1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <div class="doc_title"> LLVM Language Reference Manual </div>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a></li>
24 <li><a href="#callingconv">Calling Conventions</a></li>
25 <li><a href="#globalvars">Global Variables</a></li>
26 <li><a href="#functionstructure">Functions</a></li>
27 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
30 <li><a href="#typesystem">Type System</a>
32 <li><a href="#t_primitive">Primitive Types</a>
34 <li><a href="#t_classifications">Type Classifications</a></li>
37 <li><a href="#t_derived">Derived Types</a>
39 <li><a href="#t_array">Array Type</a></li>
40 <li><a href="#t_function">Function Type</a></li>
41 <li><a href="#t_pointer">Pointer Type</a></li>
42 <li><a href="#t_struct">Structure Type</a></li>
43 <li><a href="#t_packed">Packed Type</a></li>
44 <li><a href="#t_opaque">Opaque Type</a></li>
49 <li><a href="#constants">Constants</a>
51 <li><a href="#simpleconstants">Simple Constants</a>
52 <li><a href="#aggregateconstants">Aggregate Constants</a>
53 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
54 <li><a href="#undefvalues">Undefined Values</a>
55 <li><a href="#constantexprs">Constant Expressions</a>
58 <li><a href="#othervalues">Other Values</a>
60 <li><a href="#inlineasm">Inline Assembler Expressions</a>
63 <li><a href="#instref">Instruction Reference</a>
65 <li><a href="#terminators">Terminator Instructions</a>
67 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
68 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
69 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
70 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
71 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
72 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
75 <li><a href="#binaryops">Binary Operations</a>
77 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
78 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
79 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
80 <li><a href="#i_div">'<tt>div</tt>' Instruction</a></li>
81 <li><a href="#i_rem">'<tt>rem</tt>' Instruction</a></li>
82 <li><a href="#i_setcc">'<tt>set<i>cc</i></tt>' Instructions</a></li>
85 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
87 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
88 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
89 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
90 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
91 <li><a href="#i_shr">'<tt>shr</tt>' Instruction</a></li>
94 <li><a href="#memoryops">Memory Access Operations</a>
96 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
97 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
98 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
99 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
100 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
101 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
104 <li><a href="#otherops">Other Operations</a>
106 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
107 <li><a href="#i_cast">'<tt>cast .. to</tt>' Instruction</a></li>
108 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
109 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
110 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
111 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
112 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
117 <li><a href="#intrinsics">Intrinsic Functions</a>
119 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
121 <li><a href="#i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
122 <li><a href="#i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
123 <li><a href="#i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
126 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
128 <li><a href="#i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
129 <li><a href="#i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
130 <li><a href="#i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
133 <li><a href="#int_codegen">Code Generator Intrinsics</a>
135 <li><a href="#i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
136 <li><a href="#i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
137 <li><a href="#i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
138 <li><a href="#i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
139 <li><a href="#i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
140 <li><a href="#i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
141 <li><a href="#i_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
144 <li><a href="#int_os">Operating System Intrinsics</a>
146 <li><a href="#i_readport">'<tt>llvm.readport</tt>' Intrinsic</a></li>
147 <li><a href="#i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a></li>
148 <li><a href="#i_readio">'<tt>llvm.readio</tt>' Intrinsic</a></li>
149 <li><a href="#i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a></li>
151 <li><a href="#int_libc">Standard C Library Intrinsics</a>
153 <li><a href="#i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a></li>
154 <li><a href="#i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a></li>
155 <li><a href="#i_memset">'<tt>llvm.memset</tt>' Intrinsic</a></li>
156 <li><a href="#i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a></li>
157 <li><a href="#i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
161 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
163 <li><a href="#i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
164 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
165 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
166 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
169 <li><a href="#int_debugger">Debugger intrinsics</a></li>
174 <div class="doc_author">
175 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
176 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
179 <!-- *********************************************************************** -->
180 <div class="doc_section"> <a name="abstract">Abstract </a></div>
181 <!-- *********************************************************************** -->
183 <div class="doc_text">
184 <p>This document is a reference manual for the LLVM assembly language.
185 LLVM is an SSA based representation that provides type safety,
186 low-level operations, flexibility, and the capability of representing
187 'all' high-level languages cleanly. It is the common code
188 representation used throughout all phases of the LLVM compilation
192 <!-- *********************************************************************** -->
193 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
194 <!-- *********************************************************************** -->
196 <div class="doc_text">
198 <p>The LLVM code representation is designed to be used in three
199 different forms: as an in-memory compiler IR, as an on-disk bytecode
200 representation (suitable for fast loading by a Just-In-Time compiler),
201 and as a human readable assembly language representation. This allows
202 LLVM to provide a powerful intermediate representation for efficient
203 compiler transformations and analysis, while providing a natural means
204 to debug and visualize the transformations. The three different forms
205 of LLVM are all equivalent. This document describes the human readable
206 representation and notation.</p>
208 <p>The LLVM representation aims to be light-weight and low-level
209 while being expressive, typed, and extensible at the same time. It
210 aims to be a "universal IR" of sorts, by being at a low enough level
211 that high-level ideas may be cleanly mapped to it (similar to how
212 microprocessors are "universal IR's", allowing many source languages to
213 be mapped to them). By providing type information, LLVM can be used as
214 the target of optimizations: for example, through pointer analysis, it
215 can be proven that a C automatic variable is never accessed outside of
216 the current function... allowing it to be promoted to a simple SSA
217 value instead of a memory location.</p>
221 <!-- _______________________________________________________________________ -->
222 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
224 <div class="doc_text">
226 <p>It is important to note that this document describes 'well formed'
227 LLVM assembly language. There is a difference between what the parser
228 accepts and what is considered 'well formed'. For example, the
229 following instruction is syntactically okay, but not well formed:</p>
232 %x = <a href="#i_add">add</a> int 1, %x
235 <p>...because the definition of <tt>%x</tt> does not dominate all of
236 its uses. The LLVM infrastructure provides a verification pass that may
237 be used to verify that an LLVM module is well formed. This pass is
238 automatically run by the parser after parsing input assembly and by
239 the optimizer before it outputs bytecode. The violations pointed out
240 by the verifier pass indicate bugs in transformation passes or input to
243 <!-- Describe the typesetting conventions here. --> </div>
245 <!-- *********************************************************************** -->
246 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
247 <!-- *********************************************************************** -->
249 <div class="doc_text">
251 <p>LLVM uses three different forms of identifiers, for different
255 <li>Named values are represented as a string of characters with a '%' prefix.
256 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
257 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
258 Identifiers which require other characters in their names can be surrounded
259 with quotes. In this way, anything except a <tt>"</tt> character can be used
262 <li>Unnamed values are represented as an unsigned numeric value with a '%'
263 prefix. For example, %12, %2, %44.</li>
265 <li>Constants, which are described in a <a href="#constants">section about
266 constants</a>, below.</li>
269 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
270 don't need to worry about name clashes with reserved words, and the set of
271 reserved words may be expanded in the future without penalty. Additionally,
272 unnamed identifiers allow a compiler to quickly come up with a temporary
273 variable without having to avoid symbol table conflicts.</p>
275 <p>Reserved words in LLVM are very similar to reserved words in other
276 languages. There are keywords for different opcodes ('<tt><a
277 href="#i_add">add</a></tt>', '<tt><a href="#i_cast">cast</a></tt>', '<tt><a
278 href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
279 href="#t_void">void</a></tt>', '<tt><a href="#t_uint">uint</a></tt>', etc...),
280 and others. These reserved words cannot conflict with variable names, because
281 none of them start with a '%' character.</p>
283 <p>Here is an example of LLVM code to multiply the integer variable
284 '<tt>%X</tt>' by 8:</p>
289 %result = <a href="#i_mul">mul</a> uint %X, 8
292 <p>After strength reduction:</p>
295 %result = <a href="#i_shl">shl</a> uint %X, ubyte 3
298 <p>And the hard way:</p>
301 <a href="#i_add">add</a> uint %X, %X <i>; yields {uint}:%0</i>
302 <a href="#i_add">add</a> uint %0, %0 <i>; yields {uint}:%1</i>
303 %result = <a href="#i_add">add</a> uint %1, %1
306 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
307 important lexical features of LLVM:</p>
311 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
314 <li>Unnamed temporaries are created when the result of a computation is not
315 assigned to a named value.</li>
317 <li>Unnamed temporaries are numbered sequentially</li>
321 <p>...and it also shows a convention that we follow in this document. When
322 demonstrating instructions, we will follow an instruction with a comment that
323 defines the type and name of value produced. Comments are shown in italic
328 <!-- *********************************************************************** -->
329 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
330 <!-- *********************************************************************** -->
332 <!-- ======================================================================= -->
333 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
336 <div class="doc_text">
338 <p>LLVM programs are composed of "Module"s, each of which is a
339 translation unit of the input programs. Each module consists of
340 functions, global variables, and symbol table entries. Modules may be
341 combined together with the LLVM linker, which merges function (and
342 global variable) definitions, resolves forward declarations, and merges
343 symbol table entries. Here is an example of the "hello world" module:</p>
345 <pre><i>; Declare the string constant as a global constant...</i>
346 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
347 href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
349 <i>; External declaration of the puts function</i>
350 <a href="#functionstructure">declare</a> int %puts(sbyte*) <i>; int(sbyte*)* </i>
352 <i>; Definition of main function</i>
353 int %main() { <i>; int()* </i>
354 <i>; Convert [13x sbyte]* to sbyte *...</i>
356 href="#i_getelementptr">getelementptr</a> [13 x sbyte]* %.LC0, long 0, long 0 <i>; sbyte*</i>
358 <i>; Call puts function to write out the string to stdout...</i>
360 href="#i_call">call</a> int %puts(sbyte* %cast210) <i>; int</i>
362 href="#i_ret">ret</a> int 0<br>}<br></pre>
364 <p>This example is made up of a <a href="#globalvars">global variable</a>
365 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
366 function, and a <a href="#functionstructure">function definition</a>
367 for "<tt>main</tt>".</p>
369 <p>In general, a module is made up of a list of global values,
370 where both functions and global variables are global values. Global values are
371 represented by a pointer to a memory location (in this case, a pointer to an
372 array of char, and a pointer to a function), and have one of the following <a
373 href="#linkage">linkage types</a>.</p>
377 <!-- ======================================================================= -->
378 <div class="doc_subsection">
379 <a name="linkage">Linkage Types</a>
382 <div class="doc_text">
385 All Global Variables and Functions have one of the following types of linkage:
390 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
392 <dd>Global values with internal linkage are only directly accessible by
393 objects in the current module. In particular, linking code into a module with
394 an internal global value may cause the internal to be renamed as necessary to
395 avoid collisions. Because the symbol is internal to the module, all
396 references can be updated. This corresponds to the notion of the
397 '<tt>static</tt>' keyword in C, or the idea of "anonymous namespaces" in C++.
400 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
402 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt> linkage, with
403 the twist that linking together two modules defining the same
404 <tt>linkonce</tt> globals will cause one of the globals to be discarded. This
405 is typically used to implement inline functions. Unreferenced
406 <tt>linkonce</tt> globals are allowed to be discarded.
409 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
411 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
412 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
413 used to implement constructs in C such as "<tt>int X;</tt>" at global scope.
416 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
418 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
419 pointer to array type. When two global variables with appending linkage are
420 linked together, the two global arrays are appended together. This is the
421 LLVM, typesafe, equivalent of having the system linker append together
422 "sections" with identical names when .o files are linked.
425 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
427 <dd>If none of the above identifiers are used, the global is externally
428 visible, meaning that it participates in linkage and can be used to resolve
429 external symbol references.
433 <p><a name="linkage_external">For example, since the "<tt>.LC0</tt>"
434 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
435 variable and was linked with this one, one of the two would be renamed,
436 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
437 external (i.e., lacking any linkage declarations), they are accessible
438 outside of the current module. It is illegal for a function <i>declaration</i>
439 to have any linkage type other than "externally visible".</a></p>
443 <!-- ======================================================================= -->
444 <div class="doc_subsection">
445 <a name="callingconv">Calling Conventions</a>
448 <div class="doc_text">
450 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
451 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
452 specified for the call. The calling convention of any pair of dynamic
453 caller/callee must match, or the behavior of the program is undefined. The
454 following calling conventions are supported by LLVM, and more may be added in
458 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
460 <dd>This calling convention (the default if no other calling convention is
461 specified) matches the target C calling conventions. This calling convention
462 supports varargs function calls and tolerates some mismatch in the declared
463 prototype and implemented declaration of the function (as does normal C).
466 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
468 <dd>This calling convention attempts to make calls as fast as possible
469 (e.g. by passing things in registers). This calling convention allows the
470 target to use whatever tricks it wants to produce fast code for the target,
471 without having to conform to an externally specified ABI. Implementations of
472 this convention should allow arbitrary tail call optimization to be supported.
473 This calling convention does not support varargs and requires the prototype of
474 all callees to exactly match the prototype of the function definition.
477 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
479 <dd>This calling convention attempts to make code in the caller as efficient
480 as possible under the assumption that the call is not commonly executed. As
481 such, these calls often preserve all registers so that the call does not break
482 any live ranges in the caller side. This calling convention does not support
483 varargs and requires the prototype of all callees to exactly match the
484 prototype of the function definition.
487 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
489 <dd>Any calling convention may be specified by number, allowing
490 target-specific calling conventions to be used. Target specific calling
491 conventions start at 64.
495 <p>More calling conventions can be added/defined on an as-needed basis, to
496 support pascal conventions or any other well-known target-independent
501 <!-- ======================================================================= -->
502 <div class="doc_subsection">
503 <a name="globalvars">Global Variables</a>
506 <div class="doc_text">
508 <p>Global variables define regions of memory allocated at compilation time
509 instead of run-time. Global variables may optionally be initialized, may have
510 an explicit section to be placed in, and may
511 have an optional explicit alignment specified. A
512 variable may be defined as a global "constant," which indicates that the
513 contents of the variable will <b>never</b> be modified (enabling better
514 optimization, allowing the global data to be placed in the read-only section of
515 an executable, etc). Note that variables that need runtime initialization
516 cannot be marked "constant" as there is a store to the variable.</p>
519 LLVM explicitly allows <em>declarations</em> of global variables to be marked
520 constant, even if the final definition of the global is not. This capability
521 can be used to enable slightly better optimization of the program, but requires
522 the language definition to guarantee that optimizations based on the
523 'constantness' are valid for the translation units that do not include the
527 <p>As SSA values, global variables define pointer values that are in
528 scope (i.e. they dominate) all basic blocks in the program. Global
529 variables always define a pointer to their "content" type because they
530 describe a region of memory, and all memory objects in LLVM are
531 accessed through pointers.</p>
533 <p>LLVM allows an explicit section to be specified for globals. If the target
534 supports it, it will emit globals to the section specified.</p>
536 <p>An explicit alignment may be specified for a global. If not present, or if
537 the alignment is set to zero, the alignment of the global is set by the target
538 to whatever it feels convenient. If an explicit alignment is specified, the
539 global is forced to have at least that much alignment. All alignments must be
545 <!-- ======================================================================= -->
546 <div class="doc_subsection">
547 <a name="functionstructure">Functions</a>
550 <div class="doc_text">
552 <p>LLVM function definitions consist of an optional <a href="#linkage">linkage
553 type</a>, an optional <a href="#callingconv">calling convention</a>, a return
554 type, a function name, a (possibly empty) argument list, an optional section,
555 an optional alignment, an opening curly brace,
556 a list of basic blocks, and a closing curly brace. LLVM function declarations
557 are defined with the "<tt>declare</tt>" keyword, an optional <a
558 href="#callingconv">calling convention</a>, a return type, a function name,
559 a possibly empty list of arguments, and an optional alignment.</p>
561 <p>A function definition contains a list of basic blocks, forming the CFG for
562 the function. Each basic block may optionally start with a label (giving the
563 basic block a symbol table entry), contains a list of instructions, and ends
564 with a <a href="#terminators">terminator</a> instruction (such as a branch or
565 function return).</p>
567 <p>The first basic block in a program is special in two ways: it is immediately
568 executed on entrance to the function, and it is not allowed to have predecessor
569 basic blocks (i.e. there can not be any branches to the entry block of a
570 function). Because the block can have no predecessors, it also cannot have any
571 <a href="#i_phi">PHI nodes</a>.</p>
573 <p>LLVM functions are identified by their name and type signature. Hence, two
574 functions with the same name but different parameter lists or return values are
575 considered different functions, and LLVM will resolve references to each
578 <p>LLVM allows an explicit section to be specified for functions. If the target
579 supports it, it will emit functions to the section specified.</p>
581 <p>An explicit alignment may be specified for a function. If not present, or if
582 the alignment is set to zero, the alignment of the function is set by the target
583 to whatever it feels convenient. If an explicit alignment is specified, the
584 function is forced to have at least that much alignment. All alignments must be
589 <!-- ======================================================================= -->
590 <div class="doc_subsection">
591 <a name="moduleasm">Module-Level Inline Assembly</a></li>
594 <div class="doc_text">
596 Modules may contain "module-level inline asm" blocks, which corresponds to the
597 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
598 LLVM and treated as a single unit, but may be separated in the .ll file if
599 desired. The syntax is very simple:
602 <div class="doc_code"><pre>
603 module asm "inline asm code goes here"
604 module asm "more can go here"
607 <p>The strings can contain any character by escaping non-printable characters.
608 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
613 The inline asm code is simply printed to the machine code .s file when
614 assembly code is generated.
619 <!-- *********************************************************************** -->
620 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
621 <!-- *********************************************************************** -->
623 <div class="doc_text">
625 <p>The LLVM type system is one of the most important features of the
626 intermediate representation. Being typed enables a number of
627 optimizations to be performed on the IR directly, without having to do
628 extra analyses on the side before the transformation. A strong type
629 system makes it easier to read the generated code and enables novel
630 analyses and transformations that are not feasible to perform on normal
631 three address code representations.</p>
635 <!-- ======================================================================= -->
636 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
637 <div class="doc_text">
638 <p>The primitive types are the fundamental building blocks of the LLVM
639 system. The current set of primitive types is as follows:</p>
641 <table class="layout">
646 <tr><th>Type</th><th>Description</th></tr>
647 <tr><td><tt>void</tt></td><td>No value</td></tr>
648 <tr><td><tt>ubyte</tt></td><td>Unsigned 8-bit value</td></tr>
649 <tr><td><tt>ushort</tt></td><td>Unsigned 16-bit value</td></tr>
650 <tr><td><tt>uint</tt></td><td>Unsigned 32-bit value</td></tr>
651 <tr><td><tt>ulong</tt></td><td>Unsigned 64-bit value</td></tr>
652 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
653 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
660 <tr><th>Type</th><th>Description</th></tr>
661 <tr><td><tt>bool</tt></td><td>True or False value</td></tr>
662 <tr><td><tt>sbyte</tt></td><td>Signed 8-bit value</td></tr>
663 <tr><td><tt>short</tt></td><td>Signed 16-bit value</td></tr>
664 <tr><td><tt>int</tt></td><td>Signed 32-bit value</td></tr>
665 <tr><td><tt>long</tt></td><td>Signed 64-bit value</td></tr>
666 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
674 <!-- _______________________________________________________________________ -->
675 <div class="doc_subsubsection"> <a name="t_classifications">Type
676 Classifications</a> </div>
677 <div class="doc_text">
678 <p>These different primitive types fall into a few useful
681 <table border="1" cellspacing="0" cellpadding="4">
683 <tr><th>Classification</th><th>Types</th></tr>
685 <td><a name="t_signed">signed</a></td>
686 <td><tt>sbyte, short, int, long, float, double</tt></td>
689 <td><a name="t_unsigned">unsigned</a></td>
690 <td><tt>ubyte, ushort, uint, ulong</tt></td>
693 <td><a name="t_integer">integer</a></td>
694 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
697 <td><a name="t_integral">integral</a></td>
698 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt>
702 <td><a name="t_floating">floating point</a></td>
703 <td><tt>float, double</tt></td>
706 <td><a name="t_firstclass">first class</a></td>
707 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long,<br>
708 float, double, <a href="#t_pointer">pointer</a>,
709 <a href="#t_packed">packed</a></tt></td>
714 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
715 most important. Values of these types are the only ones which can be
716 produced by instructions, passed as arguments, or used as operands to
717 instructions. This means that all structures and arrays must be
718 manipulated either by pointer or by component.</p>
721 <!-- ======================================================================= -->
722 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
724 <div class="doc_text">
726 <p>The real power in LLVM comes from the derived types in the system.
727 This is what allows a programmer to represent arrays, functions,
728 pointers, and other useful types. Note that these derived types may be
729 recursive: For example, it is possible to have a two dimensional array.</p>
733 <!-- _______________________________________________________________________ -->
734 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
736 <div class="doc_text">
740 <p>The array type is a very simple derived type that arranges elements
741 sequentially in memory. The array type requires a size (number of
742 elements) and an underlying data type.</p>
747 [<# elements> x <elementtype>]
750 <p>The number of elements is a constant integer value; elementtype may
751 be any type with a size.</p>
754 <table class="layout">
757 <tt>[40 x int ]</tt><br/>
758 <tt>[41 x int ]</tt><br/>
759 <tt>[40 x uint]</tt><br/>
762 Array of 40 integer values.<br/>
763 Array of 41 integer values.<br/>
764 Array of 40 unsigned integer values.<br/>
768 <p>Here are some examples of multidimensional arrays:</p>
769 <table class="layout">
772 <tt>[3 x [4 x int]]</tt><br/>
773 <tt>[12 x [10 x float]]</tt><br/>
774 <tt>[2 x [3 x [4 x uint]]]</tt><br/>
777 3x4 array of integer values.<br/>
778 12x10 array of single precision floating point values.<br/>
779 2x3x4 array of unsigned integer values.<br/>
784 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
785 length array. Normally, accesses past the end of an array are undefined in
786 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
787 As a special case, however, zero length arrays are recognized to be variable
788 length. This allows implementation of 'pascal style arrays' with the LLVM
789 type "{ int, [0 x float]}", for example.</p>
793 <!-- _______________________________________________________________________ -->
794 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
795 <div class="doc_text">
797 <p>The function type can be thought of as a function signature. It
798 consists of a return type and a list of formal parameter types.
799 Function types are usually used to build virtual function tables
800 (which are structures of pointers to functions), for indirect function
801 calls, and when defining a function.</p>
803 The return type of a function type cannot be an aggregate type.
806 <pre> <returntype> (<parameter list>)<br></pre>
807 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
808 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
809 which indicates that the function takes a variable number of arguments.
810 Variable argument functions can access their arguments with the <a
811 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
813 <table class="layout">
816 <tt>int (int)</tt> <br/>
817 <tt>float (int, int *) *</tt><br/>
818 <tt>int (sbyte *, ...)</tt><br/>
821 function taking an <tt>int</tt>, returning an <tt>int</tt><br/>
822 <a href="#t_pointer">Pointer</a> to a function that takes an
823 <tt>int</tt> and a <a href="#t_pointer">pointer</a> to <tt>int</tt>,
824 returning <tt>float</tt>.<br/>
825 A vararg function that takes at least one <a href="#t_pointer">pointer</a>
826 to <tt>sbyte</tt> (signed char in C), which returns an integer. This is
827 the signature for <tt>printf</tt> in LLVM.<br/>
833 <!-- _______________________________________________________________________ -->
834 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
835 <div class="doc_text">
837 <p>The structure type is used to represent a collection of data members
838 together in memory. The packing of the field types is defined to match
839 the ABI of the underlying processor. The elements of a structure may
840 be any type that has a size.</p>
841 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
842 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
843 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
846 <pre> { <type list> }<br></pre>
848 <table class="layout">
851 <tt>{ int, int, int }</tt><br/>
852 <tt>{ float, int (int) * }</tt><br/>
855 a triple of three <tt>int</tt> values<br/>
856 A pair, where the first element is a <tt>float</tt> and the second element
857 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
858 that takes an <tt>int</tt>, returning an <tt>int</tt>.<br/>
864 <!-- _______________________________________________________________________ -->
865 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
866 <div class="doc_text">
868 <p>As in many languages, the pointer type represents a pointer or
869 reference to another object, which must live in memory.</p>
871 <pre> <type> *<br></pre>
873 <table class="layout">
876 <tt>[4x int]*</tt><br/>
877 <tt>int (int *) *</tt><br/>
880 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
881 four <tt>int</tt> values<br/>
882 A <a href="#t_pointer">pointer</a> to a <a
883 href="#t_function">function</a> that takes an <tt>int*</tt>, returning an
890 <!-- _______________________________________________________________________ -->
891 <div class="doc_subsubsection"> <a name="t_packed">Packed Type</a> </div>
892 <div class="doc_text">
896 <p>A packed type is a simple derived type that represents a vector
897 of elements. Packed types are used when multiple primitive data
898 are operated in parallel using a single instruction (SIMD).
899 A packed type requires a size (number of
900 elements) and an underlying primitive data type. Vectors must have a power
901 of two length (1, 2, 4, 8, 16 ...). Packed types are
902 considered <a href="#t_firstclass">first class</a>.</p>
907 < <# elements> x <elementtype> >
910 <p>The number of elements is a constant integer value; elementtype may
911 be any integral or floating point type.</p>
915 <table class="layout">
918 <tt><4 x int></tt><br/>
919 <tt><8 x float></tt><br/>
920 <tt><2 x uint></tt><br/>
923 Packed vector of 4 integer values.<br/>
924 Packed vector of 8 floating-point values.<br/>
925 Packed vector of 2 unsigned integer values.<br/>
931 <!-- _______________________________________________________________________ -->
932 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
933 <div class="doc_text">
937 <p>Opaque types are used to represent unknown types in the system. This
938 corresponds (for example) to the C notion of a foward declared structure type.
939 In LLVM, opaque types can eventually be resolved to any type (not just a
950 <table class="layout">
963 <!-- *********************************************************************** -->
964 <div class="doc_section"> <a name="constants">Constants</a> </div>
965 <!-- *********************************************************************** -->
967 <div class="doc_text">
969 <p>LLVM has several different basic types of constants. This section describes
970 them all and their syntax.</p>
974 <!-- ======================================================================= -->
975 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
977 <div class="doc_text">
980 <dt><b>Boolean constants</b></dt>
982 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
983 constants of the <tt><a href="#t_primitive">bool</a></tt> type.
986 <dt><b>Integer constants</b></dt>
988 <dd>Standard integers (such as '4') are constants of the <a
989 href="#t_integer">integer</a> type. Negative numbers may be used with signed
993 <dt><b>Floating point constants</b></dt>
995 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
996 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
997 notation (see below). Floating point constants must have a <a
998 href="#t_floating">floating point</a> type. </dd>
1000 <dt><b>Null pointer constants</b></dt>
1002 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1003 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1007 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1008 of floating point constants. For example, the form '<tt>double
1009 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1010 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1011 (and the only time that they are generated by the disassembler) is when a
1012 floating point constant must be emitted but it cannot be represented as a
1013 decimal floating point number. For example, NaN's, infinities, and other
1014 special values are represented in their IEEE hexadecimal format so that
1015 assembly and disassembly do not cause any bits to change in the constants.</p>
1019 <!-- ======================================================================= -->
1020 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1023 <div class="doc_text">
1024 <p>Aggregate constants arise from aggregation of simple constants
1025 and smaller aggregate constants.</p>
1028 <dt><b>Structure constants</b></dt>
1030 <dd>Structure constants are represented with notation similar to structure
1031 type definitions (a comma separated list of elements, surrounded by braces
1032 (<tt>{}</tt>)). For example: "<tt>{ int 4, float 17.0, int* %G }</tt>",
1033 where "<tt>%G</tt>" is declared as "<tt>%G = external global int</tt>". Structure constants
1034 must have <a href="#t_struct">structure type</a>, and the number and
1035 types of elements must match those specified by the type.
1038 <dt><b>Array constants</b></dt>
1040 <dd>Array constants are represented with notation similar to array type
1041 definitions (a comma separated list of elements, surrounded by square brackets
1042 (<tt>[]</tt>)). For example: "<tt>[ int 42, int 11, int 74 ]</tt>". Array
1043 constants must have <a href="#t_array">array type</a>, and the number and
1044 types of elements must match those specified by the type.
1047 <dt><b>Packed constants</b></dt>
1049 <dd>Packed constants are represented with notation similar to packed type
1050 definitions (a comma separated list of elements, surrounded by
1051 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< int 42,
1052 int 11, int 74, int 100 ></tt>". Packed constants must have <a
1053 href="#t_packed">packed type</a>, and the number and types of elements must
1054 match those specified by the type.
1057 <dt><b>Zero initialization</b></dt>
1059 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1060 value to zero of <em>any</em> type, including scalar and aggregate types.
1061 This is often used to avoid having to print large zero initializers (e.g. for
1062 large arrays) and is always exactly equivalent to using explicit zero
1069 <!-- ======================================================================= -->
1070 <div class="doc_subsection">
1071 <a name="globalconstants">Global Variable and Function Addresses</a>
1074 <div class="doc_text">
1076 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1077 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1078 constants. These constants are explicitly referenced when the <a
1079 href="#identifiers">identifier for the global</a> is used and always have <a
1080 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1086 %Z = global [2 x int*] [ int* %X, int* %Y ]
1091 <!-- ======================================================================= -->
1092 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1093 <div class="doc_text">
1094 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1095 no specific value. Undefined values may be of any type and be used anywhere
1096 a constant is permitted.</p>
1098 <p>Undefined values indicate to the compiler that the program is well defined
1099 no matter what value is used, giving the compiler more freedom to optimize.
1103 <!-- ======================================================================= -->
1104 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1107 <div class="doc_text">
1109 <p>Constant expressions are used to allow expressions involving other constants
1110 to be used as constants. Constant expressions may be of any <a
1111 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1112 that does not have side effects (e.g. load and call are not supported). The
1113 following is the syntax for constant expressions:</p>
1116 <dt><b><tt>cast ( CST to TYPE )</tt></b></dt>
1118 <dd>Cast a constant to another type.</dd>
1120 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1122 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1123 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1124 instruction, the index list may have zero or more indexes, which are required
1125 to make sense for the type of "CSTPTR".</dd>
1127 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1129 <dd>Perform the <a href="#i_select">select operation</a> on
1132 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1134 <dd>Perform the <a href="#i_extractelement">extractelement
1135 operation</a> on constants.
1137 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1139 <dd>Perform the <a href="#i_insertelement">insertelement
1140 operation</a> on constants.
1142 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1144 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1145 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1146 binary</a> operations. The constraints on operands are the same as those for
1147 the corresponding instruction (e.g. no bitwise operations on floating point
1148 values are allowed).</dd>
1152 <!-- *********************************************************************** -->
1153 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1154 <!-- *********************************************************************** -->
1156 <!-- ======================================================================= -->
1157 <div class="doc_subsection">
1158 <a name="inlineasm">Inline Assembler Expressions</a>
1161 <div class="doc_text">
1164 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1165 Module-Level Inline Assembly</a>) through the use of a special value. This
1166 value represents the inline assembler as a string (containing the instructions
1167 to emit), a list of operand constraints (stored as a string), and a flag that
1168 indicates whether or not the inline asm expression has side effects. An example
1169 inline assembler expression is:
1173 int(int) asm "bswap $0", "=r,r"
1177 Inline assembler expressions may <b>only</b> be used as the callee operand of
1178 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1182 %X = call int asm "<a href="#i_bswap">bswap</a> $0", "=r,r"(int %Y)
1186 Inline asms with side effects not visible in the constraint list must be marked
1187 as having side effects. This is done through the use of the
1188 '<tt>sideeffect</tt>' keyword, like so:
1192 call void asm sideeffect "eieio", ""()
1195 <p>TODO: The format of the asm and constraints string still need to be
1196 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1197 need to be documented).
1202 <!-- *********************************************************************** -->
1203 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1204 <!-- *********************************************************************** -->
1206 <div class="doc_text">
1208 <p>The LLVM instruction set consists of several different
1209 classifications of instructions: <a href="#terminators">terminator
1210 instructions</a>, <a href="#binaryops">binary instructions</a>,
1211 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1212 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1213 instructions</a>.</p>
1217 <!-- ======================================================================= -->
1218 <div class="doc_subsection"> <a name="terminators">Terminator
1219 Instructions</a> </div>
1221 <div class="doc_text">
1223 <p>As mentioned <a href="#functionstructure">previously</a>, every
1224 basic block in a program ends with a "Terminator" instruction, which
1225 indicates which block should be executed after the current block is
1226 finished. These terminator instructions typically yield a '<tt>void</tt>'
1227 value: they produce control flow, not values (the one exception being
1228 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1229 <p>There are six different terminator instructions: the '<a
1230 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1231 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1232 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1233 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1234 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1238 <!-- _______________________________________________________________________ -->
1239 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1240 Instruction</a> </div>
1241 <div class="doc_text">
1243 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1244 ret void <i>; Return from void function</i>
1247 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1248 value) from a function back to the caller.</p>
1249 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1250 returns a value and then causes control flow, and one that just causes
1251 control flow to occur.</p>
1253 <p>The '<tt>ret</tt>' instruction may return any '<a
1254 href="#t_firstclass">first class</a>' type. Notice that a function is
1255 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1256 instruction inside of the function that returns a value that does not
1257 match the return type of the function.</p>
1259 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1260 returns back to the calling function's context. If the caller is a "<a
1261 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1262 the instruction after the call. If the caller was an "<a
1263 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1264 at the beginning of the "normal" destination block. If the instruction
1265 returns a value, that value shall set the call or invoke instruction's
1268 <pre> ret int 5 <i>; Return an integer value of 5</i>
1269 ret void <i>; Return from a void function</i>
1272 <!-- _______________________________________________________________________ -->
1273 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1274 <div class="doc_text">
1276 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1279 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1280 transfer to a different basic block in the current function. There are
1281 two forms of this instruction, corresponding to a conditional branch
1282 and an unconditional branch.</p>
1284 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1285 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
1286 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
1287 value as a target.</p>
1289 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
1290 argument is evaluated. If the value is <tt>true</tt>, control flows
1291 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1292 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1294 <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
1295 href="#i_ret">ret</a> int 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> int 0<br></pre>
1297 <!-- _______________________________________________________________________ -->
1298 <div class="doc_subsubsection">
1299 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1302 <div class="doc_text">
1306 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1311 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1312 several different places. It is a generalization of the '<tt>br</tt>'
1313 instruction, allowing a branch to occur to one of many possible
1319 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1320 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1321 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1322 table is not allowed to contain duplicate constant entries.</p>
1326 <p>The <tt>switch</tt> instruction specifies a table of values and
1327 destinations. When the '<tt>switch</tt>' instruction is executed, this
1328 table is searched for the given value. If the value is found, control flow is
1329 transfered to the corresponding destination; otherwise, control flow is
1330 transfered to the default destination.</p>
1332 <h5>Implementation:</h5>
1334 <p>Depending on properties of the target machine and the particular
1335 <tt>switch</tt> instruction, this instruction may be code generated in different
1336 ways. For example, it could be generated as a series of chained conditional
1337 branches or with a lookup table.</p>
1342 <i>; Emulate a conditional br instruction</i>
1343 %Val = <a href="#i_cast">cast</a> bool %value to int
1344 switch int %Val, label %truedest [int 0, label %falsedest ]
1346 <i>; Emulate an unconditional br instruction</i>
1347 switch uint 0, label %dest [ ]
1349 <i>; Implement a jump table:</i>
1350 switch uint %val, label %otherwise [ uint 0, label %onzero
1351 uint 1, label %onone
1352 uint 2, label %ontwo ]
1356 <!-- _______________________________________________________________________ -->
1357 <div class="doc_subsubsection">
1358 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1361 <div class="doc_text">
1366 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1367 to label <normal label> except label <exception label>
1372 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1373 function, with the possibility of control flow transfer to either the
1374 '<tt>normal</tt>' label or the
1375 '<tt>exception</tt>' label. If the callee function returns with the
1376 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1377 "normal" label. If the callee (or any indirect callees) returns with the "<a
1378 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1379 continued at the dynamically nearest "exception" label.</p>
1383 <p>This instruction requires several arguments:</p>
1387 The optional "cconv" marker indicates which <a href="callingconv">calling
1388 convention</a> the call should use. If none is specified, the call defaults
1389 to using C calling conventions.
1391 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1392 function value being invoked. In most cases, this is a direct function
1393 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1394 an arbitrary pointer to function value.
1397 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1398 function to be invoked. </li>
1400 <li>'<tt>function args</tt>': argument list whose types match the function
1401 signature argument types. If the function signature indicates the function
1402 accepts a variable number of arguments, the extra arguments can be
1405 <li>'<tt>normal label</tt>': the label reached when the called function
1406 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1408 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1409 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1415 <p>This instruction is designed to operate as a standard '<tt><a
1416 href="#i_call">call</a></tt>' instruction in most regards. The primary
1417 difference is that it establishes an association with a label, which is used by
1418 the runtime library to unwind the stack.</p>
1420 <p>This instruction is used in languages with destructors to ensure that proper
1421 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1422 exception. Additionally, this is important for implementation of
1423 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1427 %retval = invoke int %Test(int 15) to label %Continue
1428 except label %TestCleanup <i>; {int}:retval set</i>
1429 %retval = invoke <a href="#callingconv">coldcc</a> int %Test(int 15) to label %Continue
1430 except label %TestCleanup <i>; {int}:retval set</i>
1435 <!-- _______________________________________________________________________ -->
1437 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1438 Instruction</a> </div>
1440 <div class="doc_text">
1449 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1450 at the first callee in the dynamic call stack which used an <a
1451 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1452 primarily used to implement exception handling.</p>
1456 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1457 immediately halt. The dynamic call stack is then searched for the first <a
1458 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1459 execution continues at the "exceptional" destination block specified by the
1460 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1461 dynamic call chain, undefined behavior results.</p>
1464 <!-- _______________________________________________________________________ -->
1466 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1467 Instruction</a> </div>
1469 <div class="doc_text">
1478 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1479 instruction is used to inform the optimizer that a particular portion of the
1480 code is not reachable. This can be used to indicate that the code after a
1481 no-return function cannot be reached, and other facts.</p>
1485 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1490 <!-- ======================================================================= -->
1491 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1492 <div class="doc_text">
1493 <p>Binary operators are used to do most of the computation in a
1494 program. They require two operands, execute an operation on them, and
1495 produce a single value. The operands might represent
1496 multiple data, as is the case with the <a href="#t_packed">packed</a> data type.
1497 The result value of a binary operator is not
1498 necessarily the same type as its operands.</p>
1499 <p>There are several different binary operators:</p>
1501 <!-- _______________________________________________________________________ -->
1502 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1503 Instruction</a> </div>
1504 <div class="doc_text">
1506 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1509 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1511 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1512 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1513 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1514 Both arguments must have identical types.</p>
1516 <p>The value produced is the integer or floating point sum of the two
1519 <pre> <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
1522 <!-- _______________________________________________________________________ -->
1523 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1524 Instruction</a> </div>
1525 <div class="doc_text">
1527 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1530 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1532 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1533 instruction present in most other intermediate representations.</p>
1535 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1536 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1538 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1539 Both arguments must have identical types.</p>
1541 <p>The value produced is the integer or floating point difference of
1542 the two operands.</p>
1544 <pre> <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
1545 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
1548 <!-- _______________________________________________________________________ -->
1549 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1550 Instruction</a> </div>
1551 <div class="doc_text">
1553 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1556 <p>The '<tt>mul</tt>' instruction returns the product of its two
1559 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1560 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1562 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1563 Both arguments must have identical types.</p>
1565 <p>The value produced is the integer or floating point product of the
1567 <p>There is no signed vs unsigned multiplication. The appropriate
1568 action is taken based on the type of the operand.</p>
1570 <pre> <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1573 <!-- _______________________________________________________________________ -->
1574 <div class="doc_subsubsection"> <a name="i_div">'<tt>div</tt>'
1575 Instruction</a> </div>
1576 <div class="doc_text">
1578 <pre> <result> = div <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1581 <p>The '<tt>div</tt>' instruction returns the quotient of its two
1584 <p>The two arguments to the '<tt>div</tt>' instruction must be either <a
1585 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1587 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1588 Both arguments must have identical types.</p>
1590 <p>The value produced is the integer or floating point quotient of the
1593 <pre> <result> = div int 4, %var <i>; yields {int}:result = 4 / %var</i>
1596 <!-- _______________________________________________________________________ -->
1597 <div class="doc_subsubsection"> <a name="i_rem">'<tt>rem</tt>'
1598 Instruction</a> </div>
1599 <div class="doc_text">
1601 <pre> <result> = rem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1604 <p>The '<tt>rem</tt>' instruction returns the remainder from the
1605 division of its two operands.</p>
1607 <p>The two arguments to the '<tt>rem</tt>' instruction must be either <a
1608 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1610 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1611 Both arguments must have identical types.</p>
1613 <p>This returns the <i>remainder</i> of a division (where the result
1614 has the same sign as the divisor), not the <i>modulus</i> (where the
1615 result has the same sign as the dividend) of a value. For more
1616 information about the difference, see <a
1617 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
1620 <pre> <result> = rem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1623 <!-- _______________________________________________________________________ -->
1624 <div class="doc_subsubsection"> <a name="i_setcc">'<tt>set<i>cc</i></tt>'
1625 Instructions</a> </div>
1626 <div class="doc_text">
1628 <pre> <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1629 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1630 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1631 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1632 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1633 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1636 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean
1637 value based on a comparison of their two operands.</p>
1639 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must
1640 be of <a href="#t_firstclass">first class</a> type (it is not possible
1641 to compare '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>'
1642 or '<tt>void</tt>' values, etc...). Both arguments must have identical
1645 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1646 value if both operands are equal.<br>
1647 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1648 value if both operands are unequal.<br>
1649 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1650 value if the first operand is less than the second operand.<br>
1651 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1652 value if the first operand is greater than the second operand.<br>
1653 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1654 value if the first operand is less than or equal to the second operand.<br>
1655 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1656 value if the first operand is greater than or equal to the second
1659 <pre> <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1660 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1661 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1662 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1663 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1664 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1667 <!-- ======================================================================= -->
1668 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
1669 Operations</a> </div>
1670 <div class="doc_text">
1671 <p>Bitwise binary operators are used to do various forms of
1672 bit-twiddling in a program. They are generally very efficient
1673 instructions and can commonly be strength reduced from other
1674 instructions. They require two operands, execute an operation on them,
1675 and produce a single value. The resulting value of the bitwise binary
1676 operators is always the same type as its first operand.</p>
1678 <!-- _______________________________________________________________________ -->
1679 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
1680 Instruction</a> </div>
1681 <div class="doc_text">
1683 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1686 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1687 its two operands.</p>
1689 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1690 href="#t_integral">integral</a> values. Both arguments must have
1691 identical types.</p>
1693 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1695 <div style="align: center">
1696 <table border="1" cellspacing="0" cellpadding="4">
1727 <pre> <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1728 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1729 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1732 <!-- _______________________________________________________________________ -->
1733 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
1734 <div class="doc_text">
1736 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1739 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1740 or of its two operands.</p>
1742 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1743 href="#t_integral">integral</a> values. Both arguments must have
1744 identical types.</p>
1746 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1748 <div style="align: center">
1749 <table border="1" cellspacing="0" cellpadding="4">
1780 <pre> <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1781 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1782 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1785 <!-- _______________________________________________________________________ -->
1786 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
1787 Instruction</a> </div>
1788 <div class="doc_text">
1790 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1793 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
1794 or of its two operands. The <tt>xor</tt> is used to implement the
1795 "one's complement" operation, which is the "~" operator in C.</p>
1797 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
1798 href="#t_integral">integral</a> values. Both arguments must have
1799 identical types.</p>
1801 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
1803 <div style="align: center">
1804 <table border="1" cellspacing="0" cellpadding="4">
1836 <pre> <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
1837 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
1838 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
1839 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
1842 <!-- _______________________________________________________________________ -->
1843 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
1844 Instruction</a> </div>
1845 <div class="doc_text">
1847 <pre> <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1850 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
1851 the left a specified number of bits.</p>
1853 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
1854 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1857 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
1859 <pre> <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
1860 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
1861 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
1864 <!-- _______________________________________________________________________ -->
1865 <div class="doc_subsubsection"> <a name="i_shr">'<tt>shr</tt>'
1866 Instruction</a> </div>
1867 <div class="doc_text">
1869 <pre> <result> = shr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1872 <p>The '<tt>shr</tt>' instruction returns the first operand shifted to
1873 the right a specified number of bits.</p>
1875 <p>The first argument to the '<tt>shr</tt>' instruction must be an <a
1876 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1879 <p>If the first argument is a <a href="#t_signed">signed</a> type, the
1880 most significant bit is duplicated in the newly free'd bit positions.
1881 If the first argument is unsigned, zero bits shall fill the empty
1884 <pre> <result> = shr int 4, ubyte %var <i>; yields {int}:result = 4 >> %var</i>
1885 <result> = shr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
1886 <result> = shr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
1887 <result> = shr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
1888 <result> = shr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
1892 <!-- ======================================================================= -->
1893 <div class="doc_subsection">
1894 <a name="memoryops">Memory Access Operations</a>
1897 <div class="doc_text">
1899 <p>A key design point of an SSA-based representation is how it
1900 represents memory. In LLVM, no memory locations are in SSA form, which
1901 makes things very simple. This section describes how to read, write,
1902 allocate, and free memory in LLVM.</p>
1906 <!-- _______________________________________________________________________ -->
1907 <div class="doc_subsubsection">
1908 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
1911 <div class="doc_text">
1916 <result> = malloc <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
1921 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
1922 heap and returns a pointer to it.</p>
1926 <p>The '<tt>malloc</tt>' instruction allocates
1927 <tt>sizeof(<type>)*NumElements</tt>
1928 bytes of memory from the operating system and returns a pointer of the
1929 appropriate type to the program. If "NumElements" is specified, it is the
1930 number of elements allocated. If an alignment is specified, the value result
1931 of the allocation is guaranteed to be aligned to at least that boundary. If
1932 not specified, or if zero, the target can choose to align the allocation on any
1933 convenient boundary.</p>
1935 <p>'<tt>type</tt>' must be a sized type.</p>
1939 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
1940 a pointer is returned.</p>
1945 %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
1947 %size = <a href="#i_add">add</a> uint 2, 2 <i>; yields {uint}:size = uint 4</i>
1948 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
1949 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
1950 %array3 = malloc int, uint 4, align 1024 <i>; yields {int*}:array3</i>
1951 %array4 = malloc int, align 1024 <i>; yields {int*}:array4</i>
1955 <!-- _______________________________________________________________________ -->
1956 <div class="doc_subsubsection">
1957 <a name="i_free">'<tt>free</tt>' Instruction</a>
1960 <div class="doc_text">
1965 free <type> <value> <i>; yields {void}</i>
1970 <p>The '<tt>free</tt>' instruction returns memory back to the unused
1971 memory heap to be reallocated in the future.</p>
1975 <p>'<tt>value</tt>' shall be a pointer value that points to a value
1976 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
1981 <p>Access to the memory pointed to by the pointer is no longer defined
1982 after this instruction executes.</p>
1987 %array = <a href="#i_malloc">malloc</a> [4 x ubyte] <i>; yields {[4 x ubyte]*}:array</i>
1988 free [4 x ubyte]* %array
1992 <!-- _______________________________________________________________________ -->
1993 <div class="doc_subsubsection">
1994 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
1997 <div class="doc_text">
2002 <result> = alloca <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2007 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
2008 stack frame of the procedure that is live until the current function
2009 returns to its caller.</p>
2013 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2014 bytes of memory on the runtime stack, returning a pointer of the
2015 appropriate type to the program. If "NumElements" is specified, it is the
2016 number of elements allocated. If an alignment is specified, the value result
2017 of the allocation is guaranteed to be aligned to at least that boundary. If
2018 not specified, or if zero, the target can choose to align the allocation on any
2019 convenient boundary.</p>
2021 <p>'<tt>type</tt>' may be any sized type.</p>
2025 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2026 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2027 instruction is commonly used to represent automatic variables that must
2028 have an address available. When the function returns (either with the <tt><a
2029 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2030 instructions), the memory is reclaimed.</p>
2035 %ptr = alloca int <i>; yields {int*}:ptr</i>
2036 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
2037 %ptr = alloca int, uint 4, align 1024 <i>; yields {int*}:ptr</i>
2038 %ptr = alloca int, align 1024 <i>; yields {int*}:ptr</i>
2042 <!-- _______________________________________________________________________ -->
2043 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2044 Instruction</a> </div>
2045 <div class="doc_text">
2047 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
2049 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2051 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2052 address from which to load. The pointer must point to a <a
2053 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2054 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2055 the number or order of execution of this <tt>load</tt> with other
2056 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2059 <p>The location of memory pointed to is loaded.</p>
2061 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2063 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2064 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2067 <!-- _______________________________________________________________________ -->
2068 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2069 Instruction</a> </div>
2071 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2072 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2075 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2077 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2078 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
2079 operand must be a pointer to the type of the '<tt><value></tt>'
2080 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2081 optimizer is not allowed to modify the number or order of execution of
2082 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2083 href="#i_store">store</a></tt> instructions.</p>
2085 <p>The contents of memory are updated to contain '<tt><value></tt>'
2086 at the location specified by the '<tt><pointer></tt>' operand.</p>
2088 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2090 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2091 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2093 <!-- _______________________________________________________________________ -->
2094 <div class="doc_subsubsection">
2095 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2098 <div class="doc_text">
2101 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2107 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2108 subelement of an aggregate data structure.</p>
2112 <p>This instruction takes a list of integer constants that indicate what
2113 elements of the aggregate object to index to. The actual types of the arguments
2114 provided depend on the type of the first pointer argument. The
2115 '<tt>getelementptr</tt>' instruction is used to index down through the type
2116 levels of a structure or to a specific index in an array. When indexing into a
2117 structure, only <tt>uint</tt>
2118 integer constants are allowed. When indexing into an array or pointer,
2119 <tt>int</tt> and <tt>long</tt> indexes are allowed of any sign.</p>
2121 <p>For example, let's consider a C code fragment and how it gets
2122 compiled to LLVM:</p>
2136 int *foo(struct ST *s) {
2137 return &s[1].Z.B[5][13];
2141 <p>The LLVM code generated by the GCC frontend is:</p>
2144 %RT = type { sbyte, [10 x [20 x int]], sbyte }
2145 %ST = type { int, double, %RT }
2149 int* %foo(%ST* %s) {
2151 %reg = getelementptr %ST* %s, int 1, uint 2, uint 1, int 5, int 13
2158 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2159 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2160 and <a href="#t_array">array</a> types require <tt>uint</tt>, <tt>int</tt>,
2161 <tt>ulong</tt>, or <tt>long</tt> values, and <a href="#t_struct">structure</a>
2162 types require <tt>uint</tt> <b>constants</b>.</p>
2164 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2165 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
2166 }</tt>' type, a structure. The second index indexes into the third element of
2167 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
2168 sbyte }</tt>' type, another structure. The third index indexes into the second
2169 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
2170 array. The two dimensions of the array are subscripted into, yielding an
2171 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2172 to this element, thus computing a value of '<tt>int*</tt>' type.</p>
2174 <p>Note that it is perfectly legal to index partially through a
2175 structure, returning a pointer to an inner element. Because of this,
2176 the LLVM code for the given testcase is equivalent to:</p>
2179 int* %foo(%ST* %s) {
2180 %t1 = getelementptr %ST* %s, int 1 <i>; yields %ST*:%t1</i>
2181 %t2 = getelementptr %ST* %t1, int 0, uint 2 <i>; yields %RT*:%t2</i>
2182 %t3 = getelementptr %RT* %t2, int 0, uint 1 <i>; yields [10 x [20 x int]]*:%t3</i>
2183 %t4 = getelementptr [10 x [20 x int]]* %t3, int 0, int 5 <i>; yields [20 x int]*:%t4</i>
2184 %t5 = getelementptr [20 x int]* %t4, int 0, int 13 <i>; yields int*:%t5</i>
2189 <p>Note that it is undefined to access an array out of bounds: array and
2190 pointer indexes must always be within the defined bounds of the array type.
2191 The one exception for this rules is zero length arrays. These arrays are
2192 defined to be accessible as variable length arrays, which requires access
2193 beyond the zero'th element.</p>
2198 <i>; yields [12 x ubyte]*:aptr</i>
2199 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, uint 1
2203 <!-- ======================================================================= -->
2204 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
2205 <div class="doc_text">
2206 <p>The instructions in this category are the "miscellaneous"
2207 instructions, which defy better classification.</p>
2209 <!-- _______________________________________________________________________ -->
2210 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
2211 Instruction</a> </div>
2212 <div class="doc_text">
2214 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
2216 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
2217 the SSA graph representing the function.</p>
2219 <p>The type of the incoming values are specified with the first type
2220 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
2221 as arguments, with one pair for each predecessor basic block of the
2222 current block. Only values of <a href="#t_firstclass">first class</a>
2223 type may be used as the value arguments to the PHI node. Only labels
2224 may be used as the label arguments.</p>
2225 <p>There must be no non-phi instructions between the start of a basic
2226 block and the PHI instructions: i.e. PHI instructions must be first in
2229 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
2230 value specified by the parameter, depending on which basic block we
2231 came from in the last <a href="#terminators">terminator</a> instruction.</p>
2233 <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>
2236 <!-- _______________________________________________________________________ -->
2237 <div class="doc_subsubsection">
2238 <a name="i_cast">'<tt>cast .. to</tt>' Instruction</a>
2241 <div class="doc_text">
2246 <result> = cast <ty> <value> to <ty2> <i>; yields ty2</i>
2252 The '<tt>cast</tt>' instruction is used as the primitive means to convert
2253 integers to floating point, change data type sizes, and break type safety (by
2261 The '<tt>cast</tt>' instruction takes a value to cast, which must be a first
2262 class value, and a type to cast it to, which must also be a <a
2263 href="#t_firstclass">first class</a> type.
2269 This instruction follows the C rules for explicit casts when determining how the
2270 data being cast must change to fit in its new container.
2274 When casting to bool, any value that would be considered true in the context of
2275 a C '<tt>if</tt>' condition is converted to the boolean '<tt>true</tt>' values,
2276 all else are '<tt>false</tt>'.
2280 When extending an integral value from a type of one signness to another (for
2281 example '<tt>sbyte</tt>' to '<tt>ulong</tt>'), the value is sign-extended if the
2282 <b>source</b> value is signed, and zero-extended if the source value is
2283 unsigned. <tt>bool</tt> values are always zero extended into either zero or
2290 %X = cast int 257 to ubyte <i>; yields ubyte:1</i>
2291 %Y = cast int 123 to bool <i>; yields bool:true</i>
2295 <!-- _______________________________________________________________________ -->
2296 <div class="doc_subsubsection">
2297 <a name="i_select">'<tt>select</tt>' Instruction</a>
2300 <div class="doc_text">
2305 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
2311 The '<tt>select</tt>' instruction is used to choose one value based on a
2312 condition, without branching.
2319 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.
2325 If the boolean condition evaluates to true, the instruction returns the first
2326 value argument; otherwise, it returns the second value argument.
2332 %X = select bool true, ubyte 17, ubyte 42 <i>; yields ubyte:17</i>
2337 <!-- _______________________________________________________________________ -->
2338 <div class="doc_subsubsection">
2339 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2342 <div class="doc_text">
2347 <result> = extractelement <n x <ty>> <val>, uint <idx> <i>; yields <ty></i>
2353 The '<tt>extractelement</tt>' instruction extracts a single scalar
2354 element from a packed vector at a specified index.
2361 The first operand of an '<tt>extractelement</tt>' instruction is a
2362 value of <a href="#t_packed">packed</a> type. The second operand is
2363 an index indicating the position from which to extract the element.
2364 The index may be a variable.</p>
2369 The result is a scalar of the same type as the element type of
2370 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2371 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2372 results are undefined.
2378 %result = extractelement <4 x int> %vec, uint 0 <i>; yields int</i>
2383 <!-- _______________________________________________________________________ -->
2384 <div class="doc_subsubsection">
2385 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2388 <div class="doc_text">
2393 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, uint <idx> <i>; yields <n x <ty>></i>
2399 The '<tt>insertelement</tt>' instruction inserts a scalar
2400 element into a packed vector at a specified index.
2407 The first operand of an '<tt>insertelement</tt>' instruction is a
2408 value of <a href="#t_packed">packed</a> type. The second operand is a
2409 scalar value whose type must equal the element type of the first
2410 operand. The third operand is an index indicating the position at
2411 which to insert the value. The index may be a variable.</p>
2416 The result is a packed vector of the same type as <tt>val</tt>. Its
2417 element values are those of <tt>val</tt> except at position
2418 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2419 exceeds the length of <tt>val</tt>, the results are undefined.
2425 %result = insertelement <4 x int> %vec, int 1, uint 0 <i>; yields <4 x int></i>
2430 <!-- _______________________________________________________________________ -->
2431 <div class="doc_subsubsection">
2432 <a name="i_call">'<tt>call</tt>' Instruction</a>
2435 <div class="doc_text">
2439 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
2444 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
2448 <p>This instruction requires several arguments:</p>
2452 <p>The optional "tail" marker indicates whether the callee function accesses
2453 any allocas or varargs in the caller. If the "tail" marker is present, the
2454 function call is eligible for tail call optimization. Note that calls may
2455 be marked "tail" even if they do not occur before a <a
2456 href="#i_ret"><tt>ret</tt></a> instruction.
2459 <p>The optional "cconv" marker indicates which <a href="callingconv">calling
2460 convention</a> the call should use. If none is specified, the call defaults
2461 to using C calling conventions.
2464 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
2465 being invoked. The argument types must match the types implied by this
2466 signature. This type can be omitted if the function is not varargs and
2467 if the function type does not return a pointer to a function.</p>
2470 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
2471 be invoked. In most cases, this is a direct function invocation, but
2472 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
2473 to function value.</p>
2476 <p>'<tt>function args</tt>': argument list whose types match the
2477 function signature argument types. All arguments must be of
2478 <a href="#t_firstclass">first class</a> type. If the function signature
2479 indicates the function accepts a variable number of arguments, the extra
2480 arguments can be specified.</p>
2486 <p>The '<tt>call</tt>' instruction is used to cause control flow to
2487 transfer to a specified function, with its incoming arguments bound to
2488 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
2489 instruction in the called function, control flow continues with the
2490 instruction after the function call, and the return value of the
2491 function is bound to the result argument. This is a simpler case of
2492 the <a href="#i_invoke">invoke</a> instruction.</p>
2497 %retval = call int %test(int %argc)
2498 call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);
2499 %X = tail call int %foo()
2500 %Y = tail call <a href="#callingconv">fastcc</a> int %foo()
2505 <!-- _______________________________________________________________________ -->
2506 <div class="doc_subsubsection">
2507 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
2510 <div class="doc_text">
2515 <resultval> = va_arg <va_list*> <arglist>, <argty>
2520 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
2521 the "variable argument" area of a function call. It is used to implement the
2522 <tt>va_arg</tt> macro in C.</p>
2526 <p>This instruction takes a <tt>va_list*</tt> value and the type of
2527 the argument. It returns a value of the specified argument type and
2528 increments the <tt>va_list</tt> to point to the next argument. Again, the
2529 actual type of <tt>va_list</tt> is target specific.</p>
2533 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
2534 type from the specified <tt>va_list</tt> and causes the
2535 <tt>va_list</tt> to point to the next argument. For more information,
2536 see the variable argument handling <a href="#int_varargs">Intrinsic
2539 <p>It is legal for this instruction to be called in a function which does not
2540 take a variable number of arguments, for example, the <tt>vfprintf</tt>
2543 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
2544 href="#intrinsics">intrinsic function</a> because it takes a type as an
2549 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
2553 <!-- *********************************************************************** -->
2554 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
2555 <!-- *********************************************************************** -->
2557 <div class="doc_text">
2559 <p>LLVM supports the notion of an "intrinsic function". These functions have
2560 well known names and semantics and are required to follow certain
2561 restrictions. Overall, these instructions represent an extension mechanism for
2562 the LLVM language that does not require changing all of the transformations in
2563 LLVM to add to the language (or the bytecode reader/writer, the parser,
2566 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
2567 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
2568 this. Intrinsic functions must always be external functions: you cannot define
2569 the body of intrinsic functions. Intrinsic functions may only be used in call
2570 or invoke instructions: it is illegal to take the address of an intrinsic
2571 function. Additionally, because intrinsic functions are part of the LLVM
2572 language, it is required that they all be documented here if any are added.</p>
2575 <p>To learn how to add an intrinsic function, please see the <a
2576 href="ExtendingLLVM.html">Extending LLVM Guide</a>.
2581 <!-- ======================================================================= -->
2582 <div class="doc_subsection">
2583 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
2586 <div class="doc_text">
2588 <p>Variable argument support is defined in LLVM with the <a
2589 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
2590 intrinsic functions. These functions are related to the similarly
2591 named macros defined in the <tt><stdarg.h></tt> header file.</p>
2593 <p>All of these functions operate on arguments that use a
2594 target-specific value type "<tt>va_list</tt>". The LLVM assembly
2595 language reference manual does not define what this type is, so all
2596 transformations should be prepared to handle intrinsics with any type
2599 <p>This example shows how the <a href="#i_vanext"><tt>vanext</tt></a>
2600 instruction and the variable argument handling intrinsic functions are
2604 int %test(int %X, ...) {
2605 ; Initialize variable argument processing
2607 call void %<a href="#i_va_start">llvm.va_start</a>(sbyte** %ap)
2609 ; Read a single integer argument
2610 %tmp = va_arg sbyte** %ap, int
2612 ; Demonstrate usage of llvm.va_copy and llvm.va_end
2614 call void %<a href="#i_va_copy">llvm.va_copy</a>(sbyte** %aq, sbyte** %ap)
2615 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %aq)
2617 ; Stop processing of arguments.
2618 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %ap)
2624 <!-- _______________________________________________________________________ -->
2625 <div class="doc_subsubsection">
2626 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
2630 <div class="doc_text">
2632 <pre> declare void %llvm.va_start(<va_list>* <arglist>)<br></pre>
2634 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
2635 <tt>*<arglist></tt> for subsequent use by <tt><a
2636 href="#i_va_arg">va_arg</a></tt>.</p>
2640 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
2644 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
2645 macro available in C. In a target-dependent way, it initializes the
2646 <tt>va_list</tt> element the argument points to, so that the next call to
2647 <tt>va_arg</tt> will produce the first variable argument passed to the function.
2648 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
2649 last argument of the function, the compiler can figure that out.</p>
2653 <!-- _______________________________________________________________________ -->
2654 <div class="doc_subsubsection">
2655 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
2658 <div class="doc_text">
2660 <pre> declare void %llvm.va_end(<va_list*> <arglist>)<br></pre>
2662 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
2663 which has been initialized previously with <tt><a href="#i_va_start">llvm.va_start</a></tt>
2664 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
2666 <p>The argument is a <tt>va_list</tt> to destroy.</p>
2668 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
2669 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
2670 Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and <a
2671 href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
2672 with calls to <tt>llvm.va_end</tt>.</p>
2675 <!-- _______________________________________________________________________ -->
2676 <div class="doc_subsubsection">
2677 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
2680 <div class="doc_text">
2685 declare void %llvm.va_copy(<va_list>* <destarglist>,
2686 <va_list>* <srcarglist>)
2691 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
2692 the source argument list to the destination argument list.</p>
2696 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
2697 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
2702 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
2703 available in C. In a target-dependent way, it copies the source
2704 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
2705 because the <tt><a href="i_va_begin">llvm.va_begin</a></tt> intrinsic may be
2706 arbitrarily complex and require memory allocation, for example.</p>
2710 <!-- ======================================================================= -->
2711 <div class="doc_subsection">
2712 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
2715 <div class="doc_text">
2718 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
2719 Collection</a> requires the implementation and generation of these intrinsics.
2720 These intrinsics allow identification of <a href="#i_gcroot">GC roots on the
2721 stack</a>, as well as garbage collector implementations that require <a
2722 href="#i_gcread">read</a> and <a href="#i_gcwrite">write</a> barriers.
2723 Front-ends for type-safe garbage collected languages should generate these
2724 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
2725 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
2729 <!-- _______________________________________________________________________ -->
2730 <div class="doc_subsubsection">
2731 <a name="i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
2734 <div class="doc_text">
2739 declare void %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
2744 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
2745 the code generator, and allows some metadata to be associated with it.</p>
2749 <p>The first argument specifies the address of a stack object that contains the
2750 root pointer. The second pointer (which must be either a constant or a global
2751 value address) contains the meta-data to be associated with the root.</p>
2755 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
2756 location. At compile-time, the code generator generates information to allow
2757 the runtime to find the pointer at GC safe points.
2763 <!-- _______________________________________________________________________ -->
2764 <div class="doc_subsubsection">
2765 <a name="i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
2768 <div class="doc_text">
2773 declare sbyte* %llvm.gcread(sbyte** %Ptr)
2778 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
2779 locations, allowing garbage collector implementations that require read
2784 <p>The argument is the address to read from, which should be an address
2785 allocated from the garbage collector.</p>
2789 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
2790 instruction, but may be replaced with substantially more complex code by the
2791 garbage collector runtime, as needed.</p>
2796 <!-- _______________________________________________________________________ -->
2797 <div class="doc_subsubsection">
2798 <a name="i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
2801 <div class="doc_text">
2806 declare void %llvm.gcwrite(sbyte* %P1, sbyte** %P2)
2811 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
2812 locations, allowing garbage collector implementations that require write
2813 barriers (such as generational or reference counting collectors).</p>
2817 <p>The first argument is the reference to store, and the second is the heap
2818 location to store to.</p>
2822 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
2823 instruction, but may be replaced with substantially more complex code by the
2824 garbage collector runtime, as needed.</p>
2830 <!-- ======================================================================= -->
2831 <div class="doc_subsection">
2832 <a name="int_codegen">Code Generator Intrinsics</a>
2835 <div class="doc_text">
2837 These intrinsics are provided by LLVM to expose special features that may only
2838 be implemented with code generator support.
2843 <!-- _______________________________________________________________________ -->
2844 <div class="doc_subsubsection">
2845 <a name="i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
2848 <div class="doc_text">
2852 declare sbyte *%llvm.returnaddress(uint <level>)
2858 The '<tt>llvm.returnaddress</tt>' intrinsic returns a target-specific value
2859 indicating the return address of the current function or one of its callers.
2865 The argument to this intrinsic indicates which function to return the address
2866 for. Zero indicates the calling function, one indicates its caller, etc. The
2867 argument is <b>required</b> to be a constant integer value.
2873 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
2874 the return address of the specified call frame, or zero if it cannot be
2875 identified. The value returned by this intrinsic is likely to be incorrect or 0
2876 for arguments other than zero, so it should only be used for debugging purposes.
2880 Note that calling this intrinsic does not prevent function inlining or other
2881 aggressive transformations, so the value returned may not be that of the obvious
2882 source-language caller.
2887 <!-- _______________________________________________________________________ -->
2888 <div class="doc_subsubsection">
2889 <a name="i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
2892 <div class="doc_text">
2896 declare sbyte *%llvm.frameaddress(uint <level>)
2902 The '<tt>llvm.frameaddress</tt>' intrinsic returns the target-specific frame
2903 pointer value for the specified stack frame.
2909 The argument to this intrinsic indicates which function to return the frame
2910 pointer for. Zero indicates the calling function, one indicates its caller,
2911 etc. The argument is <b>required</b> to be a constant integer value.
2917 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
2918 the frame address of the specified call frame, or zero if it cannot be
2919 identified. The value returned by this intrinsic is likely to be incorrect or 0
2920 for arguments other than zero, so it should only be used for debugging purposes.
2924 Note that calling this intrinsic does not prevent function inlining or other
2925 aggressive transformations, so the value returned may not be that of the obvious
2926 source-language caller.
2930 <!-- _______________________________________________________________________ -->
2931 <div class="doc_subsubsection">
2932 <a name="i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
2935 <div class="doc_text">
2939 declare sbyte *%llvm.stacksave()
2945 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
2946 the function stack, for use with <a href="#i_stackrestore">
2947 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
2948 features like scoped automatic variable sized arrays in C99.
2954 This intrinsic returns a opaque pointer value that can be passed to <a
2955 href="#i_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
2956 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
2957 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
2958 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
2959 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
2960 that were allocated after the <tt>llvm.stacksave</tt> was executed.
2965 <!-- _______________________________________________________________________ -->
2966 <div class="doc_subsubsection">
2967 <a name="i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
2970 <div class="doc_text">
2974 declare void %llvm.stackrestore(sbyte* %ptr)
2980 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
2981 the function stack to the state it was in when the corresponding <a
2982 href="#llvm.stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
2983 useful for implementing language features like scoped automatic variable sized
2990 See the description for <a href="#i_stacksave"><tt>llvm.stacksave</tt></a>.
2996 <!-- _______________________________________________________________________ -->
2997 <div class="doc_subsubsection">
2998 <a name="i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
3001 <div class="doc_text">
3005 declare void %llvm.prefetch(sbyte * <address>,
3006 uint <rw>, uint <locality>)
3013 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
3014 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
3016 effect on the behavior of the program but can change its performance
3023 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
3024 determining if the fetch should be for a read (0) or write (1), and
3025 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
3026 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
3027 <tt>locality</tt> arguments must be constant integers.
3033 This intrinsic does not modify the behavior of the program. In particular,
3034 prefetches cannot trap and do not produce a value. On targets that support this
3035 intrinsic, the prefetch can provide hints to the processor cache for better
3041 <!-- _______________________________________________________________________ -->
3042 <div class="doc_subsubsection">
3043 <a name="i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
3046 <div class="doc_text">
3050 declare void %llvm.pcmarker( uint <id> )
3057 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
3059 code to simulators and other tools. The method is target specific, but it is
3060 expected that the marker will use exported symbols to transmit the PC of the marker.
3061 The marker makes no guarantees that it will remain with any specific instruction
3062 after optimizations. It is possible that the presence of a marker will inhibit
3063 optimizations. The intended use is to be inserted after optmizations to allow
3064 correlations of simulation runs.
3070 <tt>id</tt> is a numerical id identifying the marker.
3076 This intrinsic does not modify the behavior of the program. Backends that do not
3077 support this intrinisic may ignore it.
3082 <!-- _______________________________________________________________________ -->
3083 <div class="doc_subsubsection">
3084 <a name="i_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
3087 <div class="doc_text">
3091 declare ulong %llvm.readcyclecounter( )
3098 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
3099 counter register (or similar low latency, high accuracy clocks) on those targets
3100 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
3101 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
3102 should only be used for small timings.
3108 When directly supported, reading the cycle counter should not modify any memory.
3109 Implementations are allowed to either return a application specific value or a
3110 system wide value. On backends without support, this is lowered to a constant 0.
3116 <!-- ======================================================================= -->
3117 <div class="doc_subsection">
3118 <a name="int_os">Operating System Intrinsics</a>
3121 <div class="doc_text">
3123 These intrinsics are provided by LLVM to support the implementation of
3124 operating system level code.
3129 <!-- _______________________________________________________________________ -->
3130 <div class="doc_subsubsection">
3131 <a name="i_readport">'<tt>llvm.readport</tt>' Intrinsic</a>
3134 <div class="doc_text">
3138 declare <integer type> %llvm.readport (<integer type> <address>)
3144 The '<tt>llvm.readport</tt>' intrinsic reads data from the specified hardware
3151 The argument to this intrinsic indicates the hardware I/O address from which
3152 to read the data. The address is in the hardware I/O address namespace (as
3153 opposed to being a memory location for memory mapped I/O).
3159 The '<tt>llvm.readport</tt>' intrinsic reads data from the hardware I/O port
3160 specified by <i>address</i> and returns the value. The address and return
3161 value must be integers, but the size is dependent upon the platform upon which
3162 the program is code generated. For example, on x86, the address must be an
3163 unsigned 16-bit value, and the return value must be 8, 16, or 32 bits.
3168 <!-- _______________________________________________________________________ -->
3169 <div class="doc_subsubsection">
3170 <a name="i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a>
3173 <div class="doc_text">
3177 call void (<integer type>, <integer type>)*
3178 %llvm.writeport (<integer type> <value>,
3179 <integer type> <address>)
3185 The '<tt>llvm.writeport</tt>' intrinsic writes data to the specified hardware
3192 The first argument is the value to write to the I/O port.
3196 The second argument indicates the hardware I/O address to which data should be
3197 written. The address is in the hardware I/O address namespace (as opposed to
3198 being a memory location for memory mapped I/O).
3204 The '<tt>llvm.writeport</tt>' intrinsic writes <i>value</i> to the I/O port
3205 specified by <i>address</i>. The address and value must be integers, but the
3206 size is dependent upon the platform upon which the program is code generated.
3207 For example, on x86, the address must be an unsigned 16-bit value, and the
3208 value written must be 8, 16, or 32 bits in length.
3213 <!-- _______________________________________________________________________ -->
3214 <div class="doc_subsubsection">
3215 <a name="i_readio">'<tt>llvm.readio</tt>' Intrinsic</a>
3218 <div class="doc_text">
3222 declare <result> %llvm.readio (<ty> * <pointer>)
3228 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
3235 The argument to this intrinsic is a pointer indicating the memory address from
3236 which to read the data. The data must be a
3237 <a href="#t_firstclass">first class</a> type.
3243 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
3244 location specified by <i>pointer</i> and returns the value. The argument must
3245 be a pointer, and the return value must be a
3246 <a href="#t_firstclass">first class</a> type. However, certain architectures
3247 may not support I/O on all first class types. For example, 32-bit processors
3248 may only support I/O on data types that are 32 bits or less.
3252 This intrinsic enforces an in-order memory model for llvm.readio and
3253 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
3254 scheduled processors may execute loads and stores out of order, re-ordering at
3255 run time accesses to memory mapped I/O registers. Using these intrinsics
3256 ensures that accesses to memory mapped I/O registers occur in program order.
3261 <!-- _______________________________________________________________________ -->
3262 <div class="doc_subsubsection">
3263 <a name="i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a>
3266 <div class="doc_text">
3270 declare void %llvm.writeio (<ty1> <value>, <ty2> * <pointer>)
3276 The '<tt>llvm.writeio</tt>' intrinsic writes data to the specified memory
3283 The first argument is the value to write to the memory mapped I/O location.
3284 The second argument is a pointer indicating the memory address to which the
3285 data should be written.
3291 The '<tt>llvm.writeio</tt>' intrinsic writes <i>value</i> to the memory mapped
3292 I/O address specified by <i>pointer</i>. The value must be a
3293 <a href="#t_firstclass">first class</a> type. However, certain architectures
3294 may not support I/O on all first class types. For example, 32-bit processors
3295 may only support I/O on data types that are 32 bits or less.
3299 This intrinsic enforces an in-order memory model for llvm.readio and
3300 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
3301 scheduled processors may execute loads and stores out of order, re-ordering at
3302 run time accesses to memory mapped I/O registers. Using these intrinsics
3303 ensures that accesses to memory mapped I/O registers occur in program order.
3308 <!-- ======================================================================= -->
3309 <div class="doc_subsection">
3310 <a name="int_libc">Standard C Library Intrinsics</a>
3313 <div class="doc_text">
3315 LLVM provides intrinsics for a few important standard C library functions.
3316 These intrinsics allow source-language front-ends to pass information about the
3317 alignment of the pointer arguments to the code generator, providing opportunity
3318 for more efficient code generation.
3323 <!-- _______________________________________________________________________ -->
3324 <div class="doc_subsubsection">
3325 <a name="i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
3328 <div class="doc_text">
3332 declare void %llvm.memcpy(sbyte* <dest>, sbyte* <src>,
3333 uint <len>, uint <align>)
3339 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
3340 location to the destination location.
3344 Note that, unlike the standard libc function, the <tt>llvm.memcpy</tt> intrinsic
3345 does not return a value, and takes an extra alignment argument.
3351 The first argument is a pointer to the destination, the second is a pointer to
3352 the source. The third argument is an (arbitrarily sized) integer argument
3353 specifying the number of bytes to copy, and the fourth argument is the alignment
3354 of the source and destination locations.
3358 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3359 the caller guarantees that the size of the copy is a multiple of the alignment
3360 and that both the source and destination pointers are aligned to that boundary.
3366 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
3367 location to the destination location, which are not allowed to overlap. It
3368 copies "len" bytes of memory over. If the argument is known to be aligned to
3369 some boundary, this can be specified as the fourth argument, otherwise it should
3375 <!-- _______________________________________________________________________ -->
3376 <div class="doc_subsubsection">
3377 <a name="i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
3380 <div class="doc_text">
3384 declare void %llvm.memmove(sbyte* <dest>, sbyte* <src>,
3385 uint <len>, uint <align>)
3391 The '<tt>llvm.memmove</tt>' intrinsic moves a block of memory from the source
3392 location to the destination location. It is similar to the '<tt>llvm.memcpy</tt>'
3393 intrinsic but allows the two memory locations to overlap.
3397 Note that, unlike the standard libc function, the <tt>llvm.memmove</tt> intrinsic
3398 does not return a value, and takes an extra alignment argument.
3404 The first argument is a pointer to the destination, the second is a pointer to
3405 the source. The third argument is an (arbitrarily sized) integer argument
3406 specifying the number of bytes to copy, and the fourth argument is the alignment
3407 of the source and destination locations.
3411 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3412 the caller guarantees that the size of the copy is a multiple of the alignment
3413 and that both the source and destination pointers are aligned to that boundary.
3419 The '<tt>llvm.memmove</tt>' intrinsic copies a block of memory from the source
3420 location to the destination location, which may overlap. It
3421 copies "len" bytes of memory over. If the argument is known to be aligned to
3422 some boundary, this can be specified as the fourth argument, otherwise it should
3428 <!-- _______________________________________________________________________ -->
3429 <div class="doc_subsubsection">
3430 <a name="i_memset">'<tt>llvm.memset</tt>' Intrinsic</a>
3433 <div class="doc_text">
3437 declare void %llvm.memset(sbyte* <dest>, ubyte <val>,
3438 uint <len>, uint <align>)
3444 The '<tt>llvm.memset</tt>' intrinsic fills a block of memory with a particular
3449 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
3450 does not return a value, and takes an extra alignment argument.
3456 The first argument is a pointer to the destination to fill, the second is the
3457 byte value to fill it with, the third argument is an (arbitrarily sized) integer
3458 argument specifying the number of bytes to fill, and the fourth argument is the
3459 known alignment of destination location.
3463 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3464 the caller guarantees that the size of the copy is a multiple of the alignment
3465 and that the destination pointer is aligned to that boundary.
3471 The '<tt>llvm.memset</tt>' intrinsic fills "len" bytes of memory starting at the
3472 destination location. If the argument is known to be aligned to some boundary,
3473 this can be specified as the fourth argument, otherwise it should be set to 0 or
3479 <!-- _______________________________________________________________________ -->
3480 <div class="doc_subsubsection">
3481 <a name="i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a>
3484 <div class="doc_text">
3488 declare bool %llvm.isunordered.f32(float Val1, float Val2)
3489 declare bool %llvm.isunordered.f64(double Val1, double Val2)
3495 The '<tt>llvm.isunordered</tt>' intrinsics return true if either or both of the
3496 specified floating point values is a NAN.
3502 The arguments are floating point numbers of the same type.
3508 If either or both of the arguments is a SNAN or QNAN, it returns true, otherwise
3514 <!-- _______________________________________________________________________ -->
3515 <div class="doc_subsubsection">
3516 <a name="i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
3519 <div class="doc_text">
3523 declare double %llvm.sqrt.f32(float Val)
3524 declare double %llvm.sqrt.f64(double Val)
3530 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
3531 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
3532 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
3533 negative numbers (which allows for better optimization).
3539 The argument and return value are floating point numbers of the same type.
3545 This function returns the sqrt of the specified operand if it is a positive
3546 floating point number.
3550 <!-- ======================================================================= -->
3551 <div class="doc_subsection">
3552 <a name="int_manip">Bit Manipulation Intrinsics</a>
3555 <div class="doc_text">
3557 LLVM provides intrinsics for a few important bit manipulation operations.
3558 These allow efficient code generation for some algorithms.
3563 <!-- _______________________________________________________________________ -->
3564 <div class="doc_subsubsection">
3565 <a name="i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
3568 <div class="doc_text">
3572 declare ushort %llvm.bswap.i16(ushort <id>)
3573 declare uint %llvm.bswap.i32(uint <id>)
3574 declare ulong %llvm.bswap.i64(ulong <id>)
3580 The '<tt>llvm.bwsap</tt>' family of intrinsics is used to byteswap a 16, 32 or
3581 64 bit quantity. These are useful for performing operations on data that is not
3582 in the target's native byte order.
3588 The <tt>llvm.bswap.16</tt> intrinsic returns a ushort value that has the high and low
3589 byte of the input ushort swapped. Similarly, the <tt>llvm.bswap.i32</tt> intrinsic
3590 returns a uint value that has the four bytes of the input uint swapped, so that
3591 if the input bytes are numbered 0, 1, 2, 3 then the returned uint will have its
3592 bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i64</tt> intrinsic extends this concept
3598 <!-- _______________________________________________________________________ -->
3599 <div class="doc_subsubsection">
3600 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
3603 <div class="doc_text">
3607 declare ubyte %llvm.ctpop.i8 (ubyte <src>)
3608 declare ushort %llvm.ctpop.i16(ushort <src>)
3609 declare uint %llvm.ctpop.i32(uint <src>)
3610 declare ulong %llvm.ctpop.i64(ulong <src>)
3616 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
3623 The only argument is the value to be counted. The argument may be of any
3624 unsigned integer type. The return type must match the argument type.
3630 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
3634 <!-- _______________________________________________________________________ -->
3635 <div class="doc_subsubsection">
3636 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
3639 <div class="doc_text">
3643 declare ubyte %llvm.ctlz.i8 (ubyte <src>)
3644 declare ushort %llvm.ctlz.i16(ushort <src>)
3645 declare uint %llvm.ctlz.i32(uint <src>)
3646 declare ulong %llvm.ctlz.i64(ulong <src>)
3652 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
3653 leading zeros in a variable.
3659 The only argument is the value to be counted. The argument may be of any
3660 unsigned integer type. The return type must match the argument type.
3666 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
3667 in a variable. If the src == 0 then the result is the size in bits of the type
3668 of src. For example, <tt>llvm.cttz(int 2) = 30</tt>.
3674 <!-- _______________________________________________________________________ -->
3675 <div class="doc_subsubsection">
3676 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
3679 <div class="doc_text">
3683 declare ubyte %llvm.cttz.i8 (ubyte <src>)
3684 declare ushort %llvm.cttz.i16(ushort <src>)
3685 declare uint %llvm.cttz.i32(uint <src>)
3686 declare ulong %llvm.cttz.i64(ulong <src>)
3692 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
3699 The only argument is the value to be counted. The argument may be of any
3700 unsigned integer type. The return type must match the argument type.
3706 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
3707 in a variable. If the src == 0 then the result is the size in bits of the type
3708 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
3712 <!-- ======================================================================= -->
3713 <div class="doc_subsection">
3714 <a name="int_debugger">Debugger Intrinsics</a>
3717 <div class="doc_text">
3719 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
3720 are described in the <a
3721 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
3722 Debugging</a> document.
3727 <!-- *********************************************************************** -->
3730 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3731 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
3732 <a href="http://validator.w3.org/check/referer"><img
3733 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
3735 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3736 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br>
3737 Last modified: $Date$