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_udiv">'<tt>udiv</tt>' Instruction</a></li>
81 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
82 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
83 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
84 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
85 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
86 <li><a href="#i_setcc">'<tt>set<i>cc</i></tt>' Instructions</a></li>
89 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
91 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
92 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
93 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
94 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
95 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
96 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
99 <li><a href="#vectorops">Vector Operations</a>
101 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
102 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
103 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
106 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
108 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
109 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
110 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
111 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
112 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
113 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
116 <li><a href="#convertops">Conversion Operations</a>
118 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
119 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
120 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
121 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
122 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
123 <li><a href="#i_fp2uint">'<tt>fp2uint .. to</tt>' Instruction</a></li>
124 <li><a href="#i_fp2sint">'<tt>fp2sint .. to</tt>' Instruction</a></li>
125 <li><a href="#i_uint2fp">'<tt>uint2fp .. to</tt>' Instruction</a></li>
126 <li><a href="#i_sint2fp">'<tt>sint2fp .. to</tt>' Instruction</a></li>
127 <li><a href="#i_bitconvert">'<tt>bitconvert .. to</tt>' Instruction</a></li>
129 <li><a href="#otherops">Other Operations</a>
131 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
132 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
133 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
134 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
139 <li><a href="#intrinsics">Intrinsic Functions</a>
141 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
143 <li><a href="#i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
144 <li><a href="#i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
145 <li><a href="#i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
148 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
150 <li><a href="#i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
151 <li><a href="#i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
152 <li><a href="#i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
155 <li><a href="#int_codegen">Code Generator Intrinsics</a>
157 <li><a href="#i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
158 <li><a href="#i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
159 <li><a href="#i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
160 <li><a href="#i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
161 <li><a href="#i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
162 <li><a href="#i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
163 <li><a href="#i_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
166 <li><a href="#int_libc">Standard C Library Intrinsics</a>
168 <li><a href="#i_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
169 <li><a href="#i_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
170 <li><a href="#i_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
171 <li><a href="#i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a></li>
172 <li><a href="#i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
173 <li><a href="#i_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
176 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
178 <li><a href="#i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
179 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
180 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
181 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
184 <li><a href="#int_debugger">Debugger intrinsics</a></li>
189 <div class="doc_author">
190 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
191 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
194 <!-- *********************************************************************** -->
195 <div class="doc_section"> <a name="abstract">Abstract </a></div>
196 <!-- *********************************************************************** -->
198 <div class="doc_text">
199 <p>This document is a reference manual for the LLVM assembly language.
200 LLVM is an SSA based representation that provides type safety,
201 low-level operations, flexibility, and the capability of representing
202 'all' high-level languages cleanly. It is the common code
203 representation used throughout all phases of the LLVM compilation
207 <!-- *********************************************************************** -->
208 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
209 <!-- *********************************************************************** -->
211 <div class="doc_text">
213 <p>The LLVM code representation is designed to be used in three
214 different forms: as an in-memory compiler IR, as an on-disk bytecode
215 representation (suitable for fast loading by a Just-In-Time compiler),
216 and as a human readable assembly language representation. This allows
217 LLVM to provide a powerful intermediate representation for efficient
218 compiler transformations and analysis, while providing a natural means
219 to debug and visualize the transformations. The three different forms
220 of LLVM are all equivalent. This document describes the human readable
221 representation and notation.</p>
223 <p>The LLVM representation aims to be light-weight and low-level
224 while being expressive, typed, and extensible at the same time. It
225 aims to be a "universal IR" of sorts, by being at a low enough level
226 that high-level ideas may be cleanly mapped to it (similar to how
227 microprocessors are "universal IR's", allowing many source languages to
228 be mapped to them). By providing type information, LLVM can be used as
229 the target of optimizations: for example, through pointer analysis, it
230 can be proven that a C automatic variable is never accessed outside of
231 the current function... allowing it to be promoted to a simple SSA
232 value instead of a memory location.</p>
236 <!-- _______________________________________________________________________ -->
237 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
239 <div class="doc_text">
241 <p>It is important to note that this document describes 'well formed'
242 LLVM assembly language. There is a difference between what the parser
243 accepts and what is considered 'well formed'. For example, the
244 following instruction is syntactically okay, but not well formed:</p>
247 %x = <a href="#i_add">add</a> int 1, %x
250 <p>...because the definition of <tt>%x</tt> does not dominate all of
251 its uses. The LLVM infrastructure provides a verification pass that may
252 be used to verify that an LLVM module is well formed. This pass is
253 automatically run by the parser after parsing input assembly and by
254 the optimizer before it outputs bytecode. The violations pointed out
255 by the verifier pass indicate bugs in transformation passes or input to
258 <!-- Describe the typesetting conventions here. --> </div>
260 <!-- *********************************************************************** -->
261 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
262 <!-- *********************************************************************** -->
264 <div class="doc_text">
266 <p>LLVM uses three different forms of identifiers, for different
270 <li>Named values are represented as a string of characters with a '%' prefix.
271 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
272 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
273 Identifiers which require other characters in their names can be surrounded
274 with quotes. In this way, anything except a <tt>"</tt> character can be used
277 <li>Unnamed values are represented as an unsigned numeric value with a '%'
278 prefix. For example, %12, %2, %44.</li>
280 <li>Constants, which are described in a <a href="#constants">section about
281 constants</a>, below.</li>
284 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
285 don't need to worry about name clashes with reserved words, and the set of
286 reserved words may be expanded in the future without penalty. Additionally,
287 unnamed identifiers allow a compiler to quickly come up with a temporary
288 variable without having to avoid symbol table conflicts.</p>
290 <p>Reserved words in LLVM are very similar to reserved words in other
291 languages. There are keywords for different opcodes ('<tt><a
292 href="#i_add">add</a></tt>', '<tt><a href="#i_cast">cast</a></tt>', '<tt><a
293 href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
294 href="#t_void">void</a></tt>', '<tt><a href="#t_uint">uint</a></tt>', etc...),
295 and others. These reserved words cannot conflict with variable names, because
296 none of them start with a '%' character.</p>
298 <p>Here is an example of LLVM code to multiply the integer variable
299 '<tt>%X</tt>' by 8:</p>
304 %result = <a href="#i_mul">mul</a> uint %X, 8
307 <p>After strength reduction:</p>
310 %result = <a href="#i_shl">shl</a> uint %X, ubyte 3
313 <p>And the hard way:</p>
316 <a href="#i_add">add</a> uint %X, %X <i>; yields {uint}:%0</i>
317 <a href="#i_add">add</a> uint %0, %0 <i>; yields {uint}:%1</i>
318 %result = <a href="#i_add">add</a> uint %1, %1
321 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
322 important lexical features of LLVM:</p>
326 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
329 <li>Unnamed temporaries are created when the result of a computation is not
330 assigned to a named value.</li>
332 <li>Unnamed temporaries are numbered sequentially</li>
336 <p>...and it also shows a convention that we follow in this document. When
337 demonstrating instructions, we will follow an instruction with a comment that
338 defines the type and name of value produced. Comments are shown in italic
343 <!-- *********************************************************************** -->
344 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
345 <!-- *********************************************************************** -->
347 <!-- ======================================================================= -->
348 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
351 <div class="doc_text">
353 <p>LLVM programs are composed of "Module"s, each of which is a
354 translation unit of the input programs. Each module consists of
355 functions, global variables, and symbol table entries. Modules may be
356 combined together with the LLVM linker, which merges function (and
357 global variable) definitions, resolves forward declarations, and merges
358 symbol table entries. Here is an example of the "hello world" module:</p>
360 <pre><i>; Declare the string constant as a global constant...</i>
361 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
362 href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
364 <i>; External declaration of the puts function</i>
365 <a href="#functionstructure">declare</a> int %puts(sbyte*) <i>; int(sbyte*)* </i>
367 <i>; Global variable / Function body section separator</i>
370 <i>; Definition of main function</i>
371 int %main() { <i>; int()* </i>
372 <i>; Convert [13x sbyte]* to sbyte *...</i>
374 href="#i_getelementptr">getelementptr</a> [13 x sbyte]* %.LC0, long 0, long 0 <i>; sbyte*</i>
376 <i>; Call puts function to write out the string to stdout...</i>
378 href="#i_call">call</a> int %puts(sbyte* %cast210) <i>; int</i>
380 href="#i_ret">ret</a> int 0<br>}<br></pre>
382 <p>This example is made up of a <a href="#globalvars">global variable</a>
383 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
384 function, and a <a href="#functionstructure">function definition</a>
385 for "<tt>main</tt>".</p>
387 <p>In general, a module is made up of a list of global values,
388 where both functions and global variables are global values. Global values are
389 represented by a pointer to a memory location (in this case, a pointer to an
390 array of char, and a pointer to a function), and have one of the following <a
391 href="#linkage">linkage types</a>.</p>
393 <p>Due to a limitation in the current LLVM assembly parser (it is limited by
394 one-token lookahead), modules are split into two pieces by the "implementation"
395 keyword. Global variable prototypes and definitions must occur before the
396 keyword, and function definitions must occur after it. Function prototypes may
397 occur either before or after it. In the future, the implementation keyword may
398 become a noop, if the parser gets smarter.</p>
402 <!-- ======================================================================= -->
403 <div class="doc_subsection">
404 <a name="linkage">Linkage Types</a>
407 <div class="doc_text">
410 All Global Variables and Functions have one of the following types of linkage:
415 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
417 <dd>Global values with internal linkage are only directly accessible by
418 objects in the current module. In particular, linking code into a module with
419 an internal global value may cause the internal to be renamed as necessary to
420 avoid collisions. Because the symbol is internal to the module, all
421 references can be updated. This corresponds to the notion of the
422 '<tt>static</tt>' keyword in C, or the idea of "anonymous namespaces" in C++.
425 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
427 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt> linkage, with
428 the twist that linking together two modules defining the same
429 <tt>linkonce</tt> globals will cause one of the globals to be discarded. This
430 is typically used to implement inline functions. Unreferenced
431 <tt>linkonce</tt> globals are allowed to be discarded.
434 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
436 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
437 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
438 used to implement constructs in C such as "<tt>int X;</tt>" at global scope.
441 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
443 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
444 pointer to array type. When two global variables with appending linkage are
445 linked together, the two global arrays are appended together. This is the
446 LLVM, typesafe, equivalent of having the system linker append together
447 "sections" with identical names when .o files are linked.
450 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
452 <dd>If none of the above identifiers are used, the global is externally
453 visible, meaning that it participates in linkage and can be used to resolve
454 external symbol references.
457 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
459 <dd>"<tt>extern_weak</tt>" TBD
463 The next two types of linkage are targeted for Microsoft Windows platform
464 only. They are designed to support importing (exporting) symbols from (to)
468 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
470 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
471 or variable via a global pointer to a pointer that is set up by the DLL
472 exporting the symbol. On Microsoft Windows targets, the pointer name is
473 formed by combining <code>_imp__</code> and the function or variable name.
476 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
478 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
479 pointer to a pointer in a DLL, so that it can be referenced with the
480 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
481 name is formed by combining <code>_imp__</code> and the function or variable
487 <p><a name="linkage_external">For example, since the "<tt>.LC0</tt>"
488 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
489 variable and was linked with this one, one of the two would be renamed,
490 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
491 external (i.e., lacking any linkage declarations), they are accessible
492 outside of the current module. It is illegal for a function <i>declaration</i>
493 to have any linkage type other than "externally visible".</a></p>
497 <!-- ======================================================================= -->
498 <div class="doc_subsection">
499 <a name="callingconv">Calling Conventions</a>
502 <div class="doc_text">
504 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
505 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
506 specified for the call. The calling convention of any pair of dynamic
507 caller/callee must match, or the behavior of the program is undefined. The
508 following calling conventions are supported by LLVM, and more may be added in
512 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
514 <dd>This calling convention (the default if no other calling convention is
515 specified) matches the target C calling conventions. This calling convention
516 supports varargs function calls and tolerates some mismatch in the declared
517 prototype and implemented declaration of the function (as does normal C).
520 <dt><b>"<tt>csretcc</tt>" - The C struct return calling convention</b>:</dt>
522 <dd>This calling convention matches the target C calling conventions, except
523 that functions with this convention are required to take a pointer as their
524 first argument, and the return type of the function must be void. This is
525 used for C functions that return aggregates by-value. In this case, the
526 function has been transformed to take a pointer to the struct as the first
527 argument to the function. For targets where the ABI specifies specific
528 behavior for structure-return calls, the calling convention can be used to
529 distinguish between struct return functions and other functions that take a
530 pointer to a struct as the first argument.
533 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
535 <dd>This calling convention attempts to make calls as fast as possible
536 (e.g. by passing things in registers). This calling convention allows the
537 target to use whatever tricks it wants to produce fast code for the target,
538 without having to conform to an externally specified ABI. Implementations of
539 this convention should allow arbitrary tail call optimization to be supported.
540 This calling convention does not support varargs and requires the prototype of
541 all callees to exactly match the prototype of the function definition.
544 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
546 <dd>This calling convention attempts to make code in the caller as efficient
547 as possible under the assumption that the call is not commonly executed. As
548 such, these calls often preserve all registers so that the call does not break
549 any live ranges in the caller side. This calling convention does not support
550 varargs and requires the prototype of all callees to exactly match the
551 prototype of the function definition.
554 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
556 <dd>Any calling convention may be specified by number, allowing
557 target-specific calling conventions to be used. Target specific calling
558 conventions start at 64.
562 <p>More calling conventions can be added/defined on an as-needed basis, to
563 support pascal conventions or any other well-known target-independent
568 <!-- ======================================================================= -->
569 <div class="doc_subsection">
570 <a name="globalvars">Global Variables</a>
573 <div class="doc_text">
575 <p>Global variables define regions of memory allocated at compilation time
576 instead of run-time. Global variables may optionally be initialized, may have
577 an explicit section to be placed in, and may
578 have an optional explicit alignment specified. A
579 variable may be defined as a global "constant," which indicates that the
580 contents of the variable will <b>never</b> be modified (enabling better
581 optimization, allowing the global data to be placed in the read-only section of
582 an executable, etc). Note that variables that need runtime initialization
583 cannot be marked "constant" as there is a store to the variable.</p>
586 LLVM explicitly allows <em>declarations</em> of global variables to be marked
587 constant, even if the final definition of the global is not. This capability
588 can be used to enable slightly better optimization of the program, but requires
589 the language definition to guarantee that optimizations based on the
590 'constantness' are valid for the translation units that do not include the
594 <p>As SSA values, global variables define pointer values that are in
595 scope (i.e. they dominate) all basic blocks in the program. Global
596 variables always define a pointer to their "content" type because they
597 describe a region of memory, and all memory objects in LLVM are
598 accessed through pointers.</p>
600 <p>LLVM allows an explicit section to be specified for globals. If the target
601 supports it, it will emit globals to the section specified.</p>
603 <p>An explicit alignment may be specified for a global. If not present, or if
604 the alignment is set to zero, the alignment of the global is set by the target
605 to whatever it feels convenient. If an explicit alignment is specified, the
606 global is forced to have at least that much alignment. All alignments must be
612 <!-- ======================================================================= -->
613 <div class="doc_subsection">
614 <a name="functionstructure">Functions</a>
617 <div class="doc_text">
619 <p>LLVM function definitions consist of an optional <a href="#linkage">linkage
620 type</a>, an optional <a href="#callingconv">calling convention</a>, a return
621 type, a function name, a (possibly empty) argument list, an optional section,
622 an optional alignment, an opening curly brace,
623 a list of basic blocks, and a closing curly brace. LLVM function declarations
624 are defined with the "<tt>declare</tt>" keyword, an optional <a
625 href="#callingconv">calling convention</a>, a return type, a function name,
626 a possibly empty list of arguments, and an optional alignment.</p>
628 <p>A function definition contains a list of basic blocks, forming the CFG for
629 the function. Each basic block may optionally start with a label (giving the
630 basic block a symbol table entry), contains a list of instructions, and ends
631 with a <a href="#terminators">terminator</a> instruction (such as a branch or
632 function return).</p>
634 <p>The first basic block in a program is special in two ways: it is immediately
635 executed on entrance to the function, and it is not allowed to have predecessor
636 basic blocks (i.e. there can not be any branches to the entry block of a
637 function). Because the block can have no predecessors, it also cannot have any
638 <a href="#i_phi">PHI nodes</a>.</p>
640 <p>LLVM functions are identified by their name and type signature. Hence, two
641 functions with the same name but different parameter lists or return values are
642 considered different functions, and LLVM will resolve references to each
645 <p>LLVM allows an explicit section to be specified for functions. If the target
646 supports it, it will emit functions to the section specified.</p>
648 <p>An explicit alignment may be specified for a function. If not present, or if
649 the alignment is set to zero, the alignment of the function is set by the target
650 to whatever it feels convenient. If an explicit alignment is specified, the
651 function is forced to have at least that much alignment. All alignments must be
656 <!-- ======================================================================= -->
657 <div class="doc_subsection">
658 <a name="moduleasm">Module-Level Inline Assembly</a>
661 <div class="doc_text">
663 Modules may contain "module-level inline asm" blocks, which corresponds to the
664 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
665 LLVM and treated as a single unit, but may be separated in the .ll file if
666 desired. The syntax is very simple:
669 <div class="doc_code"><pre>
670 module asm "inline asm code goes here"
671 module asm "more can go here"
674 <p>The strings can contain any character by escaping non-printable characters.
675 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
680 The inline asm code is simply printed to the machine code .s file when
681 assembly code is generated.
686 <!-- *********************************************************************** -->
687 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
688 <!-- *********************************************************************** -->
690 <div class="doc_text">
692 <p>The LLVM type system is one of the most important features of the
693 intermediate representation. Being typed enables a number of
694 optimizations to be performed on the IR directly, without having to do
695 extra analyses on the side before the transformation. A strong type
696 system makes it easier to read the generated code and enables novel
697 analyses and transformations that are not feasible to perform on normal
698 three address code representations.</p>
702 <!-- ======================================================================= -->
703 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
704 <div class="doc_text">
705 <p>The primitive types are the fundamental building blocks of the LLVM
706 system. The current set of primitive types is as follows:</p>
708 <table class="layout">
713 <tr><th>Type</th><th>Description</th></tr>
714 <tr><td><tt>void</tt></td><td>No value</td></tr>
715 <tr><td><tt>ubyte</tt></td><td>Unsigned 8-bit value</td></tr>
716 <tr><td><tt>ushort</tt></td><td>Unsigned 16-bit value</td></tr>
717 <tr><td><tt>uint</tt></td><td>Unsigned 32-bit value</td></tr>
718 <tr><td><tt>ulong</tt></td><td>Unsigned 64-bit value</td></tr>
719 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
720 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
727 <tr><th>Type</th><th>Description</th></tr>
728 <tr><td><tt>bool</tt></td><td>True or False value</td></tr>
729 <tr><td><tt>sbyte</tt></td><td>Signed 8-bit value</td></tr>
730 <tr><td><tt>short</tt></td><td>Signed 16-bit value</td></tr>
731 <tr><td><tt>int</tt></td><td>Signed 32-bit value</td></tr>
732 <tr><td><tt>long</tt></td><td>Signed 64-bit value</td></tr>
733 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
741 <!-- _______________________________________________________________________ -->
742 <div class="doc_subsubsection"> <a name="t_classifications">Type
743 Classifications</a> </div>
744 <div class="doc_text">
745 <p>These different primitive types fall into a few useful
748 <table border="1" cellspacing="0" cellpadding="4">
750 <tr><th>Classification</th><th>Types</th></tr>
752 <td><a name="t_signed">signed</a></td>
753 <td><tt>sbyte, short, int, long, float, double</tt></td>
756 <td><a name="t_unsigned">unsigned</a></td>
757 <td><tt>ubyte, ushort, uint, ulong</tt></td>
760 <td><a name="t_integer">integer</a></td>
761 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
764 <td><a name="t_integral">integral</a></td>
765 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt>
769 <td><a name="t_floating">floating point</a></td>
770 <td><tt>float, double</tt></td>
773 <td><a name="t_firstclass">first class</a></td>
774 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long,<br>
775 float, double, <a href="#t_pointer">pointer</a>,
776 <a href="#t_packed">packed</a></tt></td>
781 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
782 most important. Values of these types are the only ones which can be
783 produced by instructions, passed as arguments, or used as operands to
784 instructions. This means that all structures and arrays must be
785 manipulated either by pointer or by component.</p>
788 <!-- ======================================================================= -->
789 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
791 <div class="doc_text">
793 <p>The real power in LLVM comes from the derived types in the system.
794 This is what allows a programmer to represent arrays, functions,
795 pointers, and other useful types. Note that these derived types may be
796 recursive: For example, it is possible to have a two dimensional array.</p>
800 <!-- _______________________________________________________________________ -->
801 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
803 <div class="doc_text">
807 <p>The array type is a very simple derived type that arranges elements
808 sequentially in memory. The array type requires a size (number of
809 elements) and an underlying data type.</p>
814 [<# elements> x <elementtype>]
817 <p>The number of elements is a constant integer value; elementtype may
818 be any type with a size.</p>
821 <table class="layout">
824 <tt>[40 x int ]</tt><br/>
825 <tt>[41 x int ]</tt><br/>
826 <tt>[40 x uint]</tt><br/>
829 Array of 40 integer values.<br/>
830 Array of 41 integer values.<br/>
831 Array of 40 unsigned integer values.<br/>
835 <p>Here are some examples of multidimensional arrays:</p>
836 <table class="layout">
839 <tt>[3 x [4 x int]]</tt><br/>
840 <tt>[12 x [10 x float]]</tt><br/>
841 <tt>[2 x [3 x [4 x uint]]]</tt><br/>
844 3x4 array of integer values.<br/>
845 12x10 array of single precision floating point values.<br/>
846 2x3x4 array of unsigned integer values.<br/>
851 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
852 length array. Normally, accesses past the end of an array are undefined in
853 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
854 As a special case, however, zero length arrays are recognized to be variable
855 length. This allows implementation of 'pascal style arrays' with the LLVM
856 type "{ int, [0 x float]}", for example.</p>
860 <!-- _______________________________________________________________________ -->
861 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
862 <div class="doc_text">
864 <p>The function type can be thought of as a function signature. It
865 consists of a return type and a list of formal parameter types.
866 Function types are usually used to build virtual function tables
867 (which are structures of pointers to functions), for indirect function
868 calls, and when defining a function.</p>
870 The return type of a function type cannot be an aggregate type.
873 <pre> <returntype> (<parameter list>)<br></pre>
874 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
875 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
876 which indicates that the function takes a variable number of arguments.
877 Variable argument functions can access their arguments with the <a
878 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
880 <table class="layout">
883 <tt>int (int)</tt> <br/>
884 <tt>float (int, int *) *</tt><br/>
885 <tt>int (sbyte *, ...)</tt><br/>
888 function taking an <tt>int</tt>, returning an <tt>int</tt><br/>
889 <a href="#t_pointer">Pointer</a> to a function that takes an
890 <tt>int</tt> and a <a href="#t_pointer">pointer</a> to <tt>int</tt>,
891 returning <tt>float</tt>.<br/>
892 A vararg function that takes at least one <a href="#t_pointer">pointer</a>
893 to <tt>sbyte</tt> (signed char in C), which returns an integer. This is
894 the signature for <tt>printf</tt> in LLVM.<br/>
900 <!-- _______________________________________________________________________ -->
901 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
902 <div class="doc_text">
904 <p>The structure type is used to represent a collection of data members
905 together in memory. The packing of the field types is defined to match
906 the ABI of the underlying processor. The elements of a structure may
907 be any type that has a size.</p>
908 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
909 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
910 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
913 <pre> { <type list> }<br></pre>
915 <table class="layout">
918 <tt>{ int, int, int }</tt><br/>
919 <tt>{ float, int (int) * }</tt><br/>
922 a triple of three <tt>int</tt> values<br/>
923 A pair, where the first element is a <tt>float</tt> and the second element
924 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
925 that takes an <tt>int</tt>, returning an <tt>int</tt>.<br/>
931 <!-- _______________________________________________________________________ -->
932 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
933 <div class="doc_text">
935 <p>As in many languages, the pointer type represents a pointer or
936 reference to another object, which must live in memory.</p>
938 <pre> <type> *<br></pre>
940 <table class="layout">
943 <tt>[4x int]*</tt><br/>
944 <tt>int (int *) *</tt><br/>
947 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
948 four <tt>int</tt> values<br/>
949 A <a href="#t_pointer">pointer</a> to a <a
950 href="#t_function">function</a> that takes an <tt>int*</tt>, returning an
957 <!-- _______________________________________________________________________ -->
958 <div class="doc_subsubsection"> <a name="t_packed">Packed Type</a> </div>
959 <div class="doc_text">
963 <p>A packed type is a simple derived type that represents a vector
964 of elements. Packed types are used when multiple primitive data
965 are operated in parallel using a single instruction (SIMD).
966 A packed type requires a size (number of
967 elements) and an underlying primitive data type. Vectors must have a power
968 of two length (1, 2, 4, 8, 16 ...). Packed types are
969 considered <a href="#t_firstclass">first class</a>.</p>
974 < <# elements> x <elementtype> >
977 <p>The number of elements is a constant integer value; elementtype may
978 be any integral or floating point type.</p>
982 <table class="layout">
985 <tt><4 x int></tt><br/>
986 <tt><8 x float></tt><br/>
987 <tt><2 x uint></tt><br/>
990 Packed vector of 4 integer values.<br/>
991 Packed vector of 8 floating-point values.<br/>
992 Packed vector of 2 unsigned integer values.<br/>
998 <!-- _______________________________________________________________________ -->
999 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1000 <div class="doc_text">
1004 <p>Opaque types are used to represent unknown types in the system. This
1005 corresponds (for example) to the C notion of a foward declared structure type.
1006 In LLVM, opaque types can eventually be resolved to any type (not just a
1007 structure type).</p>
1017 <table class="layout">
1023 An opaque type.<br/>
1030 <!-- *********************************************************************** -->
1031 <div class="doc_section"> <a name="constants">Constants</a> </div>
1032 <!-- *********************************************************************** -->
1034 <div class="doc_text">
1036 <p>LLVM has several different basic types of constants. This section describes
1037 them all and their syntax.</p>
1041 <!-- ======================================================================= -->
1042 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1044 <div class="doc_text">
1047 <dt><b>Boolean constants</b></dt>
1049 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1050 constants of the <tt><a href="#t_primitive">bool</a></tt> type.
1053 <dt><b>Integer constants</b></dt>
1055 <dd>Standard integers (such as '4') are constants of the <a
1056 href="#t_integer">integer</a> type. Negative numbers may be used with signed
1060 <dt><b>Floating point constants</b></dt>
1062 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1063 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1064 notation (see below). Floating point constants must have a <a
1065 href="#t_floating">floating point</a> type. </dd>
1067 <dt><b>Null pointer constants</b></dt>
1069 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1070 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1074 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1075 of floating point constants. For example, the form '<tt>double
1076 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1077 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1078 (and the only time that they are generated by the disassembler) is when a
1079 floating point constant must be emitted but it cannot be represented as a
1080 decimal floating point number. For example, NaN's, infinities, and other
1081 special values are represented in their IEEE hexadecimal format so that
1082 assembly and disassembly do not cause any bits to change in the constants.</p>
1086 <!-- ======================================================================= -->
1087 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1090 <div class="doc_text">
1091 <p>Aggregate constants arise from aggregation of simple constants
1092 and smaller aggregate constants.</p>
1095 <dt><b>Structure constants</b></dt>
1097 <dd>Structure constants are represented with notation similar to structure
1098 type definitions (a comma separated list of elements, surrounded by braces
1099 (<tt>{}</tt>)). For example: "<tt>{ int 4, float 17.0, int* %G }</tt>",
1100 where "<tt>%G</tt>" is declared as "<tt>%G = external global int</tt>". Structure constants
1101 must have <a href="#t_struct">structure type</a>, and the number and
1102 types of elements must match those specified by the type.
1105 <dt><b>Array constants</b></dt>
1107 <dd>Array constants are represented with notation similar to array type
1108 definitions (a comma separated list of elements, surrounded by square brackets
1109 (<tt>[]</tt>)). For example: "<tt>[ int 42, int 11, int 74 ]</tt>". Array
1110 constants must have <a href="#t_array">array type</a>, and the number and
1111 types of elements must match those specified by the type.
1114 <dt><b>Packed constants</b></dt>
1116 <dd>Packed constants are represented with notation similar to packed type
1117 definitions (a comma separated list of elements, surrounded by
1118 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< int 42,
1119 int 11, int 74, int 100 ></tt>". Packed constants must have <a
1120 href="#t_packed">packed type</a>, and the number and types of elements must
1121 match those specified by the type.
1124 <dt><b>Zero initialization</b></dt>
1126 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1127 value to zero of <em>any</em> type, including scalar and aggregate types.
1128 This is often used to avoid having to print large zero initializers (e.g. for
1129 large arrays) and is always exactly equivalent to using explicit zero
1136 <!-- ======================================================================= -->
1137 <div class="doc_subsection">
1138 <a name="globalconstants">Global Variable and Function Addresses</a>
1141 <div class="doc_text">
1143 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1144 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1145 constants. These constants are explicitly referenced when the <a
1146 href="#identifiers">identifier for the global</a> is used and always have <a
1147 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1153 %Z = global [2 x int*] [ int* %X, int* %Y ]
1158 <!-- ======================================================================= -->
1159 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1160 <div class="doc_text">
1161 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1162 no specific value. Undefined values may be of any type and be used anywhere
1163 a constant is permitted.</p>
1165 <p>Undefined values indicate to the compiler that the program is well defined
1166 no matter what value is used, giving the compiler more freedom to optimize.
1170 <!-- ======================================================================= -->
1171 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1174 <div class="doc_text">
1176 <p>Constant expressions are used to allow expressions involving other constants
1177 to be used as constants. Constant expressions may be of any <a
1178 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1179 that does not have side effects (e.g. load and call are not supported). The
1180 following is the syntax for constant expressions:</p>
1183 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1184 <dd>Truncate a constant to another type. The bit size of CST must be larger
1185 than the bit size of TYPE. Both types must be integral.</dd>
1187 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1188 <dd>Zero extend a constant to another type. The bit size of CST must be
1189 smaller or equal to the bit size of TYPE. Both types must be integral.</dd>
1191 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1192 <dd>Sign extend a constant to another type. The bit size of CST must be
1193 smaller or equal to the bit size of TYPE. Both types must be integral.</dd>
1195 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1196 <dd>Truncate a floating point constant to another floating point type. The
1197 size of CST must be larger than the size of TYPE. Both types must be
1198 floating point.</dd>
1200 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1201 <dd>Floating point extend a constant to another type. The size of CST must be
1202 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1204 <dt><b><tt>fp2uint ( CST to TYPE )</tt></b></dt>
1205 <dd>Convert a floating point constant to the corresponding unsigned integer
1206 constant. TYPE must be an integer type. CST must be floating point. If the
1207 value won't fit in the integer type, the results are undefined.</dd>
1209 <dt><b><tt>fp2sint ( CST to TYPE )</tt></b></dt>
1210 <dd>Convert a floating point constant to the corresponding signed integer
1211 constant. TYPE must be an integer type. CST must be floating point. If the
1212 value won't fit in the integer type, the results are undefined.</dd>
1214 <dt><b><tt>uint2fp ( CST to TYPE )</tt></b></dt>
1215 <dd>Convert an unsigned integer constant to the corresponding floating point
1216 constant. TYPE must be floating point. CST must be of integer type. If the
1217 value won't fit in the floating point type, the results are undefined.</dd>
1219 <dt><b><tt>sint2fp ( CST to TYPE )</tt></b></dt>
1220 <dd>Convert a signed integer constant to the corresponding floating point
1221 constant. TYPE must be floating point. CST must be of integer type. If the
1222 value won't fit in the floating point type, the results are undefined.</dd>
1224 <dt><b><tt>bitconvert ( CST to TYPE )</tt></b></dt>
1225 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1226 identical (same number of bits). The conversion is done as if the CST value
1227 was stored to memory and read back as TYPE. In other words, no bits change
1228 with this operator, just the type. This can be used for conversion of pointer
1229 and packed types to any other type, as long as they have the same bit width.
1232 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1234 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1235 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1236 instruction, the index list may have zero or more indexes, which are required
1237 to make sense for the type of "CSTPTR".</dd>
1239 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1241 <dd>Perform the <a href="#i_select">select operation</a> on
1244 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1246 <dd>Perform the <a href="#i_extractelement">extractelement
1247 operation</a> on constants.
1249 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1251 <dd>Perform the <a href="#i_insertelement">insertelement
1252 operation</a> on constants.
1255 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1257 <dd>Perform the <a href="#i_shufflevector">shufflevector
1258 operation</a> on constants.
1260 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1262 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1263 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1264 binary</a> operations. The constraints on operands are the same as those for
1265 the corresponding instruction (e.g. no bitwise operations on floating point
1266 values are allowed).</dd>
1270 <!-- *********************************************************************** -->
1271 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1272 <!-- *********************************************************************** -->
1274 <!-- ======================================================================= -->
1275 <div class="doc_subsection">
1276 <a name="inlineasm">Inline Assembler Expressions</a>
1279 <div class="doc_text">
1282 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1283 Module-Level Inline Assembly</a>) through the use of a special value. This
1284 value represents the inline assembler as a string (containing the instructions
1285 to emit), a list of operand constraints (stored as a string), and a flag that
1286 indicates whether or not the inline asm expression has side effects. An example
1287 inline assembler expression is:
1291 int(int) asm "bswap $0", "=r,r"
1295 Inline assembler expressions may <b>only</b> be used as the callee operand of
1296 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1300 %X = call int asm "<a href="#i_bswap">bswap</a> $0", "=r,r"(int %Y)
1304 Inline asms with side effects not visible in the constraint list must be marked
1305 as having side effects. This is done through the use of the
1306 '<tt>sideeffect</tt>' keyword, like so:
1310 call void asm sideeffect "eieio", ""()
1313 <p>TODO: The format of the asm and constraints string still need to be
1314 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1315 need to be documented).
1320 <!-- *********************************************************************** -->
1321 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1322 <!-- *********************************************************************** -->
1324 <div class="doc_text">
1326 <p>The LLVM instruction set consists of several different
1327 classifications of instructions: <a href="#terminators">terminator
1328 instructions</a>, <a href="#binaryops">binary instructions</a>,
1329 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1330 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1331 instructions</a>.</p>
1335 <!-- ======================================================================= -->
1336 <div class="doc_subsection"> <a name="terminators">Terminator
1337 Instructions</a> </div>
1339 <div class="doc_text">
1341 <p>As mentioned <a href="#functionstructure">previously</a>, every
1342 basic block in a program ends with a "Terminator" instruction, which
1343 indicates which block should be executed after the current block is
1344 finished. These terminator instructions typically yield a '<tt>void</tt>'
1345 value: they produce control flow, not values (the one exception being
1346 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1347 <p>There are six different terminator instructions: the '<a
1348 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1349 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1350 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1351 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1352 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1356 <!-- _______________________________________________________________________ -->
1357 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1358 Instruction</a> </div>
1359 <div class="doc_text">
1361 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1362 ret void <i>; Return from void function</i>
1365 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1366 value) from a function back to the caller.</p>
1367 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1368 returns a value and then causes control flow, and one that just causes
1369 control flow to occur.</p>
1371 <p>The '<tt>ret</tt>' instruction may return any '<a
1372 href="#t_firstclass">first class</a>' type. Notice that a function is
1373 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1374 instruction inside of the function that returns a value that does not
1375 match the return type of the function.</p>
1377 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1378 returns back to the calling function's context. If the caller is a "<a
1379 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1380 the instruction after the call. If the caller was an "<a
1381 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1382 at the beginning of the "normal" destination block. If the instruction
1383 returns a value, that value shall set the call or invoke instruction's
1386 <pre> ret int 5 <i>; Return an integer value of 5</i>
1387 ret void <i>; Return from a void function</i>
1390 <!-- _______________________________________________________________________ -->
1391 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1392 <div class="doc_text">
1394 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1397 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1398 transfer to a different basic block in the current function. There are
1399 two forms of this instruction, corresponding to a conditional branch
1400 and an unconditional branch.</p>
1402 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1403 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
1404 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
1405 value as a target.</p>
1407 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
1408 argument is evaluated. If the value is <tt>true</tt>, control flows
1409 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1410 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1412 <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
1413 href="#i_ret">ret</a> int 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> int 0<br></pre>
1415 <!-- _______________________________________________________________________ -->
1416 <div class="doc_subsubsection">
1417 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1420 <div class="doc_text">
1424 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1429 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1430 several different places. It is a generalization of the '<tt>br</tt>'
1431 instruction, allowing a branch to occur to one of many possible
1437 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1438 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1439 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1440 table is not allowed to contain duplicate constant entries.</p>
1444 <p>The <tt>switch</tt> instruction specifies a table of values and
1445 destinations. When the '<tt>switch</tt>' instruction is executed, this
1446 table is searched for the given value. If the value is found, control flow is
1447 transfered to the corresponding destination; otherwise, control flow is
1448 transfered to the default destination.</p>
1450 <h5>Implementation:</h5>
1452 <p>Depending on properties of the target machine and the particular
1453 <tt>switch</tt> instruction, this instruction may be code generated in different
1454 ways. For example, it could be generated as a series of chained conditional
1455 branches or with a lookup table.</p>
1460 <i>; Emulate a conditional br instruction</i>
1461 %Val = <a href="#i_zext">zext</a> bool %value to int
1462 switch int %Val, label %truedest [int 0, label %falsedest ]
1464 <i>; Emulate an unconditional br instruction</i>
1465 switch uint 0, label %dest [ ]
1467 <i>; Implement a jump table:</i>
1468 switch uint %val, label %otherwise [ uint 0, label %onzero
1469 uint 1, label %onone
1470 uint 2, label %ontwo ]
1474 <!-- _______________________________________________________________________ -->
1475 <div class="doc_subsubsection">
1476 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1479 <div class="doc_text">
1484 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1485 to label <normal label> unwind label <exception label>
1490 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1491 function, with the possibility of control flow transfer to either the
1492 '<tt>normal</tt>' label or the
1493 '<tt>exception</tt>' label. If the callee function returns with the
1494 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1495 "normal" label. If the callee (or any indirect callees) returns with the "<a
1496 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1497 continued at the dynamically nearest "exception" label.</p>
1501 <p>This instruction requires several arguments:</p>
1505 The optional "cconv" marker indicates which <a href="callingconv">calling
1506 convention</a> the call should use. If none is specified, the call defaults
1507 to using C calling conventions.
1509 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1510 function value being invoked. In most cases, this is a direct function
1511 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1512 an arbitrary pointer to function value.
1515 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1516 function to be invoked. </li>
1518 <li>'<tt>function args</tt>': argument list whose types match the function
1519 signature argument types. If the function signature indicates the function
1520 accepts a variable number of arguments, the extra arguments can be
1523 <li>'<tt>normal label</tt>': the label reached when the called function
1524 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1526 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1527 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1533 <p>This instruction is designed to operate as a standard '<tt><a
1534 href="#i_call">call</a></tt>' instruction in most regards. The primary
1535 difference is that it establishes an association with a label, which is used by
1536 the runtime library to unwind the stack.</p>
1538 <p>This instruction is used in languages with destructors to ensure that proper
1539 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1540 exception. Additionally, this is important for implementation of
1541 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1545 %retval = invoke int %Test(int 15) to label %Continue
1546 unwind label %TestCleanup <i>; {int}:retval set</i>
1547 %retval = invoke <a href="#callingconv">coldcc</a> int %Test(int 15) to label %Continue
1548 unwind label %TestCleanup <i>; {int}:retval set</i>
1553 <!-- _______________________________________________________________________ -->
1555 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1556 Instruction</a> </div>
1558 <div class="doc_text">
1567 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1568 at the first callee in the dynamic call stack which used an <a
1569 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1570 primarily used to implement exception handling.</p>
1574 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1575 immediately halt. The dynamic call stack is then searched for the first <a
1576 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1577 execution continues at the "exceptional" destination block specified by the
1578 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1579 dynamic call chain, undefined behavior results.</p>
1582 <!-- _______________________________________________________________________ -->
1584 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1585 Instruction</a> </div>
1587 <div class="doc_text">
1596 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1597 instruction is used to inform the optimizer that a particular portion of the
1598 code is not reachable. This can be used to indicate that the code after a
1599 no-return function cannot be reached, and other facts.</p>
1603 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1608 <!-- ======================================================================= -->
1609 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1610 <div class="doc_text">
1611 <p>Binary operators are used to do most of the computation in a
1612 program. They require two operands, execute an operation on them, and
1613 produce a single value. The operands might represent
1614 multiple data, as is the case with the <a href="#t_packed">packed</a> data type.
1615 The result value of a binary operator is not
1616 necessarily the same type as its operands.</p>
1617 <p>There are several different binary operators:</p>
1619 <!-- _______________________________________________________________________ -->
1620 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1621 Instruction</a> </div>
1622 <div class="doc_text">
1624 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1627 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1629 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1630 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1631 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1632 Both arguments must have identical types.</p>
1634 <p>The value produced is the integer or floating point sum of the two
1637 <pre> <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
1640 <!-- _______________________________________________________________________ -->
1641 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1642 Instruction</a> </div>
1643 <div class="doc_text">
1645 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1648 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1650 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1651 instruction present in most other intermediate representations.</p>
1653 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1654 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1656 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1657 Both arguments must have identical types.</p>
1659 <p>The value produced is the integer or floating point difference of
1660 the two operands.</p>
1662 <pre> <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
1663 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
1666 <!-- _______________________________________________________________________ -->
1667 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1668 Instruction</a> </div>
1669 <div class="doc_text">
1671 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1674 <p>The '<tt>mul</tt>' instruction returns the product of its two
1677 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1678 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1680 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1681 Both arguments must have identical types.</p>
1683 <p>The value produced is the integer or floating point product of the
1685 <p>There is no signed vs unsigned multiplication. The appropriate
1686 action is taken based on the type of the operand.</p>
1688 <pre> <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1691 <!-- _______________________________________________________________________ -->
1692 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
1694 <div class="doc_text">
1696 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1699 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
1702 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
1703 <a href="#t_integer">integer</a> values. Both arguments must have identical
1704 types. This instruction can also take <a href="#t_packed">packed</a> versions
1705 of the values in which case the elements must be integers.</p>
1707 <p>The value produced is the unsigned integer quotient of the two operands. This
1708 instruction always performs an unsigned division operation, regardless of
1709 whether the arguments are unsigned or not.</p>
1711 <pre> <result> = udiv uint 4, %var <i>; yields {uint}:result = 4 / %var</i>
1714 <!-- _______________________________________________________________________ -->
1715 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
1717 <div class="doc_text">
1719 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1722 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
1725 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
1726 <a href="#t_integer">integer</a> values. Both arguments must have identical
1727 types. This instruction can also take <a href="#t_packed">packed</a> versions
1728 of the values in which case the elements must be integers.</p>
1730 <p>The value produced is the signed integer quotient of the two operands. This
1731 instruction always performs a signed division operation, regardless of whether
1732 the arguments are signed or not.</p>
1734 <pre> <result> = sdiv int 4, %var <i>; yields {int}:result = 4 / %var</i>
1737 <!-- _______________________________________________________________________ -->
1738 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
1739 Instruction</a> </div>
1740 <div class="doc_text">
1742 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1745 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
1748 <p>The two arguments to the '<tt>div</tt>' instruction must be
1749 <a href="#t_floating">floating point</a> values. Both arguments must have
1750 identical types. This instruction can also take <a href="#t_packed">packed</a>
1751 versions of the values in which case the elements must be floating point.</p>
1753 <p>The value produced is the floating point quotient of the two operands.</p>
1755 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
1758 <!-- _______________________________________________________________________ -->
1759 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
1761 <div class="doc_text">
1763 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1766 <p>The '<tt>urem</tt>' instruction returns the remainder from the
1767 unsigned division of its two arguments.</p>
1769 <p>The two arguments to the '<tt>urem</tt>' instruction must be
1770 <a href="#t_integer">integer</a> values. Both arguments must have identical
1773 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
1774 This instruction always performs an unsigned division to get the remainder,
1775 regardless of whether the arguments are unsigned or not.</p>
1777 <pre> <result> = urem uint 4, %var <i>; yields {uint}:result = 4 % %var</i>
1781 <!-- _______________________________________________________________________ -->
1782 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
1783 Instruction</a> </div>
1784 <div class="doc_text">
1786 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1789 <p>The '<tt>srem</tt>' instruction returns the remainder from the
1790 signed division of its two operands.</p>
1792 <p>The two arguments to the '<tt>srem</tt>' instruction must be
1793 <a href="#t_integer">integer</a> values. Both arguments must have identical
1796 <p>This instruction returns the <i>remainder</i> of a division (where the result
1797 has the same sign as the divisor), not the <i>modulus</i> (where the
1798 result has the same sign as the dividend) of a value. For more
1799 information about the difference, see <a
1800 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
1803 <pre> <result> = srem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1807 <!-- _______________________________________________________________________ -->
1808 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
1809 Instruction</a> </div>
1810 <div class="doc_text">
1812 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1815 <p>The '<tt>frem</tt>' instruction returns the remainder from the
1816 division of its two operands.</p>
1818 <p>The two arguments to the '<tt>frem</tt>' instruction must be
1819 <a href="#t_floating">floating point</a> values. Both arguments must have
1820 identical types.</p>
1822 <p>This instruction returns the <i>remainder</i> of a division.</p>
1824 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
1828 <!-- _______________________________________________________________________ -->
1829 <div class="doc_subsubsection"> <a name="i_setcc">'<tt>set<i>cc</i></tt>'
1830 Instructions</a> </div>
1831 <div class="doc_text">
1833 <pre> <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1834 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1835 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1836 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1837 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1838 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1841 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean
1842 value based on a comparison of their two operands.</p>
1844 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must
1845 be of <a href="#t_firstclass">first class</a> type (it is not possible
1846 to compare '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>'
1847 or '<tt>void</tt>' values, etc...). Both arguments must have identical
1850 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1851 value if both operands are equal.<br>
1852 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1853 value if both operands are unequal.<br>
1854 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1855 value if the first operand is less than the second operand.<br>
1856 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1857 value if the first operand is greater than the second operand.<br>
1858 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1859 value if the first operand is less than or equal to the second operand.<br>
1860 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1861 value if the first operand is greater than or equal to the second
1864 <pre> <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1865 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1866 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1867 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1868 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1869 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1873 <!-- ======================================================================= -->
1874 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
1875 Operations</a> </div>
1876 <div class="doc_text">
1877 <p>Bitwise binary operators are used to do various forms of
1878 bit-twiddling in a program. They are generally very efficient
1879 instructions and can commonly be strength reduced from other
1880 instructions. They require two operands, execute an operation on them,
1881 and produce a single value. The resulting value of the bitwise binary
1882 operators is always the same type as its first operand.</p>
1884 <!-- _______________________________________________________________________ -->
1885 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
1886 Instruction</a> </div>
1887 <div class="doc_text">
1889 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1892 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1893 its two operands.</p>
1895 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1896 href="#t_integral">integral</a> values. Both arguments must have
1897 identical types.</p>
1899 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1901 <div style="align: center">
1902 <table border="1" cellspacing="0" cellpadding="4">
1933 <pre> <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1934 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1935 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1938 <!-- _______________________________________________________________________ -->
1939 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
1940 <div class="doc_text">
1942 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1945 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1946 or of its two operands.</p>
1948 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1949 href="#t_integral">integral</a> values. Both arguments must have
1950 identical types.</p>
1952 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1954 <div style="align: center">
1955 <table border="1" cellspacing="0" cellpadding="4">
1986 <pre> <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1987 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1988 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1991 <!-- _______________________________________________________________________ -->
1992 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
1993 Instruction</a> </div>
1994 <div class="doc_text">
1996 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1999 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2000 or of its two operands. The <tt>xor</tt> is used to implement the
2001 "one's complement" operation, which is the "~" operator in C.</p>
2003 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2004 href="#t_integral">integral</a> values. Both arguments must have
2005 identical types.</p>
2007 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2009 <div style="align: center">
2010 <table border="1" cellspacing="0" cellpadding="4">
2042 <pre> <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
2043 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
2044 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
2045 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
2048 <!-- _______________________________________________________________________ -->
2049 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2050 Instruction</a> </div>
2051 <div class="doc_text">
2053 <pre> <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
2056 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2057 the left a specified number of bits.</p>
2059 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
2060 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
2063 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
2065 <pre> <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
2066 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
2067 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
2070 <!-- _______________________________________________________________________ -->
2071 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2072 Instruction</a> </div>
2073 <div class="doc_text">
2075 <pre> <result> = lshr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
2079 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2080 operand shifted to the right a specified number of bits.</p>
2083 <p>The first argument to the '<tt>lshr</tt>' instruction must be an <a
2084 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>' type.</p>
2087 <p>This instruction always performs a logical shift right operation, regardless
2088 of whether the arguments are unsigned or not. The <tt>var2</tt> most significant
2089 bits will be filled with zero bits after the shift.</p>
2093 <result> = lshr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
2094 <result> = lshr int 4, ubyte 2 <i>; yields {uint}:result = 1</i>
2095 <result> = lshr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
2096 <result> = lshr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = 0x7FFFFFFF </i>
2100 <!-- ======================================================================= -->
2101 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2102 Instruction</a> </div>
2103 <div class="doc_text">
2106 <pre> <result> = ashr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
2110 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2111 operand shifted to the right a specified number of bits.</p>
2114 <p>The first argument to the '<tt>ashr</tt>' instruction must be an
2115 <a href="#t_integer">integer</a> type. The second argument must be an
2116 '<tt>ubyte</tt>' type.</p>
2119 <p>This instruction always performs an arithmetic shift right operation,
2120 regardless of whether the arguments are signed or not. The <tt>var2</tt> most
2121 significant bits will be filled with the sign bit of <tt>var1</tt>.</p>
2125 <result> = ashr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
2126 <result> = ashr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
2127 <result> = ashr ubyte 4, ubyte 3 <i>; yields {ubyte}:result = 0</i>
2128 <result> = ashr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
2132 <!-- ======================================================================= -->
2133 <div class="doc_subsection">
2134 <a name="vectorops">Vector Operations</a>
2137 <div class="doc_text">
2139 <p>LLVM supports several instructions to represent vector operations in a
2140 target-independent manner. This instructions cover the element-access and
2141 vector-specific operations needed to process vectors effectively. While LLVM
2142 does directly support these vector operations, many sophisticated algorithms
2143 will want to use target-specific intrinsics to take full advantage of a specific
2148 <!-- _______________________________________________________________________ -->
2149 <div class="doc_subsubsection">
2150 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2153 <div class="doc_text">
2158 <result> = extractelement <n x <ty>> <val>, uint <idx> <i>; yields <ty></i>
2164 The '<tt>extractelement</tt>' instruction extracts a single scalar
2165 element from a packed vector at a specified index.
2172 The first operand of an '<tt>extractelement</tt>' instruction is a
2173 value of <a href="#t_packed">packed</a> type. The second operand is
2174 an index indicating the position from which to extract the element.
2175 The index may be a variable.</p>
2180 The result is a scalar of the same type as the element type of
2181 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2182 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2183 results are undefined.
2189 %result = extractelement <4 x int> %vec, uint 0 <i>; yields int</i>
2194 <!-- _______________________________________________________________________ -->
2195 <div class="doc_subsubsection">
2196 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2199 <div class="doc_text">
2204 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, uint <idx> <i>; yields <n x <ty>></i>
2210 The '<tt>insertelement</tt>' instruction inserts a scalar
2211 element into a packed vector at a specified index.
2218 The first operand of an '<tt>insertelement</tt>' instruction is a
2219 value of <a href="#t_packed">packed</a> type. The second operand is a
2220 scalar value whose type must equal the element type of the first
2221 operand. The third operand is an index indicating the position at
2222 which to insert the value. The index may be a variable.</p>
2227 The result is a packed vector of the same type as <tt>val</tt>. Its
2228 element values are those of <tt>val</tt> except at position
2229 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2230 exceeds the length of <tt>val</tt>, the results are undefined.
2236 %result = insertelement <4 x int> %vec, int 1, uint 0 <i>; yields <4 x int></i>
2240 <!-- _______________________________________________________________________ -->
2241 <div class="doc_subsubsection">
2242 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2245 <div class="doc_text">
2250 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x uint> <mask> <i>; yields <n x <ty>></i>
2256 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2257 from two input vectors, returning a vector of the same type.
2263 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2264 with types that match each other and types that match the result of the
2265 instruction. The third argument is a shuffle mask, which has the same number
2266 of elements as the other vector type, but whose element type is always 'uint'.
2270 The shuffle mask operand is required to be a constant vector with either
2271 constant integer or undef values.
2277 The elements of the two input vectors are numbered from left to right across
2278 both of the vectors. The shuffle mask operand specifies, for each element of
2279 the result vector, which element of the two input registers the result element
2280 gets. The element selector may be undef (meaning "don't care") and the second
2281 operand may be undef if performing a shuffle from only one vector.
2287 %result = shufflevector <4 x int> %v1, <4 x int> %v2,
2288 <4 x uint> <uint 0, uint 4, uint 1, uint 5> <i>; yields <4 x int></i>
2289 %result = shufflevector <4 x int> %v1, <4 x int> undef,
2290 <4 x uint> <uint 0, uint 1, uint 2, uint 3> <i>; yields <4 x int></i> - Identity shuffle.
2295 <!-- ======================================================================= -->
2296 <div class="doc_subsection">
2297 <a name="memoryops">Memory Access and Addressing Operations</a>
2300 <div class="doc_text">
2302 <p>A key design point of an SSA-based representation is how it
2303 represents memory. In LLVM, no memory locations are in SSA form, which
2304 makes things very simple. This section describes how to read, write,
2305 allocate, and free memory in LLVM.</p>
2309 <!-- _______________________________________________________________________ -->
2310 <div class="doc_subsubsection">
2311 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2314 <div class="doc_text">
2319 <result> = malloc <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2324 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2325 heap and returns a pointer to it.</p>
2329 <p>The '<tt>malloc</tt>' instruction allocates
2330 <tt>sizeof(<type>)*NumElements</tt>
2331 bytes of memory from the operating system and returns a pointer of the
2332 appropriate type to the program. If "NumElements" is specified, it is the
2333 number of elements allocated. If an alignment is specified, the value result
2334 of the allocation is guaranteed to be aligned to at least that boundary. If
2335 not specified, or if zero, the target can choose to align the allocation on any
2336 convenient boundary.</p>
2338 <p>'<tt>type</tt>' must be a sized type.</p>
2342 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2343 a pointer is returned.</p>
2348 %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
2350 %size = <a href="#i_add">add</a> uint 2, 2 <i>; yields {uint}:size = uint 4</i>
2351 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
2352 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
2353 %array3 = malloc int, uint 4, align 1024 <i>; yields {int*}:array3</i>
2354 %array4 = malloc int, align 1024 <i>; yields {int*}:array4</i>
2358 <!-- _______________________________________________________________________ -->
2359 <div class="doc_subsubsection">
2360 <a name="i_free">'<tt>free</tt>' Instruction</a>
2363 <div class="doc_text">
2368 free <type> <value> <i>; yields {void}</i>
2373 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2374 memory heap to be reallocated in the future.</p>
2378 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2379 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2384 <p>Access to the memory pointed to by the pointer is no longer defined
2385 after this instruction executes.</p>
2390 %array = <a href="#i_malloc">malloc</a> [4 x ubyte] <i>; yields {[4 x ubyte]*}:array</i>
2391 free [4 x ubyte]* %array
2395 <!-- _______________________________________________________________________ -->
2396 <div class="doc_subsubsection">
2397 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2400 <div class="doc_text">
2405 <result> = alloca <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2410 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
2411 stack frame of the procedure that is live until the current function
2412 returns to its caller.</p>
2416 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2417 bytes of memory on the runtime stack, returning a pointer of the
2418 appropriate type to the program. If "NumElements" is specified, it is the
2419 number of elements allocated. If an alignment is specified, the value result
2420 of the allocation is guaranteed to be aligned to at least that boundary. If
2421 not specified, or if zero, the target can choose to align the allocation on any
2422 convenient boundary.</p>
2424 <p>'<tt>type</tt>' may be any sized type.</p>
2428 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2429 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2430 instruction is commonly used to represent automatic variables that must
2431 have an address available. When the function returns (either with the <tt><a
2432 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2433 instructions), the memory is reclaimed.</p>
2438 %ptr = alloca int <i>; yields {int*}:ptr</i>
2439 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
2440 %ptr = alloca int, uint 4, align 1024 <i>; yields {int*}:ptr</i>
2441 %ptr = alloca int, align 1024 <i>; yields {int*}:ptr</i>
2445 <!-- _______________________________________________________________________ -->
2446 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2447 Instruction</a> </div>
2448 <div class="doc_text">
2450 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
2452 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2454 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2455 address from which to load. The pointer must point to a <a
2456 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2457 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2458 the number or order of execution of this <tt>load</tt> with other
2459 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2462 <p>The location of memory pointed to is loaded.</p>
2464 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2466 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2467 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2470 <!-- _______________________________________________________________________ -->
2471 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2472 Instruction</a> </div>
2473 <div class="doc_text">
2475 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2476 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2479 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2481 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2482 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
2483 operand must be a pointer to the type of the '<tt><value></tt>'
2484 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2485 optimizer is not allowed to modify the number or order of execution of
2486 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2487 href="#i_store">store</a></tt> instructions.</p>
2489 <p>The contents of memory are updated to contain '<tt><value></tt>'
2490 at the location specified by the '<tt><pointer></tt>' operand.</p>
2492 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2494 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2495 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2499 <!-- _______________________________________________________________________ -->
2500 <div class="doc_subsubsection">
2501 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2504 <div class="doc_text">
2507 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2513 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2514 subelement of an aggregate data structure.</p>
2518 <p>This instruction takes a list of integer constants that indicate what
2519 elements of the aggregate object to index to. The actual types of the arguments
2520 provided depend on the type of the first pointer argument. The
2521 '<tt>getelementptr</tt>' instruction is used to index down through the type
2522 levels of a structure or to a specific index in an array. When indexing into a
2523 structure, only <tt>uint</tt>
2524 integer constants are allowed. When indexing into an array or pointer,
2525 <tt>int</tt> and <tt>long</tt> indexes are allowed of any sign.</p>
2527 <p>For example, let's consider a C code fragment and how it gets
2528 compiled to LLVM:</p>
2542 int *foo(struct ST *s) {
2543 return &s[1].Z.B[5][13];
2547 <p>The LLVM code generated by the GCC frontend is:</p>
2550 %RT = type { sbyte, [10 x [20 x int]], sbyte }
2551 %ST = type { int, double, %RT }
2555 int* %foo(%ST* %s) {
2557 %reg = getelementptr %ST* %s, int 1, uint 2, uint 1, int 5, int 13
2564 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2565 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2566 and <a href="#t_array">array</a> types require <tt>uint</tt>, <tt>int</tt>,
2567 <tt>ulong</tt>, or <tt>long</tt> values, and <a href="#t_struct">structure</a>
2568 types require <tt>uint</tt> <b>constants</b>.</p>
2570 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2571 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
2572 }</tt>' type, a structure. The second index indexes into the third element of
2573 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
2574 sbyte }</tt>' type, another structure. The third index indexes into the second
2575 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
2576 array. The two dimensions of the array are subscripted into, yielding an
2577 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2578 to this element, thus computing a value of '<tt>int*</tt>' type.</p>
2580 <p>Note that it is perfectly legal to index partially through a
2581 structure, returning a pointer to an inner element. Because of this,
2582 the LLVM code for the given testcase is equivalent to:</p>
2585 int* %foo(%ST* %s) {
2586 %t1 = getelementptr %ST* %s, int 1 <i>; yields %ST*:%t1</i>
2587 %t2 = getelementptr %ST* %t1, int 0, uint 2 <i>; yields %RT*:%t2</i>
2588 %t3 = getelementptr %RT* %t2, int 0, uint 1 <i>; yields [10 x [20 x int]]*:%t3</i>
2589 %t4 = getelementptr [10 x [20 x int]]* %t3, int 0, int 5 <i>; yields [20 x int]*:%t4</i>
2590 %t5 = getelementptr [20 x int]* %t4, int 0, int 13 <i>; yields int*:%t5</i>
2595 <p>Note that it is undefined to access an array out of bounds: array and
2596 pointer indexes must always be within the defined bounds of the array type.
2597 The one exception for this rules is zero length arrays. These arrays are
2598 defined to be accessible as variable length arrays, which requires access
2599 beyond the zero'th element.</p>
2601 <p>The getelementptr instruction is often confusing. For some more insight
2602 into how it works, see <a href="GetElementPtr.html">the getelementptr
2608 <i>; yields [12 x ubyte]*:aptr</i>
2609 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, uint 1
2613 <!-- ======================================================================= -->
2614 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
2616 <div class="doc_text">
2617 <p>The instructions in this category are the conversion instructions (casting)
2618 which all take a single operand and a type. They perform various bit conversions
2622 <!-- _______________________________________________________________________ -->
2623 <div class="doc_subsubsection">
2624 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
2626 <div class="doc_text">
2630 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
2635 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
2640 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
2641 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
2642 and type of the result, which must be an <a href="#t_integral">integral</a>
2647 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
2648 and converts the reamining bits to <tt>ty2</tt>. The bit size of <tt>value</tt>
2649 must be larger than the bit size of <tt>ty2</tt>. Equal sized types are not
2650 allowed. This implies that a <tt>trunc</tt> cannot be a <i>no-op cast</i>. It
2651 will always truncate bits.</p>
2653 <p>When truncating to bool, the truncation is done as a comparison against
2654 zero. If the <tt>value</tt> was zero, the bool result will be <tt>false</tt>.
2655 If the <tt>value</tt> was non-zero, the bool result will be <tt>true</tt>.</p>
2659 %X = trunc int 257 to ubyte <i>; yields ubyte:1</i>
2660 %Y = trunc int 123 to bool <i>; yields bool:true</i>
2664 <!-- _______________________________________________________________________ -->
2665 <div class="doc_subsubsection">
2666 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
2668 <div class="doc_text">
2672 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
2676 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
2681 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
2682 <a href="#t_integral">integral</a> type, and a type to cast it to, which must
2683 also be of <a href="#t_integral">integral</a> type. The bit size of the
2684 <tt>value</tt> must be smaller than or equal to the bit size of the
2685 destination type, <tt>ty2</tt>.</p>
2688 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
2689 bits until it reaches the size of the destination type, <tt>ty2</tt>. When the
2690 the operand and the type are the same size, no bit filling is done and the
2691 cast is considered a <i>no-op cast</i> because no bits change (only the type
2694 <p>When zero extending to bool, the extension is done as a comparison against
2695 zero. If the <tt>value</tt> was zero, the bool result will be <tt>false</tt>.
2696 If the <tt>value</tt> was non-zero, the bool result will be <tt>true</tt>.</p>
2700 %X = zext int 257 to ulong <i>; yields ulong:257</i>
2701 %Y = zext bool true to int <i>; yields int:1</i>
2705 <!-- _______________________________________________________________________ -->
2706 <div class="doc_subsubsection">
2707 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
2709 <div class="doc_text">
2713 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
2717 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
2721 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
2722 <a href="#t_integral">integral</a> type, and a type to cast it to, which must
2723 also be of <a href="#t_integral">integral</a> type.</p>
2727 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
2728 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
2729 the type <tt>ty2</tt>. When the the operand and the type are the same size,
2730 no bit filling is done and the cast is considered a <i>no-op cast</i> because
2731 no bits change (only the type changes).</p>
2733 <p>When sign extending to bool, the extension is done as a comparison against
2734 zero. If the <tt>value</tt> was zero, the bool result will be <tt>false</tt>.
2735 If the <tt>value</tt> was non-zero, the bool result will be <tt>true</tt>.</p>
2740 %X = sext sbyte -1 to ushort <i>; yields ushort:65535</i>
2741 %Y = sext bool true to int <i>; yields int:-1</i>
2745 <!-- _______________________________________________________________________ -->
2746 <div class="doc_subsubsection">
2747 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
2749 <div class="doc_text">
2753 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
2757 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
2758 floating point value.</p>
2761 <p>The '<tt>fpext</tt>' instruction takes a
2762 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
2763 and a <a href="#t_floating">floating point</a> type to cast it to.</p>
2766 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from one floating
2767 point type to another. If the type of the <tt>value</tt> and <tt>ty2</tt> are
2768 the same, the instruction is considered a <i>no-op cast</i> because no bits
2773 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
2774 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
2778 <!-- _______________________________________________________________________ -->
2779 <div class="doc_subsubsection">
2780 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
2783 <div class="doc_text">
2788 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
2792 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
2797 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
2798 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
2799 cast it to. The size of <tt>value</tt> must be larger than the size of
2800 <tt>ty2</a>. This implies that <tt>fptrunc</tt> cannot be used to make a
2801 <i>no-op cast</i>.</p>
2804 <p> The '<tt>fptrunc</tt>' instruction converts a
2805 <a href="#t_floating">floating point</a> value from a larger type to a smaller
2806 type. If the value cannot fit within the destination type, <tt>ty2</tt>, then
2807 the results are undefined.</p>
2811 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
2812 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
2816 <!-- _______________________________________________________________________ -->
2817 <div class="doc_subsubsection">
2818 <a name="i_fp2uint">'<tt>fp2uint .. to</tt>' Instruction</a>
2820 <div class="doc_text">
2824 <result> = fp2uint <ty> <value> to <ty2> <i>; yields ty2</i>
2828 <p>The '<tt>fp2uint</tt>' converts a floating point <tt>value</tt> to its
2829 unsigned integer equivalent of type <tt>ty2</tt>.
2833 <p>The '<tt>fp2uint</tt>' instruction takes a value to cast, which must be a
2834 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
2835 must be an <a href="#t_integral">integral</a> type.</p>
2838 <p> The '<tt>fp2uint</tt>' instruction converts its
2839 <a href="#t_floating">floating point</a> operand into the nearest (rounding
2840 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
2841 the results are undefined.</p>
2843 <p>When converting to bool, the conversion is done as a comparison against
2844 zero. If the <tt>value</tt> was zero, the bool result will be <tt>false</tt>.
2845 If the <tt>value</tt> was non-zero, the bool result will be <tt>true</tt>.</p>
2849 %X = fp2uint double 123.0 to int <i>; yields int:123</i>
2850 %Y = fp2uint float 1.0E+300 to bool <i>; yields bool:true</i>
2851 %X = fp2uint float 1.04E+17 to ubyte <i>; yields undefined:1</i>
2855 <!-- _______________________________________________________________________ -->
2856 <div class="doc_subsubsection">
2857 <a name="i_fp2sint">'<tt>fp2sint .. to</tt>' Instruction</a>
2859 <div class="doc_text">
2863 <result> = fp2sint <ty> <value> to <ty2> <i>; yields ty2</i>
2867 <p>The '<tt>fp2sint</tt>' instruction converts
2868 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
2873 <p> The '<tt>fp2sint</tt>' instruction takes a value to cast, which must be a
2874 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
2875 must also be an <a href="#t_integral">integral</a> type.</p>
2878 <p>The '<tt>fp2sint</tt>' instruction converts its
2879 <a href="#t_floating">floating point</a> operand into the nearest (rounding
2880 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
2881 the results are undefined.</p>
2883 <p>When converting to bool, the conversion is done as a comparison against
2884 zero. If the <tt>value</tt> was zero, the bool result will be <tt>false</tt>.
2885 If the <tt>value</tt> was non-zero, the bool result will be <tt>true</tt>.</p>
2889 %X = fp2sint double -123.0 to int <i>; yields int:-123</i>
2890 %Y = fp2sint float 1.0E-247 to bool <i>; yields bool:true</i>
2891 %X = fp2sint float 1.04E+17 to sbyte <i>; yields undefined:1</i>
2895 <!-- _______________________________________________________________________ -->
2896 <div class="doc_subsubsection">
2897 <a name="i_uint2fp">'<tt>uint2fp .. to</tt>' Instruction</a>
2899 <div class="doc_text">
2903 <result> = uint2fp <ty> <value> to <ty2> <i>; yields ty2</i>
2907 <p>The '<tt>uint2fp</tt>' instruction regards <tt>value</tt> as an unsigned
2908 integer and converts that value to the <tt>ty2</tt> type.</p>
2912 <p>The '<tt>uint2fp</tt>' instruction takes a value to cast, which must be an
2913 <a href="#t_integral">integral</a> value, and a type to cast it to, which must
2914 be a <a href="#t_floating">floating point</a> type.</p>
2917 <p>The '<tt>uint2fp</tt>' instruction interprets its operand as an unsigned
2918 integer quantity and converts it to the corresponding floating point value. If
2919 the value cannot fit in the floating point value, the results are undefined.</p>
2924 %X = uint2fp int 257 to float <i>; yields float:257.0</i>
2925 %Y = uint2fp sbyte -1 to double <i>; yields double:255.0</i>
2929 <!-- _______________________________________________________________________ -->
2930 <div class="doc_subsubsection">
2931 <a name="i_sint2fp">'<tt>sint2fp .. to</tt>' Instruction</a>
2933 <div class="doc_text">
2937 <result> = sint2fp <ty> <value> to <ty2> <i>; yields ty2</i>
2941 <p>The '<tt>sint2fp</tt>' instruction regards <tt>value</tt> as a signed
2942 integer and converts that value to the <tt>ty2</tt> type.</p>
2945 <p>The '<tt>sint2fp</tt>' instruction takes a value to cast, which must be an
2946 <a href="#t_integral">integral</a> value, and a type to cast it to, which must be
2947 a <a href="#t_floating">floating point</a> type.</p>
2950 <p>The '<tt>sint2fp</tt>' instruction interprets its operand as a signed
2951 integer quantity and converts it to the corresponding floating point value. If
2952 the value cannot fit in the floating point value, the results are undefined.</p>
2956 %X = sint2fp int 257 to float <i>; yields float:257.0</i>
2957 %Y = sint2fp sbyte -1 to double <i>; yields double:-1.0</i>
2961 <!-- _______________________________________________________________________ -->
2962 <div class="doc_subsubsection">
2963 <a name="i_bitconvert">'<tt>bitconvert .. to</tt>' Instruction</a>
2965 <div class="doc_text">
2969 <result> = bitconvert <ty> <value> to <ty2> <i>; yields ty2</i>
2973 <p>The '<tt>bitconvert</tt>' instruction converts <tt>value</tt> to type
2974 <tt>ty2</tt> without changing any bits.</p>
2977 <p>The '<tt>bitconvert</tt>' instruction takes a value to cast, which must be
2978 a first class value, and a type to cast it to, which must also be a <a
2979 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
2980 and the destination type, <tt>ty2</tt>, must be identical.</p>
2983 <p>The '<tt>bitconvert</tt>' instruction converts <tt>value</tt> to type
2984 <tt>ty2</tt> as if the value had been stored to memory and read back as type
2985 <tt>ty2</tt>. That is, no bits are changed during the conversion. The
2986 <tt>bitconvert</tt> instruction may be used to construct <i>no-op casts</i> that
2987 the <tt>zext, sext, and fpext</tt> instructions do not permit.</p>
2991 %X = bitconvert ubyte 255 to sbyte <i>; yields sbyte:-1</i>
2992 %Y = bitconvert uint* %x to uint <i>; yields uint:%x</i>
2996 <!-- ======================================================================= -->
2997 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
2998 <div class="doc_text">
2999 <p>The instructions in this category are the "miscellaneous"
3000 instructions, which defy better classification.</p>
3002 <!-- _______________________________________________________________________ -->
3003 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3004 Instruction</a> </div>
3005 <div class="doc_text">
3007 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3009 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3010 the SSA graph representing the function.</p>
3012 <p>The type of the incoming values are specified with the first type
3013 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3014 as arguments, with one pair for each predecessor basic block of the
3015 current block. Only values of <a href="#t_firstclass">first class</a>
3016 type may be used as the value arguments to the PHI node. Only labels
3017 may be used as the label arguments.</p>
3018 <p>There must be no non-phi instructions between the start of a basic
3019 block and the PHI instructions: i.e. PHI instructions must be first in
3022 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
3023 value specified by the parameter, depending on which basic block we
3024 came from in the last <a href="#terminators">terminator</a> instruction.</p>
3026 <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>
3029 <!-- _______________________________________________________________________ -->
3030 <div class="doc_subsubsection">
3031 <a name="i_select">'<tt>select</tt>' Instruction</a>
3034 <div class="doc_text">
3039 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3045 The '<tt>select</tt>' instruction is used to choose one value based on a
3046 condition, without branching.
3053 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.
3059 If the boolean condition evaluates to true, the instruction returns the first
3060 value argument; otherwise, it returns the second value argument.
3066 %X = select bool true, ubyte 17, ubyte 42 <i>; yields ubyte:17</i>
3071 <!-- _______________________________________________________________________ -->
3072 <div class="doc_subsubsection">
3073 <a name="i_call">'<tt>call</tt>' Instruction</a>
3076 <div class="doc_text">
3080 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
3085 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3089 <p>This instruction requires several arguments:</p>
3093 <p>The optional "tail" marker indicates whether the callee function accesses
3094 any allocas or varargs in the caller. If the "tail" marker is present, the
3095 function call is eligible for tail call optimization. Note that calls may
3096 be marked "tail" even if they do not occur before a <a
3097 href="#i_ret"><tt>ret</tt></a> instruction.
3100 <p>The optional "cconv" marker indicates which <a href="callingconv">calling
3101 convention</a> the call should use. If none is specified, the call defaults
3102 to using C calling conventions.
3105 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
3106 being invoked. The argument types must match the types implied by this
3107 signature. This type can be omitted if the function is not varargs and
3108 if the function type does not return a pointer to a function.</p>
3111 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3112 be invoked. In most cases, this is a direct function invocation, but
3113 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3114 to function value.</p>
3117 <p>'<tt>function args</tt>': argument list whose types match the
3118 function signature argument types. All arguments must be of
3119 <a href="#t_firstclass">first class</a> type. If the function signature
3120 indicates the function accepts a variable number of arguments, the extra
3121 arguments can be specified.</p>
3127 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3128 transfer to a specified function, with its incoming arguments bound to
3129 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3130 instruction in the called function, control flow continues with the
3131 instruction after the function call, and the return value of the
3132 function is bound to the result argument. This is a simpler case of
3133 the <a href="#i_invoke">invoke</a> instruction.</p>
3138 %retval = call int %test(int %argc)
3139 call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);
3140 %X = tail call int %foo()
3141 %Y = tail call <a href="#callingconv">fastcc</a> int %foo()
3146 <!-- _______________________________________________________________________ -->
3147 <div class="doc_subsubsection">
3148 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3151 <div class="doc_text">
3156 <resultval> = va_arg <va_list*> <arglist>, <argty>
3161 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3162 the "variable argument" area of a function call. It is used to implement the
3163 <tt>va_arg</tt> macro in C.</p>
3167 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3168 the argument. It returns a value of the specified argument type and
3169 increments the <tt>va_list</tt> to point to the next argument. Again, the
3170 actual type of <tt>va_list</tt> is target specific.</p>
3174 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3175 type from the specified <tt>va_list</tt> and causes the
3176 <tt>va_list</tt> to point to the next argument. For more information,
3177 see the variable argument handling <a href="#int_varargs">Intrinsic
3180 <p>It is legal for this instruction to be called in a function which does not
3181 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3184 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3185 href="#intrinsics">intrinsic function</a> because it takes a type as an
3190 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3194 <!-- *********************************************************************** -->
3195 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3196 <!-- *********************************************************************** -->
3198 <div class="doc_text">
3200 <p>LLVM supports the notion of an "intrinsic function". These functions have
3201 well known names and semantics and are required to follow certain
3202 restrictions. Overall, these instructions represent an extension mechanism for
3203 the LLVM language that does not require changing all of the transformations in
3204 LLVM to add to the language (or the bytecode reader/writer, the parser,
3207 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3208 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
3209 this. Intrinsic functions must always be external functions: you cannot define
3210 the body of intrinsic functions. Intrinsic functions may only be used in call
3211 or invoke instructions: it is illegal to take the address of an intrinsic
3212 function. Additionally, because intrinsic functions are part of the LLVM
3213 language, it is required that they all be documented here if any are added.</p>
3216 <p>To learn how to add an intrinsic function, please see the <a
3217 href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3222 <!-- ======================================================================= -->
3223 <div class="doc_subsection">
3224 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3227 <div class="doc_text">
3229 <p>Variable argument support is defined in LLVM with the <a
3230 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3231 intrinsic functions. These functions are related to the similarly
3232 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3234 <p>All of these functions operate on arguments that use a
3235 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3236 language reference manual does not define what this type is, so all
3237 transformations should be prepared to handle intrinsics with any type
3240 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3241 instruction and the variable argument handling intrinsic functions are
3245 int %test(int %X, ...) {
3246 ; Initialize variable argument processing
3248 call void %<a href="#i_va_start">llvm.va_start</a>(sbyte** %ap)
3250 ; Read a single integer argument
3251 %tmp = va_arg sbyte** %ap, int
3253 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3255 call void %<a href="#i_va_copy">llvm.va_copy</a>(sbyte** %aq, sbyte** %ap)
3256 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %aq)
3258 ; Stop processing of arguments.
3259 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %ap)
3265 <!-- _______________________________________________________________________ -->
3266 <div class="doc_subsubsection">
3267 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3271 <div class="doc_text">
3273 <pre> declare void %llvm.va_start(<va_list>* <arglist>)<br></pre>
3275 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3276 <tt>*<arglist></tt> for subsequent use by <tt><a
3277 href="#i_va_arg">va_arg</a></tt>.</p>
3281 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3285 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3286 macro available in C. In a target-dependent way, it initializes the
3287 <tt>va_list</tt> element the argument points to, so that the next call to
3288 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3289 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3290 last argument of the function, the compiler can figure that out.</p>
3294 <!-- _______________________________________________________________________ -->
3295 <div class="doc_subsubsection">
3296 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
3299 <div class="doc_text">
3301 <pre> declare void %llvm.va_end(<va_list*> <arglist>)<br></pre>
3303 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
3304 which has been initialized previously with <tt><a href="#i_va_start">llvm.va_start</a></tt>
3305 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
3307 <p>The argument is a <tt>va_list</tt> to destroy.</p>
3309 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
3310 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
3311 Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and <a
3312 href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
3313 with calls to <tt>llvm.va_end</tt>.</p>
3316 <!-- _______________________________________________________________________ -->
3317 <div class="doc_subsubsection">
3318 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
3321 <div class="doc_text">
3326 declare void %llvm.va_copy(<va_list>* <destarglist>,
3327 <va_list>* <srcarglist>)
3332 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
3333 the source argument list to the destination argument list.</p>
3337 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
3338 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
3343 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
3344 available in C. In a target-dependent way, it copies the source
3345 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
3346 because the <tt><a href="i_va_begin">llvm.va_begin</a></tt> intrinsic may be
3347 arbitrarily complex and require memory allocation, for example.</p>
3351 <!-- ======================================================================= -->
3352 <div class="doc_subsection">
3353 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
3356 <div class="doc_text">
3359 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
3360 Collection</a> requires the implementation and generation of these intrinsics.
3361 These intrinsics allow identification of <a href="#i_gcroot">GC roots on the
3362 stack</a>, as well as garbage collector implementations that require <a
3363 href="#i_gcread">read</a> and <a href="#i_gcwrite">write</a> barriers.
3364 Front-ends for type-safe garbage collected languages should generate these
3365 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3366 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
3370 <!-- _______________________________________________________________________ -->
3371 <div class="doc_subsubsection">
3372 <a name="i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
3375 <div class="doc_text">
3380 declare void %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
3385 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
3386 the code generator, and allows some metadata to be associated with it.</p>
3390 <p>The first argument specifies the address of a stack object that contains the
3391 root pointer. The second pointer (which must be either a constant or a global
3392 value address) contains the meta-data to be associated with the root.</p>
3396 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
3397 location. At compile-time, the code generator generates information to allow
3398 the runtime to find the pointer at GC safe points.
3404 <!-- _______________________________________________________________________ -->
3405 <div class="doc_subsubsection">
3406 <a name="i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
3409 <div class="doc_text">
3414 declare sbyte* %llvm.gcread(sbyte* %ObjPtr, sbyte** %Ptr)
3419 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
3420 locations, allowing garbage collector implementations that require read
3425 <p>The second argument is the address to read from, which should be an address
3426 allocated from the garbage collector. The first object is a pointer to the
3427 start of the referenced object, if needed by the language runtime (otherwise
3432 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
3433 instruction, but may be replaced with substantially more complex code by the
3434 garbage collector runtime, as needed.</p>
3439 <!-- _______________________________________________________________________ -->
3440 <div class="doc_subsubsection">
3441 <a name="i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
3444 <div class="doc_text">
3449 declare void %llvm.gcwrite(sbyte* %P1, sbyte* %Obj, sbyte** %P2)
3454 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
3455 locations, allowing garbage collector implementations that require write
3456 barriers (such as generational or reference counting collectors).</p>
3460 <p>The first argument is the reference to store, the second is the start of the
3461 object to store it to, and the third is the address of the field of Obj to
3462 store to. If the runtime does not require a pointer to the object, Obj may be
3467 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
3468 instruction, but may be replaced with substantially more complex code by the
3469 garbage collector runtime, as needed.</p>
3475 <!-- ======================================================================= -->
3476 <div class="doc_subsection">
3477 <a name="int_codegen">Code Generator Intrinsics</a>
3480 <div class="doc_text">
3482 These intrinsics are provided by LLVM to expose special features that may only
3483 be implemented with code generator support.
3488 <!-- _______________________________________________________________________ -->
3489 <div class="doc_subsubsection">
3490 <a name="i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
3493 <div class="doc_text">
3497 declare sbyte *%llvm.returnaddress(uint <level>)
3503 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
3504 target-specific value indicating the return address of the current function
3505 or one of its callers.
3511 The argument to this intrinsic indicates which function to return the address
3512 for. Zero indicates the calling function, one indicates its caller, etc. The
3513 argument is <b>required</b> to be a constant integer value.
3519 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
3520 the return address of the specified call frame, or zero if it cannot be
3521 identified. The value returned by this intrinsic is likely to be incorrect or 0
3522 for arguments other than zero, so it should only be used for debugging purposes.
3526 Note that calling this intrinsic does not prevent function inlining or other
3527 aggressive transformations, so the value returned may not be that of the obvious
3528 source-language caller.
3533 <!-- _______________________________________________________________________ -->
3534 <div class="doc_subsubsection">
3535 <a name="i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
3538 <div class="doc_text">
3542 declare sbyte *%llvm.frameaddress(uint <level>)
3548 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
3549 target-specific frame pointer value for the specified stack frame.
3555 The argument to this intrinsic indicates which function to return the frame
3556 pointer for. Zero indicates the calling function, one indicates its caller,
3557 etc. The argument is <b>required</b> to be a constant integer value.
3563 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
3564 the frame address of the specified call frame, or zero if it cannot be
3565 identified. The value returned by this intrinsic is likely to be incorrect or 0
3566 for arguments other than zero, so it should only be used for debugging purposes.
3570 Note that calling this intrinsic does not prevent function inlining or other
3571 aggressive transformations, so the value returned may not be that of the obvious
3572 source-language caller.
3576 <!-- _______________________________________________________________________ -->
3577 <div class="doc_subsubsection">
3578 <a name="i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
3581 <div class="doc_text">
3585 declare sbyte *%llvm.stacksave()
3591 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
3592 the function stack, for use with <a href="#i_stackrestore">
3593 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
3594 features like scoped automatic variable sized arrays in C99.
3600 This intrinsic returns a opaque pointer value that can be passed to <a
3601 href="#i_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
3602 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
3603 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
3604 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
3605 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
3606 that were allocated after the <tt>llvm.stacksave</tt> was executed.
3611 <!-- _______________________________________________________________________ -->
3612 <div class="doc_subsubsection">
3613 <a name="i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
3616 <div class="doc_text">
3620 declare void %llvm.stackrestore(sbyte* %ptr)
3626 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
3627 the function stack to the state it was in when the corresponding <a
3628 href="#llvm.stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
3629 useful for implementing language features like scoped automatic variable sized
3636 See the description for <a href="#i_stacksave"><tt>llvm.stacksave</tt></a>.
3642 <!-- _______________________________________________________________________ -->
3643 <div class="doc_subsubsection">
3644 <a name="i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
3647 <div class="doc_text">
3651 declare void %llvm.prefetch(sbyte * <address>,
3652 uint <rw>, uint <locality>)
3659 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
3660 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
3662 effect on the behavior of the program but can change its performance
3669 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
3670 determining if the fetch should be for a read (0) or write (1), and
3671 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
3672 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
3673 <tt>locality</tt> arguments must be constant integers.
3679 This intrinsic does not modify the behavior of the program. In particular,
3680 prefetches cannot trap and do not produce a value. On targets that support this
3681 intrinsic, the prefetch can provide hints to the processor cache for better
3687 <!-- _______________________________________________________________________ -->
3688 <div class="doc_subsubsection">
3689 <a name="i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
3692 <div class="doc_text">
3696 declare void %llvm.pcmarker( uint <id> )
3703 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
3705 code to simulators and other tools. The method is target specific, but it is
3706 expected that the marker will use exported symbols to transmit the PC of the marker.
3707 The marker makes no guarantees that it will remain with any specific instruction
3708 after optimizations. It is possible that the presence of a marker will inhibit
3709 optimizations. The intended use is to be inserted after optimizations to allow
3710 correlations of simulation runs.
3716 <tt>id</tt> is a numerical id identifying the marker.
3722 This intrinsic does not modify the behavior of the program. Backends that do not
3723 support this intrinisic may ignore it.
3728 <!-- _______________________________________________________________________ -->
3729 <div class="doc_subsubsection">
3730 <a name="i_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
3733 <div class="doc_text">
3737 declare ulong %llvm.readcyclecounter( )
3744 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
3745 counter register (or similar low latency, high accuracy clocks) on those targets
3746 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
3747 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
3748 should only be used for small timings.
3754 When directly supported, reading the cycle counter should not modify any memory.
3755 Implementations are allowed to either return a application specific value or a
3756 system wide value. On backends without support, this is lowered to a constant 0.
3761 <!-- ======================================================================= -->
3762 <div class="doc_subsection">
3763 <a name="int_libc">Standard C Library Intrinsics</a>
3766 <div class="doc_text">
3768 LLVM provides intrinsics for a few important standard C library functions.
3769 These intrinsics allow source-language front-ends to pass information about the
3770 alignment of the pointer arguments to the code generator, providing opportunity
3771 for more efficient code generation.
3776 <!-- _______________________________________________________________________ -->
3777 <div class="doc_subsubsection">
3778 <a name="i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
3781 <div class="doc_text">
3785 declare void %llvm.memcpy.i32(sbyte* <dest>, sbyte* <src>,
3786 uint <len>, uint <align>)
3787 declare void %llvm.memcpy.i64(sbyte* <dest>, sbyte* <src>,
3788 ulong <len>, uint <align>)
3794 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
3795 location to the destination location.
3799 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
3800 intrinsics do not return a value, and takes an extra alignment argument.
3806 The first argument is a pointer to the destination, the second is a pointer to
3807 the source. The third argument is an integer argument
3808 specifying the number of bytes to copy, and the fourth argument is the alignment
3809 of the source and destination locations.
3813 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3814 the caller guarantees that both the source and destination pointers are aligned
3821 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
3822 location to the destination location, which are not allowed to overlap. It
3823 copies "len" bytes of memory over. If the argument is known to be aligned to
3824 some boundary, this can be specified as the fourth argument, otherwise it should
3830 <!-- _______________________________________________________________________ -->
3831 <div class="doc_subsubsection">
3832 <a name="i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
3835 <div class="doc_text">
3839 declare void %llvm.memmove.i32(sbyte* <dest>, sbyte* <src>,
3840 uint <len>, uint <align>)
3841 declare void %llvm.memmove.i64(sbyte* <dest>, sbyte* <src>,
3842 ulong <len>, uint <align>)
3848 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
3849 location to the destination location. It is similar to the
3850 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
3854 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
3855 intrinsics do not return a value, and takes an extra alignment argument.
3861 The first argument is a pointer to the destination, the second is a pointer to
3862 the source. The third argument is an integer argument
3863 specifying the number of bytes to copy, and the fourth argument is the alignment
3864 of the source and destination locations.
3868 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3869 the caller guarantees that the source and destination pointers are aligned to
3876 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
3877 location to the destination location, which may overlap. It
3878 copies "len" bytes of memory over. If the argument is known to be aligned to
3879 some boundary, this can be specified as the fourth argument, otherwise it should
3885 <!-- _______________________________________________________________________ -->
3886 <div class="doc_subsubsection">
3887 <a name="i_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
3890 <div class="doc_text">
3894 declare void %llvm.memset.i32(sbyte* <dest>, ubyte <val>,
3895 uint <len>, uint <align>)
3896 declare void %llvm.memset.i64(sbyte* <dest>, ubyte <val>,
3897 ulong <len>, uint <align>)
3903 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
3908 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
3909 does not return a value, and takes an extra alignment argument.
3915 The first argument is a pointer to the destination to fill, the second is the
3916 byte value to fill it with, the third argument is an integer
3917 argument specifying the number of bytes to fill, and the fourth argument is the
3918 known alignment of destination location.
3922 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3923 the caller guarantees that the destination pointer is aligned to that boundary.
3929 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
3931 destination location. If the argument is known to be aligned to some boundary,
3932 this can be specified as the fourth argument, otherwise it should be set to 0 or
3938 <!-- _______________________________________________________________________ -->
3939 <div class="doc_subsubsection">
3940 <a name="i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a>
3943 <div class="doc_text">
3947 declare bool %llvm.isunordered.f32(float Val1, float Val2)
3948 declare bool %llvm.isunordered.f64(double Val1, double Val2)
3954 The '<tt>llvm.isunordered</tt>' intrinsics return true if either or both of the
3955 specified floating point values is a NAN.
3961 The arguments are floating point numbers of the same type.
3967 If either or both of the arguments is a SNAN or QNAN, it returns true, otherwise
3973 <!-- _______________________________________________________________________ -->
3974 <div class="doc_subsubsection">
3975 <a name="i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
3978 <div class="doc_text">
3982 declare float %llvm.sqrt.f32(float %Val)
3983 declare double %llvm.sqrt.f64(double %Val)
3989 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
3990 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
3991 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
3992 negative numbers (which allows for better optimization).
3998 The argument and return value are floating point numbers of the same type.
4004 This function returns the sqrt of the specified operand if it is a positive
4005 floating point number.
4009 <!-- _______________________________________________________________________ -->
4010 <div class="doc_subsubsection">
4011 <a name="i_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4014 <div class="doc_text">
4018 declare float %llvm.powi.f32(float %Val, int %power)
4019 declare double %llvm.powi.f64(double %Val, int %power)
4025 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4026 specified (positive or negative) power. The order of evaluation of
4027 multiplications is not defined.
4033 The second argument is an integer power, and the first is a value to raise to
4040 This function returns the first value raised to the second power with an
4041 unspecified sequence of rounding operations.</p>
4045 <!-- ======================================================================= -->
4046 <div class="doc_subsection">
4047 <a name="int_manip">Bit Manipulation Intrinsics</a>
4050 <div class="doc_text">
4052 LLVM provides intrinsics for a few important bit manipulation operations.
4053 These allow efficient code generation for some algorithms.
4058 <!-- _______________________________________________________________________ -->
4059 <div class="doc_subsubsection">
4060 <a name="i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4063 <div class="doc_text">
4067 declare ushort %llvm.bswap.i16(ushort <id>)
4068 declare uint %llvm.bswap.i32(uint <id>)
4069 declare ulong %llvm.bswap.i64(ulong <id>)
4075 The '<tt>llvm.bwsap</tt>' family of intrinsics is used to byteswap a 16, 32 or
4076 64 bit quantity. These are useful for performing operations on data that is not
4077 in the target's native byte order.
4083 The <tt>llvm.bswap.16</tt> intrinsic returns a ushort value that has the high and low
4084 byte of the input ushort swapped. Similarly, the <tt>llvm.bswap.i32</tt> intrinsic
4085 returns a uint value that has the four bytes of the input uint swapped, so that
4086 if the input bytes are numbered 0, 1, 2, 3 then the returned uint will have its
4087 bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i64</tt> intrinsic extends this concept
4093 <!-- _______________________________________________________________________ -->
4094 <div class="doc_subsubsection">
4095 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4098 <div class="doc_text">
4102 declare ubyte %llvm.ctpop.i8 (ubyte <src>)
4103 declare ushort %llvm.ctpop.i16(ushort <src>)
4104 declare uint %llvm.ctpop.i32(uint <src>)
4105 declare ulong %llvm.ctpop.i64(ulong <src>)
4111 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4118 The only argument is the value to be counted. The argument may be of any
4119 unsigned integer type. The return type must match the argument type.
4125 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4129 <!-- _______________________________________________________________________ -->
4130 <div class="doc_subsubsection">
4131 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4134 <div class="doc_text">
4138 declare ubyte %llvm.ctlz.i8 (ubyte <src>)
4139 declare ushort %llvm.ctlz.i16(ushort <src>)
4140 declare uint %llvm.ctlz.i32(uint <src>)
4141 declare ulong %llvm.ctlz.i64(ulong <src>)
4147 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4148 leading zeros in a variable.
4154 The only argument is the value to be counted. The argument may be of any
4155 unsigned integer type. The return type must match the argument type.
4161 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4162 in a variable. If the src == 0 then the result is the size in bits of the type
4163 of src. For example, <tt>llvm.ctlz(int 2) = 30</tt>.
4169 <!-- _______________________________________________________________________ -->
4170 <div class="doc_subsubsection">
4171 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
4174 <div class="doc_text">
4178 declare ubyte %llvm.cttz.i8 (ubyte <src>)
4179 declare ushort %llvm.cttz.i16(ushort <src>)
4180 declare uint %llvm.cttz.i32(uint <src>)
4181 declare ulong %llvm.cttz.i64(ulong <src>)
4187 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
4194 The only argument is the value to be counted. The argument may be of any
4195 unsigned integer type. The return type must match the argument type.
4201 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
4202 in a variable. If the src == 0 then the result is the size in bits of the type
4203 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
4207 <!-- ======================================================================= -->
4208 <div class="doc_subsection">
4209 <a name="int_debugger">Debugger Intrinsics</a>
4212 <div class="doc_text">
4214 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
4215 are described in the <a
4216 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
4217 Debugging</a> document.
4222 <!-- *********************************************************************** -->
4225 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
4226 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
4227 <a href="http://validator.w3.org/check/referer"><img
4228 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
4230 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
4231 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
4232 Last modified: $Date$