %RT = type { sbyte, [10 x [20 x int]], sbyte }
2054 %ST = type { int, double, %RT }
2058 int* %foo(%ST* %s) {
2060 %reg = getelementptr %ST* %s, int 1, uint 2, uint 1, int 5, int 13
2067 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2068 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2069 and <a href="#t_array">array</a> types require <tt>uint</tt>, <tt>int</tt>,
2070 <tt>ulong</tt>, or <tt>long</tt> values, and <a href="#t_struct">structure</a>
2071 types require <tt>uint</tt> <b>constants</b>.</p>
2073 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2074 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
2075 }</tt>' type, a structure. The second index indexes into the third element of
2076 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
2077 sbyte }</tt>' type, another structure. The third index indexes into the second
2078 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
2079 array. The two dimensions of the array are subscripted into, yielding an
2080 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2081 to this element, thus computing a value of '<tt>int*</tt>' type.</p>
2083 <p>Note that it is perfectly legal to index partially through a
2084 structure, returning a pointer to an inner element. Because of this,
2085 the LLVM code for the given testcase is equivalent to:</p>
2088 int* %foo(%ST* %s) {
2089 %t1 = getelementptr %ST* %s, int 1 <i>; yields %ST*:%t1</i>
2090 %t2 = getelementptr %ST* %t1, int 0, uint 2 <i>; yields %RT*:%t2</i>
2091 %t3 = getelementptr %RT* %t2, int 0, uint 1 <i>; yields [10 x [20 x int]]*:%t3</i>
2092 %t4 = getelementptr [10 x [20 x int]]* %t3, int 0, int 5 <i>; yields [20 x int]*:%t4</i>
2093 %t5 = getelementptr [20 x int]* %t4, int 0, int 13 <i>; yields int*:%t5</i>
2098 <p>Note that it is undefined to access an array out of bounds: array and
2099 pointer indexes must always be within the defined bounds of the array type.
2100 The one exception for this rules is zero length arrays. These arrays are
2101 defined to be accessible as variable length arrays, which requires access
2102 beyond the zero'th element.</p>
2107 <i>; yields [12 x ubyte]*:aptr</i>
2108 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, uint 1
2112 <!-- ======================================================================= -->
2113 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
2114 <div class="doc_text">
2115 <p>The instructions in this category are the "miscellaneous"
2116 instructions, which defy better classification.</p>
2118 <!-- _______________________________________________________________________ -->
2119 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
2120 Instruction</a> </div>
2121 <div class="doc_text">
2123 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
2125 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
2126 the SSA graph representing the function.</p>
2128 <p>The type of the incoming values are specified with the first type
2129 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
2130 as arguments, with one pair for each predecessor basic block of the
2131 current block. Only values of <a href="#t_firstclass">first class</a>
2132 type may be used as the value arguments to the PHI node. Only labels
2133 may be used as the label arguments.</p>
2134 <p>There must be no non-phi instructions between the start of a basic
2135 block and the PHI instructions: i.e. PHI instructions must be first in
2138 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
2139 value specified by the parameter, depending on which basic block we
2140 came from in the last <a href="#terminators">terminator</a> instruction.</p>
2142 <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>
2145 <!-- _______________________________________________________________________ -->
2146 <div class="doc_subsubsection">
2147 <a name="i_cast">'<tt>cast .. to</tt>' Instruction</a>
2150 <div class="doc_text">
2155 <result> = cast <ty> <value> to <ty2> <i>; yields ty2</i>
2161 The '<tt>cast</tt>' instruction is used as the primitive means to convert
2162 integers to floating point, change data type sizes, and break type safety (by
2170 The '<tt>cast</tt>' instruction takes a value to cast, which must be a first
2171 class value, and a type to cast it to, which must also be a <a
2172 href="#t_firstclass">first class</a> type.
2178 This instruction follows the C rules for explicit casts when determining how the
2179 data being cast must change to fit in its new container.
2183 When casting to bool, any value that would be considered true in the context of
2184 a C '<tt>if</tt>' condition is converted to the boolean '<tt>true</tt>' values,
2185 all else are '<tt>false</tt>'.
2189 When extending an integral value from a type of one signness to another (for
2190 example '<tt>sbyte</tt>' to '<tt>ulong</tt>'), the value is sign-extended if the
2191 <b>source</b> value is signed, and zero-extended if the source value is
2192 unsigned. <tt>bool</tt> values are always zero extended into either zero or
2199 %X = cast int 257 to ubyte <i>; yields ubyte:1</i>
2200 %Y = cast int 123 to bool <i>; yields bool:true</i>
2204 <!-- _______________________________________________________________________ -->
2205 <div class="doc_subsubsection">
2206 <a name="i_select">'<tt>select</tt>' Instruction</a>
2209 <div class="doc_text">
2214 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
2220 The '<tt>select</tt>' instruction is used to choose one value based on a
2221 condition, without branching.
2228 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.
2234 If the boolean condition evaluates to true, the instruction returns the first
2235 value argument; otherwise, it returns the second value argument.
2241 %X = select bool true, ubyte 17, ubyte 42 <i>; yields ubyte:17</i>
2246 <!-- _______________________________________________________________________ -->
2247 <div class="doc_subsubsection">
2248 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2251 <div class="doc_text">
2256 <result> = extractelement <n x <ty>> <val>, uint <idx> <i>; yields <ty></i>
2262 The '<tt>extractelement</tt>' instruction extracts a single scalar
2263 element from a vector at a specified index.
2270 The first operand of an '<tt>extractelement</tt>' instruction is a
2271 value of <a href="#t_packed">packed</a> type. The second operand is
2272 an index indicating the position from which to extract the element.
2273 The index may be a variable.</p>
2278 The result is a scalar of the same type as the element type of
2279 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2280 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2281 results are undefined.
2287 %result = extractelement <4 x int> %vec, uint 0 <i>; yields int</i>
2292 <!-- _______________________________________________________________________ -->
2293 <div class="doc_subsubsection">
2294 <a name="i_call">'<tt>call</tt>' Instruction</a>
2297 <div class="doc_text">
2301 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
2306 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
2310 <p>This instruction requires several arguments:</p>
2314 <p>The optional "tail" marker indicates whether the callee function accesses
2315 any allocas or varargs in the caller. If the "tail" marker is present, the
2316 function call is eligible for tail call optimization. Note that calls may
2317 be marked "tail" even if they do not occur before a <a
2318 href="#i_ret"><tt>ret</tt></a> instruction.
2321 <p>The optional "cconv" marker indicates which <a href="callingconv">calling
2322 convention</a> the call should use. If none is specified, the call defaults
2323 to using C calling conventions.
2326 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
2327 being invoked. The argument types must match the types implied by this
2328 signature. This type can be omitted if the function is not varargs and
2329 if the function type does not return a pointer to a function.</p>
2332 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
2333 be invoked. In most cases, this is a direct function invocation, but
2334 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
2335 to function value.</p>
2338 <p>'<tt>function args</tt>': argument list whose types match the
2339 function signature argument types. All arguments must be of
2340 <a href="#t_firstclass">first class</a> type. If the function signature
2341 indicates the function accepts a variable number of arguments, the extra
2342 arguments can be specified.</p>
2348 <p>The '<tt>call</tt>' instruction is used to cause control flow to
2349 transfer to a specified function, with its incoming arguments bound to
2350 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
2351 instruction in the called function, control flow continues with the
2352 instruction after the function call, and the return value of the
2353 function is bound to the result argument. This is a simpler case of
2354 the <a href="#i_invoke">invoke</a> instruction.</p>
2359 %retval = call int %test(int %argc)
2360 call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);
2361 %X = tail call int %foo()
2362 %Y = tail call <a href="#callingconv">fastcc</a> int %foo()
2367 <!-- _______________________________________________________________________ -->
2368 <div class="doc_subsubsection">
2369 <a name="i_vaarg">'<tt>vaarg</tt>' Instruction</a>
2372 <div class="doc_text">
2377 <resultval> = va_arg <va_list*> <arglist>, <argty>
2382 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
2383 the "variable argument" area of a function call. It is used to implement the
2384 <tt>va_arg</tt> macro in C.</p>
2388 <p>This instruction takes a <tt>va_list*</tt> value and the type of
2389 the argument. It returns a value of the specified argument type and
2390 increments the <tt>va_list</tt> to point to the next argument. Again, the
2391 actual type of <tt>va_list</tt> is target specific.</p>
2395 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
2396 type from the specified <tt>va_list</tt> and causes the
2397 <tt>va_list</tt> to point to the next argument. For more information,
2398 see the variable argument handling <a href="#int_varargs">Intrinsic
2401 <p>It is legal for this instruction to be called in a function which does not
2402 take a variable number of arguments, for example, the <tt>vfprintf</tt>
2405 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
2406 href="#intrinsics">intrinsic function</a> because it takes a type as an
2411 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
2415 <!-- *********************************************************************** -->
2416 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
2417 <!-- *********************************************************************** -->
2419 <div class="doc_text">
2421 <p>LLVM supports the notion of an "intrinsic function". These functions have
2422 well known names and semantics and are required to follow certain
2423 restrictions. Overall, these instructions represent an extension mechanism for
2424 the LLVM language that does not require changing all of the transformations in
2425 LLVM to add to the language (or the bytecode reader/writer, the parser,
2428 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
2429 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
2430 this. Intrinsic functions must always be external functions: you cannot define
2431 the body of intrinsic functions. Intrinsic functions may only be used in call
2432 or invoke instructions: it is illegal to take the address of an intrinsic
2433 function. Additionally, because intrinsic functions are part of the LLVM
2434 language, it is required that they all be documented here if any are added.</p>
2437 <p>To learn how to add an intrinsic function, please see the <a
2438 href="ExtendingLLVM.html">Extending LLVM Guide</a>.
2443 <!-- ======================================================================= -->
2444 <div class="doc_subsection">
2445 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
2448 <div class="doc_text">
2450 <p>Variable argument support is defined in LLVM with the <a
2451 href="#i_vanext"><tt>vanext</tt></a> instruction and these three
2452 intrinsic functions. These functions are related to the similarly
2453 named macros defined in the <tt><stdarg.h></tt> header file.</p>
2455 <p>All of these functions operate on arguments that use a
2456 target-specific value type "<tt>va_list</tt>". The LLVM assembly
2457 language reference manual does not define what this type is, so all
2458 transformations should be prepared to handle intrinsics with any type
2461 <p>This example shows how the <a href="#i_vanext"><tt>vanext</tt></a>
2462 instruction and the variable argument handling intrinsic functions are
2466 int %test(int %X, ...) {
2467 ; Initialize variable argument processing
2469 call void %<a href="#i_va_start">llvm.va_start</a>(sbyte** %ap)
2471 ; Read a single integer argument
2472 %tmp = va_arg sbyte** %ap, int
2474 ; Demonstrate usage of llvm.va_copy and llvm.va_end
2476 call void %<a href="#i_va_copy">llvm.va_copy</a>(sbyte** %aq, sbyte** %ap)
2477 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %aq)
2479 ; Stop processing of arguments.
2480 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %ap)
2486 <!-- _______________________________________________________________________ -->
2487 <div class="doc_subsubsection">
2488 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
2492 <div class="doc_text">
2494 <pre> declare void %llvm.va_start(<va_list>* <arglist>)<br></pre>
2496 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
2497 <tt>*<arglist></tt> for subsequent use by <tt><a
2498 href="#i_va_arg">va_arg</a></tt>.</p>
2502 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
2506 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
2507 macro available in C. In a target-dependent way, it initializes the
2508 <tt>va_list</tt> element the argument points to, so that the next call to
2509 <tt>va_arg</tt> will produce the first variable argument passed to the function.
2510 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
2511 last argument of the function, the compiler can figure that out.</p>
2515 <!-- _______________________________________________________________________ -->
2516 <div class="doc_subsubsection">
2517 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
2520 <div class="doc_text">
2522 <pre> declare void %llvm.va_end(<va_list*> <arglist>)<br></pre>
2524 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
2525 which has been initialized previously with <tt><a href="#i_va_start">llvm.va_start</a></tt>
2526 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
2528 <p>The argument is a <tt>va_list</tt> to destroy.</p>
2530 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
2531 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
2532 Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and <a
2533 href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
2534 with calls to <tt>llvm.va_end</tt>.</p>
2537 <!-- _______________________________________________________________________ -->
2538 <div class="doc_subsubsection">
2539 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
2542 <div class="doc_text">
2547 declare void %llvm.va_copy(<va_list>* <destarglist>,
2548 <va_list>* <srcarglist>)
2553 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
2554 the source argument list to the destination argument list.</p>
2558 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
2559 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
2564 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
2565 available in C. In a target-dependent way, it copies the source
2566 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
2567 because the <tt><a href="i_va_begin">llvm.va_begin</a></tt> intrinsic may be
2568 arbitrarily complex and require memory allocation, for example.</p>
2572 <!-- ======================================================================= -->
2573 <div class="doc_subsection">
2574 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
2577 <div class="doc_text">
2580 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
2581 Collection</a> requires the implementation and generation of these intrinsics.
2582 These intrinsics allow identification of <a href="#i_gcroot">GC roots on the
2583 stack</a>, as well as garbage collector implementations that require <a
2584 href="#i_gcread">read</a> and <a href="#i_gcwrite">write</a> barriers.
2585 Front-ends for type-safe garbage collected languages should generate these
2586 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
2587 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
2591 <!-- _______________________________________________________________________ -->
2592 <div class="doc_subsubsection">
2593 <a name="i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
2596 <div class="doc_text">
2601 declare void %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
2606 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
2607 the code generator, and allows some metadata to be associated with it.</p>
2611 <p>The first argument specifies the address of a stack object that contains the
2612 root pointer. The second pointer (which must be either a constant or a global
2613 value address) contains the meta-data to be associated with the root.</p>
2617 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
2618 location. At compile-time, the code generator generates information to allow
2619 the runtime to find the pointer at GC safe points.
2625 <!-- _______________________________________________________________________ -->
2626 <div class="doc_subsubsection">
2627 <a name="i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
2630 <div class="doc_text">
2635 declare sbyte* %llvm.gcread(sbyte** %Ptr)
2640 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
2641 locations, allowing garbage collector implementations that require read
2646 <p>The argument is the address to read from, which should be an address
2647 allocated from the garbage collector.</p>
2651 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
2652 instruction, but may be replaced with substantially more complex code by the
2653 garbage collector runtime, as needed.</p>
2658 <!-- _______________________________________________________________________ -->
2659 <div class="doc_subsubsection">
2660 <a name="i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
2663 <div class="doc_text">
2668 declare void %llvm.gcwrite(sbyte* %P1, sbyte** %P2)
2673 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
2674 locations, allowing garbage collector implementations that require write
2675 barriers (such as generational or reference counting collectors).</p>
2679 <p>The first argument is the reference to store, and the second is the heap
2680 location to store to.</p>
2684 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
2685 instruction, but may be replaced with substantially more complex code by the
2686 garbage collector runtime, as needed.</p>
2692 <!-- ======================================================================= -->
2693 <div class="doc_subsection">
2694 <a name="int_codegen">Code Generator Intrinsics</a>
2697 <div class="doc_text">
2699 These intrinsics are provided by LLVM to expose special features that may only
2700 be implemented with code generator support.
2705 <!-- _______________________________________________________________________ -->
2706 <div class="doc_subsubsection">
2707 <a name="i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
2710 <div class="doc_text">
2714 declare sbyte *%llvm.returnaddress(uint <level>)
2720 The '<tt>llvm.returnaddress</tt>' intrinsic returns a target-specific value
2721 indicating the return address of the current function or one of its callers.
2727 The argument to this intrinsic indicates which function to return the address
2728 for. Zero indicates the calling function, one indicates its caller, etc. The
2729 argument is <b>required</b> to be a constant integer value.
2735 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
2736 the return address of the specified call frame, or zero if it cannot be
2737 identified. The value returned by this intrinsic is likely to be incorrect or 0
2738 for arguments other than zero, so it should only be used for debugging purposes.
2742 Note that calling this intrinsic does not prevent function inlining or other
2743 aggressive transformations, so the value returned may not be that of the obvious
2744 source-language caller.
2749 <!-- _______________________________________________________________________ -->
2750 <div class="doc_subsubsection">
2751 <a name="i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
2754 <div class="doc_text">
2758 declare sbyte *%llvm.frameaddress(uint <level>)
2764 The '<tt>llvm.frameaddress</tt>' intrinsic returns the target-specific frame
2765 pointer value for the specified stack frame.
2771 The argument to this intrinsic indicates which function to return the frame
2772 pointer for. Zero indicates the calling function, one indicates its caller,
2773 etc. The argument is <b>required</b> to be a constant integer value.
2779 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
2780 the frame address of the specified call frame, or zero if it cannot be
2781 identified. The value returned by this intrinsic is likely to be incorrect or 0
2782 for arguments other than zero, so it should only be used for debugging purposes.
2786 Note that calling this intrinsic does not prevent function inlining or other
2787 aggressive transformations, so the value returned may not be that of the obvious
2788 source-language caller.
2792 <!-- _______________________________________________________________________ -->
2793 <div class="doc_subsubsection">
2794 <a name="i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
2797 <div class="doc_text">
2801 declare sbyte *%llvm.stacksave()
2807 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
2808 the function stack, for use with <a href="#i_stackrestore">
2809 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
2810 features like scoped automatic variable sized arrays in C99.
2816 This intrinsic returns a opaque pointer value that can be passed to <a
2817 href="#i_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
2818 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
2819 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
2820 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
2821 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
2822 that were allocated after the <tt>llvm.stacksave</tt> was executed.
2827 <!-- _______________________________________________________________________ -->
2828 <div class="doc_subsubsection">
2829 <a name="i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
2832 <div class="doc_text">
2836 declare void %llvm.stackrestore(sbyte* %ptr)
2842 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
2843 the function stack to the state it was in when the corresponding <a
2844 href="#llvm.stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
2845 useful for implementing language features like scoped automatic variable sized
2852 See the description for <a href="#i_stacksave"><tt>llvm.stacksave</tt></a>.
2858 <!-- _______________________________________________________________________ -->
2859 <div class="doc_subsubsection">
2860 <a name="i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
2863 <div class="doc_text">
2867 declare void %llvm.prefetch(sbyte * <address>,
2868 uint <rw>, uint <locality>)
2875 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
2876 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
2878 effect on the behavior of the program but can change its performance
2885 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
2886 determining if the fetch should be for a read (0) or write (1), and
2887 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
2888 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
2889 <tt>locality</tt> arguments must be constant integers.
2895 This intrinsic does not modify the behavior of the program. In particular,
2896 prefetches cannot trap and do not produce a value. On targets that support this
2897 intrinsic, the prefetch can provide hints to the processor cache for better
2903 <!-- _______________________________________________________________________ -->
2904 <div class="doc_subsubsection">
2905 <a name="i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
2908 <div class="doc_text">
2912 declare void %llvm.pcmarker( uint <id> )
2919 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
2921 code to simulators and other tools. The method is target specific, but it is
2922 expected that the marker will use exported symbols to transmit the PC of the marker.
2923 The marker makes no guarantees that it will remain with any specific instruction
2924 after optimizations. It is possible that the presence of a marker will inhibit
2925 optimizations. The intended use is to be inserted after optmizations to allow
2926 correlations of simulation runs.
2932 <tt>id</tt> is a numerical id identifying the marker.
2938 This intrinsic does not modify the behavior of the program. Backends that do not
2939 support this intrinisic may ignore it.
2944 <!-- _______________________________________________________________________ -->
2945 <div class="doc_subsubsection">
2946 <a name="i_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
2949 <div class="doc_text">
2953 declare ulong %llvm.readcyclecounter( )
2960 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
2961 counter register (or similar low latency, high accuracy clocks) on those targets
2962 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
2963 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
2964 should only be used for small timings.
2970 When directly supported, reading the cycle counter should not modify any memory.
2971 Implementations are allowed to either return a application specific value or a
2972 system wide value. On backends without support, this is lowered to a constant 0.
2978 <!-- ======================================================================= -->
2979 <div class="doc_subsection">
2980 <a name="int_os">Operating System Intrinsics</a>
2983 <div class="doc_text">
2985 These intrinsics are provided by LLVM to support the implementation of
2986 operating system level code.
2991 <!-- _______________________________________________________________________ -->
2992 <div class="doc_subsubsection">
2993 <a name="i_readport">'<tt>llvm.readport</tt>' Intrinsic</a>
2996 <div class="doc_text">
3000 declare <integer type> %llvm.readport (<integer type> <address>)
3006 The '<tt>llvm.readport</tt>' intrinsic reads data from the specified hardware
3013 The argument to this intrinsic indicates the hardware I/O address from which
3014 to read the data. The address is in the hardware I/O address namespace (as
3015 opposed to being a memory location for memory mapped I/O).
3021 The '<tt>llvm.readport</tt>' intrinsic reads data from the hardware I/O port
3022 specified by <i>address</i> and returns the value. The address and return
3023 value must be integers, but the size is dependent upon the platform upon which
3024 the program is code generated. For example, on x86, the address must be an
3025 unsigned 16-bit value, and the return value must be 8, 16, or 32 bits.
3030 <!-- _______________________________________________________________________ -->
3031 <div class="doc_subsubsection">
3032 <a name="i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a>
3035 <div class="doc_text">
3039 call void (<integer type>, <integer type>)*
3040 %llvm.writeport (<integer type> <value>,
3041 <integer type> <address>)
3047 The '<tt>llvm.writeport</tt>' intrinsic writes data to the specified hardware
3054 The first argument is the value to write to the I/O port.
3058 The second argument indicates the hardware I/O address to which data should be
3059 written. The address is in the hardware I/O address namespace (as opposed to
3060 being a memory location for memory mapped I/O).
3066 The '<tt>llvm.writeport</tt>' intrinsic writes <i>value</i> to the I/O port
3067 specified by <i>address</i>. The address and value must be integers, but the
3068 size is dependent upon the platform upon which the program is code generated.
3069 For example, on x86, the address must be an unsigned 16-bit value, and the
3070 value written must be 8, 16, or 32 bits in length.
3075 <!-- _______________________________________________________________________ -->
3076 <div class="doc_subsubsection">
3077 <a name="i_readio">'<tt>llvm.readio</tt>' Intrinsic</a>
3080 <div class="doc_text">
3084 declare <result> %llvm.readio (<ty> * <pointer>)
3090 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
3097 The argument to this intrinsic is a pointer indicating the memory address from
3098 which to read the data. The data must be a
3099 <a href="#t_firstclass">first class</a> type.
3105 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
3106 location specified by <i>pointer</i> and returns the value. The argument must
3107 be a pointer, and the return value must be a
3108 <a href="#t_firstclass">first class</a> type. However, certain architectures
3109 may not support I/O on all first class types. For example, 32-bit processors
3110 may only support I/O on data types that are 32 bits or less.
3114 This intrinsic enforces an in-order memory model for llvm.readio and
3115 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
3116 scheduled processors may execute loads and stores out of order, re-ordering at
3117 run time accesses to memory mapped I/O registers. Using these intrinsics
3118 ensures that accesses to memory mapped I/O registers occur in program order.
3123 <!-- _______________________________________________________________________ -->
3124 <div class="doc_subsubsection">
3125 <a name="i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a>
3128 <div class="doc_text">
3132 declare void %llvm.writeio (<ty1> <value>, <ty2> * <pointer>)
3138 The '<tt>llvm.writeio</tt>' intrinsic writes data to the specified memory
3145 The first argument is the value to write to the memory mapped I/O location.
3146 The second argument is a pointer indicating the memory address to which the
3147 data should be written.
3153 The '<tt>llvm.writeio</tt>' intrinsic writes <i>value</i> to the memory mapped
3154 I/O address specified by <i>pointer</i>. The value must be a
3155 <a href="#t_firstclass">first class</a> type. However, certain architectures
3156 may not support I/O on all first class types. For example, 32-bit processors
3157 may only support I/O on data types that are 32 bits or less.
3161 This intrinsic enforces an in-order memory model for llvm.readio and
3162 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
3163 scheduled processors may execute loads and stores out of order, re-ordering at
3164 run time accesses to memory mapped I/O registers. Using these intrinsics
3165 ensures that accesses to memory mapped I/O registers occur in program order.
3170 <!-- ======================================================================= -->
3171 <div class="doc_subsection">
3172 <a name="int_libc">Standard C Library Intrinsics</a>
3175 <div class="doc_text">
3177 LLVM provides intrinsics for a few important standard C library functions.
3178 These intrinsics allow source-language front-ends to pass information about the
3179 alignment of the pointer arguments to the code generator, providing opportunity
3180 for more efficient code generation.
3185 <!-- _______________________________________________________________________ -->
3186 <div class="doc_subsubsection">
3187 <a name="i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
3190 <div class="doc_text">
3194 declare void %llvm.memcpy(sbyte* <dest>, sbyte* <src>,
3195 uint <len>, uint <align>)
3201 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
3202 location to the destination location.
3206 Note that, unlike the standard libc function, the <tt>llvm.memcpy</tt> intrinsic
3207 does not return a value, and takes an extra alignment argument.
3213 The first argument is a pointer to the destination, the second is a pointer to
3214 the source. The third argument is an (arbitrarily sized) integer argument
3215 specifying the number of bytes to copy, and the fourth argument is the alignment
3216 of the source and destination locations.
3220 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3221 the caller guarantees that the size of the copy is a multiple of the alignment
3222 and that both the source and destination pointers are aligned to that boundary.
3228 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
3229 location to the destination location, which are not allowed to overlap. It
3230 copies "len" bytes of memory over. If the argument is known to be aligned to
3231 some boundary, this can be specified as the fourth argument, otherwise it should
3237 <!-- _______________________________________________________________________ -->
3238 <div class="doc_subsubsection">
3239 <a name="i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
3242 <div class="doc_text">
3246 declare void %llvm.memmove(sbyte* <dest>, sbyte* <src>,
3247 uint <len>, uint <align>)
3253 The '<tt>llvm.memmove</tt>' intrinsic moves a block of memory from the source
3254 location to the destination location. It is similar to the '<tt>llvm.memcpy</tt>'
3255 intrinsic but allows the two memory locations to overlap.
3259 Note that, unlike the standard libc function, the <tt>llvm.memmove</tt> intrinsic
3260 does not return a value, and takes an extra alignment argument.
3266 The first argument is a pointer to the destination, the second is a pointer to
3267 the source. The third argument is an (arbitrarily sized) integer argument
3268 specifying the number of bytes to copy, and the fourth argument is the alignment
3269 of the source and destination locations.
3273 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3274 the caller guarantees that the size of the copy is a multiple of the alignment
3275 and that both the source and destination pointers are aligned to that boundary.
3281 The '<tt>llvm.memmove</tt>' intrinsic copies a block of memory from the source
3282 location to the destination location, which may overlap. It
3283 copies "len" bytes of memory over. If the argument is known to be aligned to
3284 some boundary, this can be specified as the fourth argument, otherwise it should
3290 <!-- _______________________________________________________________________ -->
3291 <div class="doc_subsubsection">
3292 <a name="i_memset">'<tt>llvm.memset</tt>' Intrinsic</a>
3295 <div class="doc_text">
3299 declare void %llvm.memset(sbyte* <dest>, ubyte <val>,
3300 uint <len>, uint <align>)
3306 The '<tt>llvm.memset</tt>' intrinsic fills a block of memory with a particular
3311 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
3312 does not return a value, and takes an extra alignment argument.
3318 The first argument is a pointer to the destination to fill, the second is the
3319 byte value to fill it with, the third argument is an (arbitrarily sized) integer
3320 argument specifying the number of bytes to fill, and the fourth argument is the
3321 known alignment of destination location.
3325 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3326 the caller guarantees that the size of the copy is a multiple of the alignment
3327 and that the destination pointer is aligned to that boundary.
3333 The '<tt>llvm.memset</tt>' intrinsic fills "len" bytes of memory starting at the
3334 destination location. If the argument is known to be aligned to some boundary,
3335 this can be specified as the fourth argument, otherwise it should be set to 0 or
3341 <!-- _______________________________________________________________________ -->
3342 <div class="doc_subsubsection">
3343 <a name="i_isunordered">'<tt>llvm.isunordered</tt>' Intrinsic</a>
3346 <div class="doc_text">
3350 declare bool %llvm.isunordered(<float or double> Val1, <float or double> Val2)
3356 The '<tt>llvm.isunordered</tt>' intrinsic returns true if either or both of the
3357 specified floating point values is a NAN.
3363 The arguments are floating point numbers of the same type.
3369 If either or both of the arguments is a SNAN or QNAN, it returns true, otherwise
3375 <!-- _______________________________________________________________________ -->
3376 <div class="doc_subsubsection">
3377 <a name="i_sqrt">'<tt>llvm.sqrt</tt>' Intrinsic</a>
3380 <div class="doc_text">
3384 declare <float or double> %llvm.sqrt(<float or double> Val)
3390 The '<tt>llvm.sqrt</tt>' intrinsic returns the sqrt of the specified operand,
3391 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
3392 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
3393 negative numbers (which allows for better optimization).
3399 The argument and return value are floating point numbers of the same type.
3405 This function returns the sqrt of the specified operand if it is a positive
3406 floating point number.
3410 <!-- ======================================================================= -->
3411 <div class="doc_subsection">
3412 <a name="int_count">Bit Counting Intrinsics</a>
3415 <div class="doc_text">
3417 LLVM provides intrinsics for a few important bit counting operations.
3418 These allow efficient code generation for some algorithms.
3423 <!-- _______________________________________________________________________ -->
3424 <div class="doc_subsubsection">
3425 <a name="int_ctpop">'<tt>llvm.ctpop</tt>' Intrinsic</a>
3428 <div class="doc_text">
3432 declare int %llvm.ctpop(int <src>)
3438 The '<tt>llvm.ctpop</tt>' intrinsic counts the number of ones in a variable.
3444 The only argument is the value to be counted. The argument may be of any
3445 integer type. The return type must match the argument type.
3451 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
3455 <!-- _______________________________________________________________________ -->
3456 <div class="doc_subsubsection">
3457 <a name="int_ctlz">'<tt>llvm.ctlz</tt>' Intrinsic</a>
3460 <div class="doc_text">
3464 declare int %llvm.ctlz(int <src>)
3471 The '<tt>llvm.ctlz</tt>' intrinsic counts the number of leading zeros in a
3478 The only argument is the value to be counted. The argument may be of any
3479 integer type. The return type must match the argument type.
3485 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
3486 in a variable. If the src == 0 then the result is the size in bits of the type
3487 of src. For example, <tt>llvm.cttz(int 2) = 30</tt>.
3493 <!-- _______________________________________________________________________ -->
3494 <div class="doc_subsubsection">
3495 <a name="int_cttz">'<tt>llvm.cttz</tt>' Intrinsic</a>
3498 <div class="doc_text">
3502 declare int %llvm.cttz(int <src>)
3508 The '<tt>llvm.cttz</tt>' intrinsic counts the number of trailing zeros.
3514 The only argument is the value to be counted. The argument may be of any
3515 integer type. The return type must match the argument type.
3521 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
3522 in a variable. If the src == 0 then the result is the size in bits of the type
3523 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
3527 <!-- ======================================================================= -->
3528 <div class="doc_subsection">
3529 <a name="int_debugger">Debugger Intrinsics</a>
3532 <div class="doc_text">
3534 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
3535 are described in the <a
3536 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
3537 Debugging</a> document.
3542 <!-- *********************************************************************** -->
3545 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3546 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
3547 <a href="http://validator.w3.org/check/referer"><img
3548 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
3550 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3551 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br>
3552 Last modified: $Date$