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_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
124 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
125 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
126 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
127 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
128 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
129 <li><a href="#i_bitconvert">'<tt>bitconvert .. to</tt>' Instruction</a></li>
131 <li><a href="#otherops">Other Operations</a>
133 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
134 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
135 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
136 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
141 <li><a href="#intrinsics">Intrinsic Functions</a>
143 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
145 <li><a href="#i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
146 <li><a href="#i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
147 <li><a href="#i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
150 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
152 <li><a href="#i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
153 <li><a href="#i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
154 <li><a href="#i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
157 <li><a href="#int_codegen">Code Generator Intrinsics</a>
159 <li><a href="#i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
160 <li><a href="#i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
161 <li><a href="#i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
162 <li><a href="#i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
163 <li><a href="#i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
164 <li><a href="#i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
165 <li><a href="#i_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
168 <li><a href="#int_libc">Standard C Library Intrinsics</a>
170 <li><a href="#i_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
171 <li><a href="#i_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
172 <li><a href="#i_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
173 <li><a href="#i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a></li>
174 <li><a href="#i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
175 <li><a href="#i_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
178 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
180 <li><a href="#i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
181 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
182 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
183 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
186 <li><a href="#int_debugger">Debugger intrinsics</a></li>
191 <div class="doc_author">
192 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
193 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
196 <!-- *********************************************************************** -->
197 <div class="doc_section"> <a name="abstract">Abstract </a></div>
198 <!-- *********************************************************************** -->
200 <div class="doc_text">
201 <p>This document is a reference manual for the LLVM assembly language.
202 LLVM is an SSA based representation that provides type safety,
203 low-level operations, flexibility, and the capability of representing
204 'all' high-level languages cleanly. It is the common code
205 representation used throughout all phases of the LLVM compilation
209 <!-- *********************************************************************** -->
210 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
211 <!-- *********************************************************************** -->
213 <div class="doc_text">
215 <p>The LLVM code representation is designed to be used in three
216 different forms: as an in-memory compiler IR, as an on-disk bytecode
217 representation (suitable for fast loading by a Just-In-Time compiler),
218 and as a human readable assembly language representation. This allows
219 LLVM to provide a powerful intermediate representation for efficient
220 compiler transformations and analysis, while providing a natural means
221 to debug and visualize the transformations. The three different forms
222 of LLVM are all equivalent. This document describes the human readable
223 representation and notation.</p>
225 <p>The LLVM representation aims to be light-weight and low-level
226 while being expressive, typed, and extensible at the same time. It
227 aims to be a "universal IR" of sorts, by being at a low enough level
228 that high-level ideas may be cleanly mapped to it (similar to how
229 microprocessors are "universal IR's", allowing many source languages to
230 be mapped to them). By providing type information, LLVM can be used as
231 the target of optimizations: for example, through pointer analysis, it
232 can be proven that a C automatic variable is never accessed outside of
233 the current function... allowing it to be promoted to a simple SSA
234 value instead of a memory location.</p>
238 <!-- _______________________________________________________________________ -->
239 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
241 <div class="doc_text">
243 <p>It is important to note that this document describes 'well formed'
244 LLVM assembly language. There is a difference between what the parser
245 accepts and what is considered 'well formed'. For example, the
246 following instruction is syntactically okay, but not well formed:</p>
249 %x = <a href="#i_add">add</a> int 1, %x
252 <p>...because the definition of <tt>%x</tt> does not dominate all of
253 its uses. The LLVM infrastructure provides a verification pass that may
254 be used to verify that an LLVM module is well formed. This pass is
255 automatically run by the parser after parsing input assembly and by
256 the optimizer before it outputs bytecode. The violations pointed out
257 by the verifier pass indicate bugs in transformation passes or input to
260 <!-- Describe the typesetting conventions here. --> </div>
262 <!-- *********************************************************************** -->
263 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
264 <!-- *********************************************************************** -->
266 <div class="doc_text">
268 <p>LLVM uses three different forms of identifiers, for different
272 <li>Named values are represented as a string of characters with a '%' prefix.
273 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
274 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
275 Identifiers which require other characters in their names can be surrounded
276 with quotes. In this way, anything except a <tt>"</tt> character can be used
279 <li>Unnamed values are represented as an unsigned numeric value with a '%'
280 prefix. For example, %12, %2, %44.</li>
282 <li>Constants, which are described in a <a href="#constants">section about
283 constants</a>, below.</li>
286 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
287 don't need to worry about name clashes with reserved words, and the set of
288 reserved words may be expanded in the future without penalty. Additionally,
289 unnamed identifiers allow a compiler to quickly come up with a temporary
290 variable without having to avoid symbol table conflicts.</p>
292 <p>Reserved words in LLVM are very similar to reserved words in other
293 languages. There are keywords for different opcodes ('<tt><a
294 href="#i_add">add</a></tt>', '<tt><a href="#i_cast">cast</a></tt>', '<tt><a
295 href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
296 href="#t_void">void</a></tt>', '<tt><a href="#t_uint">uint</a></tt>', etc...),
297 and others. These reserved words cannot conflict with variable names, because
298 none of them start with a '%' character.</p>
300 <p>Here is an example of LLVM code to multiply the integer variable
301 '<tt>%X</tt>' by 8:</p>
306 %result = <a href="#i_mul">mul</a> uint %X, 8
309 <p>After strength reduction:</p>
312 %result = <a href="#i_shl">shl</a> uint %X, ubyte 3
315 <p>And the hard way:</p>
318 <a href="#i_add">add</a> uint %X, %X <i>; yields {uint}:%0</i>
319 <a href="#i_add">add</a> uint %0, %0 <i>; yields {uint}:%1</i>
320 %result = <a href="#i_add">add</a> uint %1, %1
323 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
324 important lexical features of LLVM:</p>
328 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
331 <li>Unnamed temporaries are created when the result of a computation is not
332 assigned to a named value.</li>
334 <li>Unnamed temporaries are numbered sequentially</li>
338 <p>...and it also shows a convention that we follow in this document. When
339 demonstrating instructions, we will follow an instruction with a comment that
340 defines the type and name of value produced. Comments are shown in italic
345 <!-- *********************************************************************** -->
346 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
347 <!-- *********************************************************************** -->
349 <!-- ======================================================================= -->
350 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
353 <div class="doc_text">
355 <p>LLVM programs are composed of "Module"s, each of which is a
356 translation unit of the input programs. Each module consists of
357 functions, global variables, and symbol table entries. Modules may be
358 combined together with the LLVM linker, which merges function (and
359 global variable) definitions, resolves forward declarations, and merges
360 symbol table entries. Here is an example of the "hello world" module:</p>
362 <pre><i>; Declare the string constant as a global constant...</i>
363 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
364 href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
366 <i>; External declaration of the puts function</i>
367 <a href="#functionstructure">declare</a> int %puts(sbyte*) <i>; int(sbyte*)* </i>
369 <i>; Global variable / Function body section separator</i>
372 <i>; Definition of main function</i>
373 int %main() { <i>; int()* </i>
374 <i>; Convert [13x sbyte]* to sbyte *...</i>
376 href="#i_getelementptr">getelementptr</a> [13 x sbyte]* %.LC0, long 0, long 0 <i>; sbyte*</i>
378 <i>; Call puts function to write out the string to stdout...</i>
380 href="#i_call">call</a> int %puts(sbyte* %cast210) <i>; int</i>
382 href="#i_ret">ret</a> int 0<br>}<br></pre>
384 <p>This example is made up of a <a href="#globalvars">global variable</a>
385 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
386 function, and a <a href="#functionstructure">function definition</a>
387 for "<tt>main</tt>".</p>
389 <p>In general, a module is made up of a list of global values,
390 where both functions and global variables are global values. Global values are
391 represented by a pointer to a memory location (in this case, a pointer to an
392 array of char, and a pointer to a function), and have one of the following <a
393 href="#linkage">linkage types</a>.</p>
395 <p>Due to a limitation in the current LLVM assembly parser (it is limited by
396 one-token lookahead), modules are split into two pieces by the "implementation"
397 keyword. Global variable prototypes and definitions must occur before the
398 keyword, and function definitions must occur after it. Function prototypes may
399 occur either before or after it. In the future, the implementation keyword may
400 become a noop, if the parser gets smarter.</p>
404 <!-- ======================================================================= -->
405 <div class="doc_subsection">
406 <a name="linkage">Linkage Types</a>
409 <div class="doc_text">
412 All Global Variables and Functions have one of the following types of linkage:
417 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
419 <dd>Global values with internal linkage are only directly accessible by
420 objects in the current module. In particular, linking code into a module with
421 an internal global value may cause the internal to be renamed as necessary to
422 avoid collisions. Because the symbol is internal to the module, all
423 references can be updated. This corresponds to the notion of the
424 '<tt>static</tt>' keyword in C, or the idea of "anonymous namespaces" in C++.
427 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
429 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt> linkage, with
430 the twist that linking together two modules defining the same
431 <tt>linkonce</tt> globals will cause one of the globals to be discarded. This
432 is typically used to implement inline functions. Unreferenced
433 <tt>linkonce</tt> globals are allowed to be discarded.
436 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
438 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
439 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
440 used to implement constructs in C such as "<tt>int X;</tt>" at global scope.
443 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
445 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
446 pointer to array type. When two global variables with appending linkage are
447 linked together, the two global arrays are appended together. This is the
448 LLVM, typesafe, equivalent of having the system linker append together
449 "sections" with identical names when .o files are linked.
452 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
454 <dd>If none of the above identifiers are used, the global is externally
455 visible, meaning that it participates in linkage and can be used to resolve
456 external symbol references.
459 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
461 <dd>"<tt>extern_weak</tt>" TBD
465 The next two types of linkage are targeted for Microsoft Windows platform
466 only. They are designed to support importing (exporting) symbols from (to)
470 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
472 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
473 or variable via a global pointer to a pointer that is set up by the DLL
474 exporting the symbol. On Microsoft Windows targets, the pointer name is
475 formed by combining <code>_imp__</code> and the function or variable name.
478 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
480 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
481 pointer to a pointer in a DLL, so that it can be referenced with the
482 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
483 name is formed by combining <code>_imp__</code> and the function or variable
489 <p><a name="linkage_external">For example, since the "<tt>.LC0</tt>"
490 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
491 variable and was linked with this one, one of the two would be renamed,
492 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
493 external (i.e., lacking any linkage declarations), they are accessible
494 outside of the current module. It is illegal for a function <i>declaration</i>
495 to have any linkage type other than "externally visible".</a></p>
499 <!-- ======================================================================= -->
500 <div class="doc_subsection">
501 <a name="callingconv">Calling Conventions</a>
504 <div class="doc_text">
506 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
507 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
508 specified for the call. The calling convention of any pair of dynamic
509 caller/callee must match, or the behavior of the program is undefined. The
510 following calling conventions are supported by LLVM, and more may be added in
514 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
516 <dd>This calling convention (the default if no other calling convention is
517 specified) matches the target C calling conventions. This calling convention
518 supports varargs function calls and tolerates some mismatch in the declared
519 prototype and implemented declaration of the function (as does normal C).
522 <dt><b>"<tt>csretcc</tt>" - The C struct return calling convention</b>:</dt>
524 <dd>This calling convention matches the target C calling conventions, except
525 that functions with this convention are required to take a pointer as their
526 first argument, and the return type of the function must be void. This is
527 used for C functions that return aggregates by-value. In this case, the
528 function has been transformed to take a pointer to the struct as the first
529 argument to the function. For targets where the ABI specifies specific
530 behavior for structure-return calls, the calling convention can be used to
531 distinguish between struct return functions and other functions that take a
532 pointer to a struct as the first argument.
535 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
537 <dd>This calling convention attempts to make calls as fast as possible
538 (e.g. by passing things in registers). This calling convention allows the
539 target to use whatever tricks it wants to produce fast code for the target,
540 without having to conform to an externally specified ABI. Implementations of
541 this convention should allow arbitrary tail call optimization to be supported.
542 This calling convention does not support varargs and requires the prototype of
543 all callees to exactly match the prototype of the function definition.
546 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
548 <dd>This calling convention attempts to make code in the caller as efficient
549 as possible under the assumption that the call is not commonly executed. As
550 such, these calls often preserve all registers so that the call does not break
551 any live ranges in the caller side. This calling convention does not support
552 varargs and requires the prototype of all callees to exactly match the
553 prototype of the function definition.
556 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
558 <dd>Any calling convention may be specified by number, allowing
559 target-specific calling conventions to be used. Target specific calling
560 conventions start at 64.
564 <p>More calling conventions can be added/defined on an as-needed basis, to
565 support pascal conventions or any other well-known target-independent
570 <!-- ======================================================================= -->
571 <div class="doc_subsection">
572 <a name="globalvars">Global Variables</a>
575 <div class="doc_text">
577 <p>Global variables define regions of memory allocated at compilation time
578 instead of run-time. Global variables may optionally be initialized, may have
579 an explicit section to be placed in, and may
580 have an optional explicit alignment specified. A
581 variable may be defined as a global "constant," which indicates that the
582 contents of the variable will <b>never</b> be modified (enabling better
583 optimization, allowing the global data to be placed in the read-only section of
584 an executable, etc). Note that variables that need runtime initialization
585 cannot be marked "constant" as there is a store to the variable.</p>
588 LLVM explicitly allows <em>declarations</em> of global variables to be marked
589 constant, even if the final definition of the global is not. This capability
590 can be used to enable slightly better optimization of the program, but requires
591 the language definition to guarantee that optimizations based on the
592 'constantness' are valid for the translation units that do not include the
596 <p>As SSA values, global variables define pointer values that are in
597 scope (i.e. they dominate) all basic blocks in the program. Global
598 variables always define a pointer to their "content" type because they
599 describe a region of memory, and all memory objects in LLVM are
600 accessed through pointers.</p>
602 <p>LLVM allows an explicit section to be specified for globals. If the target
603 supports it, it will emit globals to the section specified.</p>
605 <p>An explicit alignment may be specified for a global. If not present, or if
606 the alignment is set to zero, the alignment of the global is set by the target
607 to whatever it feels convenient. If an explicit alignment is specified, the
608 global is forced to have at least that much alignment. All alignments must be
614 <!-- ======================================================================= -->
615 <div class="doc_subsection">
616 <a name="functionstructure">Functions</a>
619 <div class="doc_text">
621 <p>LLVM function definitions consist of an optional <a href="#linkage">linkage
622 type</a>, an optional <a href="#callingconv">calling convention</a>, a return
623 type, a function name, a (possibly empty) argument list, an optional section,
624 an optional alignment, an opening curly brace,
625 a list of basic blocks, and a closing curly brace. LLVM function declarations
626 are defined with the "<tt>declare</tt>" keyword, an optional <a
627 href="#callingconv">calling convention</a>, a return type, a function name,
628 a possibly empty list of arguments, and an optional alignment.</p>
630 <p>A function definition contains a list of basic blocks, forming the CFG for
631 the function. Each basic block may optionally start with a label (giving the
632 basic block a symbol table entry), contains a list of instructions, and ends
633 with a <a href="#terminators">terminator</a> instruction (such as a branch or
634 function return).</p>
636 <p>The first basic block in a program is special in two ways: it is immediately
637 executed on entrance to the function, and it is not allowed to have predecessor
638 basic blocks (i.e. there can not be any branches to the entry block of a
639 function). Because the block can have no predecessors, it also cannot have any
640 <a href="#i_phi">PHI nodes</a>.</p>
642 <p>LLVM functions are identified by their name and type signature. Hence, two
643 functions with the same name but different parameter lists or return values are
644 considered different functions, and LLVM will resolve references to each
647 <p>LLVM allows an explicit section to be specified for functions. If the target
648 supports it, it will emit functions to the section specified.</p>
650 <p>An explicit alignment may be specified for a function. If not present, or if
651 the alignment is set to zero, the alignment of the function is set by the target
652 to whatever it feels convenient. If an explicit alignment is specified, the
653 function is forced to have at least that much alignment. All alignments must be
658 <!-- ======================================================================= -->
659 <div class="doc_subsection">
660 <a name="moduleasm">Module-Level Inline Assembly</a>
663 <div class="doc_text">
665 Modules may contain "module-level inline asm" blocks, which corresponds to the
666 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
667 LLVM and treated as a single unit, but may be separated in the .ll file if
668 desired. The syntax is very simple:
671 <div class="doc_code"><pre>
672 module asm "inline asm code goes here"
673 module asm "more can go here"
676 <p>The strings can contain any character by escaping non-printable characters.
677 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
682 The inline asm code is simply printed to the machine code .s file when
683 assembly code is generated.
688 <!-- *********************************************************************** -->
689 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
690 <!-- *********************************************************************** -->
692 <div class="doc_text">
694 <p>The LLVM type system is one of the most important features of the
695 intermediate representation. Being typed enables a number of
696 optimizations to be performed on the IR directly, without having to do
697 extra analyses on the side before the transformation. A strong type
698 system makes it easier to read the generated code and enables novel
699 analyses and transformations that are not feasible to perform on normal
700 three address code representations.</p>
704 <!-- ======================================================================= -->
705 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
706 <div class="doc_text">
707 <p>The primitive types are the fundamental building blocks of the LLVM
708 system. The current set of primitive types is as follows:</p>
710 <table class="layout">
715 <tr><th>Type</th><th>Description</th></tr>
716 <tr><td><tt>void</tt></td><td>No value</td></tr>
717 <tr><td><tt>ubyte</tt></td><td>Unsigned 8-bit value</td></tr>
718 <tr><td><tt>ushort</tt></td><td>Unsigned 16-bit value</td></tr>
719 <tr><td><tt>uint</tt></td><td>Unsigned 32-bit value</td></tr>
720 <tr><td><tt>ulong</tt></td><td>Unsigned 64-bit value</td></tr>
721 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
722 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
729 <tr><th>Type</th><th>Description</th></tr>
730 <tr><td><tt>bool</tt></td><td>True or False value</td></tr>
731 <tr><td><tt>sbyte</tt></td><td>Signed 8-bit value</td></tr>
732 <tr><td><tt>short</tt></td><td>Signed 16-bit value</td></tr>
733 <tr><td><tt>int</tt></td><td>Signed 32-bit value</td></tr>
734 <tr><td><tt>long</tt></td><td>Signed 64-bit value</td></tr>
735 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
743 <!-- _______________________________________________________________________ -->
744 <div class="doc_subsubsection"> <a name="t_classifications">Type
745 Classifications</a> </div>
746 <div class="doc_text">
747 <p>These different primitive types fall into a few useful
750 <table border="1" cellspacing="0" cellpadding="4">
752 <tr><th>Classification</th><th>Types</th></tr>
754 <td><a name="t_signed">signed</a></td>
755 <td><tt>sbyte, short, int, long, float, double</tt></td>
758 <td><a name="t_unsigned">unsigned</a></td>
759 <td><tt>ubyte, ushort, uint, ulong</tt></td>
762 <td><a name="t_integer">integer</a></td>
763 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
766 <td><a name="t_integral">integral</a></td>
767 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt>
771 <td><a name="t_floating">floating point</a></td>
772 <td><tt>float, double</tt></td>
775 <td><a name="t_firstclass">first class</a></td>
776 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long,<br>
777 float, double, <a href="#t_pointer">pointer</a>,
778 <a href="#t_packed">packed</a></tt></td>
783 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
784 most important. Values of these types are the only ones which can be
785 produced by instructions, passed as arguments, or used as operands to
786 instructions. This means that all structures and arrays must be
787 manipulated either by pointer or by component.</p>
790 <!-- ======================================================================= -->
791 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
793 <div class="doc_text">
795 <p>The real power in LLVM comes from the derived types in the system.
796 This is what allows a programmer to represent arrays, functions,
797 pointers, and other useful types. Note that these derived types may be
798 recursive: For example, it is possible to have a two dimensional array.</p>
802 <!-- _______________________________________________________________________ -->
803 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
805 <div class="doc_text">
809 <p>The array type is a very simple derived type that arranges elements
810 sequentially in memory. The array type requires a size (number of
811 elements) and an underlying data type.</p>
816 [<# elements> x <elementtype>]
819 <p>The number of elements is a constant integer value; elementtype may
820 be any type with a size.</p>
823 <table class="layout">
826 <tt>[40 x int ]</tt><br/>
827 <tt>[41 x int ]</tt><br/>
828 <tt>[40 x uint]</tt><br/>
831 Array of 40 integer values.<br/>
832 Array of 41 integer values.<br/>
833 Array of 40 unsigned integer values.<br/>
837 <p>Here are some examples of multidimensional arrays:</p>
838 <table class="layout">
841 <tt>[3 x [4 x int]]</tt><br/>
842 <tt>[12 x [10 x float]]</tt><br/>
843 <tt>[2 x [3 x [4 x uint]]]</tt><br/>
846 3x4 array of integer values.<br/>
847 12x10 array of single precision floating point values.<br/>
848 2x3x4 array of unsigned integer values.<br/>
853 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
854 length array. Normally, accesses past the end of an array are undefined in
855 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
856 As a special case, however, zero length arrays are recognized to be variable
857 length. This allows implementation of 'pascal style arrays' with the LLVM
858 type "{ int, [0 x float]}", for example.</p>
862 <!-- _______________________________________________________________________ -->
863 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
864 <div class="doc_text">
866 <p>The function type can be thought of as a function signature. It
867 consists of a return type and a list of formal parameter types.
868 Function types are usually used to build virtual function tables
869 (which are structures of pointers to functions), for indirect function
870 calls, and when defining a function.</p>
872 The return type of a function type cannot be an aggregate type.
875 <pre> <returntype> (<parameter list>)<br></pre>
876 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
877 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
878 which indicates that the function takes a variable number of arguments.
879 Variable argument functions can access their arguments with the <a
880 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
882 <table class="layout">
885 <tt>int (int)</tt> <br/>
886 <tt>float (int, int *) *</tt><br/>
887 <tt>int (sbyte *, ...)</tt><br/>
890 function taking an <tt>int</tt>, returning an <tt>int</tt><br/>
891 <a href="#t_pointer">Pointer</a> to a function that takes an
892 <tt>int</tt> and a <a href="#t_pointer">pointer</a> to <tt>int</tt>,
893 returning <tt>float</tt>.<br/>
894 A vararg function that takes at least one <a href="#t_pointer">pointer</a>
895 to <tt>sbyte</tt> (signed char in C), which returns an integer. This is
896 the signature for <tt>printf</tt> in LLVM.<br/>
902 <!-- _______________________________________________________________________ -->
903 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
904 <div class="doc_text">
906 <p>The structure type is used to represent a collection of data members
907 together in memory. The packing of the field types is defined to match
908 the ABI of the underlying processor. The elements of a structure may
909 be any type that has a size.</p>
910 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
911 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
912 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
915 <pre> { <type list> }<br></pre>
917 <table class="layout">
920 <tt>{ int, int, int }</tt><br/>
921 <tt>{ float, int (int) * }</tt><br/>
924 a triple of three <tt>int</tt> values<br/>
925 A pair, where the first element is a <tt>float</tt> and the second element
926 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
927 that takes an <tt>int</tt>, returning an <tt>int</tt>.<br/>
933 <!-- _______________________________________________________________________ -->
934 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
935 <div class="doc_text">
937 <p>As in many languages, the pointer type represents a pointer or
938 reference to another object, which must live in memory.</p>
940 <pre> <type> *<br></pre>
942 <table class="layout">
945 <tt>[4x int]*</tt><br/>
946 <tt>int (int *) *</tt><br/>
949 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
950 four <tt>int</tt> values<br/>
951 A <a href="#t_pointer">pointer</a> to a <a
952 href="#t_function">function</a> that takes an <tt>int*</tt>, returning an
959 <!-- _______________________________________________________________________ -->
960 <div class="doc_subsubsection"> <a name="t_packed">Packed Type</a> </div>
961 <div class="doc_text">
965 <p>A packed type is a simple derived type that represents a vector
966 of elements. Packed types are used when multiple primitive data
967 are operated in parallel using a single instruction (SIMD).
968 A packed type requires a size (number of
969 elements) and an underlying primitive data type. Vectors must have a power
970 of two length (1, 2, 4, 8, 16 ...). Packed types are
971 considered <a href="#t_firstclass">first class</a>.</p>
976 < <# elements> x <elementtype> >
979 <p>The number of elements is a constant integer value; elementtype may
980 be any integral or floating point type.</p>
984 <table class="layout">
987 <tt><4 x int></tt><br/>
988 <tt><8 x float></tt><br/>
989 <tt><2 x uint></tt><br/>
992 Packed vector of 4 integer values.<br/>
993 Packed vector of 8 floating-point values.<br/>
994 Packed vector of 2 unsigned integer values.<br/>
1000 <!-- _______________________________________________________________________ -->
1001 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1002 <div class="doc_text">
1006 <p>Opaque types are used to represent unknown types in the system. This
1007 corresponds (for example) to the C notion of a foward declared structure type.
1008 In LLVM, opaque types can eventually be resolved to any type (not just a
1009 structure type).</p>
1019 <table class="layout">
1025 An opaque type.<br/>
1032 <!-- *********************************************************************** -->
1033 <div class="doc_section"> <a name="constants">Constants</a> </div>
1034 <!-- *********************************************************************** -->
1036 <div class="doc_text">
1038 <p>LLVM has several different basic types of constants. This section describes
1039 them all and their syntax.</p>
1043 <!-- ======================================================================= -->
1044 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1046 <div class="doc_text">
1049 <dt><b>Boolean constants</b></dt>
1051 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1052 constants of the <tt><a href="#t_primitive">bool</a></tt> type.
1055 <dt><b>Integer constants</b></dt>
1057 <dd>Standard integers (such as '4') are constants of the <a
1058 href="#t_integer">integer</a> type. Negative numbers may be used with signed
1062 <dt><b>Floating point constants</b></dt>
1064 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1065 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1066 notation (see below). Floating point constants must have a <a
1067 href="#t_floating">floating point</a> type. </dd>
1069 <dt><b>Null pointer constants</b></dt>
1071 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1072 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1076 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1077 of floating point constants. For example, the form '<tt>double
1078 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1079 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1080 (and the only time that they are generated by the disassembler) is when a
1081 floating point constant must be emitted but it cannot be represented as a
1082 decimal floating point number. For example, NaN's, infinities, and other
1083 special values are represented in their IEEE hexadecimal format so that
1084 assembly and disassembly do not cause any bits to change in the constants.</p>
1088 <!-- ======================================================================= -->
1089 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1092 <div class="doc_text">
1093 <p>Aggregate constants arise from aggregation of simple constants
1094 and smaller aggregate constants.</p>
1097 <dt><b>Structure constants</b></dt>
1099 <dd>Structure constants are represented with notation similar to structure
1100 type definitions (a comma separated list of elements, surrounded by braces
1101 (<tt>{}</tt>)). For example: "<tt>{ int 4, float 17.0, int* %G }</tt>",
1102 where "<tt>%G</tt>" is declared as "<tt>%G = external global int</tt>". Structure constants
1103 must have <a href="#t_struct">structure type</a>, and the number and
1104 types of elements must match those specified by the type.
1107 <dt><b>Array constants</b></dt>
1109 <dd>Array constants are represented with notation similar to array type
1110 definitions (a comma separated list of elements, surrounded by square brackets
1111 (<tt>[]</tt>)). For example: "<tt>[ int 42, int 11, int 74 ]</tt>". Array
1112 constants must have <a href="#t_array">array type</a>, and the number and
1113 types of elements must match those specified by the type.
1116 <dt><b>Packed constants</b></dt>
1118 <dd>Packed constants are represented with notation similar to packed type
1119 definitions (a comma separated list of elements, surrounded by
1120 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< int 42,
1121 int 11, int 74, int 100 ></tt>". Packed constants must have <a
1122 href="#t_packed">packed type</a>, and the number and types of elements must
1123 match those specified by the type.
1126 <dt><b>Zero initialization</b></dt>
1128 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1129 value to zero of <em>any</em> type, including scalar and aggregate types.
1130 This is often used to avoid having to print large zero initializers (e.g. for
1131 large arrays) and is always exactly equivalent to using explicit zero
1138 <!-- ======================================================================= -->
1139 <div class="doc_subsection">
1140 <a name="globalconstants">Global Variable and Function Addresses</a>
1143 <div class="doc_text">
1145 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1146 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1147 constants. These constants are explicitly referenced when the <a
1148 href="#identifiers">identifier for the global</a> is used and always have <a
1149 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1155 %Z = global [2 x int*] [ int* %X, int* %Y ]
1160 <!-- ======================================================================= -->
1161 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1162 <div class="doc_text">
1163 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1164 no specific value. Undefined values may be of any type and be used anywhere
1165 a constant is permitted.</p>
1167 <p>Undefined values indicate to the compiler that the program is well defined
1168 no matter what value is used, giving the compiler more freedom to optimize.
1172 <!-- ======================================================================= -->
1173 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1176 <div class="doc_text">
1178 <p>Constant expressions are used to allow expressions involving other constants
1179 to be used as constants. Constant expressions may be of any <a
1180 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1181 that does not have side effects (e.g. load and call are not supported). The
1182 following is the syntax for constant expressions:</p>
1185 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1186 <dd>Truncate a constant to another type. The bit size of CST must be larger
1187 than the bit size of TYPE. Both types must be integral.</dd>
1189 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1190 <dd>Zero extend a constant to another type. The bit size of CST must be
1191 smaller or equal to the bit size of TYPE. Both types must be integral.</dd>
1193 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1194 <dd>Sign extend a constant to another type. The bit size of CST must be
1195 smaller or equal to the bit size of TYPE. Both types must be integral.</dd>
1197 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1198 <dd>Truncate a floating point constant to another floating point type. The
1199 size of CST must be larger than the size of TYPE. Both types must be
1200 floating point.</dd>
1202 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1203 <dd>Floating point extend a constant to another type. The size of CST must be
1204 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1206 <dt><b><tt>fp2uint ( CST to TYPE )</tt></b></dt>
1207 <dd>Convert a floating point constant to the corresponding unsigned integer
1208 constant. TYPE must be an integer type. CST must be floating point. If the
1209 value won't fit in the integer type, the results are undefined.</dd>
1211 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1212 <dd>Convert a floating point constant to the corresponding signed integer
1213 constant. TYPE must be an integer type. CST must be floating point. If the
1214 value won't fit in the integer type, the results are undefined.</dd>
1216 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1217 <dd>Convert an unsigned integer constant to the corresponding floating point
1218 constant. TYPE must be floating point. CST must be of integer type. If the
1219 value won't fit in the floating point type, the results are undefined.</dd>
1221 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1222 <dd>Convert a signed integer constant to the corresponding floating point
1223 constant. TYPE must be floating point. CST must be of integer type. If the
1224 value won't fit in the floating point type, the results are undefined.</dd>
1226 <dt><b><tt>bitconvert ( CST to TYPE )</tt></b></dt>
1227 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1228 identical (same number of bits). The conversion is done as if the CST value
1229 was stored to memory and read back as TYPE. In other words, no bits change
1230 with this operator, just the type. This can be used for conversion of pointer
1231 and packed types to any other type, as long as they have the same bit width.
1234 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1236 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1237 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1238 instruction, the index list may have zero or more indexes, which are required
1239 to make sense for the type of "CSTPTR".</dd>
1241 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1243 <dd>Perform the <a href="#i_select">select operation</a> on
1246 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1248 <dd>Perform the <a href="#i_extractelement">extractelement
1249 operation</a> on constants.
1251 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1253 <dd>Perform the <a href="#i_insertelement">insertelement
1254 operation</a> on constants.
1257 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1259 <dd>Perform the <a href="#i_shufflevector">shufflevector
1260 operation</a> on constants.
1262 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1264 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1265 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1266 binary</a> operations. The constraints on operands are the same as those for
1267 the corresponding instruction (e.g. no bitwise operations on floating point
1268 values are allowed).</dd>
1272 <!-- *********************************************************************** -->
1273 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1274 <!-- *********************************************************************** -->
1276 <!-- ======================================================================= -->
1277 <div class="doc_subsection">
1278 <a name="inlineasm">Inline Assembler Expressions</a>
1281 <div class="doc_text">
1284 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1285 Module-Level Inline Assembly</a>) through the use of a special value. This
1286 value represents the inline assembler as a string (containing the instructions
1287 to emit), a list of operand constraints (stored as a string), and a flag that
1288 indicates whether or not the inline asm expression has side effects. An example
1289 inline assembler expression is:
1293 int(int) asm "bswap $0", "=r,r"
1297 Inline assembler expressions may <b>only</b> be used as the callee operand of
1298 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1302 %X = call int asm "<a href="#i_bswap">bswap</a> $0", "=r,r"(int %Y)
1306 Inline asms with side effects not visible in the constraint list must be marked
1307 as having side effects. This is done through the use of the
1308 '<tt>sideeffect</tt>' keyword, like so:
1312 call void asm sideeffect "eieio", ""()
1315 <p>TODO: The format of the asm and constraints string still need to be
1316 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1317 need to be documented).
1322 <!-- *********************************************************************** -->
1323 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1324 <!-- *********************************************************************** -->
1326 <div class="doc_text">
1328 <p>The LLVM instruction set consists of several different
1329 classifications of instructions: <a href="#terminators">terminator
1330 instructions</a>, <a href="#binaryops">binary instructions</a>,
1331 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1332 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1333 instructions</a>.</p>
1337 <!-- ======================================================================= -->
1338 <div class="doc_subsection"> <a name="terminators">Terminator
1339 Instructions</a> </div>
1341 <div class="doc_text">
1343 <p>As mentioned <a href="#functionstructure">previously</a>, every
1344 basic block in a program ends with a "Terminator" instruction, which
1345 indicates which block should be executed after the current block is
1346 finished. These terminator instructions typically yield a '<tt>void</tt>'
1347 value: they produce control flow, not values (the one exception being
1348 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1349 <p>There are six different terminator instructions: the '<a
1350 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1351 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1352 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1353 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1354 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1358 <!-- _______________________________________________________________________ -->
1359 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1360 Instruction</a> </div>
1361 <div class="doc_text">
1363 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1364 ret void <i>; Return from void function</i>
1367 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1368 value) from a function back to the caller.</p>
1369 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1370 returns a value and then causes control flow, and one that just causes
1371 control flow to occur.</p>
1373 <p>The '<tt>ret</tt>' instruction may return any '<a
1374 href="#t_firstclass">first class</a>' type. Notice that a function is
1375 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1376 instruction inside of the function that returns a value that does not
1377 match the return type of the function.</p>
1379 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1380 returns back to the calling function's context. If the caller is a "<a
1381 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1382 the instruction after the call. If the caller was an "<a
1383 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1384 at the beginning of the "normal" destination block. If the instruction
1385 returns a value, that value shall set the call or invoke instruction's
1388 <pre> ret int 5 <i>; Return an integer value of 5</i>
1389 ret void <i>; Return from a void function</i>
1392 <!-- _______________________________________________________________________ -->
1393 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1394 <div class="doc_text">
1396 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1399 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1400 transfer to a different basic block in the current function. There are
1401 two forms of this instruction, corresponding to a conditional branch
1402 and an unconditional branch.</p>
1404 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1405 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
1406 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
1407 value as a target.</p>
1409 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
1410 argument is evaluated. If the value is <tt>true</tt>, control flows
1411 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1412 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1414 <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
1415 href="#i_ret">ret</a> int 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> int 0<br></pre>
1417 <!-- _______________________________________________________________________ -->
1418 <div class="doc_subsubsection">
1419 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1422 <div class="doc_text">
1426 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1431 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1432 several different places. It is a generalization of the '<tt>br</tt>'
1433 instruction, allowing a branch to occur to one of many possible
1439 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1440 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1441 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1442 table is not allowed to contain duplicate constant entries.</p>
1446 <p>The <tt>switch</tt> instruction specifies a table of values and
1447 destinations. When the '<tt>switch</tt>' instruction is executed, this
1448 table is searched for the given value. If the value is found, control flow is
1449 transfered to the corresponding destination; otherwise, control flow is
1450 transfered to the default destination.</p>
1452 <h5>Implementation:</h5>
1454 <p>Depending on properties of the target machine and the particular
1455 <tt>switch</tt> instruction, this instruction may be code generated in different
1456 ways. For example, it could be generated as a series of chained conditional
1457 branches or with a lookup table.</p>
1462 <i>; Emulate a conditional br instruction</i>
1463 %Val = <a href="#i_zext">zext</a> bool %value to int
1464 switch int %Val, label %truedest [int 0, label %falsedest ]
1466 <i>; Emulate an unconditional br instruction</i>
1467 switch uint 0, label %dest [ ]
1469 <i>; Implement a jump table:</i>
1470 switch uint %val, label %otherwise [ uint 0, label %onzero
1471 uint 1, label %onone
1472 uint 2, label %ontwo ]
1476 <!-- _______________________________________________________________________ -->
1477 <div class="doc_subsubsection">
1478 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1481 <div class="doc_text">
1486 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1487 to label <normal label> unwind label <exception label>
1492 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1493 function, with the possibility of control flow transfer to either the
1494 '<tt>normal</tt>' label or the
1495 '<tt>exception</tt>' label. If the callee function returns with the
1496 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1497 "normal" label. If the callee (or any indirect callees) returns with the "<a
1498 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1499 continued at the dynamically nearest "exception" label.</p>
1503 <p>This instruction requires several arguments:</p>
1507 The optional "cconv" marker indicates which <a href="callingconv">calling
1508 convention</a> the call should use. If none is specified, the call defaults
1509 to using C calling conventions.
1511 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1512 function value being invoked. In most cases, this is a direct function
1513 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1514 an arbitrary pointer to function value.
1517 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1518 function to be invoked. </li>
1520 <li>'<tt>function args</tt>': argument list whose types match the function
1521 signature argument types. If the function signature indicates the function
1522 accepts a variable number of arguments, the extra arguments can be
1525 <li>'<tt>normal label</tt>': the label reached when the called function
1526 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1528 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1529 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1535 <p>This instruction is designed to operate as a standard '<tt><a
1536 href="#i_call">call</a></tt>' instruction in most regards. The primary
1537 difference is that it establishes an association with a label, which is used by
1538 the runtime library to unwind the stack.</p>
1540 <p>This instruction is used in languages with destructors to ensure that proper
1541 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1542 exception. Additionally, this is important for implementation of
1543 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1547 %retval = invoke int %Test(int 15) to label %Continue
1548 unwind label %TestCleanup <i>; {int}:retval set</i>
1549 %retval = invoke <a href="#callingconv">coldcc</a> int %Test(int 15) to label %Continue
1550 unwind label %TestCleanup <i>; {int}:retval set</i>
1555 <!-- _______________________________________________________________________ -->
1557 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1558 Instruction</a> </div>
1560 <div class="doc_text">
1569 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1570 at the first callee in the dynamic call stack which used an <a
1571 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1572 primarily used to implement exception handling.</p>
1576 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1577 immediately halt. The dynamic call stack is then searched for the first <a
1578 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1579 execution continues at the "exceptional" destination block specified by the
1580 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1581 dynamic call chain, undefined behavior results.</p>
1584 <!-- _______________________________________________________________________ -->
1586 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1587 Instruction</a> </div>
1589 <div class="doc_text">
1598 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1599 instruction is used to inform the optimizer that a particular portion of the
1600 code is not reachable. This can be used to indicate that the code after a
1601 no-return function cannot be reached, and other facts.</p>
1605 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1610 <!-- ======================================================================= -->
1611 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1612 <div class="doc_text">
1613 <p>Binary operators are used to do most of the computation in a
1614 program. They require two operands, execute an operation on them, and
1615 produce a single value. The operands might represent
1616 multiple data, as is the case with the <a href="#t_packed">packed</a> data type.
1617 The result value of a binary operator is not
1618 necessarily the same type as its operands.</p>
1619 <p>There are several different binary operators:</p>
1621 <!-- _______________________________________________________________________ -->
1622 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1623 Instruction</a> </div>
1624 <div class="doc_text">
1626 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1629 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1631 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1632 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1633 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1634 Both arguments must have identical types.</p>
1636 <p>The value produced is the integer or floating point sum of the two
1639 <pre> <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
1642 <!-- _______________________________________________________________________ -->
1643 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1644 Instruction</a> </div>
1645 <div class="doc_text">
1647 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1650 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1652 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1653 instruction present in most other intermediate representations.</p>
1655 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1656 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1658 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1659 Both arguments must have identical types.</p>
1661 <p>The value produced is the integer or floating point difference of
1662 the two operands.</p>
1664 <pre> <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
1665 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
1668 <!-- _______________________________________________________________________ -->
1669 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1670 Instruction</a> </div>
1671 <div class="doc_text">
1673 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1676 <p>The '<tt>mul</tt>' instruction returns the product of its two
1679 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1680 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1682 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1683 Both arguments must have identical types.</p>
1685 <p>The value produced is the integer or floating point product of the
1687 <p>There is no signed vs unsigned multiplication. The appropriate
1688 action is taken based on the type of the operand.</p>
1690 <pre> <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1693 <!-- _______________________________________________________________________ -->
1694 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
1696 <div class="doc_text">
1698 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1701 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
1704 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
1705 <a href="#t_integer">integer</a> values. Both arguments must have identical
1706 types. This instruction can also take <a href="#t_packed">packed</a> versions
1707 of the values in which case the elements must be integers.</p>
1709 <p>The value produced is the unsigned integer quotient of the two operands. This
1710 instruction always performs an unsigned division operation, regardless of
1711 whether the arguments are unsigned or not.</p>
1713 <pre> <result> = udiv uint 4, %var <i>; yields {uint}:result = 4 / %var</i>
1716 <!-- _______________________________________________________________________ -->
1717 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
1719 <div class="doc_text">
1721 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1724 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
1727 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
1728 <a href="#t_integer">integer</a> values. Both arguments must have identical
1729 types. This instruction can also take <a href="#t_packed">packed</a> versions
1730 of the values in which case the elements must be integers.</p>
1732 <p>The value produced is the signed integer quotient of the two operands. This
1733 instruction always performs a signed division operation, regardless of whether
1734 the arguments are signed or not.</p>
1736 <pre> <result> = sdiv int 4, %var <i>; yields {int}:result = 4 / %var</i>
1739 <!-- _______________________________________________________________________ -->
1740 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
1741 Instruction</a> </div>
1742 <div class="doc_text">
1744 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1747 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
1750 <p>The two arguments to the '<tt>div</tt>' instruction must be
1751 <a href="#t_floating">floating point</a> values. Both arguments must have
1752 identical types. This instruction can also take <a href="#t_packed">packed</a>
1753 versions of the values in which case the elements must be floating point.</p>
1755 <p>The value produced is the floating point quotient of the two operands.</p>
1757 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
1760 <!-- _______________________________________________________________________ -->
1761 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
1763 <div class="doc_text">
1765 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1768 <p>The '<tt>urem</tt>' instruction returns the remainder from the
1769 unsigned division of its two arguments.</p>
1771 <p>The two arguments to the '<tt>urem</tt>' instruction must be
1772 <a href="#t_integer">integer</a> values. Both arguments must have identical
1775 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
1776 This instruction always performs an unsigned division to get the remainder,
1777 regardless of whether the arguments are unsigned or not.</p>
1779 <pre> <result> = urem uint 4, %var <i>; yields {uint}:result = 4 % %var</i>
1783 <!-- _______________________________________________________________________ -->
1784 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
1785 Instruction</a> </div>
1786 <div class="doc_text">
1788 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1791 <p>The '<tt>srem</tt>' instruction returns the remainder from the
1792 signed division of its two operands.</p>
1794 <p>The two arguments to the '<tt>srem</tt>' instruction must be
1795 <a href="#t_integer">integer</a> values. Both arguments must have identical
1798 <p>This instruction returns the <i>remainder</i> of a division (where the result
1799 has the same sign as the divisor), not the <i>modulus</i> (where the
1800 result has the same sign as the dividend) of a value. For more
1801 information about the difference, see <a
1802 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
1805 <pre> <result> = srem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1809 <!-- _______________________________________________________________________ -->
1810 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
1811 Instruction</a> </div>
1812 <div class="doc_text">
1814 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1817 <p>The '<tt>frem</tt>' instruction returns the remainder from the
1818 division of its two operands.</p>
1820 <p>The two arguments to the '<tt>frem</tt>' instruction must be
1821 <a href="#t_floating">floating point</a> values. Both arguments must have
1822 identical types.</p>
1824 <p>This instruction returns the <i>remainder</i> of a division.</p>
1826 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
1830 <!-- _______________________________________________________________________ -->
1831 <div class="doc_subsubsection"> <a name="i_setcc">'<tt>set<i>cc</i></tt>'
1832 Instructions</a> </div>
1833 <div class="doc_text">
1835 <pre> <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1836 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1837 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1838 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1839 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1840 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1843 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean
1844 value based on a comparison of their two operands.</p>
1846 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must
1847 be of <a href="#t_firstclass">first class</a> type (it is not possible
1848 to compare '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>'
1849 or '<tt>void</tt>' values, etc...). Both arguments must have identical
1852 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1853 value if both operands are equal.<br>
1854 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1855 value if both operands are unequal.<br>
1856 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1857 value if the first operand is less than the second operand.<br>
1858 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1859 value if the first operand is greater than the second operand.<br>
1860 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1861 value if the first operand is less than or equal to the second operand.<br>
1862 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1863 value if the first operand is greater than or equal to the second
1866 <pre> <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1867 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1868 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1869 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1870 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1871 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1875 <!-- ======================================================================= -->
1876 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
1877 Operations</a> </div>
1878 <div class="doc_text">
1879 <p>Bitwise binary operators are used to do various forms of
1880 bit-twiddling in a program. They are generally very efficient
1881 instructions and can commonly be strength reduced from other
1882 instructions. They require two operands, execute an operation on them,
1883 and produce a single value. The resulting value of the bitwise binary
1884 operators is always the same type as its first operand.</p>
1886 <!-- _______________________________________________________________________ -->
1887 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
1888 Instruction</a> </div>
1889 <div class="doc_text">
1891 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1894 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1895 its two operands.</p>
1897 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1898 href="#t_integral">integral</a> values. Both arguments must have
1899 identical types.</p>
1901 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1903 <div style="align: center">
1904 <table border="1" cellspacing="0" cellpadding="4">
1935 <pre> <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1936 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1937 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1940 <!-- _______________________________________________________________________ -->
1941 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
1942 <div class="doc_text">
1944 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1947 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1948 or of its two operands.</p>
1950 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1951 href="#t_integral">integral</a> values. Both arguments must have
1952 identical types.</p>
1954 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1956 <div style="align: center">
1957 <table border="1" cellspacing="0" cellpadding="4">
1988 <pre> <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1989 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1990 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1993 <!-- _______________________________________________________________________ -->
1994 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
1995 Instruction</a> </div>
1996 <div class="doc_text">
1998 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2001 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2002 or of its two operands. The <tt>xor</tt> is used to implement the
2003 "one's complement" operation, which is the "~" operator in C.</p>
2005 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2006 href="#t_integral">integral</a> values. Both arguments must have
2007 identical types.</p>
2009 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2011 <div style="align: center">
2012 <table border="1" cellspacing="0" cellpadding="4">
2044 <pre> <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
2045 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
2046 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
2047 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
2050 <!-- _______________________________________________________________________ -->
2051 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2052 Instruction</a> </div>
2053 <div class="doc_text">
2055 <pre> <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
2058 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2059 the left a specified number of bits.</p>
2061 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
2062 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
2065 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
2067 <pre> <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
2068 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
2069 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
2072 <!-- _______________________________________________________________________ -->
2073 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2074 Instruction</a> </div>
2075 <div class="doc_text">
2077 <pre> <result> = lshr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
2081 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2082 operand shifted to the right a specified number of bits.</p>
2085 <p>The first argument to the '<tt>lshr</tt>' instruction must be an <a
2086 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>' type.</p>
2089 <p>This instruction always performs a logical shift right operation, regardless
2090 of whether the arguments are unsigned or not. The <tt>var2</tt> most significant
2091 bits will be filled with zero bits after the shift.</p>
2095 <result> = lshr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
2096 <result> = lshr int 4, ubyte 2 <i>; yields {uint}:result = 1</i>
2097 <result> = lshr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
2098 <result> = lshr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = 0x7FFFFFFF </i>
2102 <!-- ======================================================================= -->
2103 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2104 Instruction</a> </div>
2105 <div class="doc_text">
2108 <pre> <result> = ashr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
2112 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2113 operand shifted to the right a specified number of bits.</p>
2116 <p>The first argument to the '<tt>ashr</tt>' instruction must be an
2117 <a href="#t_integer">integer</a> type. The second argument must be an
2118 '<tt>ubyte</tt>' type.</p>
2121 <p>This instruction always performs an arithmetic shift right operation,
2122 regardless of whether the arguments are signed or not. The <tt>var2</tt> most
2123 significant bits will be filled with the sign bit of <tt>var1</tt>.</p>
2127 <result> = ashr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
2128 <result> = ashr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
2129 <result> = ashr ubyte 4, ubyte 3 <i>; yields {ubyte}:result = 0</i>
2130 <result> = ashr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
2134 <!-- ======================================================================= -->
2135 <div class="doc_subsection">
2136 <a name="vectorops">Vector Operations</a>
2139 <div class="doc_text">
2141 <p>LLVM supports several instructions to represent vector operations in a
2142 target-independent manner. This instructions cover the element-access and
2143 vector-specific operations needed to process vectors effectively. While LLVM
2144 does directly support these vector operations, many sophisticated algorithms
2145 will want to use target-specific intrinsics to take full advantage of a specific
2150 <!-- _______________________________________________________________________ -->
2151 <div class="doc_subsubsection">
2152 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2155 <div class="doc_text">
2160 <result> = extractelement <n x <ty>> <val>, uint <idx> <i>; yields <ty></i>
2166 The '<tt>extractelement</tt>' instruction extracts a single scalar
2167 element from a packed vector at a specified index.
2174 The first operand of an '<tt>extractelement</tt>' instruction is a
2175 value of <a href="#t_packed">packed</a> type. The second operand is
2176 an index indicating the position from which to extract the element.
2177 The index may be a variable.</p>
2182 The result is a scalar of the same type as the element type of
2183 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2184 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2185 results are undefined.
2191 %result = extractelement <4 x int> %vec, uint 0 <i>; yields int</i>
2196 <!-- _______________________________________________________________________ -->
2197 <div class="doc_subsubsection">
2198 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2201 <div class="doc_text">
2206 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, uint <idx> <i>; yields <n x <ty>></i>
2212 The '<tt>insertelement</tt>' instruction inserts a scalar
2213 element into a packed vector at a specified index.
2220 The first operand of an '<tt>insertelement</tt>' instruction is a
2221 value of <a href="#t_packed">packed</a> type. The second operand is a
2222 scalar value whose type must equal the element type of the first
2223 operand. The third operand is an index indicating the position at
2224 which to insert the value. The index may be a variable.</p>
2229 The result is a packed vector of the same type as <tt>val</tt>. Its
2230 element values are those of <tt>val</tt> except at position
2231 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2232 exceeds the length of <tt>val</tt>, the results are undefined.
2238 %result = insertelement <4 x int> %vec, int 1, uint 0 <i>; yields <4 x int></i>
2242 <!-- _______________________________________________________________________ -->
2243 <div class="doc_subsubsection">
2244 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2247 <div class="doc_text">
2252 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x uint> <mask> <i>; yields <n x <ty>></i>
2258 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2259 from two input vectors, returning a vector of the same type.
2265 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2266 with types that match each other and types that match the result of the
2267 instruction. The third argument is a shuffle mask, which has the same number
2268 of elements as the other vector type, but whose element type is always 'uint'.
2272 The shuffle mask operand is required to be a constant vector with either
2273 constant integer or undef values.
2279 The elements of the two input vectors are numbered from left to right across
2280 both of the vectors. The shuffle mask operand specifies, for each element of
2281 the result vector, which element of the two input registers the result element
2282 gets. The element selector may be undef (meaning "don't care") and the second
2283 operand may be undef if performing a shuffle from only one vector.
2289 %result = shufflevector <4 x int> %v1, <4 x int> %v2,
2290 <4 x uint> <uint 0, uint 4, uint 1, uint 5> <i>; yields <4 x int></i>
2291 %result = shufflevector <4 x int> %v1, <4 x int> undef,
2292 <4 x uint> <uint 0, uint 1, uint 2, uint 3> <i>; yields <4 x int></i> - Identity shuffle.
2297 <!-- ======================================================================= -->
2298 <div class="doc_subsection">
2299 <a name="memoryops">Memory Access and Addressing Operations</a>
2302 <div class="doc_text">
2304 <p>A key design point of an SSA-based representation is how it
2305 represents memory. In LLVM, no memory locations are in SSA form, which
2306 makes things very simple. This section describes how to read, write,
2307 allocate, and free memory in LLVM.</p>
2311 <!-- _______________________________________________________________________ -->
2312 <div class="doc_subsubsection">
2313 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2316 <div class="doc_text">
2321 <result> = malloc <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2326 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2327 heap and returns a pointer to it.</p>
2331 <p>The '<tt>malloc</tt>' instruction allocates
2332 <tt>sizeof(<type>)*NumElements</tt>
2333 bytes of memory from the operating system and returns a pointer of the
2334 appropriate type to the program. If "NumElements" is specified, it is the
2335 number of elements allocated. If an alignment is specified, the value result
2336 of the allocation is guaranteed to be aligned to at least that boundary. If
2337 not specified, or if zero, the target can choose to align the allocation on any
2338 convenient boundary.</p>
2340 <p>'<tt>type</tt>' must be a sized type.</p>
2344 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2345 a pointer is returned.</p>
2350 %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
2352 %size = <a href="#i_add">add</a> uint 2, 2 <i>; yields {uint}:size = uint 4</i>
2353 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
2354 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
2355 %array3 = malloc int, uint 4, align 1024 <i>; yields {int*}:array3</i>
2356 %array4 = malloc int, align 1024 <i>; yields {int*}:array4</i>
2360 <!-- _______________________________________________________________________ -->
2361 <div class="doc_subsubsection">
2362 <a name="i_free">'<tt>free</tt>' Instruction</a>
2365 <div class="doc_text">
2370 free <type> <value> <i>; yields {void}</i>
2375 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2376 memory heap to be reallocated in the future.</p>
2380 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2381 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2386 <p>Access to the memory pointed to by the pointer is no longer defined
2387 after this instruction executes.</p>
2392 %array = <a href="#i_malloc">malloc</a> [4 x ubyte] <i>; yields {[4 x ubyte]*}:array</i>
2393 free [4 x ubyte]* %array
2397 <!-- _______________________________________________________________________ -->
2398 <div class="doc_subsubsection">
2399 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2402 <div class="doc_text">
2407 <result> = alloca <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2412 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
2413 stack frame of the procedure that is live until the current function
2414 returns to its caller.</p>
2418 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2419 bytes of memory on the runtime stack, returning a pointer of the
2420 appropriate type to the program. If "NumElements" is specified, it is the
2421 number of elements allocated. If an alignment is specified, the value result
2422 of the allocation is guaranteed to be aligned to at least that boundary. If
2423 not specified, or if zero, the target can choose to align the allocation on any
2424 convenient boundary.</p>
2426 <p>'<tt>type</tt>' may be any sized type.</p>
2430 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2431 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2432 instruction is commonly used to represent automatic variables that must
2433 have an address available. When the function returns (either with the <tt><a
2434 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2435 instructions), the memory is reclaimed.</p>
2440 %ptr = alloca int <i>; yields {int*}:ptr</i>
2441 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
2442 %ptr = alloca int, uint 4, align 1024 <i>; yields {int*}:ptr</i>
2443 %ptr = alloca int, align 1024 <i>; yields {int*}:ptr</i>
2447 <!-- _______________________________________________________________________ -->
2448 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2449 Instruction</a> </div>
2450 <div class="doc_text">
2452 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
2454 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2456 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2457 address from which to load. The pointer must point to a <a
2458 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2459 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2460 the number or order of execution of this <tt>load</tt> with other
2461 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2464 <p>The location of memory pointed to is loaded.</p>
2466 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2468 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2469 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2472 <!-- _______________________________________________________________________ -->
2473 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2474 Instruction</a> </div>
2475 <div class="doc_text">
2477 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2478 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2481 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2483 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2484 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
2485 operand must be a pointer to the type of the '<tt><value></tt>'
2486 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2487 optimizer is not allowed to modify the number or order of execution of
2488 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2489 href="#i_store">store</a></tt> instructions.</p>
2491 <p>The contents of memory are updated to contain '<tt><value></tt>'
2492 at the location specified by the '<tt><pointer></tt>' operand.</p>
2494 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2496 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2497 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2501 <!-- _______________________________________________________________________ -->
2502 <div class="doc_subsubsection">
2503 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2506 <div class="doc_text">
2509 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2515 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2516 subelement of an aggregate data structure.</p>
2520 <p>This instruction takes a list of integer constants that indicate what
2521 elements of the aggregate object to index to. The actual types of the arguments
2522 provided depend on the type of the first pointer argument. The
2523 '<tt>getelementptr</tt>' instruction is used to index down through the type
2524 levels of a structure or to a specific index in an array. When indexing into a
2525 structure, only <tt>uint</tt>
2526 integer constants are allowed. When indexing into an array or pointer,
2527 <tt>int</tt> and <tt>long</tt> indexes are allowed of any sign.</p>
2529 <p>For example, let's consider a C code fragment and how it gets
2530 compiled to LLVM:</p>
2544 int *foo(struct ST *s) {
2545 return &s[1].Z.B[5][13];
2549 <p>The LLVM code generated by the GCC frontend is:</p>
2552 %RT = type { sbyte, [10 x [20 x int]], sbyte }
2553 %ST = type { int, double, %RT }
2557 int* %foo(%ST* %s) {
2559 %reg = getelementptr %ST* %s, int 1, uint 2, uint 1, int 5, int 13
2566 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2567 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2568 and <a href="#t_array">array</a> types require <tt>uint</tt>, <tt>int</tt>,
2569 <tt>ulong</tt>, or <tt>long</tt> values, and <a href="#t_struct">structure</a>
2570 types require <tt>uint</tt> <b>constants</b>.</p>
2572 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2573 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
2574 }</tt>' type, a structure. The second index indexes into the third element of
2575 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
2576 sbyte }</tt>' type, another structure. The third index indexes into the second
2577 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
2578 array. The two dimensions of the array are subscripted into, yielding an
2579 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2580 to this element, thus computing a value of '<tt>int*</tt>' type.</p>
2582 <p>Note that it is perfectly legal to index partially through a
2583 structure, returning a pointer to an inner element. Because of this,
2584 the LLVM code for the given testcase is equivalent to:</p>
2587 int* %foo(%ST* %s) {
2588 %t1 = getelementptr %ST* %s, int 1 <i>; yields %ST*:%t1</i>
2589 %t2 = getelementptr %ST* %t1, int 0, uint 2 <i>; yields %RT*:%t2</i>
2590 %t3 = getelementptr %RT* %t2, int 0, uint 1 <i>; yields [10 x [20 x int]]*:%t3</i>
2591 %t4 = getelementptr [10 x [20 x int]]* %t3, int 0, int 5 <i>; yields [20 x int]*:%t4</i>
2592 %t5 = getelementptr [20 x int]* %t4, int 0, int 13 <i>; yields int*:%t5</i>
2597 <p>Note that it is undefined to access an array out of bounds: array and
2598 pointer indexes must always be within the defined bounds of the array type.
2599 The one exception for this rules is zero length arrays. These arrays are
2600 defined to be accessible as variable length arrays, which requires access
2601 beyond the zero'th element.</p>
2603 <p>The getelementptr instruction is often confusing. For some more insight
2604 into how it works, see <a href="GetElementPtr.html">the getelementptr
2610 <i>; yields [12 x ubyte]*:aptr</i>
2611 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, uint 1
2615 <!-- ======================================================================= -->
2616 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
2618 <div class="doc_text">
2619 <p>The instructions in this category are the conversion instructions (casting)
2620 which all take a single operand and a type. They perform various bit conversions
2624 <!-- _______________________________________________________________________ -->
2625 <div class="doc_subsubsection">
2626 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
2628 <div class="doc_text">
2632 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
2637 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
2642 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
2643 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
2644 and type of the result, which must be an <a href="#t_integral">integral</a>
2645 type. The bit size of <tt>value</tt> must be larger than the bit size of
2646 <tt>ty2</tt>. Equal sized types are not allowed.</p>
2650 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
2651 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
2652 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
2653 It will always truncate bits.</p>
2657 %X = trunc int 257 to ubyte <i>; yields ubyte:1</i>
2658 %Y = trunc int 123 to bool <i>; yields bool:true</i>
2662 <!-- _______________________________________________________________________ -->
2663 <div class="doc_subsubsection">
2664 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
2666 <div class="doc_text">
2670 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
2674 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
2679 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
2680 <a href="#t_integral">integral</a> type, and a type to cast it to, which must
2681 also be of <a href="#t_integral">integral</a> type. The bit size of the
2682 <tt>value</tt> must be smaller than the bit size of the destination type,
2686 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
2687 bits until it reaches the size of the destination type, <tt>ty2</tt>. When the
2688 the operand and the type are the same size, no bit filling is done and the
2689 cast is considered a <i>no-op cast</i> because no bits change (only the type
2692 <p>When zero extending from bool, the result will alwasy be either 0 or 1.</p>
2696 %X = zext int 257 to ulong <i>; yields ulong:257</i>
2697 %Y = zext bool true to int <i>; yields int:1</i>
2701 <!-- _______________________________________________________________________ -->
2702 <div class="doc_subsubsection">
2703 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
2705 <div class="doc_text">
2709 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
2713 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
2717 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
2718 <a href="#t_integral">integral</a> type, and a type to cast it to, which must
2719 also be of <a href="#t_integral">integral</a> type. The bit size of the
2720 <tt>value</tt> must be smaller than the bit size of the destination type,
2725 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
2726 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
2727 the type <tt>ty2</tt>. When the the operand and the type are the same size,
2728 no bit filling is done and the cast is considered a <i>no-op cast</i> because
2729 no bits change (only the type changes).</p>
2731 <p>When sign extending from bool, the extension always results in -1 or 0.</p>
2735 %X = sext sbyte -1 to ushort <i>; yields ushort:65535</i>
2736 %Y = sext bool true to int <i>; yields int:-1</i>
2740 <!-- _______________________________________________________________________ -->
2741 <div class="doc_subsubsection">
2742 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
2745 <div class="doc_text">
2750 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
2754 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
2759 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
2760 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
2761 cast it to. The size of <tt>value</tt> must be larger than the size of
2762 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
2763 <i>no-op cast</i>.</p>
2766 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
2767 <a href="#t_floating">floating point</a> type to a smaller
2768 <a href="#t_floating">floating point</a> type. If the value cannot fit within
2769 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
2773 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
2774 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
2778 <!-- _______________________________________________________________________ -->
2779 <div class="doc_subsubsection">
2780 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
2782 <div class="doc_text">
2786 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
2790 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
2791 floating point value.</p>
2794 <p>The '<tt>fpext</tt>' instruction takes a
2795 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
2796 and a <a href="#t_floating">floating point</a> type to cast it to. The source
2797 type must be smaller than the destination type.</p>
2800 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
2801 <a href="t_floating">floating point</a> type to a larger
2802 <a href="t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
2803 used to make a <i>no-op cast</i> because it always changes bits. Use
2804 <tt>bitconvert</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
2808 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
2809 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
2813 <!-- _______________________________________________________________________ -->
2814 <div class="doc_subsubsection">
2815 <a name="i_fp2uint">'<tt>fptoui .. to</tt>' Instruction</a>
2817 <div class="doc_text">
2821 <result> = fp2uint <ty> <value> to <ty2> <i>; yields ty2</i>
2825 <p>The '<tt>fp2uint</tt>' converts a floating point <tt>value</tt> to its
2826 unsigned integer equivalent of type <tt>ty2</tt>.
2830 <p>The '<tt>fp2uint</tt>' instruction takes a value to cast, which must be a
2831 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
2832 must be an <a href="#t_integral">integral</a> type.</p>
2835 <p> The '<tt>fp2uint</tt>' instruction converts its
2836 <a href="#t_floating">floating point</a> operand into the nearest (rounding
2837 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
2838 the results are undefined.</p>
2840 <p>When converting to bool, the conversion is done as a comparison against
2841 zero. If the <tt>value</tt> was zero, the bool result will be <tt>false</tt>.
2842 If the <tt>value</tt> was non-zero, the bool result will be <tt>true</tt>.</p>
2846 %X = fp2uint double 123.0 to int <i>; yields int:123</i>
2847 %Y = fp2uint float 1.0E+300 to bool <i>; yields bool:true</i>
2848 %X = fp2uint float 1.04E+17 to ubyte <i>; yields undefined:1</i>
2852 <!-- _______________________________________________________________________ -->
2853 <div class="doc_subsubsection">
2854 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
2856 <div class="doc_text">
2860 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
2864 <p>The '<tt>fptosi</tt>' instruction converts
2865 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
2870 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
2871 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
2872 must also be an <a href="#t_integral">integral</a> type.</p>
2875 <p>The '<tt>fptosi</tt>' instruction converts its
2876 <a href="#t_floating">floating point</a> operand into the nearest (rounding
2877 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
2878 the results are undefined.</p>
2880 <p>When converting to bool, the conversion is done as a comparison against
2881 zero. If the <tt>value</tt> was zero, the bool result will be <tt>false</tt>.
2882 If the <tt>value</tt> was non-zero, the bool result will be <tt>true</tt>.</p>
2886 %X = fptosi double -123.0 to int <i>; yields int:-123</i>
2887 %Y = fptosi float 1.0E-247 to bool <i>; yields bool:true</i>
2888 %X = fptosi float 1.04E+17 to sbyte <i>; yields undefined:1</i>
2892 <!-- _______________________________________________________________________ -->
2893 <div class="doc_subsubsection">
2894 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
2896 <div class="doc_text">
2900 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
2904 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
2905 integer and converts that value to the <tt>ty2</tt> type.</p>
2909 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be an
2910 <a href="#t_integral">integral</a> value, and a type to cast it to, which must
2911 be a <a href="#t_floating">floating point</a> type.</p>
2914 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
2915 integer quantity and converts it to the corresponding floating point value. If
2916 the value cannot fit in the floating point value, the results are undefined.</p>
2921 %X = uitofp int 257 to float <i>; yields float:257.0</i>
2922 %Y = uitofp sbyte -1 to double <i>; yields double:255.0</i>
2926 <!-- _______________________________________________________________________ -->
2927 <div class="doc_subsubsection">
2928 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
2930 <div class="doc_text">
2934 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
2938 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
2939 integer and converts that value to the <tt>ty2</tt> type.</p>
2942 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be an
2943 <a href="#t_integral">integral</a> value, and a type to cast it to, which must be
2944 a <a href="#t_floating">floating point</a> type.</p>
2947 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
2948 integer quantity and converts it to the corresponding floating point value. If
2949 the value cannot fit in the floating point value, the results are undefined.</p>
2953 %X = sitofp int 257 to float <i>; yields float:257.0</i>
2954 %Y = sitofp sbyte -1 to double <i>; yields double:-1.0</i>
2958 <!-- _______________________________________________________________________ -->
2959 <div class="doc_subsubsection">
2960 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
2962 <div class="doc_text">
2966 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
2970 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
2971 the integer type <tt>ty2</tt>.</p>
2974 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
2975 must be a <a href="t_pointer">pointer</a> value, and a type to cast it to
2976 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
2979 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
2980 <tt>ty2</tt> by interpreting the pointer value as an integer and either
2981 truncating or zero extending that value to the size of the integer type. If
2982 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
2983 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
2984 are the same size, then nothing is done (<i>no-op cast</i>).</p>
2988 %X = ptrtoint int* %X to sbyte <i>; yields truncation on 32-bit</i>
2989 %Y = ptrtoint int* %x to ulong <i>; yields zero extend on 32-bit</i>
2993 <!-- _______________________________________________________________________ -->
2994 <div class="doc_subsubsection">
2995 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
2997 <div class="doc_text">
3001 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3005 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3006 a pointer type, <tt>ty2</tt>.</p>
3009 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="i_integer">integer</a>
3010 value to cast, and a type to cast it to, which must be a
3011 <a href="#t_pointer">pointer</a> type. </tt>
3014 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3015 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3016 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3017 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3018 the size of a pointer then a zero extension is done. If they are the same size,
3019 nothing is done (<i>no-op cast</i>).</p>
3023 %X = inttoptr int 255 to int* <i>; yields zero extend on 64-bit</i>
3024 %X = inttoptr int 255 to int* <i>; yields no-op on 32-bit </i>
3025 %Y = inttoptr short 0 to int* <i>; yields zero extend on 32-bit</i>
3029 <!-- _______________________________________________________________________ -->
3030 <div class="doc_subsubsection">
3031 <a name="i_bitconvert">'<tt>bitconvert .. to</tt>' Instruction</a>
3033 <div class="doc_text">
3037 <result> = bitconvert <ty> <value> to <ty2> <i>; yields ty2</i>
3041 <p>The '<tt>bitconvert</tt>' instruction converts <tt>value</tt> to type
3042 <tt>ty2</tt> without changing any bits.</p>
3045 <p>The '<tt>bitconvert</tt>' instruction takes a value to cast, which must be
3046 a first class value, and a type to cast it to, which must also be a <a
3047 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3048 and the destination type, <tt>ty2</tt>, must be identical.</p>
3051 <p>The '<tt>bitconvert</tt>' instruction converts <tt>value</tt> to type
3052 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3053 this conversion. The conversion is done as if the <tt>value</tt> had been
3054 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3055 converted to other pointer types with this instruction. To convert pointers to
3056 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3057 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3061 %X = bitconvert ubyte 255 to sbyte <i>; yields sbyte:-1</i>
3062 %Y = bitconvert uint* %x to uint <i>; yields uint:%x</i>
3066 <!-- ======================================================================= -->
3067 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3068 <div class="doc_text">
3069 <p>The instructions in this category are the "miscellaneous"
3070 instructions, which defy better classification.</p>
3072 <!-- _______________________________________________________________________ -->
3073 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3074 Instruction</a> </div>
3075 <div class="doc_text">
3077 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3079 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3080 the SSA graph representing the function.</p>
3082 <p>The type of the incoming values are specified with the first type
3083 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3084 as arguments, with one pair for each predecessor basic block of the
3085 current block. Only values of <a href="#t_firstclass">first class</a>
3086 type may be used as the value arguments to the PHI node. Only labels
3087 may be used as the label arguments.</p>
3088 <p>There must be no non-phi instructions between the start of a basic
3089 block and the PHI instructions: i.e. PHI instructions must be first in
3092 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
3093 value specified by the parameter, depending on which basic block we
3094 came from in the last <a href="#terminators">terminator</a> instruction.</p>
3096 <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>
3099 <!-- _______________________________________________________________________ -->
3100 <div class="doc_subsubsection">
3101 <a name="i_select">'<tt>select</tt>' Instruction</a>
3104 <div class="doc_text">
3109 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3115 The '<tt>select</tt>' instruction is used to choose one value based on a
3116 condition, without branching.
3123 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.
3129 If the boolean condition evaluates to true, the instruction returns the first
3130 value argument; otherwise, it returns the second value argument.
3136 %X = select bool true, ubyte 17, ubyte 42 <i>; yields ubyte:17</i>
3141 <!-- _______________________________________________________________________ -->
3142 <div class="doc_subsubsection">
3143 <a name="i_call">'<tt>call</tt>' Instruction</a>
3146 <div class="doc_text">
3150 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
3155 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3159 <p>This instruction requires several arguments:</p>
3163 <p>The optional "tail" marker indicates whether the callee function accesses
3164 any allocas or varargs in the caller. If the "tail" marker is present, the
3165 function call is eligible for tail call optimization. Note that calls may
3166 be marked "tail" even if they do not occur before a <a
3167 href="#i_ret"><tt>ret</tt></a> instruction.
3170 <p>The optional "cconv" marker indicates which <a href="callingconv">calling
3171 convention</a> the call should use. If none is specified, the call defaults
3172 to using C calling conventions.
3175 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
3176 being invoked. The argument types must match the types implied by this
3177 signature. This type can be omitted if the function is not varargs and
3178 if the function type does not return a pointer to a function.</p>
3181 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3182 be invoked. In most cases, this is a direct function invocation, but
3183 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3184 to function value.</p>
3187 <p>'<tt>function args</tt>': argument list whose types match the
3188 function signature argument types. All arguments must be of
3189 <a href="#t_firstclass">first class</a> type. If the function signature
3190 indicates the function accepts a variable number of arguments, the extra
3191 arguments can be specified.</p>
3197 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3198 transfer to a specified function, with its incoming arguments bound to
3199 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3200 instruction in the called function, control flow continues with the
3201 instruction after the function call, and the return value of the
3202 function is bound to the result argument. This is a simpler case of
3203 the <a href="#i_invoke">invoke</a> instruction.</p>
3208 %retval = call int %test(int %argc)
3209 call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);
3210 %X = tail call int %foo()
3211 %Y = tail call <a href="#callingconv">fastcc</a> int %foo()
3216 <!-- _______________________________________________________________________ -->
3217 <div class="doc_subsubsection">
3218 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3221 <div class="doc_text">
3226 <resultval> = va_arg <va_list*> <arglist>, <argty>
3231 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3232 the "variable argument" area of a function call. It is used to implement the
3233 <tt>va_arg</tt> macro in C.</p>
3237 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3238 the argument. It returns a value of the specified argument type and
3239 increments the <tt>va_list</tt> to point to the next argument. Again, the
3240 actual type of <tt>va_list</tt> is target specific.</p>
3244 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3245 type from the specified <tt>va_list</tt> and causes the
3246 <tt>va_list</tt> to point to the next argument. For more information,
3247 see the variable argument handling <a href="#int_varargs">Intrinsic
3250 <p>It is legal for this instruction to be called in a function which does not
3251 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3254 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3255 href="#intrinsics">intrinsic function</a> because it takes a type as an
3260 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3264 <!-- *********************************************************************** -->
3265 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3266 <!-- *********************************************************************** -->
3268 <div class="doc_text">
3270 <p>LLVM supports the notion of an "intrinsic function". These functions have
3271 well known names and semantics and are required to follow certain
3272 restrictions. Overall, these instructions represent an extension mechanism for
3273 the LLVM language that does not require changing all of the transformations in
3274 LLVM to add to the language (or the bytecode reader/writer, the parser,
3277 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3278 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
3279 this. Intrinsic functions must always be external functions: you cannot define
3280 the body of intrinsic functions. Intrinsic functions may only be used in call
3281 or invoke instructions: it is illegal to take the address of an intrinsic
3282 function. Additionally, because intrinsic functions are part of the LLVM
3283 language, it is required that they all be documented here if any are added.</p>
3286 <p>To learn how to add an intrinsic function, please see the <a
3287 href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3292 <!-- ======================================================================= -->
3293 <div class="doc_subsection">
3294 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3297 <div class="doc_text">
3299 <p>Variable argument support is defined in LLVM with the <a
3300 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3301 intrinsic functions. These functions are related to the similarly
3302 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3304 <p>All of these functions operate on arguments that use a
3305 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3306 language reference manual does not define what this type is, so all
3307 transformations should be prepared to handle intrinsics with any type
3310 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3311 instruction and the variable argument handling intrinsic functions are
3315 int %test(int %X, ...) {
3316 ; Initialize variable argument processing
3318 call void %<a href="#i_va_start">llvm.va_start</a>(sbyte** %ap)
3320 ; Read a single integer argument
3321 %tmp = va_arg sbyte** %ap, int
3323 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3325 call void %<a href="#i_va_copy">llvm.va_copy</a>(sbyte** %aq, sbyte** %ap)
3326 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %aq)
3328 ; Stop processing of arguments.
3329 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %ap)
3335 <!-- _______________________________________________________________________ -->
3336 <div class="doc_subsubsection">
3337 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3341 <div class="doc_text">
3343 <pre> declare void %llvm.va_start(<va_list>* <arglist>)<br></pre>
3345 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3346 <tt>*<arglist></tt> for subsequent use by <tt><a
3347 href="#i_va_arg">va_arg</a></tt>.</p>
3351 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3355 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3356 macro available in C. In a target-dependent way, it initializes the
3357 <tt>va_list</tt> element the argument points to, so that the next call to
3358 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3359 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3360 last argument of the function, the compiler can figure that out.</p>
3364 <!-- _______________________________________________________________________ -->
3365 <div class="doc_subsubsection">
3366 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
3369 <div class="doc_text">
3371 <pre> declare void %llvm.va_end(<va_list*> <arglist>)<br></pre>
3373 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
3374 which has been initialized previously with <tt><a href="#i_va_start">llvm.va_start</a></tt>
3375 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
3377 <p>The argument is a <tt>va_list</tt> to destroy.</p>
3379 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
3380 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
3381 Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and <a
3382 href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
3383 with calls to <tt>llvm.va_end</tt>.</p>
3386 <!-- _______________________________________________________________________ -->
3387 <div class="doc_subsubsection">
3388 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
3391 <div class="doc_text">
3396 declare void %llvm.va_copy(<va_list>* <destarglist>,
3397 <va_list>* <srcarglist>)
3402 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
3403 the source argument list to the destination argument list.</p>
3407 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
3408 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
3413 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
3414 available in C. In a target-dependent way, it copies the source
3415 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
3416 because the <tt><a href="i_va_begin">llvm.va_begin</a></tt> intrinsic may be
3417 arbitrarily complex and require memory allocation, for example.</p>
3421 <!-- ======================================================================= -->
3422 <div class="doc_subsection">
3423 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
3426 <div class="doc_text">
3429 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
3430 Collection</a> requires the implementation and generation of these intrinsics.
3431 These intrinsics allow identification of <a href="#i_gcroot">GC roots on the
3432 stack</a>, as well as garbage collector implementations that require <a
3433 href="#i_gcread">read</a> and <a href="#i_gcwrite">write</a> barriers.
3434 Front-ends for type-safe garbage collected languages should generate these
3435 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3436 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
3440 <!-- _______________________________________________________________________ -->
3441 <div class="doc_subsubsection">
3442 <a name="i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
3445 <div class="doc_text">
3450 declare void %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
3455 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
3456 the code generator, and allows some metadata to be associated with it.</p>
3460 <p>The first argument specifies the address of a stack object that contains the
3461 root pointer. The second pointer (which must be either a constant or a global
3462 value address) contains the meta-data to be associated with the root.</p>
3466 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
3467 location. At compile-time, the code generator generates information to allow
3468 the runtime to find the pointer at GC safe points.
3474 <!-- _______________________________________________________________________ -->
3475 <div class="doc_subsubsection">
3476 <a name="i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
3479 <div class="doc_text">
3484 declare sbyte* %llvm.gcread(sbyte* %ObjPtr, sbyte** %Ptr)
3489 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
3490 locations, allowing garbage collector implementations that require read
3495 <p>The second argument is the address to read from, which should be an address
3496 allocated from the garbage collector. The first object is a pointer to the
3497 start of the referenced object, if needed by the language runtime (otherwise
3502 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
3503 instruction, but may be replaced with substantially more complex code by the
3504 garbage collector runtime, as needed.</p>
3509 <!-- _______________________________________________________________________ -->
3510 <div class="doc_subsubsection">
3511 <a name="i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
3514 <div class="doc_text">
3519 declare void %llvm.gcwrite(sbyte* %P1, sbyte* %Obj, sbyte** %P2)
3524 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
3525 locations, allowing garbage collector implementations that require write
3526 barriers (such as generational or reference counting collectors).</p>
3530 <p>The first argument is the reference to store, the second is the start of the
3531 object to store it to, and the third is the address of the field of Obj to
3532 store to. If the runtime does not require a pointer to the object, Obj may be
3537 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
3538 instruction, but may be replaced with substantially more complex code by the
3539 garbage collector runtime, as needed.</p>
3545 <!-- ======================================================================= -->
3546 <div class="doc_subsection">
3547 <a name="int_codegen">Code Generator Intrinsics</a>
3550 <div class="doc_text">
3552 These intrinsics are provided by LLVM to expose special features that may only
3553 be implemented with code generator support.
3558 <!-- _______________________________________________________________________ -->
3559 <div class="doc_subsubsection">
3560 <a name="i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
3563 <div class="doc_text">
3567 declare sbyte *%llvm.returnaddress(uint <level>)
3573 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
3574 target-specific value indicating the return address of the current function
3575 or one of its callers.
3581 The argument to this intrinsic indicates which function to return the address
3582 for. Zero indicates the calling function, one indicates its caller, etc. The
3583 argument is <b>required</b> to be a constant integer value.
3589 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
3590 the return address of the specified call frame, or zero if it cannot be
3591 identified. The value returned by this intrinsic is likely to be incorrect or 0
3592 for arguments other than zero, so it should only be used for debugging purposes.
3596 Note that calling this intrinsic does not prevent function inlining or other
3597 aggressive transformations, so the value returned may not be that of the obvious
3598 source-language caller.
3603 <!-- _______________________________________________________________________ -->
3604 <div class="doc_subsubsection">
3605 <a name="i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
3608 <div class="doc_text">
3612 declare sbyte *%llvm.frameaddress(uint <level>)
3618 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
3619 target-specific frame pointer value for the specified stack frame.
3625 The argument to this intrinsic indicates which function to return the frame
3626 pointer for. Zero indicates the calling function, one indicates its caller,
3627 etc. The argument is <b>required</b> to be a constant integer value.
3633 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
3634 the frame address of the specified call frame, or zero if it cannot be
3635 identified. The value returned by this intrinsic is likely to be incorrect or 0
3636 for arguments other than zero, so it should only be used for debugging purposes.
3640 Note that calling this intrinsic does not prevent function inlining or other
3641 aggressive transformations, so the value returned may not be that of the obvious
3642 source-language caller.
3646 <!-- _______________________________________________________________________ -->
3647 <div class="doc_subsubsection">
3648 <a name="i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
3651 <div class="doc_text">
3655 declare sbyte *%llvm.stacksave()
3661 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
3662 the function stack, for use with <a href="#i_stackrestore">
3663 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
3664 features like scoped automatic variable sized arrays in C99.
3670 This intrinsic returns a opaque pointer value that can be passed to <a
3671 href="#i_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
3672 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
3673 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
3674 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
3675 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
3676 that were allocated after the <tt>llvm.stacksave</tt> was executed.
3681 <!-- _______________________________________________________________________ -->
3682 <div class="doc_subsubsection">
3683 <a name="i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
3686 <div class="doc_text">
3690 declare void %llvm.stackrestore(sbyte* %ptr)
3696 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
3697 the function stack to the state it was in when the corresponding <a
3698 href="#llvm.stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
3699 useful for implementing language features like scoped automatic variable sized
3706 See the description for <a href="#i_stacksave"><tt>llvm.stacksave</tt></a>.
3712 <!-- _______________________________________________________________________ -->
3713 <div class="doc_subsubsection">
3714 <a name="i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
3717 <div class="doc_text">
3721 declare void %llvm.prefetch(sbyte * <address>,
3722 uint <rw>, uint <locality>)
3729 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
3730 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
3732 effect on the behavior of the program but can change its performance
3739 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
3740 determining if the fetch should be for a read (0) or write (1), and
3741 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
3742 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
3743 <tt>locality</tt> arguments must be constant integers.
3749 This intrinsic does not modify the behavior of the program. In particular,
3750 prefetches cannot trap and do not produce a value. On targets that support this
3751 intrinsic, the prefetch can provide hints to the processor cache for better
3757 <!-- _______________________________________________________________________ -->
3758 <div class="doc_subsubsection">
3759 <a name="i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
3762 <div class="doc_text">
3766 declare void %llvm.pcmarker( uint <id> )
3773 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
3775 code to simulators and other tools. The method is target specific, but it is
3776 expected that the marker will use exported symbols to transmit the PC of the marker.
3777 The marker makes no guarantees that it will remain with any specific instruction
3778 after optimizations. It is possible that the presence of a marker will inhibit
3779 optimizations. The intended use is to be inserted after optimizations to allow
3780 correlations of simulation runs.
3786 <tt>id</tt> is a numerical id identifying the marker.
3792 This intrinsic does not modify the behavior of the program. Backends that do not
3793 support this intrinisic may ignore it.
3798 <!-- _______________________________________________________________________ -->
3799 <div class="doc_subsubsection">
3800 <a name="i_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
3803 <div class="doc_text">
3807 declare ulong %llvm.readcyclecounter( )
3814 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
3815 counter register (or similar low latency, high accuracy clocks) on those targets
3816 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
3817 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
3818 should only be used for small timings.
3824 When directly supported, reading the cycle counter should not modify any memory.
3825 Implementations are allowed to either return a application specific value or a
3826 system wide value. On backends without support, this is lowered to a constant 0.
3831 <!-- ======================================================================= -->
3832 <div class="doc_subsection">
3833 <a name="int_libc">Standard C Library Intrinsics</a>
3836 <div class="doc_text">
3838 LLVM provides intrinsics for a few important standard C library functions.
3839 These intrinsics allow source-language front-ends to pass information about the
3840 alignment of the pointer arguments to the code generator, providing opportunity
3841 for more efficient code generation.
3846 <!-- _______________________________________________________________________ -->
3847 <div class="doc_subsubsection">
3848 <a name="i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
3851 <div class="doc_text">
3855 declare void %llvm.memcpy.i32(sbyte* <dest>, sbyte* <src>,
3856 uint <len>, uint <align>)
3857 declare void %llvm.memcpy.i64(sbyte* <dest>, sbyte* <src>,
3858 ulong <len>, uint <align>)
3864 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
3865 location to the destination location.
3869 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
3870 intrinsics do not return a value, and takes an extra alignment argument.
3876 The first argument is a pointer to the destination, the second is a pointer to
3877 the source. The third argument is an integer argument
3878 specifying the number of bytes to copy, and the fourth argument is the alignment
3879 of the source and destination locations.
3883 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3884 the caller guarantees that both the source and destination pointers are aligned
3891 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
3892 location to the destination location, which are not allowed to overlap. It
3893 copies "len" bytes of memory over. If the argument is known to be aligned to
3894 some boundary, this can be specified as the fourth argument, otherwise it should
3900 <!-- _______________________________________________________________________ -->
3901 <div class="doc_subsubsection">
3902 <a name="i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
3905 <div class="doc_text">
3909 declare void %llvm.memmove.i32(sbyte* <dest>, sbyte* <src>,
3910 uint <len>, uint <align>)
3911 declare void %llvm.memmove.i64(sbyte* <dest>, sbyte* <src>,
3912 ulong <len>, uint <align>)
3918 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
3919 location to the destination location. It is similar to the
3920 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
3924 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
3925 intrinsics do not return a value, and takes an extra alignment argument.
3931 The first argument is a pointer to the destination, the second is a pointer to
3932 the source. The third argument is an integer argument
3933 specifying the number of bytes to copy, and the fourth argument is the alignment
3934 of the source and destination locations.
3938 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3939 the caller guarantees that the source and destination pointers are aligned to
3946 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
3947 location to the destination location, which may overlap. It
3948 copies "len" bytes of memory over. If the argument is known to be aligned to
3949 some boundary, this can be specified as the fourth argument, otherwise it should
3955 <!-- _______________________________________________________________________ -->
3956 <div class="doc_subsubsection">
3957 <a name="i_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
3960 <div class="doc_text">
3964 declare void %llvm.memset.i32(sbyte* <dest>, ubyte <val>,
3965 uint <len>, uint <align>)
3966 declare void %llvm.memset.i64(sbyte* <dest>, ubyte <val>,
3967 ulong <len>, uint <align>)
3973 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
3978 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
3979 does not return a value, and takes an extra alignment argument.
3985 The first argument is a pointer to the destination to fill, the second is the
3986 byte value to fill it with, the third argument is an integer
3987 argument specifying the number of bytes to fill, and the fourth argument is the
3988 known alignment of destination location.
3992 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3993 the caller guarantees that the destination pointer is aligned to that boundary.
3999 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4001 destination location. If the argument is known to be aligned to some boundary,
4002 this can be specified as the fourth argument, otherwise it should be set to 0 or
4008 <!-- _______________________________________________________________________ -->
4009 <div class="doc_subsubsection">
4010 <a name="i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a>
4013 <div class="doc_text">
4017 declare bool %llvm.isunordered.f32(float Val1, float Val2)
4018 declare bool %llvm.isunordered.f64(double Val1, double Val2)
4024 The '<tt>llvm.isunordered</tt>' intrinsics return true if either or both of the
4025 specified floating point values is a NAN.
4031 The arguments are floating point numbers of the same type.
4037 If either or both of the arguments is a SNAN or QNAN, it returns true, otherwise
4043 <!-- _______________________________________________________________________ -->
4044 <div class="doc_subsubsection">
4045 <a name="i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4048 <div class="doc_text">
4052 declare float %llvm.sqrt.f32(float %Val)
4053 declare double %llvm.sqrt.f64(double %Val)
4059 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4060 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
4061 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4062 negative numbers (which allows for better optimization).
4068 The argument and return value are floating point numbers of the same type.
4074 This function returns the sqrt of the specified operand if it is a positive
4075 floating point number.
4079 <!-- _______________________________________________________________________ -->
4080 <div class="doc_subsubsection">
4081 <a name="i_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4084 <div class="doc_text">
4088 declare float %llvm.powi.f32(float %Val, int %power)
4089 declare double %llvm.powi.f64(double %Val, int %power)
4095 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4096 specified (positive or negative) power. The order of evaluation of
4097 multiplications is not defined.
4103 The second argument is an integer power, and the first is a value to raise to
4110 This function returns the first value raised to the second power with an
4111 unspecified sequence of rounding operations.</p>
4115 <!-- ======================================================================= -->
4116 <div class="doc_subsection">
4117 <a name="int_manip">Bit Manipulation Intrinsics</a>
4120 <div class="doc_text">
4122 LLVM provides intrinsics for a few important bit manipulation operations.
4123 These allow efficient code generation for some algorithms.
4128 <!-- _______________________________________________________________________ -->
4129 <div class="doc_subsubsection">
4130 <a name="i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4133 <div class="doc_text">
4137 declare ushort %llvm.bswap.i16(ushort <id>)
4138 declare uint %llvm.bswap.i32(uint <id>)
4139 declare ulong %llvm.bswap.i64(ulong <id>)
4145 The '<tt>llvm.bwsap</tt>' family of intrinsics is used to byteswap a 16, 32 or
4146 64 bit quantity. These are useful for performing operations on data that is not
4147 in the target's native byte order.
4153 The <tt>llvm.bswap.16</tt> intrinsic returns a ushort value that has the high and low
4154 byte of the input ushort swapped. Similarly, the <tt>llvm.bswap.i32</tt> intrinsic
4155 returns a uint value that has the four bytes of the input uint swapped, so that
4156 if the input bytes are numbered 0, 1, 2, 3 then the returned uint will have its
4157 bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i64</tt> intrinsic extends this concept
4163 <!-- _______________________________________________________________________ -->
4164 <div class="doc_subsubsection">
4165 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4168 <div class="doc_text">
4172 declare ubyte %llvm.ctpop.i8 (ubyte <src>)
4173 declare ushort %llvm.ctpop.i16(ushort <src>)
4174 declare uint %llvm.ctpop.i32(uint <src>)
4175 declare ulong %llvm.ctpop.i64(ulong <src>)
4181 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4188 The only argument is the value to be counted. The argument may be of any
4189 unsigned integer type. The return type must match the argument type.
4195 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4199 <!-- _______________________________________________________________________ -->
4200 <div class="doc_subsubsection">
4201 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4204 <div class="doc_text">
4208 declare ubyte %llvm.ctlz.i8 (ubyte <src>)
4209 declare ushort %llvm.ctlz.i16(ushort <src>)
4210 declare uint %llvm.ctlz.i32(uint <src>)
4211 declare ulong %llvm.ctlz.i64(ulong <src>)
4217 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4218 leading zeros in a variable.
4224 The only argument is the value to be counted. The argument may be of any
4225 unsigned integer type. The return type must match the argument type.
4231 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4232 in a variable. If the src == 0 then the result is the size in bits of the type
4233 of src. For example, <tt>llvm.ctlz(int 2) = 30</tt>.
4239 <!-- _______________________________________________________________________ -->
4240 <div class="doc_subsubsection">
4241 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
4244 <div class="doc_text">
4248 declare ubyte %llvm.cttz.i8 (ubyte <src>)
4249 declare ushort %llvm.cttz.i16(ushort <src>)
4250 declare uint %llvm.cttz.i32(uint <src>)
4251 declare ulong %llvm.cttz.i64(ulong <src>)
4257 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
4264 The only argument is the value to be counted. The argument may be of any
4265 unsigned integer type. The return type must match the argument type.
4271 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
4272 in a variable. If the src == 0 then the result is the size in bits of the type
4273 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
4277 <!-- ======================================================================= -->
4278 <div class="doc_subsection">
4279 <a name="int_debugger">Debugger Intrinsics</a>
4282 <div class="doc_text">
4284 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
4285 are described in the <a
4286 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
4287 Debugging</a> document.
4292 <!-- *********************************************************************** -->
4295 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
4296 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
4297 <a href="http://validator.w3.org/check/referer"><img
4298 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
4300 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
4301 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
4302 Last modified: $Date$