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_call">'<tt>call</tt>' Instruction</a></li>
104 <li><a href="#i_vaarg">'<tt>vaarg</tt>' Instruction</a></li>
109 <li><a href="#intrinsics">Intrinsic Functions</a>
111 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
113 <li><a href="#i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
114 <li><a href="#i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
115 <li><a href="#i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
118 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
120 <li><a href="#i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
121 <li><a href="#i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
122 <li><a href="#i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
125 <li><a href="#int_codegen">Code Generator Intrinsics</a>
127 <li><a href="#i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
128 <li><a href="#i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
129 <li><a href="#i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
130 <li><a href="#i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
133 <li><a href="#int_os">Operating System Intrinsics</a>
135 <li><a href="#i_readport">'<tt>llvm.readport</tt>' Intrinsic</a></li>
136 <li><a href="#i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a></li>
137 <li><a href="#i_readio">'<tt>llvm.readio</tt>' Intrinsic</a></li>
138 <li><a href="#i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a></li>
140 <li><a href="#int_libc">Standard C Library Intrinsics</a>
142 <li><a href="#i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a></li>
143 <li><a href="#i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a></li>
144 <li><a href="#i_memset">'<tt>llvm.memset</tt>' Intrinsic</a></li>
145 <li><a href="#i_isunordered">'<tt>llvm.isunordered</tt>' Intrinsic</a></li>
146 <li><a href="#i_sqrt">'<tt>llvm.sqrt</tt>' Intrinsic</a></li>
150 <li><a href="#int_count">Bit counting Intrinsics</a>
152 <li><a href="#int_ctpop">'<tt>llvm.ctpop</tt>' Intrinsic </a></li>
153 <li><a href="#int_ctlz">'<tt>llvm.ctlz</tt>' Intrinsic </a></li>
154 <li><a href="#int_cttz">'<tt>llvm.cttz</tt>' Intrinsic </a></li>
157 <li><a href="#int_debugger">Debugger intrinsics</a></li>
162 <div class="doc_author">
163 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
164 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
167 <!-- *********************************************************************** -->
168 <div class="doc_section"> <a name="abstract">Abstract </a></div>
169 <!-- *********************************************************************** -->
171 <div class="doc_text">
172 <p>This document is a reference manual for the LLVM assembly language.
173 LLVM is an SSA based representation that provides type safety,
174 low-level operations, flexibility, and the capability of representing
175 'all' high-level languages cleanly. It is the common code
176 representation used throughout all phases of the LLVM compilation
180 <!-- *********************************************************************** -->
181 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
182 <!-- *********************************************************************** -->
184 <div class="doc_text">
186 <p>The LLVM code representation is designed to be used in three
187 different forms: as an in-memory compiler IR, as an on-disk bytecode
188 representation (suitable for fast loading by a Just-In-Time compiler),
189 and as a human readable assembly language representation. This allows
190 LLVM to provide a powerful intermediate representation for efficient
191 compiler transformations and analysis, while providing a natural means
192 to debug and visualize the transformations. The three different forms
193 of LLVM are all equivalent. This document describes the human readable
194 representation and notation.</p>
196 <p>The LLVM representation aims to be light-weight and low-level
197 while being expressive, typed, and extensible at the same time. It
198 aims to be a "universal IR" of sorts, by being at a low enough level
199 that high-level ideas may be cleanly mapped to it (similar to how
200 microprocessors are "universal IR's", allowing many source languages to
201 be mapped to them). By providing type information, LLVM can be used as
202 the target of optimizations: for example, through pointer analysis, it
203 can be proven that a C automatic variable is never accessed outside of
204 the current function... allowing it to be promoted to a simple SSA
205 value instead of a memory location.</p>
209 <!-- _______________________________________________________________________ -->
210 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
212 <div class="doc_text">
214 <p>It is important to note that this document describes 'well formed'
215 LLVM assembly language. There is a difference between what the parser
216 accepts and what is considered 'well formed'. For example, the
217 following instruction is syntactically okay, but not well formed:</p>
220 %x = <a href="#i_add">add</a> int 1, %x
223 <p>...because the definition of <tt>%x</tt> does not dominate all of
224 its uses. The LLVM infrastructure provides a verification pass that may
225 be used to verify that an LLVM module is well formed. This pass is
226 automatically run by the parser after parsing input assembly and by
227 the optimizer before it outputs bytecode. The violations pointed out
228 by the verifier pass indicate bugs in transformation passes or input to
231 <!-- Describe the typesetting conventions here. --> </div>
233 <!-- *********************************************************************** -->
234 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
235 <!-- *********************************************************************** -->
237 <div class="doc_text">
239 <p>LLVM uses three different forms of identifiers, for different
243 <li>Named values are represented as a string of characters with a '%' prefix.
244 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
245 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
246 Identifiers which require other characters in their names can be surrounded
247 with quotes. In this way, anything except a <tt>"</tt> character can be used
250 <li>Unnamed values are represented as an unsigned numeric value with a '%'
251 prefix. For example, %12, %2, %44.</li>
253 <li>Constants, which are described in a <a href="#constants">section about
254 constants</a>, below.</li>
257 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
258 don't need to worry about name clashes with reserved words, and the set of
259 reserved words may be expanded in the future without penalty. Additionally,
260 unnamed identifiers allow a compiler to quickly come up with a temporary
261 variable without having to avoid symbol table conflicts.</p>
263 <p>Reserved words in LLVM are very similar to reserved words in other
264 languages. There are keywords for different opcodes ('<tt><a
265 href="#i_add">add</a></tt>', '<tt><a href="#i_cast">cast</a></tt>', '<tt><a
266 href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
267 href="#t_void">void</a></tt>', '<tt><a href="#t_uint">uint</a></tt>', etc...),
268 and others. These reserved words cannot conflict with variable names, because
269 none of them start with a '%' character.</p>
271 <p>Here is an example of LLVM code to multiply the integer variable
272 '<tt>%X</tt>' by 8:</p>
277 %result = <a href="#i_mul">mul</a> uint %X, 8
280 <p>After strength reduction:</p>
283 %result = <a href="#i_shl">shl</a> uint %X, ubyte 3
286 <p>And the hard way:</p>
289 <a href="#i_add">add</a> uint %X, %X <i>; yields {uint}:%0</i>
290 <a href="#i_add">add</a> uint %0, %0 <i>; yields {uint}:%1</i>
291 %result = <a href="#i_add">add</a> uint %1, %1
294 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
295 important lexical features of LLVM:</p>
299 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
302 <li>Unnamed temporaries are created when the result of a computation is not
303 assigned to a named value.</li>
305 <li>Unnamed temporaries are numbered sequentially</li>
309 <p>...and it also shows a convention that we follow in this document. When
310 demonstrating instructions, we will follow an instruction with a comment that
311 defines the type and name of value produced. Comments are shown in italic
316 <!-- *********************************************************************** -->
317 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
318 <!-- *********************************************************************** -->
320 <!-- ======================================================================= -->
321 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
324 <div class="doc_text">
326 <p>LLVM programs are composed of "Module"s, each of which is a
327 translation unit of the input programs. Each module consists of
328 functions, global variables, and symbol table entries. Modules may be
329 combined together with the LLVM linker, which merges function (and
330 global variable) definitions, resolves forward declarations, and merges
331 symbol table entries. Here is an example of the "hello world" module:</p>
333 <pre><i>; Declare the string constant as a global constant...</i>
334 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
335 href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
337 <i>; External declaration of the puts function</i>
338 <a href="#functionstructure">declare</a> int %puts(sbyte*) <i>; int(sbyte*)* </i>
340 <i>; Definition of main function</i>
341 int %main() { <i>; int()* </i>
342 <i>; Convert [13x sbyte]* to sbyte *...</i>
344 href="#i_getelementptr">getelementptr</a> [13 x sbyte]* %.LC0, long 0, long 0 <i>; sbyte*</i>
346 <i>; Call puts function to write out the string to stdout...</i>
348 href="#i_call">call</a> int %puts(sbyte* %cast210) <i>; int</i>
350 href="#i_ret">ret</a> int 0<br>}<br></pre>
352 <p>This example is made up of a <a href="#globalvars">global variable</a>
353 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
354 function, and a <a href="#functionstructure">function definition</a>
355 for "<tt>main</tt>".</p>
357 <p>In general, a module is made up of a list of global values,
358 where both functions and global variables are global values. Global values are
359 represented by a pointer to a memory location (in this case, a pointer to an
360 array of char, and a pointer to a function), and have one of the following <a
361 href="#linkage">linkage types</a>.</p>
365 <!-- ======================================================================= -->
366 <div class="doc_subsection">
367 <a name="linkage">Linkage Types</a>
370 <div class="doc_text">
373 All Global Variables and Functions have one of the following types of linkage:
378 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
380 <dd>Global values with internal linkage are only directly accessible by
381 objects in the current module. In particular, linking code into a module with
382 an internal global value may cause the internal to be renamed as necessary to
383 avoid collisions. Because the symbol is internal to the module, all
384 references can be updated. This corresponds to the notion of the
385 '<tt>static</tt>' keyword in C, or the idea of "anonymous namespaces" in C++.
388 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
390 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt> linkage, with
391 the twist that linking together two modules defining the same
392 <tt>linkonce</tt> globals will cause one of the globals to be discarded. This
393 is typically used to implement inline functions. Unreferenced
394 <tt>linkonce</tt> globals are allowed to be discarded.
397 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
399 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
400 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
401 used to implement constructs in C such as "<tt>int X;</tt>" at global scope.
404 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
406 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
407 pointer to array type. When two global variables with appending linkage are
408 linked together, the two global arrays are appended together. This is the
409 LLVM, typesafe, equivalent of having the system linker append together
410 "sections" with identical names when .o files are linked.
413 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
415 <dd>If none of the above identifiers are used, the global is externally
416 visible, meaning that it participates in linkage and can be used to resolve
417 external symbol references.
421 <p><a name="linkage_external">For example, since the "<tt>.LC0</tt>"
422 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
423 variable and was linked with this one, one of the two would be renamed,
424 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
425 external (i.e., lacking any linkage declarations), they are accessible
426 outside of the current module. It is illegal for a function <i>declaration</i>
427 to have any linkage type other than "externally visible".</a></p>
431 <!-- ======================================================================= -->
432 <div class="doc_subsection">
433 <a name="callingconv">Calling Conventions</a>
436 <div class="doc_text">
438 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
439 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
440 specified for the call. The calling convention of any pair of dynamic
441 caller/callee must match, or the behavior of the program is undefined. The
442 following calling conventions are supported by LLVM, and more may be added in
446 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
448 <dd>This calling convention (the default if no other calling convention is
449 specified) matches the target C calling conventions. This calling convention
450 supports varargs function calls and tolerates some mismatch in the declared
451 prototype and implemented declaration of the function (as does normal C).
454 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
456 <dd>This calling convention attempts to make calls as fast as possible
457 (e.g. by passing things in registers). This calling convention allows the
458 target to use whatever tricks it wants to produce fast code for the target,
459 without having to conform to an externally specified ABI. Implementations of
460 this convention should allow arbitrary tail call optimization to be supported.
461 This calling convention does not support varargs and requires the prototype of
462 all callees to exactly match the prototype of the function definition.
465 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
467 <dd>This calling convention attempts to make code in the caller as efficient
468 as possible under the assumption that the call is not commonly executed. As
469 such, these calls often preserve all registers so that the call does not break
470 any live ranges in the caller side. This calling convention does not support
471 varargs and requires the prototype of all callees to exactly match the
472 prototype of the function definition.
475 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
477 <dd>Any calling convention may be specified by number, allowing
478 target-specific calling conventions to be used. Target specific calling
479 conventions start at 64.
483 <p>More calling conventions can be added/defined on an as-needed basis, to
484 support pascal conventions or any other well-known target-independent
489 <!-- ======================================================================= -->
490 <div class="doc_subsection">
491 <a name="globalvars">Global Variables</a>
494 <div class="doc_text">
496 <p>Global variables define regions of memory allocated at compilation time
497 instead of run-time. Global variables may optionally be initialized, and may
498 have an optional explicit alignment specified. A
499 variable may be defined as a global "constant," which indicates that the
500 contents of the variable will <b>never</b> be modified (enabling better
501 optimization, allowing the global data to be placed in the read-only section of
502 an executable, etc). Note that variables that need runtime initialization
503 cannot be marked "constant" as there is a store to the variable.</p>
506 LLVM explicitly allows <em>declarations</em> of global variables to be marked
507 constant, even if the final definition of the global is not. This capability
508 can be used to enable slightly better optimization of the program, but requires
509 the language definition to guarantee that optimizations based on the
510 'constantness' are valid for the translation units that do not include the
514 <p>As SSA values, global variables define pointer values that are in
515 scope (i.e. they dominate) all basic blocks in the program. Global
516 variables always define a pointer to their "content" type because they
517 describe a region of memory, and all memory objects in LLVM are
518 accessed through pointers.</p>
520 <p>An explicit alignment may be specified for a global. If not present, or if
521 the alignment is set to zero, the alignment of the global is set by the target
522 to whatever it feels convenient. If an explicit alignment is specified, the
523 global is forced to have at least that much alignment. All alignments must be
529 <!-- ======================================================================= -->
530 <div class="doc_subsection">
531 <a name="functionstructure">Functions</a>
534 <div class="doc_text">
536 <p>LLVM function definitions consist of an optional <a href="#linkage">linkage
537 type</a>, an optional <a href="#callingconv">calling convention</a>, a return
538 type, a function name, a (possibly empty) argument list, an optional alignment,
539 an opening curly brace,
540 a list of basic blocks, and a closing curly brace. LLVM function declarations
541 are defined with the "<tt>declare</tt>" keyword, an optional <a
542 href="#callingconv">calling convention</a>, a return type, a function name,
543 a possibly empty list of arguments, and an optional alignment.</p>
545 <p>A function definition contains a list of basic blocks, forming the CFG for
546 the function. Each basic block may optionally start with a label (giving the
547 basic block a symbol table entry), contains a list of instructions, and ends
548 with a <a href="#terminators">terminator</a> instruction (such as a branch or
549 function return).</p>
551 <p>The first basic block in a program is special in two ways: it is immediately
552 executed on entrance to the function, and it is not allowed to have predecessor
553 basic blocks (i.e. there can not be any branches to the entry block of a
554 function). Because the block can have no predecessors, it also cannot have any
555 <a href="#i_phi">PHI nodes</a>.</p>
557 <p>LLVM functions are identified by their name and type signature. Hence, two
558 functions with the same name but different parameter lists or return values are
559 considered different functions, and LLVM will resolve references to each
562 <p>An explicit alignment may be specified for a function. If not present, or if
563 the alignment is set to zero, the alignment of the function is set by the target
564 to whatever it feels convenient. If an explicit alignment is specified, the
565 function is forced to have at least that much alignment. All alignments must be
572 <!-- *********************************************************************** -->
573 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
574 <!-- *********************************************************************** -->
576 <div class="doc_text">
578 <p>The LLVM type system is one of the most important features of the
579 intermediate representation. Being typed enables a number of
580 optimizations to be performed on the IR directly, without having to do
581 extra analyses on the side before the transformation. A strong type
582 system makes it easier to read the generated code and enables novel
583 analyses and transformations that are not feasible to perform on normal
584 three address code representations.</p>
588 <!-- ======================================================================= -->
589 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
590 <div class="doc_text">
591 <p>The primitive types are the fundamental building blocks of the LLVM
592 system. The current set of primitive types is as follows:</p>
594 <table class="layout">
599 <tr><th>Type</th><th>Description</th></tr>
600 <tr><td><tt>void</tt></td><td>No value</td></tr>
601 <tr><td><tt>ubyte</tt></td><td>Unsigned 8-bit value</td></tr>
602 <tr><td><tt>ushort</tt></td><td>Unsigned 16-bit value</td></tr>
603 <tr><td><tt>uint</tt></td><td>Unsigned 32-bit value</td></tr>
604 <tr><td><tt>ulong</tt></td><td>Unsigned 64-bit value</td></tr>
605 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
606 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
613 <tr><th>Type</th><th>Description</th></tr>
614 <tr><td><tt>bool</tt></td><td>True or False value</td></tr>
615 <tr><td><tt>sbyte</tt></td><td>Signed 8-bit value</td></tr>
616 <tr><td><tt>short</tt></td><td>Signed 16-bit value</td></tr>
617 <tr><td><tt>int</tt></td><td>Signed 32-bit value</td></tr>
618 <tr><td><tt>long</tt></td><td>Signed 64-bit value</td></tr>
619 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
627 <!-- _______________________________________________________________________ -->
628 <div class="doc_subsubsection"> <a name="t_classifications">Type
629 Classifications</a> </div>
630 <div class="doc_text">
631 <p>These different primitive types fall into a few useful
634 <table border="1" cellspacing="0" cellpadding="4">
636 <tr><th>Classification</th><th>Types</th></tr>
638 <td><a name="t_signed">signed</a></td>
639 <td><tt>sbyte, short, int, long, float, double</tt></td>
642 <td><a name="t_unsigned">unsigned</a></td>
643 <td><tt>ubyte, ushort, uint, ulong</tt></td>
646 <td><a name="t_integer">integer</a></td>
647 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
650 <td><a name="t_integral">integral</a></td>
651 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt>
655 <td><a name="t_floating">floating point</a></td>
656 <td><tt>float, double</tt></td>
659 <td><a name="t_firstclass">first class</a></td>
660 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long,<br>
661 float, double, <a href="#t_pointer">pointer</a>,
662 <a href="#t_packed">packed</a></tt></td>
667 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
668 most important. Values of these types are the only ones which can be
669 produced by instructions, passed as arguments, or used as operands to
670 instructions. This means that all structures and arrays must be
671 manipulated either by pointer or by component.</p>
674 <!-- ======================================================================= -->
675 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
677 <div class="doc_text">
679 <p>The real power in LLVM comes from the derived types in the system.
680 This is what allows a programmer to represent arrays, functions,
681 pointers, and other useful types. Note that these derived types may be
682 recursive: For example, it is possible to have a two dimensional array.</p>
686 <!-- _______________________________________________________________________ -->
687 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
689 <div class="doc_text">
693 <p>The array type is a very simple derived type that arranges elements
694 sequentially in memory. The array type requires a size (number of
695 elements) and an underlying data type.</p>
700 [<# elements> x <elementtype>]
703 <p>The number of elements is a constant integer value; elementtype may
704 be any type with a size.</p>
707 <table class="layout">
710 <tt>[40 x int ]</tt><br/>
711 <tt>[41 x int ]</tt><br/>
712 <tt>[40 x uint]</tt><br/>
715 Array of 40 integer values.<br/>
716 Array of 41 integer values.<br/>
717 Array of 40 unsigned integer values.<br/>
721 <p>Here are some examples of multidimensional arrays:</p>
722 <table class="layout">
725 <tt>[3 x [4 x int]]</tt><br/>
726 <tt>[12 x [10 x float]]</tt><br/>
727 <tt>[2 x [3 x [4 x uint]]]</tt><br/>
730 3x4 array of integer values.<br/>
731 12x10 array of single precision floating point values.<br/>
732 2x3x4 array of unsigned integer values.<br/>
737 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
738 length array. Normally, accesses past the end of an array are undefined in
739 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
740 As a special case, however, zero length arrays are recognized to be variable
741 length. This allows implementation of 'pascal style arrays' with the LLVM
742 type "{ int, [0 x float]}", for example.</p>
746 <!-- _______________________________________________________________________ -->
747 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
748 <div class="doc_text">
750 <p>The function type can be thought of as a function signature. It
751 consists of a return type and a list of formal parameter types.
752 Function types are usually used to build virtual function tables
753 (which are structures of pointers to functions), for indirect function
754 calls, and when defining a function.</p>
756 The return type of a function type cannot be an aggregate type.
759 <pre> <returntype> (<parameter list>)<br></pre>
760 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
761 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
762 which indicates that the function takes a variable number of arguments.
763 Variable argument functions can access their arguments with the <a
764 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
766 <table class="layout">
769 <tt>int (int)</tt> <br/>
770 <tt>float (int, int *) *</tt><br/>
771 <tt>int (sbyte *, ...)</tt><br/>
774 function taking an <tt>int</tt>, returning an <tt>int</tt><br/>
775 <a href="#t_pointer">Pointer</a> to a function that takes an
776 <tt>int</tt> and a <a href="#t_pointer">pointer</a> to <tt>int</tt>,
777 returning <tt>float</tt>.<br/>
778 A vararg function that takes at least one <a href="#t_pointer">pointer</a>
779 to <tt>sbyte</tt> (signed char in C), which returns an integer. This is
780 the signature for <tt>printf</tt> in LLVM.<br/>
786 <!-- _______________________________________________________________________ -->
787 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
788 <div class="doc_text">
790 <p>The structure type is used to represent a collection of data members
791 together in memory. The packing of the field types is defined to match
792 the ABI of the underlying processor. The elements of a structure may
793 be any type that has a size.</p>
794 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
795 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
796 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
799 <pre> { <type list> }<br></pre>
801 <table class="layout">
804 <tt>{ int, int, int }</tt><br/>
805 <tt>{ float, int (int) * }</tt><br/>
808 a triple of three <tt>int</tt> values<br/>
809 A pair, where the first element is a <tt>float</tt> and the second element
810 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
811 that takes an <tt>int</tt>, returning an <tt>int</tt>.<br/>
817 <!-- _______________________________________________________________________ -->
818 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
819 <div class="doc_text">
821 <p>As in many languages, the pointer type represents a pointer or
822 reference to another object, which must live in memory.</p>
824 <pre> <type> *<br></pre>
826 <table class="layout">
829 <tt>[4x int]*</tt><br/>
830 <tt>int (int *) *</tt><br/>
833 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
834 four <tt>int</tt> values<br/>
835 A <a href="#t_pointer">pointer</a> to a <a
836 href="#t_function">function</a> that takes an <tt>int*</tt>, returning an
843 <!-- _______________________________________________________________________ -->
844 <div class="doc_subsubsection"> <a name="t_packed">Packed Type</a> </div>
845 <div class="doc_text">
849 <p>A packed type is a simple derived type that represents a vector
850 of elements. Packed types are used when multiple primitive data
851 are operated in parallel using a single instruction (SIMD).
852 A packed type requires a size (number of
853 elements) and an underlying primitive data type. Packed types are
854 considered <a href="#t_firstclass">first class</a>.</p>
859 < <# elements> x <elementtype> >
862 <p>The number of elements is a constant integer value; elementtype may
863 be any integral or floating point type.</p>
867 <table class="layout">
870 <tt><4 x int></tt><br/>
871 <tt><8 x float></tt><br/>
872 <tt><2 x uint></tt><br/>
875 Packed vector of 4 integer values.<br/>
876 Packed vector of 8 floating-point values.<br/>
877 Packed vector of 2 unsigned integer values.<br/>
883 <!-- _______________________________________________________________________ -->
884 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
885 <div class="doc_text">
889 <p>Opaque types are used to represent unknown types in the system. This
890 corresponds (for example) to the C notion of a foward declared structure type.
891 In LLVM, opaque types can eventually be resolved to any type (not just a
902 <table class="layout">
915 <!-- *********************************************************************** -->
916 <div class="doc_section"> <a name="constants">Constants</a> </div>
917 <!-- *********************************************************************** -->
919 <div class="doc_text">
921 <p>LLVM has several different basic types of constants. This section describes
922 them all and their syntax.</p>
926 <!-- ======================================================================= -->
927 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
929 <div class="doc_text">
932 <dt><b>Boolean constants</b></dt>
934 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
935 constants of the <tt><a href="#t_primitive">bool</a></tt> type.
938 <dt><b>Integer constants</b></dt>
940 <dd>Standard integers (such as '4') are constants of the <a
941 href="#t_integer">integer</a> type. Negative numbers may be used with signed
945 <dt><b>Floating point constants</b></dt>
947 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
948 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
949 notation (see below). Floating point constants must have a <a
950 href="#t_floating">floating point</a> type. </dd>
952 <dt><b>Null pointer constants</b></dt>
954 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
955 and must be of <a href="#t_pointer">pointer type</a>.</dd>
959 <p>The one non-intuitive notation for constants is the optional hexadecimal form
960 of floating point constants. For example, the form '<tt>double
961 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
962 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
963 (and the only time that they are generated by the disassembler) is when a
964 floating point constant must be emitted but it cannot be represented as a
965 decimal floating point number. For example, NaN's, infinities, and other
966 special values are represented in their IEEE hexadecimal format so that
967 assembly and disassembly do not cause any bits to change in the constants.</p>
971 <!-- ======================================================================= -->
972 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
975 <div class="doc_text">
976 <p>Aggregate constants arise from aggregation of simple constants
977 and smaller aggregate constants.</p>
980 <dt><b>Structure constants</b></dt>
982 <dd>Structure constants are represented with notation similar to structure
983 type definitions (a comma separated list of elements, surrounded by braces
984 (<tt>{}</tt>)). For example: "<tt>{ int 4, float 17.0, int* %G }</tt>",
985 where "<tt>%G</tt>" is declared as "<tt>%G = external global int</tt>". Structure constants
986 must have <a href="#t_struct">structure type</a>, and the number and
987 types of elements must match those specified by the type.
990 <dt><b>Array constants</b></dt>
992 <dd>Array constants are represented with notation similar to array type
993 definitions (a comma separated list of elements, surrounded by square brackets
994 (<tt>[]</tt>)). For example: "<tt>[ int 42, int 11, int 74 ]</tt>". Array
995 constants must have <a href="#t_array">array type</a>, and the number and
996 types of elements must match those specified by the type.
999 <dt><b>Packed constants</b></dt>
1001 <dd>Packed constants are represented with notation similar to packed type
1002 definitions (a comma separated list of elements, surrounded by
1003 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< int 42,
1004 int 11, int 74, int 100 ></tt>". Packed constants must have <a
1005 href="#t_packed">packed type</a>, and the number and types of elements must
1006 match those specified by the type.
1009 <dt><b>Zero initialization</b></dt>
1011 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1012 value to zero of <em>any</em> type, including scalar and aggregate types.
1013 This is often used to avoid having to print large zero initializers (e.g. for
1014 large arrays) and is always exactly equivalent to using explicit zero
1021 <!-- ======================================================================= -->
1022 <div class="doc_subsection">
1023 <a name="globalconstants">Global Variable and Function Addresses</a>
1026 <div class="doc_text">
1028 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1029 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1030 constants. These constants are explicitly referenced when the <a
1031 href="#identifiers">identifier for the global</a> is used and always have <a
1032 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1038 %Z = global [2 x int*] [ int* %X, int* %Y ]
1043 <!-- ======================================================================= -->
1044 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1045 <div class="doc_text">
1046 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1047 no specific value. Undefined values may be of any type and be used anywhere
1048 a constant is permitted.</p>
1050 <p>Undefined values indicate to the compiler that the program is well defined
1051 no matter what value is used, giving the compiler more freedom to optimize.
1055 <!-- ======================================================================= -->
1056 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1059 <div class="doc_text">
1061 <p>Constant expressions are used to allow expressions involving other constants
1062 to be used as constants. Constant expressions may be of any <a
1063 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1064 that does not have side effects (e.g. load and call are not supported). The
1065 following is the syntax for constant expressions:</p>
1068 <dt><b><tt>cast ( CST to TYPE )</tt></b></dt>
1070 <dd>Cast a constant to another type.</dd>
1072 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1074 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1075 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1076 instruction, the index list may have zero or more indexes, which are required
1077 to make sense for the type of "CSTPTR".</dd>
1079 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1081 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1082 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1083 binary</a> operations. The constraints on operands are the same as those for
1084 the corresponding instruction (e.g. no bitwise operations on floating point
1085 values are allowed).</dd>
1089 <!-- *********************************************************************** -->
1090 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1091 <!-- *********************************************************************** -->
1093 <div class="doc_text">
1095 <p>The LLVM instruction set consists of several different
1096 classifications of instructions: <a href="#terminators">terminator
1097 instructions</a>, <a href="#binaryops">binary instructions</a>,
1098 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1099 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1100 instructions</a>.</p>
1104 <!-- ======================================================================= -->
1105 <div class="doc_subsection"> <a name="terminators">Terminator
1106 Instructions</a> </div>
1108 <div class="doc_text">
1110 <p>As mentioned <a href="#functionstructure">previously</a>, every
1111 basic block in a program ends with a "Terminator" instruction, which
1112 indicates which block should be executed after the current block is
1113 finished. These terminator instructions typically yield a '<tt>void</tt>'
1114 value: they produce control flow, not values (the one exception being
1115 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1116 <p>There are six different terminator instructions: the '<a
1117 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1118 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1119 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1120 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1121 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1125 <!-- _______________________________________________________________________ -->
1126 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1127 Instruction</a> </div>
1128 <div class="doc_text">
1130 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1131 ret void <i>; Return from void function</i>
1134 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1135 value) from a function back to the caller.</p>
1136 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1137 returns a value and then causes control flow, and one that just causes
1138 control flow to occur.</p>
1140 <p>The '<tt>ret</tt>' instruction may return any '<a
1141 href="#t_firstclass">first class</a>' type. Notice that a function is
1142 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1143 instruction inside of the function that returns a value that does not
1144 match the return type of the function.</p>
1146 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1147 returns back to the calling function's context. If the caller is a "<a
1148 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1149 the instruction after the call. If the caller was an "<a
1150 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1151 at the beginning of the "normal" destination block. If the instruction
1152 returns a value, that value shall set the call or invoke instruction's
1155 <pre> ret int 5 <i>; Return an integer value of 5</i>
1156 ret void <i>; Return from a void function</i>
1159 <!-- _______________________________________________________________________ -->
1160 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1161 <div class="doc_text">
1163 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1166 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1167 transfer to a different basic block in the current function. There are
1168 two forms of this instruction, corresponding to a conditional branch
1169 and an unconditional branch.</p>
1171 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1172 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
1173 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
1174 value as a target.</p>
1176 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
1177 argument is evaluated. If the value is <tt>true</tt>, control flows
1178 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1179 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1181 <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
1182 href="#i_ret">ret</a> int 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> int 0<br></pre>
1184 <!-- _______________________________________________________________________ -->
1185 <div class="doc_subsubsection">
1186 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1189 <div class="doc_text">
1193 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1198 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1199 several different places. It is a generalization of the '<tt>br</tt>'
1200 instruction, allowing a branch to occur to one of many possible
1206 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1207 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1208 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1209 table is not allowed to contain duplicate constant entries.</p>
1213 <p>The <tt>switch</tt> instruction specifies a table of values and
1214 destinations. When the '<tt>switch</tt>' instruction is executed, this
1215 table is searched for the given value. If the value is found, control flow is
1216 transfered to the corresponding destination; otherwise, control flow is
1217 transfered to the default destination.</p>
1219 <h5>Implementation:</h5>
1221 <p>Depending on properties of the target machine and the particular
1222 <tt>switch</tt> instruction, this instruction may be code generated in different
1223 ways. For example, it could be generated as a series of chained conditional
1224 branches or with a lookup table.</p>
1229 <i>; Emulate a conditional br instruction</i>
1230 %Val = <a href="#i_cast">cast</a> bool %value to int
1231 switch int %Val, label %truedest [int 0, label %falsedest ]
1233 <i>; Emulate an unconditional br instruction</i>
1234 switch uint 0, label %dest [ ]
1236 <i>; Implement a jump table:</i>
1237 switch uint %val, label %otherwise [ uint 0, label %onzero
1238 uint 1, label %onone
1239 uint 2, label %ontwo ]
1243 <!-- _______________________________________________________________________ -->
1244 <div class="doc_subsubsection">
1245 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1248 <div class="doc_text">
1253 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1254 to label <normal label> except label <exception label>
1259 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1260 function, with the possibility of control flow transfer to either the
1261 '<tt>normal</tt>' label or the
1262 '<tt>exception</tt>' label. If the callee function returns with the
1263 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1264 "normal" label. If the callee (or any indirect callees) returns with the "<a
1265 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1266 continued at the dynamically nearest "exception" label.</p>
1270 <p>This instruction requires several arguments:</p>
1274 The optional "cconv" marker indicates which <a href="callingconv">calling
1275 convention</a> the call should use. If none is specified, the call defaults
1276 to using C calling conventions.
1278 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1279 function value being invoked. In most cases, this is a direct function
1280 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1281 an arbitrary pointer to function value.
1284 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1285 function to be invoked. </li>
1287 <li>'<tt>function args</tt>': argument list whose types match the function
1288 signature argument types. If the function signature indicates the function
1289 accepts a variable number of arguments, the extra arguments can be
1292 <li>'<tt>normal label</tt>': the label reached when the called function
1293 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1295 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1296 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1302 <p>This instruction is designed to operate as a standard '<tt><a
1303 href="#i_call">call</a></tt>' instruction in most regards. The primary
1304 difference is that it establishes an association with a label, which is used by
1305 the runtime library to unwind the stack.</p>
1307 <p>This instruction is used in languages with destructors to ensure that proper
1308 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1309 exception. Additionally, this is important for implementation of
1310 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1314 %retval = invoke int %Test(int 15) to label %Continue
1315 except label %TestCleanup <i>; {int}:retval set</i>
1316 %retval = invoke <a href="#callingconv">coldcc</a> int %Test(int 15) to label %Continue
1317 except label %TestCleanup <i>; {int}:retval set</i>
1322 <!-- _______________________________________________________________________ -->
1324 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1325 Instruction</a> </div>
1327 <div class="doc_text">
1336 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1337 at the first callee in the dynamic call stack which used an <a
1338 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1339 primarily used to implement exception handling.</p>
1343 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1344 immediately halt. The dynamic call stack is then searched for the first <a
1345 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1346 execution continues at the "exceptional" destination block specified by the
1347 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1348 dynamic call chain, undefined behavior results.</p>
1351 <!-- _______________________________________________________________________ -->
1353 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1354 Instruction</a> </div>
1356 <div class="doc_text">
1365 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1366 instruction is used to inform the optimizer that a particular portion of the
1367 code is not reachable. This can be used to indicate that the code after a
1368 no-return function cannot be reached, and other facts.</p>
1372 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1377 <!-- ======================================================================= -->
1378 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1379 <div class="doc_text">
1380 <p>Binary operators are used to do most of the computation in a
1381 program. They require two operands, execute an operation on them, and
1382 produce a single value. The operands might represent
1383 multiple data, as is the case with the <a href="#t_packed">packed</a> data type.
1384 The result value of a binary operator is not
1385 necessarily the same type as its operands.</p>
1386 <p>There are several different binary operators:</p>
1388 <!-- _______________________________________________________________________ -->
1389 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1390 Instruction</a> </div>
1391 <div class="doc_text">
1393 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1396 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1398 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1399 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1400 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1401 Both arguments must have identical types.</p>
1403 <p>The value produced is the integer or floating point sum of the two
1406 <pre> <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
1409 <!-- _______________________________________________________________________ -->
1410 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1411 Instruction</a> </div>
1412 <div class="doc_text">
1414 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1417 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1419 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1420 instruction present in most other intermediate representations.</p>
1422 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1423 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1425 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1426 Both arguments must have identical types.</p>
1428 <p>The value produced is the integer or floating point difference of
1429 the two operands.</p>
1431 <pre> <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
1432 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
1435 <!-- _______________________________________________________________________ -->
1436 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1437 Instruction</a> </div>
1438 <div class="doc_text">
1440 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1443 <p>The '<tt>mul</tt>' instruction returns the product of its two
1446 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1447 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1449 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1450 Both arguments must have identical types.</p>
1452 <p>The value produced is the integer or floating point product of the
1454 <p>There is no signed vs unsigned multiplication. The appropriate
1455 action is taken based on the type of the operand.</p>
1457 <pre> <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1460 <!-- _______________________________________________________________________ -->
1461 <div class="doc_subsubsection"> <a name="i_div">'<tt>div</tt>'
1462 Instruction</a> </div>
1463 <div class="doc_text">
1465 <pre> <result> = div <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1468 <p>The '<tt>div</tt>' instruction returns the quotient of its two
1471 <p>The two arguments to the '<tt>div</tt>' instruction must be either <a
1472 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1474 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1475 Both arguments must have identical types.</p>
1477 <p>The value produced is the integer or floating point quotient of the
1480 <pre> <result> = div int 4, %var <i>; yields {int}:result = 4 / %var</i>
1483 <!-- _______________________________________________________________________ -->
1484 <div class="doc_subsubsection"> <a name="i_rem">'<tt>rem</tt>'
1485 Instruction</a> </div>
1486 <div class="doc_text">
1488 <pre> <result> = rem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1491 <p>The '<tt>rem</tt>' instruction returns the remainder from the
1492 division of its two operands.</p>
1494 <p>The two arguments to the '<tt>rem</tt>' instruction must be either <a
1495 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1497 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1498 Both arguments must have identical types.</p>
1500 <p>This returns the <i>remainder</i> of a division (where the result
1501 has the same sign as the divisor), not the <i>modulus</i> (where the
1502 result has the same sign as the dividend) of a value. For more
1503 information about the difference, see <a
1504 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
1507 <pre> <result> = rem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1510 <!-- _______________________________________________________________________ -->
1511 <div class="doc_subsubsection"> <a name="i_setcc">'<tt>set<i>cc</i></tt>'
1512 Instructions</a> </div>
1513 <div class="doc_text">
1515 <pre> <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1516 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1517 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1518 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1519 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1520 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1523 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean
1524 value based on a comparison of their two operands.</p>
1526 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must
1527 be of <a href="#t_firstclass">first class</a> type (it is not possible
1528 to compare '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>'
1529 or '<tt>void</tt>' values, etc...). Both arguments must have identical
1532 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1533 value if both operands are equal.<br>
1534 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1535 value if both operands are unequal.<br>
1536 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1537 value if the first operand is less than the second operand.<br>
1538 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1539 value if the first operand is greater than the second operand.<br>
1540 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1541 value if the first operand is less than or equal to the second operand.<br>
1542 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1543 value if the first operand is greater than or equal to the second
1546 <pre> <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1547 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1548 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1549 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1550 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1551 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1554 <!-- ======================================================================= -->
1555 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
1556 Operations</a> </div>
1557 <div class="doc_text">
1558 <p>Bitwise binary operators are used to do various forms of
1559 bit-twiddling in a program. They are generally very efficient
1560 instructions and can commonly be strength reduced from other
1561 instructions. They require two operands, execute an operation on them,
1562 and produce a single value. The resulting value of the bitwise binary
1563 operators is always the same type as its first operand.</p>
1565 <!-- _______________________________________________________________________ -->
1566 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
1567 Instruction</a> </div>
1568 <div class="doc_text">
1570 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1573 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1574 its two operands.</p>
1576 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1577 href="#t_integral">integral</a> values. Both arguments must have
1578 identical types.</p>
1580 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1582 <div style="align: center">
1583 <table border="1" cellspacing="0" cellpadding="4">
1614 <pre> <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1615 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1616 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1619 <!-- _______________________________________________________________________ -->
1620 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
1621 <div class="doc_text">
1623 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1626 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1627 or of its two operands.</p>
1629 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1630 href="#t_integral">integral</a> values. Both arguments must have
1631 identical types.</p>
1633 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1635 <div style="align: center">
1636 <table border="1" cellspacing="0" cellpadding="4">
1667 <pre> <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1668 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1669 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1672 <!-- _______________________________________________________________________ -->
1673 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
1674 Instruction</a> </div>
1675 <div class="doc_text">
1677 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1680 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
1681 or of its two operands. The <tt>xor</tt> is used to implement the
1682 "one's complement" operation, which is the "~" operator in C.</p>
1684 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
1685 href="#t_integral">integral</a> values. Both arguments must have
1686 identical types.</p>
1688 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
1690 <div style="align: center">
1691 <table border="1" cellspacing="0" cellpadding="4">
1723 <pre> <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
1724 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
1725 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
1726 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
1729 <!-- _______________________________________________________________________ -->
1730 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
1731 Instruction</a> </div>
1732 <div class="doc_text">
1734 <pre> <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1737 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
1738 the left a specified number of bits.</p>
1740 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
1741 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1744 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
1746 <pre> <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
1747 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
1748 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
1751 <!-- _______________________________________________________________________ -->
1752 <div class="doc_subsubsection"> <a name="i_shr">'<tt>shr</tt>'
1753 Instruction</a> </div>
1754 <div class="doc_text">
1756 <pre> <result> = shr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1759 <p>The '<tt>shr</tt>' instruction returns the first operand shifted to
1760 the right a specified number of bits.</p>
1762 <p>The first argument to the '<tt>shr</tt>' instruction must be an <a
1763 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1766 <p>If the first argument is a <a href="#t_signed">signed</a> type, the
1767 most significant bit is duplicated in the newly free'd bit positions.
1768 If the first argument is unsigned, zero bits shall fill the empty
1771 <pre> <result> = shr int 4, ubyte %var <i>; yields {int}:result = 4 >> %var</i>
1772 <result> = shr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
1773 <result> = shr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
1774 <result> = shr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
1775 <result> = shr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
1779 <!-- ======================================================================= -->
1780 <div class="doc_subsection">
1781 <a name="memoryops">Memory Access Operations</a>
1784 <div class="doc_text">
1786 <p>A key design point of an SSA-based representation is how it
1787 represents memory. In LLVM, no memory locations are in SSA form, which
1788 makes things very simple. This section describes how to read, write,
1789 allocate, and free memory in LLVM.</p>
1793 <!-- _______________________________________________________________________ -->
1794 <div class="doc_subsubsection">
1795 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
1798 <div class="doc_text">
1803 <result> = malloc <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
1808 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
1809 heap and returns a pointer to it.</p>
1813 <p>The '<tt>malloc</tt>' instruction allocates
1814 <tt>sizeof(<type>)*NumElements</tt>
1815 bytes of memory from the operating system and returns a pointer of the
1816 appropriate type to the program. If "NumElements" is specified, it is the
1817 number of elements allocated. If an alignment is specified, the value result
1818 of the allocation is guaranteed to be aligned to at least that boundary. If
1819 not specified, or if zero, the target can choose to align the allocation on any
1820 convenient boundary.</p>
1822 <p>'<tt>type</tt>' must be a sized type.</p>
1826 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
1827 a pointer is returned.</p>
1832 %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
1834 %size = <a href="#i_add">add</a> uint 2, 2 <i>; yields {uint}:size = uint 4</i>
1835 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
1836 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
1837 %array3 = malloc int, uint 4, align 1024 <i>; yields {int*}:array3</i>
1838 %array4 = malloc int, align 1024 <i>; yields {int*}:array4</i>
1842 <!-- _______________________________________________________________________ -->
1843 <div class="doc_subsubsection">
1844 <a name="i_free">'<tt>free</tt>' Instruction</a>
1847 <div class="doc_text">
1852 free <type> <value> <i>; yields {void}</i>
1857 <p>The '<tt>free</tt>' instruction returns memory back to the unused
1858 memory heap to be reallocated in the future.</p>
1862 <p>'<tt>value</tt>' shall be a pointer value that points to a value
1863 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
1868 <p>Access to the memory pointed to by the pointer is no longer defined
1869 after this instruction executes.</p>
1874 %array = <a href="#i_malloc">malloc</a> [4 x ubyte] <i>; yields {[4 x ubyte]*}:array</i>
1875 free [4 x ubyte]* %array
1879 <!-- _______________________________________________________________________ -->
1880 <div class="doc_subsubsection">
1881 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
1884 <div class="doc_text">
1889 <result> = alloca <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
1894 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
1895 stack frame of the procedure that is live until the current function
1896 returns to its caller.</p>
1900 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
1901 bytes of memory on the runtime stack, returning a pointer of the
1902 appropriate type to the program. If "NumElements" is specified, it is the
1903 number of elements allocated. If an alignment is specified, the value result
1904 of the allocation is guaranteed to be aligned to at least that boundary. If
1905 not specified, or if zero, the target can choose to align the allocation on any
1906 convenient boundary.</p>
1908 <p>'<tt>type</tt>' may be any sized type.</p>
1912 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
1913 memory is automatically released when the function returns. The '<tt>alloca</tt>'
1914 instruction is commonly used to represent automatic variables that must
1915 have an address available. When the function returns (either with the <tt><a
1916 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
1917 instructions), the memory is reclaimed.</p>
1922 %ptr = alloca int <i>; yields {int*}:ptr</i>
1923 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
1924 %ptr = alloca int, uint 4, align 1024 <i>; yields {int*}:ptr</i>
1925 %ptr = alloca int, align 1024 <i>; yields {int*}:ptr</i>
1929 <!-- _______________________________________________________________________ -->
1930 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
1931 Instruction</a> </div>
1932 <div class="doc_text">
1934 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
1936 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
1938 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
1939 address from which to load. The pointer must point to a <a
1940 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
1941 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
1942 the number or order of execution of this <tt>load</tt> with other
1943 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
1946 <p>The location of memory pointed to is loaded.</p>
1948 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
1950 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
1951 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
1954 <!-- _______________________________________________________________________ -->
1955 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
1956 Instruction</a> </div>
1958 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
1959 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
1962 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
1964 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
1965 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
1966 operand must be a pointer to the type of the '<tt><value></tt>'
1967 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
1968 optimizer is not allowed to modify the number or order of execution of
1969 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
1970 href="#i_store">store</a></tt> instructions.</p>
1972 <p>The contents of memory are updated to contain '<tt><value></tt>'
1973 at the location specified by the '<tt><pointer></tt>' operand.</p>
1975 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
1977 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
1978 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
1980 <!-- _______________________________________________________________________ -->
1981 <div class="doc_subsubsection">
1982 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
1985 <div class="doc_text">
1988 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
1994 The '<tt>getelementptr</tt>' instruction is used to get the address of a
1995 subelement of an aggregate data structure.</p>
1999 <p>This instruction takes a list of integer constants that indicate what
2000 elements of the aggregate object to index to. The actual types of the arguments
2001 provided depend on the type of the first pointer argument. The
2002 '<tt>getelementptr</tt>' instruction is used to index down through the type
2003 levels of a structure or to a specific index in an array. When indexing into a
2004 structure, only <tt>uint</tt>
2005 integer constants are allowed. When indexing into an array or pointer,
2006 <tt>int</tt> and <tt>long</tt> indexes are allowed of any sign.</p>
2008 <p>For example, let's consider a C code fragment and how it gets
2009 compiled to LLVM:</p>
2023 int *foo(struct ST *s) {
2024 return &s[1].Z.B[5][13];
2028 <p>The LLVM code generated by the GCC frontend is:</p>
2031 %RT = type { sbyte, [10 x [20 x int]], sbyte }
2032 %ST = type { int, double, %RT }
2036 int* %foo(%ST* %s) {
2038 %reg = getelementptr %ST* %s, int 1, uint 2, uint 1, int 5, int 13
2045 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2046 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2047 and <a href="#t_array">array</a> types require <tt>uint</tt>, <tt>int</tt>,
2048 <tt>ulong</tt>, or <tt>long</tt> values, and <a href="#t_struct">structure</a>
2049 types require <tt>uint</tt> <b>constants</b>.</p>
2051 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2052 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
2053 }</tt>' type, a structure. The second index indexes into the third element of
2054 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
2055 sbyte }</tt>' type, another structure. The third index indexes into the second
2056 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
2057 array. The two dimensions of the array are subscripted into, yielding an
2058 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2059 to this element, thus computing a value of '<tt>int*</tt>' type.</p>
2061 <p>Note that it is perfectly legal to index partially through a
2062 structure, returning a pointer to an inner element. Because of this,
2063 the LLVM code for the given testcase is equivalent to:</p>
2066 int* %foo(%ST* %s) {
2067 %t1 = getelementptr %ST* %s, int 1 <i>; yields %ST*:%t1</i>
2068 %t2 = getelementptr %ST* %t1, int 0, uint 2 <i>; yields %RT*:%t2</i>
2069 %t3 = getelementptr %RT* %t2, int 0, uint 1 <i>; yields [10 x [20 x int]]*:%t3</i>
2070 %t4 = getelementptr [10 x [20 x int]]* %t3, int 0, int 5 <i>; yields [20 x int]*:%t4</i>
2071 %t5 = getelementptr [20 x int]* %t4, int 0, int 13 <i>; yields int*:%t5</i>
2076 <p>Note that it is undefined to access an array out of bounds: array and
2077 pointer indexes must always be within the defined bounds of the array type.
2078 The one exception for this rules is zero length arrays. These arrays are
2079 defined to be accessible as variable length arrays, which requires access
2080 beyond the zero'th element.</p>
2085 <i>; yields [12 x ubyte]*:aptr</i>
2086 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, uint 1
2090 <!-- ======================================================================= -->
2091 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
2092 <div class="doc_text">
2093 <p>The instructions in this category are the "miscellaneous"
2094 instructions, which defy better classification.</p>
2096 <!-- _______________________________________________________________________ -->
2097 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
2098 Instruction</a> </div>
2099 <div class="doc_text">
2101 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
2103 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
2104 the SSA graph representing the function.</p>
2106 <p>The type of the incoming values are specified with the first type
2107 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
2108 as arguments, with one pair for each predecessor basic block of the
2109 current block. Only values of <a href="#t_firstclass">first class</a>
2110 type may be used as the value arguments to the PHI node. Only labels
2111 may be used as the label arguments.</p>
2112 <p>There must be no non-phi instructions between the start of a basic
2113 block and the PHI instructions: i.e. PHI instructions must be first in
2116 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
2117 value specified by the parameter, depending on which basic block we
2118 came from in the last <a href="#terminators">terminator</a> instruction.</p>
2120 <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>
2123 <!-- _______________________________________________________________________ -->
2124 <div class="doc_subsubsection">
2125 <a name="i_cast">'<tt>cast .. to</tt>' Instruction</a>
2128 <div class="doc_text">
2133 <result> = cast <ty> <value> to <ty2> <i>; yields ty2</i>
2139 The '<tt>cast</tt>' instruction is used as the primitive means to convert
2140 integers to floating point, change data type sizes, and break type safety (by
2148 The '<tt>cast</tt>' instruction takes a value to cast, which must be a first
2149 class value, and a type to cast it to, which must also be a <a
2150 href="#t_firstclass">first class</a> type.
2156 This instruction follows the C rules for explicit casts when determining how the
2157 data being cast must change to fit in its new container.
2161 When casting to bool, any value that would be considered true in the context of
2162 a C '<tt>if</tt>' condition is converted to the boolean '<tt>true</tt>' values,
2163 all else are '<tt>false</tt>'.
2167 When extending an integral value from a type of one signness to another (for
2168 example '<tt>sbyte</tt>' to '<tt>ulong</tt>'), the value is sign-extended if the
2169 <b>source</b> value is signed, and zero-extended if the source value is
2170 unsigned. <tt>bool</tt> values are always zero extended into either zero or
2177 %X = cast int 257 to ubyte <i>; yields ubyte:1</i>
2178 %Y = cast int 123 to bool <i>; yields bool:true</i>
2182 <!-- _______________________________________________________________________ -->
2183 <div class="doc_subsubsection">
2184 <a name="i_select">'<tt>select</tt>' Instruction</a>
2187 <div class="doc_text">
2192 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
2198 The '<tt>select</tt>' instruction is used to choose one value based on a
2199 condition, without branching.
2206 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.
2212 If the boolean condition evaluates to true, the instruction returns the first
2213 value argument; otherwise, it returns the second value argument.
2219 %X = select bool true, ubyte 17, ubyte 42 <i>; yields ubyte:17</i>
2227 <!-- _______________________________________________________________________ -->
2228 <div class="doc_subsubsection">
2229 <a name="i_call">'<tt>call</tt>' Instruction</a>
2232 <div class="doc_text">
2236 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
2241 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
2245 <p>This instruction requires several arguments:</p>
2249 <p>The optional "tail" marker indicates whether the callee function accesses
2250 any allocas or varargs in the caller. If the "tail" marker is present, the
2251 function call is eligible for tail call optimization. Note that calls may
2252 be marked "tail" even if they do not occur before a <a
2253 href="#i_ret"><tt>ret</tt></a> instruction.
2256 <p>The optional "cconv" marker indicates which <a href="callingconv">calling
2257 convention</a> the call should use. If none is specified, the call defaults
2258 to using C calling conventions.
2261 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
2262 being invoked. The argument types must match the types implied by this
2263 signature. This type can be omitted if the function is not varargs and
2264 if the function type does not return a pointer to a function.</p>
2267 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
2268 be invoked. In most cases, this is a direct function invocation, but
2269 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
2270 to function value.</p>
2273 <p>'<tt>function args</tt>': argument list whose types match the
2274 function signature argument types. All arguments must be of
2275 <a href="#t_firstclass">first class</a> type. If the function signature
2276 indicates the function accepts a variable number of arguments, the extra
2277 arguments can be specified.</p>
2283 <p>The '<tt>call</tt>' instruction is used to cause control flow to
2284 transfer to a specified function, with its incoming arguments bound to
2285 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
2286 instruction in the called function, control flow continues with the
2287 instruction after the function call, and the return value of the
2288 function is bound to the result argument. This is a simpler case of
2289 the <a href="#i_invoke">invoke</a> instruction.</p>
2294 %retval = call int %test(int %argc)
2295 call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);
2296 %X = tail call int %foo()
2297 %Y = tail call <a href="#callingconv">fastcc</a> int %foo()
2302 <!-- _______________________________________________________________________ -->
2303 <div class="doc_subsubsection">
2304 <a name="i_vaarg">'<tt>vaarg</tt>' Instruction</a>
2307 <div class="doc_text">
2312 <resultval> = va_arg <va_list*> <arglist>, <argty>
2317 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
2318 the "variable argument" area of a function call. It is used to implement the
2319 <tt>va_arg</tt> macro in C.</p>
2323 <p>This instruction takes a <tt>va_list*</tt> value and the type of
2324 the argument. It returns a value of the specified argument type and
2325 increments the <tt>va_list</tt> to poin to the next argument. Again, the
2326 actual type of <tt>va_list</tt> is target specific.</p>
2330 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
2331 type from the specified <tt>va_list</tt> and causes the
2332 <tt>va_list</tt> to point to the next argument. For more information,
2333 see the variable argument handling <a href="#int_varargs">Intrinsic
2336 <p>It is legal for this instruction to be called in a function which does not
2337 take a variable number of arguments, for example, the <tt>vfprintf</tt>
2340 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
2341 href="#intrinsics">intrinsic function</a> because it takes a type as an
2346 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
2350 <!-- *********************************************************************** -->
2351 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
2352 <!-- *********************************************************************** -->
2354 <div class="doc_text">
2356 <p>LLVM supports the notion of an "intrinsic function". These functions have
2357 well known names and semantics and are required to follow certain
2358 restrictions. Overall, these instructions represent an extension mechanism for
2359 the LLVM language that does not require changing all of the transformations in
2360 LLVM to add to the language (or the bytecode reader/writer, the parser,
2363 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
2364 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
2365 this. Intrinsic functions must always be external functions: you cannot define
2366 the body of intrinsic functions. Intrinsic functions may only be used in call
2367 or invoke instructions: it is illegal to take the address of an intrinsic
2368 function. Additionally, because intrinsic functions are part of the LLVM
2369 language, it is required that they all be documented here if any are added.</p>
2372 <p>To learn how to add an intrinsic function, please see the <a
2373 href="ExtendingLLVM.html">Extending LLVM Guide</a>.
2378 <!-- ======================================================================= -->
2379 <div class="doc_subsection">
2380 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
2383 <div class="doc_text">
2385 <p>Variable argument support is defined in LLVM with the <a
2386 href="#i_vanext"><tt>vanext</tt></a> instruction and these three
2387 intrinsic functions. These functions are related to the similarly
2388 named macros defined in the <tt><stdarg.h></tt> header file.</p>
2390 <p>All of these functions operate on arguments that use a
2391 target-specific value type "<tt>va_list</tt>". The LLVM assembly
2392 language reference manual does not define what this type is, so all
2393 transformations should be prepared to handle intrinsics with any type
2396 <p>This example shows how the <a href="#i_vanext"><tt>vanext</tt></a>
2397 instruction and the variable argument handling intrinsic functions are
2401 int %test(int %X, ...) {
2402 ; Initialize variable argument processing
2404 call void %<a href="#i_va_start">llvm.va_start</a>(sbyte** %ap)
2406 ; Read a single integer argument
2407 %tmp = va_arg sbyte** %ap, int
2409 ; Demonstrate usage of llvm.va_copy and llvm.va_end
2411 call void %<a href="#i_va_copy">llvm.va_copy</a>(sbyte** %aq, sbyte** %ap)
2412 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %aq)
2414 ; Stop processing of arguments.
2415 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %ap)
2421 <!-- _______________________________________________________________________ -->
2422 <div class="doc_subsubsection">
2423 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
2427 <div class="doc_text">
2429 <pre> declare void %llvm.va_start(<va_list>* <arglist>)<br></pre>
2431 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
2432 <tt>*<arglist></tt> for subsequent use by <tt><a
2433 href="#i_va_arg">va_arg</a></tt>.</p>
2437 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
2441 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
2442 macro available in C. In a target-dependent way, it initializes the
2443 <tt>va_list</tt> element the argument points to, so that the next call to
2444 <tt>va_arg</tt> will produce the first variable argument passed to the function.
2445 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
2446 last argument of the function, the compiler can figure that out.</p>
2450 <!-- _______________________________________________________________________ -->
2451 <div class="doc_subsubsection">
2452 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
2455 <div class="doc_text">
2457 <pre> declare void %llvm.va_end(<va_list*> <arglist>)<br></pre>
2459 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
2460 which has been initialized previously with <tt><a href="#i_va_start">llvm.va_start</a></tt>
2461 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
2463 <p>The argument is a <tt>va_list</tt> to destroy.</p>
2465 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
2466 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
2467 Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and <a
2468 href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
2469 with calls to <tt>llvm.va_end</tt>.</p>
2472 <!-- _______________________________________________________________________ -->
2473 <div class="doc_subsubsection">
2474 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
2477 <div class="doc_text">
2482 declare void %llvm.va_copy(<va_list>* <destarglist>,
2483 <va_list>* <srcarglist>)
2488 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
2489 the source argument list to the destination argument list.</p>
2493 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
2494 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
2499 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
2500 available in C. In a target-dependent way, it copies the source
2501 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
2502 because the <tt><a href="i_va_begin">llvm.va_begin</a></tt> intrinsic may be
2503 arbitrarily complex and require memory allocation, for example.</p>
2507 <!-- ======================================================================= -->
2508 <div class="doc_subsection">
2509 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
2512 <div class="doc_text">
2515 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
2516 Collection</a> requires the implementation and generation of these intrinsics.
2517 These intrinsics allow identification of <a href="#i_gcroot">GC roots on the
2518 stack</a>, as well as garbage collector implementations that require <a
2519 href="#i_gcread">read</a> and <a href="#i_gcwrite">write</a> barriers.
2520 Front-ends for type-safe garbage collected languages should generate these
2521 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
2522 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
2526 <!-- _______________________________________________________________________ -->
2527 <div class="doc_subsubsection">
2528 <a name="i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
2531 <div class="doc_text">
2536 declare void %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
2541 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
2542 the code generator, and allows some metadata to be associated with it.</p>
2546 <p>The first argument specifies the address of a stack object that contains the
2547 root pointer. The second pointer (which must be either a constant or a global
2548 value address) contains the meta-data to be associated with the root.</p>
2552 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
2553 location. At compile-time, the code generator generates information to allow
2554 the runtime to find the pointer at GC safe points.
2560 <!-- _______________________________________________________________________ -->
2561 <div class="doc_subsubsection">
2562 <a name="i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
2565 <div class="doc_text">
2570 declare sbyte* %llvm.gcread(sbyte** %Ptr)
2575 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
2576 locations, allowing garbage collector implementations that require read
2581 <p>The argument is the address to read from, which should be an address
2582 allocated from the garbage collector.</p>
2586 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
2587 instruction, but may be replaced with substantially more complex code by the
2588 garbage collector runtime, as needed.</p>
2593 <!-- _______________________________________________________________________ -->
2594 <div class="doc_subsubsection">
2595 <a name="i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
2598 <div class="doc_text">
2603 declare void %llvm.gcwrite(sbyte* %P1, sbyte** %P2)
2608 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
2609 locations, allowing garbage collector implementations that require write
2610 barriers (such as generational or reference counting collectors).</p>
2614 <p>The first argument is the reference to store, and the second is the heap
2615 location to store to.</p>
2619 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
2620 instruction, but may be replaced with substantially more complex code by the
2621 garbage collector runtime, as needed.</p>
2627 <!-- ======================================================================= -->
2628 <div class="doc_subsection">
2629 <a name="int_codegen">Code Generator Intrinsics</a>
2632 <div class="doc_text">
2634 These intrinsics are provided by LLVM to expose special features that may only
2635 be implemented with code generator support.
2640 <!-- _______________________________________________________________________ -->
2641 <div class="doc_subsubsection">
2642 <a name="i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
2645 <div class="doc_text">
2649 declare void* %llvm.returnaddress(uint <level>)
2655 The '<tt>llvm.returnaddress</tt>' intrinsic returns a target-specific value
2656 indicating the return address of the current function or one of its callers.
2662 The argument to this intrinsic indicates which function to return the address
2663 for. Zero indicates the calling function, one indicates its caller, etc. The
2664 argument is <b>required</b> to be a constant integer value.
2670 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
2671 the return address of the specified call frame, or zero if it cannot be
2672 identified. The value returned by this intrinsic is likely to be incorrect or 0
2673 for arguments other than zero, so it should only be used for debugging purposes.
2677 Note that calling this intrinsic does not prevent function inlining or other
2678 aggressive transformations, so the value returned may not be that of the obvious
2679 source-language caller.
2684 <!-- _______________________________________________________________________ -->
2685 <div class="doc_subsubsection">
2686 <a name="i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
2689 <div class="doc_text">
2693 declare void* %llvm.frameaddress(uint <level>)
2699 The '<tt>llvm.frameaddress</tt>' intrinsic returns the target-specific frame
2700 pointer value for the specified stack frame.
2706 The argument to this intrinsic indicates which function to return the frame
2707 pointer for. Zero indicates the calling function, one indicates its caller,
2708 etc. The argument is <b>required</b> to be a constant integer value.
2714 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
2715 the frame address of the specified call frame, or zero if it cannot be
2716 identified. The value returned by this intrinsic is likely to be incorrect or 0
2717 for arguments other than zero, so it should only be used for debugging purposes.
2721 Note that calling this intrinsic does not prevent function inlining or other
2722 aggressive transformations, so the value returned may not be that of the obvious
2723 source-language caller.
2727 <!-- _______________________________________________________________________ -->
2728 <div class="doc_subsubsection">
2729 <a name="i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
2732 <div class="doc_text">
2736 declare void %llvm.prefetch(sbyte * <address>,
2737 uint <rw>, uint <locality>)
2744 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
2745 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
2747 effect on the behavior of the program but can change its performance
2754 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
2755 determining if the fetch should be for a read (0) or write (1), and
2756 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
2757 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
2758 <tt>locality</tt> arguments must be constant integers.
2764 This intrinsic does not modify the behavior of the program. In particular,
2765 prefetches cannot trap and do not produce a value. On targets that support this
2766 intrinsic, the prefetch can provide hints to the processor cache for better
2772 <!-- _______________________________________________________________________ -->
2773 <div class="doc_subsubsection">
2774 <a name="i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
2777 <div class="doc_text">
2781 declare void %llvm.pcmarker( uint <id> )
2788 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
2790 code to simulators and other tools. The method is target specific, but it is
2791 expected that the marker will use exported symbols to transmit the PC of the marker.
2792 The marker makes no guaranties that it will remain with any specific instruction
2793 after optimizations. It is possible that the presense of a marker will inhibit
2794 optimizations. The intended use is to be inserted after optmizations to allow
2795 correlations of simulation runs.
2801 <tt>id</tt> is a numerical id identifying the marker.
2807 This intrinsic does not modify the behavior of the program. Backends that do not
2808 support this intrinisic may ignore it.
2814 <!-- ======================================================================= -->
2815 <div class="doc_subsection">
2816 <a name="int_os">Operating System Intrinsics</a>
2819 <div class="doc_text">
2821 These intrinsics are provided by LLVM to support the implementation of
2822 operating system level code.
2827 <!-- _______________________________________________________________________ -->
2828 <div class="doc_subsubsection">
2829 <a name="i_readport">'<tt>llvm.readport</tt>' Intrinsic</a>
2832 <div class="doc_text">
2836 declare <integer type> %llvm.readport (<integer type> <address>)
2842 The '<tt>llvm.readport</tt>' intrinsic reads data from the specified hardware
2849 The argument to this intrinsic indicates the hardware I/O address from which
2850 to read the data. The address is in the hardware I/O address namespace (as
2851 opposed to being a memory location for memory mapped I/O).
2857 The '<tt>llvm.readport</tt>' intrinsic reads data from the hardware I/O port
2858 specified by <i>address</i> and returns the value. The address and return
2859 value must be integers, but the size is dependent upon the platform upon which
2860 the program is code generated. For example, on x86, the address must be an
2861 unsigned 16-bit value, and the return value must be 8, 16, or 32 bits.
2866 <!-- _______________________________________________________________________ -->
2867 <div class="doc_subsubsection">
2868 <a name="i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a>
2871 <div class="doc_text">
2875 call void (<integer type>, <integer type>)*
2876 %llvm.writeport (<integer type> <value>,
2877 <integer type> <address>)
2883 The '<tt>llvm.writeport</tt>' intrinsic writes data to the specified hardware
2890 The first argument is the value to write to the I/O port.
2894 The second argument indicates the hardware I/O address to which data should be
2895 written. The address is in the hardware I/O address namespace (as opposed to
2896 being a memory location for memory mapped I/O).
2902 The '<tt>llvm.writeport</tt>' intrinsic writes <i>value</i> to the I/O port
2903 specified by <i>address</i>. The address and value must be integers, but the
2904 size is dependent upon the platform upon which the program is code generated.
2905 For example, on x86, the address must be an unsigned 16-bit value, and the
2906 value written must be 8, 16, or 32 bits in length.
2911 <!-- _______________________________________________________________________ -->
2912 <div class="doc_subsubsection">
2913 <a name="i_readio">'<tt>llvm.readio</tt>' Intrinsic</a>
2916 <div class="doc_text">
2920 declare <result> %llvm.readio (<ty> * <pointer>)
2926 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
2933 The argument to this intrinsic is a pointer indicating the memory address from
2934 which to read the data. The data must be a
2935 <a href="#t_firstclass">first class</a> type.
2941 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
2942 location specified by <i>pointer</i> and returns the value. The argument must
2943 be a pointer, and the return value must be a
2944 <a href="#t_firstclass">first class</a> type. However, certain architectures
2945 may not support I/O on all first class types. For example, 32-bit processors
2946 may only support I/O on data types that are 32 bits or less.
2950 This intrinsic enforces an in-order memory model for llvm.readio and
2951 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
2952 scheduled processors may execute loads and stores out of order, re-ordering at
2953 run time accesses to memory mapped I/O registers. Using these intrinsics
2954 ensures that accesses to memory mapped I/O registers occur in program order.
2959 <!-- _______________________________________________________________________ -->
2960 <div class="doc_subsubsection">
2961 <a name="i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a>
2964 <div class="doc_text">
2968 declare void %llvm.writeio (<ty1> <value>, <ty2> * <pointer>)
2974 The '<tt>llvm.writeio</tt>' intrinsic writes data to the specified memory
2981 The first argument is the value to write to the memory mapped I/O location.
2982 The second argument is a pointer indicating the memory address to which the
2983 data should be written.
2989 The '<tt>llvm.writeio</tt>' intrinsic writes <i>value</i> to the memory mapped
2990 I/O address specified by <i>pointer</i>. The value must be a
2991 <a href="#t_firstclass">first class</a> type. However, certain architectures
2992 may not support I/O on all first class types. For example, 32-bit processors
2993 may only support I/O on data types that are 32 bits or less.
2997 This intrinsic enforces an in-order memory model for llvm.readio and
2998 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
2999 scheduled processors may execute loads and stores out of order, re-ordering at
3000 run time accesses to memory mapped I/O registers. Using these intrinsics
3001 ensures that accesses to memory mapped I/O registers occur in program order.
3006 <!-- ======================================================================= -->
3007 <div class="doc_subsection">
3008 <a name="int_libc">Standard C Library Intrinsics</a>
3011 <div class="doc_text">
3013 LLVM provides intrinsics for a few important standard C library functions.
3014 These intrinsics allow source-language front-ends to pass information about the
3015 alignment of the pointer arguments to the code generator, providing opportunity
3016 for more efficient code generation.
3021 <!-- _______________________________________________________________________ -->
3022 <div class="doc_subsubsection">
3023 <a name="i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
3026 <div class="doc_text">
3030 declare void %llvm.memcpy(sbyte* <dest>, sbyte* <src>,
3031 uint <len>, uint <align>)
3037 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
3038 location to the destination location.
3042 Note that, unlike the standard libc function, the <tt>llvm.memcpy</tt> intrinsic
3043 does not return a value, and takes an extra alignment argument.
3049 The first argument is a pointer to the destination, the second is a pointer to
3050 the source. The third argument is an (arbitrarily sized) integer argument
3051 specifying the number of bytes to copy, and the fourth argument is the alignment
3052 of the source and destination locations.
3056 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3057 the caller guarantees that the size of the copy is a multiple of the alignment
3058 and that both the source and destination pointers are aligned to that boundary.
3064 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
3065 location to the destination location, which are not allowed to overlap. It
3066 copies "len" bytes of memory over. If the argument is known to be aligned to
3067 some boundary, this can be specified as the fourth argument, otherwise it should
3073 <!-- _______________________________________________________________________ -->
3074 <div class="doc_subsubsection">
3075 <a name="i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
3078 <div class="doc_text">
3082 declare void %llvm.memmove(sbyte* <dest>, sbyte* <src>,
3083 uint <len>, uint <align>)
3089 The '<tt>llvm.memmove</tt>' intrinsic moves a block of memory from the source
3090 location to the destination location. It is similar to the '<tt>llvm.memcpy</tt>'
3091 intrinsic but allows the two memory locations to overlap.
3095 Note that, unlike the standard libc function, the <tt>llvm.memmove</tt> intrinsic
3096 does not return a value, and takes an extra alignment argument.
3102 The first argument is a pointer to the destination, the second is a pointer to
3103 the source. The third argument is an (arbitrarily sized) integer argument
3104 specifying the number of bytes to copy, and the fourth argument is the alignment
3105 of the source and destination locations.
3109 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3110 the caller guarantees that the size of the copy is a multiple of the alignment
3111 and that both the source and destination pointers are aligned to that boundary.
3117 The '<tt>llvm.memmove</tt>' intrinsic copies a block of memory from the source
3118 location to the destination location, which may overlap. It
3119 copies "len" bytes of memory over. If the argument is known to be aligned to
3120 some boundary, this can be specified as the fourth argument, otherwise it should
3126 <!-- _______________________________________________________________________ -->
3127 <div class="doc_subsubsection">
3128 <a name="i_memset">'<tt>llvm.memset</tt>' Intrinsic</a>
3131 <div class="doc_text">
3135 declare void %llvm.memset(sbyte* <dest>, ubyte <val>,
3136 uint <len>, uint <align>)
3142 The '<tt>llvm.memset</tt>' intrinsic fills a block of memory with a particular
3147 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
3148 does not return a value, and takes an extra alignment argument.
3154 The first argument is a pointer to the destination to fill, the second is the
3155 byte value to fill it with, the third argument is an (arbitrarily sized) integer
3156 argument specifying the number of bytes to fill, and the fourth argument is the
3157 known alignment of destination location.
3161 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3162 the caller guarantees that the size of the copy is a multiple of the alignment
3163 and that the destination pointer is aligned to that boundary.
3169 The '<tt>llvm.memset</tt>' intrinsic fills "len" bytes of memory starting at the
3170 destination location. If the argument is known to be aligned to some boundary,
3171 this can be specified as the fourth argument, otherwise it should be set to 0 or
3177 <!-- _______________________________________________________________________ -->
3178 <div class="doc_subsubsection">
3179 <a name="i_isunordered">'<tt>llvm.isunordered</tt>' Intrinsic</a>
3182 <div class="doc_text">
3186 declare bool %llvm.isunordered(<float or double> Val1, <float or double> Val2)
3192 The '<tt>llvm.isunordered</tt>' intrinsic returns true if either or both of the
3193 specified floating point values is a NAN.
3199 The arguments are floating point numbers of the same type.
3205 If either or both of the arguments is a SNAN or QNAN, it returns true, otherwise
3211 <!-- _______________________________________________________________________ -->
3212 <div class="doc_subsubsection">
3213 <a name="i_sqrt">'<tt>llvm.sqrt</tt>' Intrinsic</a>
3216 <div class="doc_text">
3220 declare <float or double> %llvm.sqrt(<float or double> Val)
3226 The '<tt>llvm.sqrt</tt>' intrinsic returns the sqrt of the specified operand,
3227 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
3228 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
3229 negative numbers (which allows for better optimization).
3235 The argument and return value are floating point numbers of the same type.
3241 This function returns the sqrt of the specified operand if it is a positive
3242 floating point number.
3246 <!-- ======================================================================= -->
3247 <div class="doc_subsection">
3248 <a name="int_count">Bit Counting Intrinsics</a>
3251 <div class="doc_text">
3253 LLVM provides intrinsics for a few important bit counting operations.
3254 These allow efficient code generation for some algorithms.
3259 <!-- _______________________________________________________________________ -->
3260 <div class="doc_subsubsection">
3261 <a name="int_ctpop">'<tt>llvm.ctpop</tt>' Intrinsic</a>
3264 <div class="doc_text">
3268 declare int %llvm.ctpop(int <src>)
3275 The '<tt>llvm.ctpop</tt>' intrinsic counts the number of ones in a variable.
3281 The only argument is the value to be counted. The argument may be of any
3282 integer type. The return type must match the argument type.
3288 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
3292 <!-- _______________________________________________________________________ -->
3293 <div class="doc_subsubsection">
3294 <a name="int_ctlz">'<tt>llvm.ctlz</tt>' Intrinsic</a>
3297 <div class="doc_text">
3301 declare int %llvm.ctlz(int <src>)
3308 The '<tt>llvm.ctlz</tt>' intrinsic counts the number of leading zeros in a
3315 The only argument is the value to be counted. The argument may be of any
3316 integer type. The return type must match the argument type.
3322 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
3323 in a variable. If the src == 0 then the result is the size in bits of the type
3324 of src. For example, <tt>llvm.cttz(int 2) = 30</tt>.
3330 <!-- _______________________________________________________________________ -->
3331 <div class="doc_subsubsection">
3332 <a name="int_cttz">'<tt>llvm.cttz</tt>' Intrinsic</a>
3335 <div class="doc_text">
3339 declare int %llvm.cttz(int <src>)
3346 The '<tt>llvm.cttz</tt>' intrinsic counts the number of trailing zeros.
3352 The only argument is the value to be counted. The argument may be of any
3353 integer type. The return type must match the argument type.
3359 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
3360 in a variable. If the src == 0 then the result is the size in bits of the type
3361 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
3365 <!-- ======================================================================= -->
3366 <div class="doc_subsection">
3367 <a name="int_debugger">Debugger Intrinsics</a>
3370 <div class="doc_text">
3372 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
3373 are described in the <a
3374 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
3375 Debugging</a> document.
3380 <!-- *********************************************************************** -->
3383 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3384 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
3385 <a href="http://validator.w3.org/check/referer"><img
3386 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
3388 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3389 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br>
3390 Last modified: $Date$