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="#paramattrs">Parameter Attributes</a></li>
28 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
31 <li><a href="#typesystem">Type System</a>
33 <li><a href="#t_primitive">Primitive Types</a>
35 <li><a href="#t_classifications">Type Classifications</a></li>
38 <li><a href="#t_derived">Derived Types</a>
40 <li><a href="#t_array">Array Type</a></li>
41 <li><a href="#t_function">Function Type</a></li>
42 <li><a href="#t_pointer">Pointer Type</a></li>
43 <li><a href="#t_struct">Structure Type</a></li>
44 <li><a href="#t_pstruct">Packed Structure Type</a></li>
45 <li><a href="#t_packed">Packed Type</a></li>
46 <li><a href="#t_opaque">Opaque Type</a></li>
51 <li><a href="#constants">Constants</a>
53 <li><a href="#simpleconstants">Simple Constants</a>
54 <li><a href="#aggregateconstants">Aggregate Constants</a>
55 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
56 <li><a href="#undefvalues">Undefined Values</a>
57 <li><a href="#constantexprs">Constant Expressions</a>
60 <li><a href="#othervalues">Other Values</a>
62 <li><a href="#inlineasm">Inline Assembler Expressions</a>
65 <li><a href="#instref">Instruction Reference</a>
67 <li><a href="#terminators">Terminator Instructions</a>
69 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
70 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
71 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
72 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
73 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
74 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
77 <li><a href="#binaryops">Binary Operations</a>
79 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
80 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
81 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
82 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
83 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
84 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
85 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
86 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
87 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
90 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
92 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
93 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
94 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
95 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
96 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
97 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
100 <li><a href="#vectorops">Vector Operations</a>
102 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
103 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
104 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
107 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
109 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
110 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
111 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
112 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
113 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
114 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
117 <li><a href="#convertops">Conversion Operations</a>
119 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
120 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
121 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
122 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
123 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
124 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
125 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
126 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
127 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
128 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
129 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
130 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
132 <li><a href="#otherops">Other Operations</a>
134 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
135 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
136 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
137 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
138 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
139 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
144 <li><a href="#intrinsics">Intrinsic Functions</a>
146 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
148 <li><a href="#i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
149 <li><a href="#i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
150 <li><a href="#i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
153 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
155 <li><a href="#i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
156 <li><a href="#i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
157 <li><a href="#i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
160 <li><a href="#int_codegen">Code Generator Intrinsics</a>
162 <li><a href="#i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
163 <li><a href="#i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
164 <li><a href="#i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
165 <li><a href="#i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
166 <li><a href="#i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
167 <li><a href="#i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
168 <li><a href="#i_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
171 <li><a href="#int_libc">Standard C Library Intrinsics</a>
173 <li><a href="#i_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
174 <li><a href="#i_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
175 <li><a href="#i_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
176 <li><a href="#i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a></li>
177 <li><a href="#i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
178 <li><a href="#i_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
181 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
183 <li><a href="#i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
184 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
185 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
186 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
189 <li><a href="#int_debugger">Debugger intrinsics</a></li>
194 <div class="doc_author">
195 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
196 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
199 <!-- *********************************************************************** -->
200 <div class="doc_section"> <a name="abstract">Abstract </a></div>
201 <!-- *********************************************************************** -->
203 <div class="doc_text">
204 <p>This document is a reference manual for the LLVM assembly language.
205 LLVM is an SSA based representation that provides type safety,
206 low-level operations, flexibility, and the capability of representing
207 'all' high-level languages cleanly. It is the common code
208 representation used throughout all phases of the LLVM compilation
212 <!-- *********************************************************************** -->
213 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
214 <!-- *********************************************************************** -->
216 <div class="doc_text">
218 <p>The LLVM code representation is designed to be used in three
219 different forms: as an in-memory compiler IR, as an on-disk bytecode
220 representation (suitable for fast loading by a Just-In-Time compiler),
221 and as a human readable assembly language representation. This allows
222 LLVM to provide a powerful intermediate representation for efficient
223 compiler transformations and analysis, while providing a natural means
224 to debug and visualize the transformations. The three different forms
225 of LLVM are all equivalent. This document describes the human readable
226 representation and notation.</p>
228 <p>The LLVM representation aims to be light-weight and low-level
229 while being expressive, typed, and extensible at the same time. It
230 aims to be a "universal IR" of sorts, by being at a low enough level
231 that high-level ideas may be cleanly mapped to it (similar to how
232 microprocessors are "universal IR's", allowing many source languages to
233 be mapped to them). By providing type information, LLVM can be used as
234 the target of optimizations: for example, through pointer analysis, it
235 can be proven that a C automatic variable is never accessed outside of
236 the current function... allowing it to be promoted to a simple SSA
237 value instead of a memory location.</p>
241 <!-- _______________________________________________________________________ -->
242 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
244 <div class="doc_text">
246 <p>It is important to note that this document describes 'well formed'
247 LLVM assembly language. There is a difference between what the parser
248 accepts and what is considered 'well formed'. For example, the
249 following instruction is syntactically okay, but not well formed:</p>
252 %x = <a href="#i_add">add</a> i32 1, %x
255 <p>...because the definition of <tt>%x</tt> does not dominate all of
256 its uses. The LLVM infrastructure provides a verification pass that may
257 be used to verify that an LLVM module is well formed. This pass is
258 automatically run by the parser after parsing input assembly and by
259 the optimizer before it outputs bytecode. The violations pointed out
260 by the verifier pass indicate bugs in transformation passes or input to
263 <!-- Describe the typesetting conventions here. --> </div>
265 <!-- *********************************************************************** -->
266 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
267 <!-- *********************************************************************** -->
269 <div class="doc_text">
271 <p>LLVM uses three different forms of identifiers, for different
275 <li>Named values are represented as a string of characters with a '%' prefix.
276 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
277 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
278 Identifiers which require other characters in their names can be surrounded
279 with quotes. In this way, anything except a <tt>"</tt> character can be used
282 <li>Unnamed values are represented as an unsigned numeric value with a '%'
283 prefix. For example, %12, %2, %44.</li>
285 <li>Constants, which are described in a <a href="#constants">section about
286 constants</a>, below.</li>
289 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
290 don't need to worry about name clashes with reserved words, and the set of
291 reserved words may be expanded in the future without penalty. Additionally,
292 unnamed identifiers allow a compiler to quickly come up with a temporary
293 variable without having to avoid symbol table conflicts.</p>
295 <p>Reserved words in LLVM are very similar to reserved words in other
296 languages. There are keywords for different opcodes
297 ('<tt><a href="#i_add">add</a></tt>',
298 '<tt><a href="#i_bitcast">bitcast</a></tt>',
299 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
300 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
301 and others. These reserved words cannot conflict with variable names, because
302 none of them start with a '%' character.</p>
304 <p>Here is an example of LLVM code to multiply the integer variable
305 '<tt>%X</tt>' by 8:</p>
310 %result = <a href="#i_mul">mul</a> i32 %X, 8
313 <p>After strength reduction:</p>
316 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
319 <p>And the hard way:</p>
322 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
323 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
324 %result = <a href="#i_add">add</a> i32 %1, %1
327 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
328 important lexical features of LLVM:</p>
332 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
335 <li>Unnamed temporaries are created when the result of a computation is not
336 assigned to a named value.</li>
338 <li>Unnamed temporaries are numbered sequentially</li>
342 <p>...and it also shows a convention that we follow in this document. When
343 demonstrating instructions, we will follow an instruction with a comment that
344 defines the type and name of value produced. Comments are shown in italic
349 <!-- *********************************************************************** -->
350 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
351 <!-- *********************************************************************** -->
353 <!-- ======================================================================= -->
354 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
357 <div class="doc_text">
359 <p>LLVM programs are composed of "Module"s, each of which is a
360 translation unit of the input programs. Each module consists of
361 functions, global variables, and symbol table entries. Modules may be
362 combined together with the LLVM linker, which merges function (and
363 global variable) definitions, resolves forward declarations, and merges
364 symbol table entries. Here is an example of the "hello world" module:</p>
366 <pre><i>; Declare the string constant as a global constant...</i>
367 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
368 href="#globalvars">constant</a> <a href="#t_array">[13 x i8 ]</a> c"hello world\0A\00" <i>; [13 x i8 ]*</i>
370 <i>; External declaration of the puts function</i>
371 <a href="#functionstructure">declare</a> i32 %puts(i8 *) <i>; i32(i8 *)* </i>
373 <i>; Global variable / Function body section separator</i>
376 <i>; Definition of main function</i>
377 define i32 %main() { <i>; i32()* </i>
378 <i>; Convert [13x i8 ]* to i8 *...</i>
380 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* %.LC0, i64 0, i64 0 <i>; i8 *</i>
382 <i>; Call puts function to write out the string to stdout...</i>
384 href="#i_call">call</a> i32 %puts(i8 * %cast210) <i>; i32</i>
386 href="#i_ret">ret</a> i32 0<br>}<br></pre>
388 <p>This example is made up of a <a href="#globalvars">global variable</a>
389 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
390 function, and a <a href="#functionstructure">function definition</a>
391 for "<tt>main</tt>".</p>
393 <p>In general, a module is made up of a list of global values,
394 where both functions and global variables are global values. Global values are
395 represented by a pointer to a memory location (in this case, a pointer to an
396 array of char, and a pointer to a function), and have one of the following <a
397 href="#linkage">linkage types</a>.</p>
399 <p>Due to a limitation in the current LLVM assembly parser (it is limited by
400 one-token lookahead), modules are split into two pieces by the "implementation"
401 keyword. Global variable prototypes and definitions must occur before the
402 keyword, and function definitions must occur after it. Function prototypes may
403 occur either before or after it. In the future, the implementation keyword may
404 become a noop, if the parser gets smarter.</p>
408 <!-- ======================================================================= -->
409 <div class="doc_subsection">
410 <a name="linkage">Linkage Types</a>
413 <div class="doc_text">
416 All Global Variables and Functions have one of the following types of linkage:
421 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
423 <dd>Global values with internal linkage are only directly accessible by
424 objects in the current module. In particular, linking code into a module with
425 an internal global value may cause the internal to be renamed as necessary to
426 avoid collisions. Because the symbol is internal to the module, all
427 references can be updated. This corresponds to the notion of the
428 '<tt>static</tt>' keyword in C, or the idea of "anonymous namespaces" in C++.
431 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
433 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt> linkage, with
434 the twist that linking together two modules defining the same
435 <tt>linkonce</tt> globals will cause one of the globals to be discarded. This
436 is typically used to implement inline functions. Unreferenced
437 <tt>linkonce</tt> globals are allowed to be discarded.
440 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
442 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
443 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
444 used to implement constructs in C such as "<tt>i32 X;</tt>" at global scope.
447 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
449 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
450 pointer to array type. When two global variables with appending linkage are
451 linked together, the two global arrays are appended together. This is the
452 LLVM, typesafe, equivalent of having the system linker append together
453 "sections" with identical names when .o files are linked.
456 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
458 <dd>If none of the above identifiers are used, the global is externally
459 visible, meaning that it participates in linkage and can be used to resolve
460 external symbol references.
463 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
465 <dd>"<tt>extern_weak</tt>" TBD
469 The next two types of linkage are targeted for Microsoft Windows platform
470 only. They are designed to support importing (exporting) symbols from (to)
474 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
476 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
477 or variable via a global pointer to a pointer that is set up by the DLL
478 exporting the symbol. On Microsoft Windows targets, the pointer name is
479 formed by combining <code>_imp__</code> and the function or variable name.
482 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
484 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
485 pointer to a pointer in a DLL, so that it can be referenced with the
486 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
487 name is formed by combining <code>_imp__</code> and the function or variable
493 <p><a name="linkage_external">For example, since the "<tt>.LC0</tt>"
494 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
495 variable and was linked with this one, one of the two would be renamed,
496 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
497 external (i.e., lacking any linkage declarations), they are accessible
498 outside of the current module. It is illegal for a function <i>declaration</i>
499 to have any linkage type other than "externally visible".</a></p>
503 <!-- ======================================================================= -->
504 <div class="doc_subsection">
505 <a name="callingconv">Calling Conventions</a>
508 <div class="doc_text">
510 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
511 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
512 specified for the call. The calling convention of any pair of dynamic
513 caller/callee must match, or the behavior of the program is undefined. The
514 following calling conventions are supported by LLVM, and more may be added in
518 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
520 <dd>This calling convention (the default if no other calling convention is
521 specified) matches the target C calling conventions. This calling convention
522 supports varargs function calls and tolerates some mismatch in the declared
523 prototype and implemented declaration of the function (as does normal C).
526 <dt><b>"<tt>csretcc</tt>" - The C struct return calling convention</b>:</dt>
528 <dd>This calling convention matches the target C calling conventions, except
529 that functions with this convention are required to take a pointer as their
530 first argument, and the return type of the function must be void. This is
531 used for C functions that return aggregates by-value. In this case, the
532 function has been transformed to take a pointer to the struct as the first
533 argument to the function. For targets where the ABI specifies specific
534 behavior for structure-return calls, the calling convention can be used to
535 distinguish between struct return functions and other functions that take a
536 pointer to a struct as the first argument.
539 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
541 <dd>This calling convention attempts to make calls as fast as possible
542 (e.g. by passing things in registers). This calling convention allows the
543 target to use whatever tricks it wants to produce fast code for the target,
544 without having to conform to an externally specified ABI. Implementations of
545 this convention should allow arbitrary tail call optimization to be supported.
546 This calling convention does not support varargs and requires the prototype of
547 all callees to exactly match the prototype of the function definition.
550 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
552 <dd>This calling convention attempts to make code in the caller as efficient
553 as possible under the assumption that the call is not commonly executed. As
554 such, these calls often preserve all registers so that the call does not break
555 any live ranges in the caller side. This calling convention does not support
556 varargs and requires the prototype of all callees to exactly match the
557 prototype of the function definition.
560 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
562 <dd>Any calling convention may be specified by number, allowing
563 target-specific calling conventions to be used. Target specific calling
564 conventions start at 64.
568 <p>More calling conventions can be added/defined on an as-needed basis, to
569 support pascal conventions or any other well-known target-independent
574 <!-- ======================================================================= -->
575 <div class="doc_subsection">
576 <a name="globalvars">Global Variables</a>
579 <div class="doc_text">
581 <p>Global variables define regions of memory allocated at compilation time
582 instead of run-time. Global variables may optionally be initialized, may have
583 an explicit section to be placed in, and may
584 have an optional explicit alignment specified. A
585 variable may be defined as a global "constant," which indicates that the
586 contents of the variable will <b>never</b> be modified (enabling better
587 optimization, allowing the global data to be placed in the read-only section of
588 an executable, etc). Note that variables that need runtime initialization
589 cannot be marked "constant" as there is a store to the variable.</p>
592 LLVM explicitly allows <em>declarations</em> of global variables to be marked
593 constant, even if the final definition of the global is not. This capability
594 can be used to enable slightly better optimization of the program, but requires
595 the language definition to guarantee that optimizations based on the
596 'constantness' are valid for the translation units that do not include the
600 <p>As SSA values, global variables define pointer values that are in
601 scope (i.e. they dominate) all basic blocks in the program. Global
602 variables always define a pointer to their "content" type because they
603 describe a region of memory, and all memory objects in LLVM are
604 accessed through pointers.</p>
606 <p>LLVM allows an explicit section to be specified for globals. If the target
607 supports it, it will emit globals to the section specified.</p>
609 <p>An explicit alignment may be specified for a global. If not present, or if
610 the alignment is set to zero, the alignment of the global is set by the target
611 to whatever it feels convenient. If an explicit alignment is specified, the
612 global is forced to have at least that much alignment. All alignments must be
618 <!-- ======================================================================= -->
619 <div class="doc_subsection">
620 <a name="functionstructure">Functions</a>
623 <div class="doc_text">
625 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
626 an optional <a href="#linkage">linkage type</a>, an optional
627 <a href="#callingconv">calling convention</a>, a return type, an optional
628 <a href="#paramattrs">parameter attribute</a> for the return type, a function
629 name, a (possibly empty) argument list (each with optional
630 <a href="#paramattrs">parameter attributes</a>), an optional section, an
631 optional alignment, an opening curly brace, a list of basic blocks, and a
632 closing curly brace. LLVM function declarations
633 consist of the "<tt>declare</tt>" keyword, an optional <a
634 href="#callingconv">calling convention</a>, a return type, an optional
635 <a href="#paramattrs">parameter attribute</a> for the return type, a function
636 name, a possibly empty list of arguments, and an optional alignment.</p>
638 <p>A function definition contains a list of basic blocks, forming the CFG for
639 the function. Each basic block may optionally start with a label (giving the
640 basic block a symbol table entry), contains a list of instructions, and ends
641 with a <a href="#terminators">terminator</a> instruction (such as a branch or
642 function return).</p>
644 <p>The first basic block in a program is special in two ways: it is immediately
645 executed on entrance to the function, and it is not allowed to have predecessor
646 basic blocks (i.e. there can not be any branches to the entry block of a
647 function). Because the block can have no predecessors, it also cannot have any
648 <a href="#i_phi">PHI nodes</a>.</p>
650 <p>LLVM functions are identified by their name and type signature. Hence, two
651 functions with the same name but different parameter lists or return values are
652 considered different functions, and LLVM will resolve references to each
655 <p>LLVM allows an explicit section to be specified for functions. If the target
656 supports it, it will emit functions to the section specified.</p>
658 <p>An explicit alignment may be specified for a function. If not present, or if
659 the alignment is set to zero, the alignment of the function is set by the target
660 to whatever it feels convenient. If an explicit alignment is specified, the
661 function is forced to have at least that much alignment. All alignments must be
666 <!-- ======================================================================= -->
667 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
668 <div class="doc_text">
669 <p>The return type and each parameter of a function type may have a set of
670 <i>parameter attributes</i> associated with them. Parameter attributes are
671 used to communicate additional information about the result or parameters of
672 a function. Parameter attributes are considered to be part of the function
673 type so two functions types that differ only by the parameter attributes
674 are different function types.</p>
676 <p>Parameter attributes consist of an at sign (@) followed by either a single
677 keyword or a comma separate list of keywords enclosed in parentheses. For
679 %someFunc = i16 @zext (i8 @(sext) %someParam)
680 %someFunc = i16 @zext (i8 @zext %someParam)</pre>
681 Note that the two function types above are unique because the parameter
682 has a different attribute (@sext in the first one, @zext in the second).</p>
684 <p>Currently, only the following parameter attributes are defined:
686 <dt><tt>@zext</tt></dt>
687 <dd>This indicates that the parameter should be zero extended just before
688 a call to this function.</dd>
689 <dt><tt>@sext</tt></dt>
690 <dd>This indicates that the parameter should be sign extended just before
691 a call to this function.</dd>
694 <p>The current motivation for parameter attributes is to enable the sign and
695 zero extend information necessary for the C calling convention to be passed
696 from the front end to LLVM. The <tt>@zext</tt> and <tt>@sext</tt> attributes
697 are used by the code generator to perform the required extension. However,
698 parameter attributes are an orthogonal feature to calling conventions and
699 may be used for other purposes in the future.</p>
702 <!-- ======================================================================= -->
703 <div class="doc_subsection">
704 <a name="moduleasm">Module-Level Inline Assembly</a>
707 <div class="doc_text">
709 Modules may contain "module-level inline asm" blocks, which corresponds to the
710 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
711 LLVM and treated as a single unit, but may be separated in the .ll file if
712 desired. The syntax is very simple:
715 <div class="doc_code"><pre>
716 module asm "inline asm code goes here"
717 module asm "more can go here"
720 <p>The strings can contain any character by escaping non-printable characters.
721 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
726 The inline asm code is simply printed to the machine code .s file when
727 assembly code is generated.
732 <!-- *********************************************************************** -->
733 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
734 <!-- *********************************************************************** -->
736 <div class="doc_text">
738 <p>The LLVM type system is one of the most important features of the
739 intermediate representation. Being typed enables a number of
740 optimizations to be performed on the IR directly, without having to do
741 extra analyses on the side before the transformation. A strong type
742 system makes it easier to read the generated code and enables novel
743 analyses and transformations that are not feasible to perform on normal
744 three address code representations.</p>
748 <!-- ======================================================================= -->
749 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
750 <div class="doc_text">
751 <p>The primitive types are the fundamental building blocks of the LLVM
752 system. The current set of primitive types is as follows:</p>
754 <table class="layout">
759 <tr><th>Type</th><th>Description</th></tr>
760 <tr><td><tt>void</tt></td><td>No value</td></tr>
761 <tr><td><tt>i8</tt></td><td>Signless 8-bit value</td></tr>
762 <tr><td><tt>i32</tt></td><td>Signless 32-bit value</td></tr>
763 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
764 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
771 <tr><th>Type</th><th>Description</th></tr>
772 <tr><td><tt>bool</tt></td><td>True or False value</td></tr>
773 <tr><td><tt>i16</tt></td><td>Signless 16-bit value</td></tr>
774 <tr><td><tt>i64</tt></td><td>Signless 64-bit value</td></tr>
775 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
783 <!-- _______________________________________________________________________ -->
784 <div class="doc_subsubsection"> <a name="t_classifications">Type
785 Classifications</a> </div>
786 <div class="doc_text">
787 <p>These different primitive types fall into a few useful
790 <table border="1" cellspacing="0" cellpadding="4">
792 <tr><th>Classification</th><th>Types</th></tr>
794 <td><a name="t_integer">integer</a></td>
795 <td><tt>i8, i16, i32, i64</tt></td>
798 <td><a name="t_integral">integral</a></td>
799 <td><tt>bool, i8, i16, i32, i64</tt>
803 <td><a name="t_floating">floating point</a></td>
804 <td><tt>float, double</tt></td>
807 <td><a name="t_firstclass">first class</a></td>
808 <td><tt>bool, i8, i16, i32, i64, float, double, <br/>
809 <a href="#t_pointer">pointer</a>,<a href="#t_packed">packed</a></tt>
815 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
816 most important. Values of these types are the only ones which can be
817 produced by instructions, passed as arguments, or used as operands to
818 instructions. This means that all structures and arrays must be
819 manipulated either by pointer or by component.</p>
822 <!-- ======================================================================= -->
823 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
825 <div class="doc_text">
827 <p>The real power in LLVM comes from the derived types in the system.
828 This is what allows a programmer to represent arrays, functions,
829 pointers, and other useful types. Note that these derived types may be
830 recursive: For example, it is possible to have a two dimensional array.</p>
834 <!-- _______________________________________________________________________ -->
835 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
837 <div class="doc_text">
841 <p>The array type is a very simple derived type that arranges elements
842 sequentially in memory. The array type requires a size (number of
843 elements) and an underlying data type.</p>
848 [<# elements> x <elementtype>]
851 <p>The number of elements is a constant integer value; elementtype may
852 be any type with a size.</p>
855 <table class="layout">
858 <tt>[40 x i32 ]</tt><br/>
859 <tt>[41 x i32 ]</tt><br/>
860 <tt>[40 x i32]</tt><br/>
863 Array of 40 integer values.<br/>
864 Array of 41 integer values.<br/>
865 Array of 40 unsigned integer values.<br/>
869 <p>Here are some examples of multidimensional arrays:</p>
870 <table class="layout">
873 <tt>[3 x [4 x i32]]</tt><br/>
874 <tt>[12 x [10 x float]]</tt><br/>
875 <tt>[2 x [3 x [4 x i32]]]</tt><br/>
878 3x4 array of integer values.<br/>
879 12x10 array of single precision floating point values.<br/>
880 2x3x4 array of unsigned integer values.<br/>
885 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
886 length array. Normally, accesses past the end of an array are undefined in
887 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
888 As a special case, however, zero length arrays are recognized to be variable
889 length. This allows implementation of 'pascal style arrays' with the LLVM
890 type "{ i32, [0 x float]}", for example.</p>
894 <!-- _______________________________________________________________________ -->
895 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
896 <div class="doc_text">
898 <p>The function type can be thought of as a function signature. It
899 consists of a return type and a list of formal parameter types.
900 Function types are usually used to build virtual function tables
901 (which are structures of pointers to functions), for indirect function
902 calls, and when defining a function.</p>
904 The return type of a function type cannot be an aggregate type.
907 <pre> <returntype> (<parameter list>)<br></pre>
908 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
909 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
910 which indicates that the function takes a variable number of arguments.
911 Variable argument functions can access their arguments with the <a
912 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
914 <table class="layout">
916 <td class="left"><tt>i32 (i32)</tt></td>
917 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
919 </tr><tr class="layout">
920 <td class="left"><tt>float (i16 @sext, i32 *) *
922 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
923 an <tt>i16</tt> that should be sign extended and a
924 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
927 </tr><tr class="layout">
928 <td class="left"><tt>i32 (i8*, ...)</tt></td>
929 <td class="left">A vararg function that takes at least one
930 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (signed char in C),
931 which returns an integer. This is the signature for <tt>printf</tt> in
938 <!-- _______________________________________________________________________ -->
939 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
940 <div class="doc_text">
942 <p>The structure type is used to represent a collection of data members
943 together in memory. The packing of the field types is defined to match
944 the ABI of the underlying processor. The elements of a structure may
945 be any type that has a size.</p>
946 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
947 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
948 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
951 <pre> { <type list> }<br></pre>
953 <table class="layout">
956 <tt>{ i32, i32, i32 }</tt><br/>
957 <tt>{ float, i32 (i32) * }</tt><br/>
960 a triple of three <tt>i32</tt> values<br/>
961 A pair, where the first element is a <tt>float</tt> and the second element
962 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
963 that takes an <tt>i32</tt>, returning an <tt>i32</tt>.<br/>
969 <!-- _______________________________________________________________________ -->
970 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
972 <div class="doc_text">
974 <p>The packed structure type is used to represent a collection of data members
975 together in memory. There is no padding between fields. Further, the alignment
976 of a packed structure is 1 byte. The elements of a packed structure may
977 be any type that has a size.</p>
978 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
979 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
980 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
983 <pre> < { <type list> } > <br></pre>
985 <table class="layout">
988 <tt> < { i32, i32, i32 } > </tt><br/>
989 <tt> < { float, i32 (i32) * } > </tt><br/>
992 a triple of three <tt>i32</tt> values<br/>
993 A pair, where the first element is a <tt>float</tt> and the second element
994 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
995 that takes an <tt>i32</tt>, returning an <tt>i32</tt>.<br/>
1001 <!-- _______________________________________________________________________ -->
1002 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1003 <div class="doc_text">
1005 <p>As in many languages, the pointer type represents a pointer or
1006 reference to another object, which must live in memory.</p>
1008 <pre> <type> *<br></pre>
1010 <table class="layout">
1013 <tt>[4x i32]*</tt><br/>
1014 <tt>i32 (i32 *) *</tt><br/>
1017 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
1018 four <tt>i32</tt> values<br/>
1019 A <a href="#t_pointer">pointer</a> to a <a
1020 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1027 <!-- _______________________________________________________________________ -->
1028 <div class="doc_subsubsection"> <a name="t_packed">Packed Type</a> </div>
1029 <div class="doc_text">
1033 <p>A packed type is a simple derived type that represents a vector
1034 of elements. Packed types are used when multiple primitive data
1035 are operated in parallel using a single instruction (SIMD).
1036 A packed type requires a size (number of
1037 elements) and an underlying primitive data type. Vectors must have a power
1038 of two length (1, 2, 4, 8, 16 ...). Packed types are
1039 considered <a href="#t_firstclass">first class</a>.</p>
1044 < <# elements> x <elementtype> >
1047 <p>The number of elements is a constant integer value; elementtype may
1048 be any integral or floating point type.</p>
1052 <table class="layout">
1055 <tt><4 x i32></tt><br/>
1056 <tt><8 x float></tt><br/>
1057 <tt><2 x i32></tt><br/>
1060 Packed vector of 4 integer values.<br/>
1061 Packed vector of 8 floating-point values.<br/>
1062 Packed vector of 2 unsigned integer values.<br/>
1068 <!-- _______________________________________________________________________ -->
1069 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1070 <div class="doc_text">
1074 <p>Opaque types are used to represent unknown types in the system. This
1075 corresponds (for example) to the C notion of a foward declared structure type.
1076 In LLVM, opaque types can eventually be resolved to any type (not just a
1077 structure type).</p>
1087 <table class="layout">
1093 An opaque type.<br/>
1100 <!-- *********************************************************************** -->
1101 <div class="doc_section"> <a name="constants">Constants</a> </div>
1102 <!-- *********************************************************************** -->
1104 <div class="doc_text">
1106 <p>LLVM has several different basic types of constants. This section describes
1107 them all and their syntax.</p>
1111 <!-- ======================================================================= -->
1112 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1114 <div class="doc_text">
1117 <dt><b>Boolean constants</b></dt>
1119 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1120 constants of the <tt><a href="#t_primitive">bool</a></tt> type.
1123 <dt><b>Integer constants</b></dt>
1125 <dd>Standard integers (such as '4') are constants of the <a
1126 href="#t_integer">integer</a> type. Negative numbers may be used with signed
1130 <dt><b>Floating point constants</b></dt>
1132 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1133 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1134 notation (see below). Floating point constants must have a <a
1135 href="#t_floating">floating point</a> type. </dd>
1137 <dt><b>Null pointer constants</b></dt>
1139 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1140 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1144 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1145 of floating point constants. For example, the form '<tt>double
1146 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1147 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1148 (and the only time that they are generated by the disassembler) is when a
1149 floating point constant must be emitted but it cannot be represented as a
1150 decimal floating point number. For example, NaN's, infinities, and other
1151 special values are represented in their IEEE hexadecimal format so that
1152 assembly and disassembly do not cause any bits to change in the constants.</p>
1156 <!-- ======================================================================= -->
1157 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1160 <div class="doc_text">
1161 <p>Aggregate constants arise from aggregation of simple constants
1162 and smaller aggregate constants.</p>
1165 <dt><b>Structure constants</b></dt>
1167 <dd>Structure constants are represented with notation similar to structure
1168 type definitions (a comma separated list of elements, surrounded by braces
1169 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* %G }</tt>",
1170 where "<tt>%G</tt>" is declared as "<tt>%G = external global i32</tt>". Structure constants
1171 must have <a href="#t_struct">structure type</a>, and the number and
1172 types of elements must match those specified by the type.
1175 <dt><b>Array constants</b></dt>
1177 <dd>Array constants are represented with notation similar to array type
1178 definitions (a comma separated list of elements, surrounded by square brackets
1179 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1180 constants must have <a href="#t_array">array type</a>, and the number and
1181 types of elements must match those specified by the type.
1184 <dt><b>Packed constants</b></dt>
1186 <dd>Packed constants are represented with notation similar to packed type
1187 definitions (a comma separated list of elements, surrounded by
1188 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1189 i32 11, i32 74, i32 100 ></tt>". Packed constants must have <a
1190 href="#t_packed">packed type</a>, and the number and types of elements must
1191 match those specified by the type.
1194 <dt><b>Zero initialization</b></dt>
1196 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1197 value to zero of <em>any</em> type, including scalar and aggregate types.
1198 This is often used to avoid having to print large zero initializers (e.g. for
1199 large arrays) and is always exactly equivalent to using explicit zero
1206 <!-- ======================================================================= -->
1207 <div class="doc_subsection">
1208 <a name="globalconstants">Global Variable and Function Addresses</a>
1211 <div class="doc_text">
1213 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1214 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1215 constants. These constants are explicitly referenced when the <a
1216 href="#identifiers">identifier for the global</a> is used and always have <a
1217 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1223 %Z = global [2 x i32*] [ i32* %X, i32* %Y ]
1228 <!-- ======================================================================= -->
1229 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1230 <div class="doc_text">
1231 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1232 no specific value. Undefined values may be of any type and be used anywhere
1233 a constant is permitted.</p>
1235 <p>Undefined values indicate to the compiler that the program is well defined
1236 no matter what value is used, giving the compiler more freedom to optimize.
1240 <!-- ======================================================================= -->
1241 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1244 <div class="doc_text">
1246 <p>Constant expressions are used to allow expressions involving other constants
1247 to be used as constants. Constant expressions may be of any <a
1248 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1249 that does not have side effects (e.g. load and call are not supported). The
1250 following is the syntax for constant expressions:</p>
1253 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1254 <dd>Truncate a constant to another type. The bit size of CST must be larger
1255 than the bit size of TYPE. Both types must be integral.</dd>
1257 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1258 <dd>Zero extend a constant to another type. The bit size of CST must be
1259 smaller or equal to the bit size of TYPE. Both types must be integral.</dd>
1261 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1262 <dd>Sign extend a constant to another type. The bit size of CST must be
1263 smaller or equal to the bit size of TYPE. Both types must be integral.</dd>
1265 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1266 <dd>Truncate a floating point constant to another floating point type. The
1267 size of CST must be larger than the size of TYPE. Both types must be
1268 floating point.</dd>
1270 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1271 <dd>Floating point extend a constant to another type. The size of CST must be
1272 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1274 <dt><b><tt>fp2uint ( CST to TYPE )</tt></b></dt>
1275 <dd>Convert a floating point constant to the corresponding unsigned integer
1276 constant. TYPE must be an integer type. CST must be floating point. If the
1277 value won't fit in the integer type, the results are undefined.</dd>
1279 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1280 <dd>Convert a floating point constant to the corresponding signed integer
1281 constant. TYPE must be an integer type. CST must be floating point. If the
1282 value won't fit in the integer type, the results are undefined.</dd>
1284 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1285 <dd>Convert an unsigned integer constant to the corresponding floating point
1286 constant. TYPE must be floating point. CST must be of integer type. If the
1287 value won't fit in the floating point type, the results are undefined.</dd>
1289 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1290 <dd>Convert a signed integer constant to the corresponding floating point
1291 constant. TYPE must be floating point. CST must be of integer type. If the
1292 value won't fit in the floating point type, the results are undefined.</dd>
1294 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1295 <dd>Convert a pointer typed constant to the corresponding integer constant
1296 TYPE must be an integer type. CST must be of pointer type. The CST value is
1297 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1299 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1300 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1301 pointer type. CST must be of integer type. The CST value is zero extended,
1302 truncated, or unchanged to make it fit in a pointer size. This one is
1303 <i>really</i> dangerous!</dd>
1305 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1306 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1307 identical (same number of bits). The conversion is done as if the CST value
1308 was stored to memory and read back as TYPE. In other words, no bits change
1309 with this operator, just the type. This can be used for conversion of
1310 packed types to any other type, as long as they have the same bit width. For
1311 pointers it is only valid to cast to another pointer type.
1314 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1316 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1317 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1318 instruction, the index list may have zero or more indexes, which are required
1319 to make sense for the type of "CSTPTR".</dd>
1321 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1323 <dd>Perform the <a href="#i_select">select operation</a> on
1326 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1327 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1329 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1330 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1332 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1334 <dd>Perform the <a href="#i_extractelement">extractelement
1335 operation</a> on constants.
1337 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1339 <dd>Perform the <a href="#i_insertelement">insertelement
1340 operation</a> on constants.</dd>
1343 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1345 <dd>Perform the <a href="#i_shufflevector">shufflevector
1346 operation</a> on constants.</dd>
1348 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1350 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1351 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1352 binary</a> operations. The constraints on operands are the same as those for
1353 the corresponding instruction (e.g. no bitwise operations on floating point
1354 values are allowed).</dd>
1358 <!-- *********************************************************************** -->
1359 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1360 <!-- *********************************************************************** -->
1362 <!-- ======================================================================= -->
1363 <div class="doc_subsection">
1364 <a name="inlineasm">Inline Assembler Expressions</a>
1367 <div class="doc_text">
1370 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1371 Module-Level Inline Assembly</a>) through the use of a special value. This
1372 value represents the inline assembler as a string (containing the instructions
1373 to emit), a list of operand constraints (stored as a string), and a flag that
1374 indicates whether or not the inline asm expression has side effects. An example
1375 inline assembler expression is:
1379 i32 (i32) asm "bswap $0", "=r,r"
1383 Inline assembler expressions may <b>only</b> be used as the callee operand of
1384 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1388 %X = call i32 asm "<a href="#i_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1392 Inline asms with side effects not visible in the constraint list must be marked
1393 as having side effects. This is done through the use of the
1394 '<tt>sideeffect</tt>' keyword, like so:
1398 call void asm sideeffect "eieio", ""()
1401 <p>TODO: The format of the asm and constraints string still need to be
1402 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1403 need to be documented).
1408 <!-- *********************************************************************** -->
1409 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1410 <!-- *********************************************************************** -->
1412 <div class="doc_text">
1414 <p>The LLVM instruction set consists of several different
1415 classifications of instructions: <a href="#terminators">terminator
1416 instructions</a>, <a href="#binaryops">binary instructions</a>,
1417 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1418 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1419 instructions</a>.</p>
1423 <!-- ======================================================================= -->
1424 <div class="doc_subsection"> <a name="terminators">Terminator
1425 Instructions</a> </div>
1427 <div class="doc_text">
1429 <p>As mentioned <a href="#functionstructure">previously</a>, every
1430 basic block in a program ends with a "Terminator" instruction, which
1431 indicates which block should be executed after the current block is
1432 finished. These terminator instructions typically yield a '<tt>void</tt>'
1433 value: they produce control flow, not values (the one exception being
1434 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1435 <p>There are six different terminator instructions: the '<a
1436 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1437 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1438 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1439 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1440 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1444 <!-- _______________________________________________________________________ -->
1445 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1446 Instruction</a> </div>
1447 <div class="doc_text">
1449 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1450 ret void <i>; Return from void function</i>
1453 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1454 value) from a function back to the caller.</p>
1455 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1456 returns a value and then causes control flow, and one that just causes
1457 control flow to occur.</p>
1459 <p>The '<tt>ret</tt>' instruction may return any '<a
1460 href="#t_firstclass">first class</a>' type. Notice that a function is
1461 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1462 instruction inside of the function that returns a value that does not
1463 match the return type of the function.</p>
1465 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1466 returns back to the calling function's context. If the caller is a "<a
1467 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1468 the instruction after the call. If the caller was an "<a
1469 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1470 at the beginning of the "normal" destination block. If the instruction
1471 returns a value, that value shall set the call or invoke instruction's
1474 <pre> ret i32 5 <i>; Return an integer value of 5</i>
1475 ret void <i>; Return from a void function</i>
1478 <!-- _______________________________________________________________________ -->
1479 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1480 <div class="doc_text">
1482 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1485 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1486 transfer to a different basic block in the current function. There are
1487 two forms of this instruction, corresponding to a conditional branch
1488 and an unconditional branch.</p>
1490 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1491 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
1492 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
1493 value as a target.</p>
1495 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
1496 argument is evaluated. If the value is <tt>true</tt>, control flows
1497 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1498 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1500 <pre>Test:<br> %cond = <a href="#i_icmp">icmp</a> eq, i32 %a, %b<br> br bool %cond, label %IfEqual, label %IfUnequal<br>IfEqual:<br> <a
1501 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1503 <!-- _______________________________________________________________________ -->
1504 <div class="doc_subsubsection">
1505 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1508 <div class="doc_text">
1512 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1517 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1518 several different places. It is a generalization of the '<tt>br</tt>'
1519 instruction, allowing a branch to occur to one of many possible
1525 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1526 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1527 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1528 table is not allowed to contain duplicate constant entries.</p>
1532 <p>The <tt>switch</tt> instruction specifies a table of values and
1533 destinations. When the '<tt>switch</tt>' instruction is executed, this
1534 table is searched for the given value. If the value is found, control flow is
1535 transfered to the corresponding destination; otherwise, control flow is
1536 transfered to the default destination.</p>
1538 <h5>Implementation:</h5>
1540 <p>Depending on properties of the target machine and the particular
1541 <tt>switch</tt> instruction, this instruction may be code generated in different
1542 ways. For example, it could be generated as a series of chained conditional
1543 branches or with a lookup table.</p>
1548 <i>; Emulate a conditional br instruction</i>
1549 %Val = <a href="#i_zext">zext</a> bool %value to i32
1550 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1552 <i>; Emulate an unconditional br instruction</i>
1553 switch i32 0, label %dest [ ]
1555 <i>; Implement a jump table:</i>
1556 switch i32 %val, label %otherwise [ i32 0, label %onzero
1558 i32 2, label %ontwo ]
1562 <!-- _______________________________________________________________________ -->
1563 <div class="doc_subsubsection">
1564 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1567 <div class="doc_text">
1572 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1573 to label <normal label> unwind label <exception label>
1578 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1579 function, with the possibility of control flow transfer to either the
1580 '<tt>normal</tt>' label or the
1581 '<tt>exception</tt>' label. If the callee function returns with the
1582 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1583 "normal" label. If the callee (or any indirect callees) returns with the "<a
1584 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1585 continued at the dynamically nearest "exception" label.</p>
1589 <p>This instruction requires several arguments:</p>
1593 The optional "cconv" marker indicates which <a href="callingconv">calling
1594 convention</a> the call should use. If none is specified, the call defaults
1595 to using C calling conventions.
1597 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1598 function value being invoked. In most cases, this is a direct function
1599 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1600 an arbitrary pointer to function value.
1603 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1604 function to be invoked. </li>
1606 <li>'<tt>function args</tt>': argument list whose types match the function
1607 signature argument types. If the function signature indicates the function
1608 accepts a variable number of arguments, the extra arguments can be
1611 <li>'<tt>normal label</tt>': the label reached when the called function
1612 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1614 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1615 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1621 <p>This instruction is designed to operate as a standard '<tt><a
1622 href="#i_call">call</a></tt>' instruction in most regards. The primary
1623 difference is that it establishes an association with a label, which is used by
1624 the runtime library to unwind the stack.</p>
1626 <p>This instruction is used in languages with destructors to ensure that proper
1627 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1628 exception. Additionally, this is important for implementation of
1629 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1633 %retval = invoke i32 %Test(i32 15) to label %Continue
1634 unwind label %TestCleanup <i>; {i32}:retval set</i>
1635 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Test(i32 15) to label %Continue
1636 unwind label %TestCleanup <i>; {i32}:retval set</i>
1641 <!-- _______________________________________________________________________ -->
1643 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1644 Instruction</a> </div>
1646 <div class="doc_text">
1655 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1656 at the first callee in the dynamic call stack which used an <a
1657 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1658 primarily used to implement exception handling.</p>
1662 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1663 immediately halt. The dynamic call stack is then searched for the first <a
1664 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1665 execution continues at the "exceptional" destination block specified by the
1666 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1667 dynamic call chain, undefined behavior results.</p>
1670 <!-- _______________________________________________________________________ -->
1672 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1673 Instruction</a> </div>
1675 <div class="doc_text">
1684 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1685 instruction is used to inform the optimizer that a particular portion of the
1686 code is not reachable. This can be used to indicate that the code after a
1687 no-return function cannot be reached, and other facts.</p>
1691 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1696 <!-- ======================================================================= -->
1697 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1698 <div class="doc_text">
1699 <p>Binary operators are used to do most of the computation in a
1700 program. They require two operands, execute an operation on them, and
1701 produce a single value. The operands might represent
1702 multiple data, as is the case with the <a href="#t_packed">packed</a> data type.
1703 The result value of a binary operator is not
1704 necessarily the same type as its operands.</p>
1705 <p>There are several different binary operators:</p>
1707 <!-- _______________________________________________________________________ -->
1708 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1709 Instruction</a> </div>
1710 <div class="doc_text">
1712 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1715 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1717 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1718 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1719 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1720 Both arguments must have identical types.</p>
1722 <p>The value produced is the integer or floating point sum of the two
1725 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
1728 <!-- _______________________________________________________________________ -->
1729 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1730 Instruction</a> </div>
1731 <div class="doc_text">
1733 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1736 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1738 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1739 instruction present in most other intermediate representations.</p>
1741 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1742 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1744 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1745 Both arguments must have identical types.</p>
1747 <p>The value produced is the integer or floating point difference of
1748 the two operands.</p>
1750 <pre> <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
1751 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
1754 <!-- _______________________________________________________________________ -->
1755 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1756 Instruction</a> </div>
1757 <div class="doc_text">
1759 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1762 <p>The '<tt>mul</tt>' instruction returns the product of its two
1765 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1766 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1768 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1769 Both arguments must have identical types.</p>
1771 <p>The value produced is the integer or floating point product of the
1773 <p>There is no signed vs unsigned multiplication. The appropriate
1774 action is taken based on the type of the operand.</p>
1776 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
1779 <!-- _______________________________________________________________________ -->
1780 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
1782 <div class="doc_text">
1784 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1787 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
1790 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
1791 <a href="#t_integer">integer</a> values. Both arguments must have identical
1792 types. This instruction can also take <a href="#t_packed">packed</a> versions
1793 of the values in which case the elements must be integers.</p>
1795 <p>The value produced is the unsigned integer quotient of the two operands. This
1796 instruction always performs an unsigned division operation, regardless of
1797 whether the arguments are unsigned or not.</p>
1799 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
1802 <!-- _______________________________________________________________________ -->
1803 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
1805 <div class="doc_text">
1807 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1810 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
1813 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
1814 <a href="#t_integer">integer</a> values. Both arguments must have identical
1815 types. This instruction can also take <a href="#t_packed">packed</a> versions
1816 of the values in which case the elements must be integers.</p>
1818 <p>The value produced is the signed integer quotient of the two operands. This
1819 instruction always performs a signed division operation, regardless of whether
1820 the arguments are signed or not.</p>
1822 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
1825 <!-- _______________________________________________________________________ -->
1826 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
1827 Instruction</a> </div>
1828 <div class="doc_text">
1830 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1833 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
1836 <p>The two arguments to the '<tt>div</tt>' instruction must be
1837 <a href="#t_floating">floating point</a> values. Both arguments must have
1838 identical types. This instruction can also take <a href="#t_packed">packed</a>
1839 versions of the values in which case the elements must be floating point.</p>
1841 <p>The value produced is the floating point quotient of the two operands.</p>
1843 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
1846 <!-- _______________________________________________________________________ -->
1847 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
1849 <div class="doc_text">
1851 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1854 <p>The '<tt>urem</tt>' instruction returns the remainder from the
1855 unsigned division of its two arguments.</p>
1857 <p>The two arguments to the '<tt>urem</tt>' instruction must be
1858 <a href="#t_integer">integer</a> values. Both arguments must have identical
1861 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
1862 This instruction always performs an unsigned division to get the remainder,
1863 regardless of whether the arguments are unsigned or not.</p>
1865 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
1869 <!-- _______________________________________________________________________ -->
1870 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
1871 Instruction</a> </div>
1872 <div class="doc_text">
1874 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1877 <p>The '<tt>srem</tt>' instruction returns the remainder from the
1878 signed division of its two operands.</p>
1880 <p>The two arguments to the '<tt>srem</tt>' instruction must be
1881 <a href="#t_integer">integer</a> values. Both arguments must have identical
1884 <p>This instruction returns the <i>remainder</i> of a division (where the result
1885 has the same sign as the divisor), not the <i>modulus</i> (where the
1886 result has the same sign as the dividend) of a value. For more
1887 information about the difference, see <a
1888 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
1891 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
1895 <!-- _______________________________________________________________________ -->
1896 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
1897 Instruction</a> </div>
1898 <div class="doc_text">
1900 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1903 <p>The '<tt>frem</tt>' instruction returns the remainder from the
1904 division of its two operands.</p>
1906 <p>The two arguments to the '<tt>frem</tt>' instruction must be
1907 <a href="#t_floating">floating point</a> values. Both arguments must have
1908 identical types.</p>
1910 <p>This instruction returns the <i>remainder</i> of a division.</p>
1912 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
1916 <!-- ======================================================================= -->
1917 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
1918 Operations</a> </div>
1919 <div class="doc_text">
1920 <p>Bitwise binary operators are used to do various forms of
1921 bit-twiddling in a program. They are generally very efficient
1922 instructions and can commonly be strength reduced from other
1923 instructions. They require two operands, execute an operation on them,
1924 and produce a single value. The resulting value of the bitwise binary
1925 operators is always the same type as its first operand.</p>
1927 <!-- _______________________________________________________________________ -->
1928 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
1929 Instruction</a> </div>
1930 <div class="doc_text">
1932 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1935 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1936 its two operands.</p>
1938 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1939 href="#t_integral">integral</a> values. Both arguments must have
1940 identical types.</p>
1942 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1944 <div style="align: center">
1945 <table border="1" cellspacing="0" cellpadding="4">
1976 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
1977 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
1978 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
1981 <!-- _______________________________________________________________________ -->
1982 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
1983 <div class="doc_text">
1985 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1988 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1989 or of its two operands.</p>
1991 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1992 href="#t_integral">integral</a> values. Both arguments must have
1993 identical types.</p>
1995 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1997 <div style="align: center">
1998 <table border="1" cellspacing="0" cellpadding="4">
2029 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2030 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2031 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2034 <!-- _______________________________________________________________________ -->
2035 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2036 Instruction</a> </div>
2037 <div class="doc_text">
2039 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2042 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2043 or of its two operands. The <tt>xor</tt> is used to implement the
2044 "one's complement" operation, which is the "~" operator in C.</p>
2046 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2047 href="#t_integral">integral</a> values. Both arguments must have
2048 identical types.</p>
2050 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2052 <div style="align: center">
2053 <table border="1" cellspacing="0" cellpadding="4">
2085 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2086 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2087 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2088 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2091 <!-- _______________________________________________________________________ -->
2092 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2093 Instruction</a> </div>
2094 <div class="doc_text">
2096 <pre> <result> = shl <ty> <var1>, i8 <var2> <i>; yields {ty}:result</i>
2099 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2100 the left a specified number of bits.</p>
2102 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
2103 href="#t_integer">integer</a> type. The second argument must be an '<tt>i8</tt>'
2106 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
2108 <pre> <result> = shl i32 4, i8 %var <i>; yields {i32}:result = 4 << %var</i>
2109 <result> = shl i32 4, i8 2 <i>; yields {i32}:result = 16</i>
2110 <result> = shl i32 1, i8 10 <i>; yields {i32}:result = 1024</i>
2113 <!-- _______________________________________________________________________ -->
2114 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2115 Instruction</a> </div>
2116 <div class="doc_text">
2118 <pre> <result> = lshr <ty> <var1>, i8 <var2> <i>; yields {ty}:result</i>
2122 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2123 operand shifted to the right a specified number of bits.</p>
2126 <p>The first argument to the '<tt>lshr</tt>' instruction must be an <a
2127 href="#t_integer">integer</a> type. The second argument must be an '<tt>i8</tt>' type.</p>
2130 <p>This instruction always performs a logical shift right operation, regardless
2131 of whether the arguments are unsigned or not. The <tt>var2</tt> most significant
2132 bits will be filled with zero bits after the shift.</p>
2136 <result> = lshr i32 4, i8 1 <i>; yields {i32}:result = 2</i>
2137 <result> = lshr i32 4, i8 2 <i>; yields {i32}:result = 1</i>
2138 <result> = lshr i8 4, i8 3 <i>; yields {i8 }:result = 0</i>
2139 <result> = lshr i8 -2, i8 1 <i>; yields {i8 }:result = 0x7FFFFFFF </i>
2143 <!-- ======================================================================= -->
2144 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2145 Instruction</a> </div>
2146 <div class="doc_text">
2149 <pre> <result> = ashr <ty> <var1>, i8 <var2> <i>; yields {ty}:result</i>
2153 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2154 operand shifted to the right a specified number of bits.</p>
2157 <p>The first argument to the '<tt>ashr</tt>' instruction must be an
2158 <a href="#t_integer">integer</a> type. The second argument must be an
2159 '<tt>i8</tt>' type.</p>
2162 <p>This instruction always performs an arithmetic shift right operation,
2163 regardless of whether the arguments are signed or not. The <tt>var2</tt> most
2164 significant bits will be filled with the sign bit of <tt>var1</tt>.</p>
2168 <result> = ashr i32 4, i8 1 <i>; yields {i32}:result = 2</i>
2169 <result> = ashr i32 4, i8 2 <i>; yields {i32}:result = 1</i>
2170 <result> = ashr i8 4, i8 3 <i>; yields {i8}:result = 0</i>
2171 <result> = ashr i8 -2, i8 1 <i>; yields {i8 }:result = -1</i>
2175 <!-- ======================================================================= -->
2176 <div class="doc_subsection">
2177 <a name="vectorops">Vector Operations</a>
2180 <div class="doc_text">
2182 <p>LLVM supports several instructions to represent vector operations in a
2183 target-independent manner. This instructions cover the element-access and
2184 vector-specific operations needed to process vectors effectively. While LLVM
2185 does directly support these vector operations, many sophisticated algorithms
2186 will want to use target-specific intrinsics to take full advantage of a specific
2191 <!-- _______________________________________________________________________ -->
2192 <div class="doc_subsubsection">
2193 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2196 <div class="doc_text">
2201 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2207 The '<tt>extractelement</tt>' instruction extracts a single scalar
2208 element from a packed vector at a specified index.
2215 The first operand of an '<tt>extractelement</tt>' instruction is a
2216 value of <a href="#t_packed">packed</a> type. The second operand is
2217 an index indicating the position from which to extract the element.
2218 The index may be a variable.</p>
2223 The result is a scalar of the same type as the element type of
2224 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2225 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2226 results are undefined.
2232 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2237 <!-- _______________________________________________________________________ -->
2238 <div class="doc_subsubsection">
2239 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2242 <div class="doc_text">
2247 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2253 The '<tt>insertelement</tt>' instruction inserts a scalar
2254 element into a packed vector at a specified index.
2261 The first operand of an '<tt>insertelement</tt>' instruction is a
2262 value of <a href="#t_packed">packed</a> type. The second operand is a
2263 scalar value whose type must equal the element type of the first
2264 operand. The third operand is an index indicating the position at
2265 which to insert the value. The index may be a variable.</p>
2270 The result is a packed vector of the same type as <tt>val</tt>. Its
2271 element values are those of <tt>val</tt> except at position
2272 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2273 exceeds the length of <tt>val</tt>, the results are undefined.
2279 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2283 <!-- _______________________________________________________________________ -->
2284 <div class="doc_subsubsection">
2285 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2288 <div class="doc_text">
2293 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2299 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2300 from two input vectors, returning a vector of the same type.
2306 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2307 with types that match each other and types that match the result of the
2308 instruction. The third argument is a shuffle mask, which has the same number
2309 of elements as the other vector type, but whose element type is always 'i32'.
2313 The shuffle mask operand is required to be a constant vector with either
2314 constant integer or undef values.
2320 The elements of the two input vectors are numbered from left to right across
2321 both of the vectors. The shuffle mask operand specifies, for each element of
2322 the result vector, which element of the two input registers the result element
2323 gets. The element selector may be undef (meaning "don't care") and the second
2324 operand may be undef if performing a shuffle from only one vector.
2330 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2331 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2332 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2333 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2338 <!-- ======================================================================= -->
2339 <div class="doc_subsection">
2340 <a name="memoryops">Memory Access and Addressing Operations</a>
2343 <div class="doc_text">
2345 <p>A key design point of an SSA-based representation is how it
2346 represents memory. In LLVM, no memory locations are in SSA form, which
2347 makes things very simple. This section describes how to read, write,
2348 allocate, and free memory in LLVM.</p>
2352 <!-- _______________________________________________________________________ -->
2353 <div class="doc_subsubsection">
2354 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2357 <div class="doc_text">
2362 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2367 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2368 heap and returns a pointer to it.</p>
2372 <p>The '<tt>malloc</tt>' instruction allocates
2373 <tt>sizeof(<type>)*NumElements</tt>
2374 bytes of memory from the operating system and returns a pointer of the
2375 appropriate type to the program. If "NumElements" is specified, it is the
2376 number of elements allocated. If an alignment is specified, the value result
2377 of the allocation is guaranteed to be aligned to at least that boundary. If
2378 not specified, or if zero, the target can choose to align the allocation on any
2379 convenient boundary.</p>
2381 <p>'<tt>type</tt>' must be a sized type.</p>
2385 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2386 a pointer is returned.</p>
2391 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2393 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2394 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2395 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2396 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2397 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2401 <!-- _______________________________________________________________________ -->
2402 <div class="doc_subsubsection">
2403 <a name="i_free">'<tt>free</tt>' Instruction</a>
2406 <div class="doc_text">
2411 free <type> <value> <i>; yields {void}</i>
2416 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2417 memory heap to be reallocated in the future.</p>
2421 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2422 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2427 <p>Access to the memory pointed to by the pointer is no longer defined
2428 after this instruction executes.</p>
2433 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2434 free [4 x i8]* %array
2438 <!-- _______________________________________________________________________ -->
2439 <div class="doc_subsubsection">
2440 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2443 <div class="doc_text">
2448 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2453 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
2454 stack frame of the procedure that is live until the current function
2455 returns to its caller.</p>
2459 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2460 bytes of memory on the runtime stack, returning a pointer of the
2461 appropriate type to the program. If "NumElements" is specified, it is the
2462 number of elements allocated. If an alignment is specified, the value result
2463 of the allocation is guaranteed to be aligned to at least that boundary. If
2464 not specified, or if zero, the target can choose to align the allocation on any
2465 convenient boundary.</p>
2467 <p>'<tt>type</tt>' may be any sized type.</p>
2471 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2472 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2473 instruction is commonly used to represent automatic variables that must
2474 have an address available. When the function returns (either with the <tt><a
2475 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2476 instructions), the memory is reclaimed.</p>
2481 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2482 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2483 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2484 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2488 <!-- _______________________________________________________________________ -->
2489 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2490 Instruction</a> </div>
2491 <div class="doc_text">
2493 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
2495 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2497 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2498 address from which to load. The pointer must point to a <a
2499 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2500 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2501 the number or order of execution of this <tt>load</tt> with other
2502 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2505 <p>The location of memory pointed to is loaded.</p>
2507 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2509 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2510 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2513 <!-- _______________________________________________________________________ -->
2514 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2515 Instruction</a> </div>
2516 <div class="doc_text">
2518 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2519 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2522 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2524 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2525 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
2526 operand must be a pointer to the type of the '<tt><value></tt>'
2527 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2528 optimizer is not allowed to modify the number or order of execution of
2529 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2530 href="#i_store">store</a></tt> instructions.</p>
2532 <p>The contents of memory are updated to contain '<tt><value></tt>'
2533 at the location specified by the '<tt><pointer></tt>' operand.</p>
2535 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2537 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2538 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2542 <!-- _______________________________________________________________________ -->
2543 <div class="doc_subsubsection">
2544 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2547 <div class="doc_text">
2550 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2556 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2557 subelement of an aggregate data structure.</p>
2561 <p>This instruction takes a list of integer operands that indicate what
2562 elements of the aggregate object to index to. The actual types of the arguments
2563 provided depend on the type of the first pointer argument. The
2564 '<tt>getelementptr</tt>' instruction is used to index down through the type
2565 levels of a structure or to a specific index in an array. When indexing into a
2566 structure, only <tt>i32</tt> integer constants are allowed. When indexing
2567 into an array or pointer, only integers of 32 or 64 bits are allowed, and will
2568 be sign extended to 64-bit values.</p>
2570 <p>For example, let's consider a C code fragment and how it gets
2571 compiled to LLVM:</p>
2585 define i32 *foo(struct ST *s) {
2586 return &s[1].Z.B[5][13];
2590 <p>The LLVM code generated by the GCC frontend is:</p>
2593 %RT = type { i8 , [10 x [20 x i32]], i8 }
2594 %ST = type { i32, double, %RT }
2598 define i32* %foo(%ST* %s) {
2600 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
2607 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2608 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2609 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
2610 <a href="#t_integer">integer</a> type but the value will always be sign extended
2611 to 64-bits. <a href="#t_struct">Structure</a> types, require <tt>i32</tt>
2612 <b>constants</b>.</p>
2614 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2615 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
2616 }</tt>' type, a structure. The second index indexes into the third element of
2617 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
2618 i8 }</tt>' type, another structure. The third index indexes into the second
2619 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
2620 array. The two dimensions of the array are subscripted into, yielding an
2621 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2622 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
2624 <p>Note that it is perfectly legal to index partially through a
2625 structure, returning a pointer to an inner element. Because of this,
2626 the LLVM code for the given testcase is equivalent to:</p>
2629 define i32* %foo(%ST* %s) {
2630 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
2631 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
2632 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
2633 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
2634 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
2639 <p>Note that it is undefined to access an array out of bounds: array and
2640 pointer indexes must always be within the defined bounds of the array type.
2641 The one exception for this rules is zero length arrays. These arrays are
2642 defined to be accessible as variable length arrays, which requires access
2643 beyond the zero'th element.</p>
2645 <p>The getelementptr instruction is often confusing. For some more insight
2646 into how it works, see <a href="GetElementPtr.html">the getelementptr
2652 <i>; yields [12 x i8]*:aptr</i>
2653 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
2657 <!-- ======================================================================= -->
2658 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
2660 <div class="doc_text">
2661 <p>The instructions in this category are the conversion instructions (casting)
2662 which all take a single operand and a type. They perform various bit conversions
2666 <!-- _______________________________________________________________________ -->
2667 <div class="doc_subsubsection">
2668 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
2670 <div class="doc_text">
2674 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
2679 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
2684 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
2685 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
2686 and type of the result, which must be an <a href="#t_integral">integral</a>
2687 type. The bit size of <tt>value</tt> must be larger than the bit size of
2688 <tt>ty2</tt>. Equal sized types are not allowed.</p>
2692 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
2693 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
2694 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
2695 It will always truncate bits.</p>
2699 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
2700 %Y = trunc i32 123 to bool <i>; yields bool:true</i>
2704 <!-- _______________________________________________________________________ -->
2705 <div class="doc_subsubsection">
2706 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
2708 <div class="doc_text">
2712 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
2716 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
2721 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
2722 <a href="#t_integral">integral</a> type, and a type to cast it to, which must
2723 also be of <a href="#t_integral">integral</a> type. The bit size of the
2724 <tt>value</tt> must be smaller than the bit size of the destination type,
2728 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
2729 bits until it reaches the size of the destination type, <tt>ty2</tt>. When the
2730 the operand and the type are the same size, no bit filling is done and the
2731 cast is considered a <i>no-op cast</i> because no bits change (only the type
2734 <p>When zero extending from bool, the result will alwasy be either 0 or 1.</p>
2738 %X = zext i32 257 to i64 <i>; yields i64:257</i>
2739 %Y = zext bool true to i32 <i>; yields i32:1</i>
2743 <!-- _______________________________________________________________________ -->
2744 <div class="doc_subsubsection">
2745 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
2747 <div class="doc_text">
2751 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
2755 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
2759 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
2760 <a href="#t_integral">integral</a> type, and a type to cast it to, which must
2761 also be of <a href="#t_integral">integral</a> type. The bit size of the
2762 <tt>value</tt> must be smaller than the bit size of the destination type,
2767 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
2768 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
2769 the type <tt>ty2</tt>. When the the operand and the type are the same size,
2770 no bit filling is done and the cast is considered a <i>no-op cast</i> because
2771 no bits change (only the type changes).</p>
2773 <p>When sign extending from bool, the extension always results in -1 or 0.</p>
2777 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
2778 %Y = sext bool true to i32 <i>; yields i32:-1</i>
2782 <!-- _______________________________________________________________________ -->
2783 <div class="doc_subsubsection">
2784 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
2787 <div class="doc_text">
2792 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
2796 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
2801 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
2802 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
2803 cast it to. The size of <tt>value</tt> must be larger than the size of
2804 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
2805 <i>no-op cast</i>.</p>
2808 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
2809 <a href="#t_floating">floating point</a> type to a smaller
2810 <a href="#t_floating">floating point</a> type. If the value cannot fit within
2811 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
2815 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
2816 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
2820 <!-- _______________________________________________________________________ -->
2821 <div class="doc_subsubsection">
2822 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
2824 <div class="doc_text">
2828 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
2832 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
2833 floating point value.</p>
2836 <p>The '<tt>fpext</tt>' instruction takes a
2837 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
2838 and a <a href="#t_floating">floating point</a> type to cast it to. The source
2839 type must be smaller than the destination type.</p>
2842 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
2843 <a href="t_floating">floating point</a> type to a larger
2844 <a href="t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
2845 used to make a <i>no-op cast</i> because it always changes bits. Use
2846 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
2850 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
2851 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
2855 <!-- _______________________________________________________________________ -->
2856 <div class="doc_subsubsection">
2857 <a name="i_fp2uint">'<tt>fptoui .. to</tt>' Instruction</a>
2859 <div class="doc_text">
2863 <result> = fp2uint <ty> <value> to <ty2> <i>; yields ty2</i>
2867 <p>The '<tt>fp2uint</tt>' converts a floating point <tt>value</tt> to its
2868 unsigned integer equivalent of type <tt>ty2</tt>.
2872 <p>The '<tt>fp2uint</tt>' instruction takes a value to cast, which must be a
2873 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
2874 must be an <a href="#t_integral">integral</a> type.</p>
2877 <p> The '<tt>fp2uint</tt>' instruction converts its
2878 <a href="#t_floating">floating point</a> operand into the nearest (rounding
2879 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
2880 the results are undefined.</p>
2882 <p>When converting to bool, the conversion is done as a comparison against
2883 zero. If the <tt>value</tt> was zero, the bool result will be <tt>false</tt>.
2884 If the <tt>value</tt> was non-zero, the bool result will be <tt>true</tt>.</p>
2888 %X = fp2uint double 123.0 to i32 <i>; yields i32:123</i>
2889 %Y = fp2uint float 1.0E+300 to bool <i>; yields bool:true</i>
2890 %X = fp2uint float 1.04E+17 to i8 <i>; yields undefined:1</i>
2894 <!-- _______________________________________________________________________ -->
2895 <div class="doc_subsubsection">
2896 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
2898 <div class="doc_text">
2902 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
2906 <p>The '<tt>fptosi</tt>' instruction converts
2907 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
2912 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
2913 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
2914 must also be an <a href="#t_integral">integral</a> type.</p>
2917 <p>The '<tt>fptosi</tt>' instruction converts its
2918 <a href="#t_floating">floating point</a> operand into the nearest (rounding
2919 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
2920 the results are undefined.</p>
2922 <p>When converting to bool, the conversion is done as a comparison against
2923 zero. If the <tt>value</tt> was zero, the bool result will be <tt>false</tt>.
2924 If the <tt>value</tt> was non-zero, the bool result will be <tt>true</tt>.</p>
2928 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
2929 %Y = fptosi float 1.0E-247 to bool <i>; yields bool:true</i>
2930 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
2934 <!-- _______________________________________________________________________ -->
2935 <div class="doc_subsubsection">
2936 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
2938 <div class="doc_text">
2942 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
2946 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
2947 integer and converts that value to the <tt>ty2</tt> type.</p>
2951 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be an
2952 <a href="#t_integral">integral</a> value, and a type to cast it to, which must
2953 be a <a href="#t_floating">floating point</a> type.</p>
2956 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
2957 integer quantity and converts it to the corresponding floating point value. If
2958 the value cannot fit in the floating point value, the results are undefined.</p>
2963 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
2964 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
2968 <!-- _______________________________________________________________________ -->
2969 <div class="doc_subsubsection">
2970 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
2972 <div class="doc_text">
2976 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
2980 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
2981 integer and converts that value to the <tt>ty2</tt> type.</p>
2984 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be an
2985 <a href="#t_integral">integral</a> value, and a type to cast it to, which must be
2986 a <a href="#t_floating">floating point</a> type.</p>
2989 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
2990 integer quantity and converts it to the corresponding floating point value. If
2991 the value cannot fit in the floating point value, the results are undefined.</p>
2995 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
2996 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3000 <!-- _______________________________________________________________________ -->
3001 <div class="doc_subsubsection">
3002 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3004 <div class="doc_text">
3008 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3012 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3013 the integer type <tt>ty2</tt>.</p>
3016 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3017 must be a <a href="t_pointer">pointer</a> value, and a type to cast it to
3018 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3021 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3022 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3023 truncating or zero extending that value to the size of the integer type. If
3024 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3025 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3026 are the same size, then nothing is done (<i>no-op cast</i>).</p>
3030 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit</i>
3031 %Y = ptrtoint i32* %x to i64 <i>; yields zero extend on 32-bit</i>
3035 <!-- _______________________________________________________________________ -->
3036 <div class="doc_subsubsection">
3037 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3039 <div class="doc_text">
3043 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3047 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3048 a pointer type, <tt>ty2</tt>.</p>
3051 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="i_integer">integer</a>
3052 value to cast, and a type to cast it to, which must be a
3053 <a href="#t_pointer">pointer</a> type. </tt>
3056 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3057 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3058 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3059 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3060 the size of a pointer then a zero extension is done. If they are the same size,
3061 nothing is done (<i>no-op cast</i>).</p>
3065 %X = inttoptr i32 255 to i32* <i>; yields zero extend on 64-bit</i>
3066 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit </i>
3067 %Y = inttoptr i16 0 to i32* <i>; yields zero extend on 32-bit</i>
3071 <!-- _______________________________________________________________________ -->
3072 <div class="doc_subsubsection">
3073 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3075 <div class="doc_text">
3079 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3083 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3084 <tt>ty2</tt> without changing any bits.</p>
3087 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3088 a first class value, and a type to cast it to, which must also be a <a
3089 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3090 and the destination type, <tt>ty2</tt>, must be identical.</p>
3093 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3094 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3095 this conversion. The conversion is done as if the <tt>value</tt> had been
3096 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3097 converted to other pointer types with this instruction. To convert pointers to
3098 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3099 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3103 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3104 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3105 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3109 <!-- ======================================================================= -->
3110 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3111 <div class="doc_text">
3112 <p>The instructions in this category are the "miscellaneous"
3113 instructions, which defy better classification.</p>
3116 <!-- _______________________________________________________________________ -->
3117 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3119 <div class="doc_text">
3121 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {bool}:result</i>
3124 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3125 of its two integer operands.</p>
3127 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3128 the condition code which indicates the kind of comparison to perform. It is not
3129 a value, just a keyword. The possibilities for the condition code are:
3131 <li><tt>eq</tt>: equal</li>
3132 <li><tt>ne</tt>: not equal </li>
3133 <li><tt>ugt</tt>: unsigned greater than</li>
3134 <li><tt>uge</tt>: unsigned greater or equal</li>
3135 <li><tt>ult</tt>: unsigned less than</li>
3136 <li><tt>ule</tt>: unsigned less or equal</li>
3137 <li><tt>sgt</tt>: signed greater than</li>
3138 <li><tt>sge</tt>: signed greater or equal</li>
3139 <li><tt>slt</tt>: signed less than</li>
3140 <li><tt>sle</tt>: signed less or equal</li>
3142 <p>The remaining two arguments must be <a href="#t_integral">integral</a> or
3143 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3145 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3146 the condition code given as <tt>cond</tt>. The comparison performed always
3147 yields a <a href="#t_bool">bool</a> result, as follows:
3149 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3150 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3152 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3153 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3154 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3155 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3156 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3157 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3158 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3159 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3160 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3161 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3162 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3163 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3164 <li><tt>sge</tt>: interprets the operands as signed values and yields
3165 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3166 <li><tt>slt</tt>: interprets the operands as signed values and yields
3167 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3168 <li><tt>sle</tt>: interprets the operands as signed values and yields
3169 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3172 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3173 values are treated as integers and then compared.</p>
3174 <p>If the operands are <a href="#t_packed">packed</a> typed, the elements of
3175 the vector are compared in turn and the predicate must hold for all
3179 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3180 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3181 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3182 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3183 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3184 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3188 <!-- _______________________________________________________________________ -->
3189 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3191 <div class="doc_text">
3193 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {bool}:result</i>
3196 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3197 of its floating point operands.</p>
3199 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3200 the condition code which indicates the kind of comparison to perform. It is not
3201 a value, just a keyword. The possibilities for the condition code are:
3203 <li><tt>false</tt>: no comparison, always returns false</li>
3204 <li><tt>oeq</tt>: ordered and equal</li>
3205 <li><tt>ogt</tt>: ordered and greater than </li>
3206 <li><tt>oge</tt>: ordered and greater than or equal</li>
3207 <li><tt>olt</tt>: ordered and less than </li>
3208 <li><tt>ole</tt>: ordered and less than or equal</li>
3209 <li><tt>one</tt>: ordered and not equal</li>
3210 <li><tt>ord</tt>: ordered (no nans)</li>
3211 <li><tt>ueq</tt>: unordered or equal</li>
3212 <li><tt>ugt</tt>: unordered or greater than </li>
3213 <li><tt>uge</tt>: unordered or greater than or equal</li>
3214 <li><tt>ult</tt>: unordered or less than </li>
3215 <li><tt>ule</tt>: unordered or less than or equal</li>
3216 <li><tt>une</tt>: unordered or not equal</li>
3217 <li><tt>uno</tt>: unordered (either nans)</li>
3218 <li><tt>true</tt>: no comparison, always returns true</li>
3220 <p>In the preceding, <i>ordered</i> means that neither operand is a QNAN while
3221 <i>unordered</i> means that either operand may be a QNAN.</p>
3222 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3223 <a href="#t_floating">floating point</a> typed. They must have identical
3225 <p>In the foregoing, <i>ordered</i> means that neither operand is a QNAN and
3226 <i>unordered</i> means that either operand is a QNAN.</p>
3228 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3229 the condition code given as <tt>cond</tt>. The comparison performed always
3230 yields a <a href="#t_bool">bool</a> result, as follows:
3232 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3233 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3234 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3235 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3236 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3237 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3238 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3239 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3240 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3241 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3242 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3243 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3244 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3245 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3246 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3247 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3248 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3249 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3250 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3251 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3252 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3253 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3254 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3255 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3256 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3257 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3258 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3259 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3261 <p>If the operands are <a href="#t_packed">packed</a> typed, the elements of
3262 the vector are compared in turn and the predicate must hold for all elements.
3266 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3267 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3268 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3269 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3273 <!-- _______________________________________________________________________ -->
3274 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3275 Instruction</a> </div>
3276 <div class="doc_text">
3278 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3280 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3281 the SSA graph representing the function.</p>
3283 <p>The type of the incoming values are specified with the first type
3284 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3285 as arguments, with one pair for each predecessor basic block of the
3286 current block. Only values of <a href="#t_firstclass">first class</a>
3287 type may be used as the value arguments to the PHI node. Only labels
3288 may be used as the label arguments.</p>
3289 <p>There must be no non-phi instructions between the start of a basic
3290 block and the PHI instructions: i.e. PHI instructions must be first in
3293 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
3294 value specified by the parameter, depending on which basic block we
3295 came from in the last <a href="#terminators">terminator</a> instruction.</p>
3297 <pre>Loop: ; Infinite loop that counts from 0 on up...<br> %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]<br> %nextindvar = add i32 %indvar, 1<br> br label %Loop<br></pre>
3300 <!-- _______________________________________________________________________ -->
3301 <div class="doc_subsubsection">
3302 <a name="i_select">'<tt>select</tt>' Instruction</a>
3305 <div class="doc_text">
3310 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3316 The '<tt>select</tt>' instruction is used to choose one value based on a
3317 condition, without branching.
3324 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.
3330 If the boolean condition evaluates to true, the instruction returns the first
3331 value argument; otherwise, it returns the second value argument.
3337 %X = select bool true, i8 17, i8 42 <i>; yields i8:17</i>
3342 <!-- _______________________________________________________________________ -->
3343 <div class="doc_subsubsection">
3344 <a name="i_call">'<tt>call</tt>' Instruction</a>
3347 <div class="doc_text">
3351 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
3356 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3360 <p>This instruction requires several arguments:</p>
3364 <p>The optional "tail" marker indicates whether the callee function accesses
3365 any allocas or varargs in the caller. If the "tail" marker is present, the
3366 function call is eligible for tail call optimization. Note that calls may
3367 be marked "tail" even if they do not occur before a <a
3368 href="#i_ret"><tt>ret</tt></a> instruction.
3371 <p>The optional "cconv" marker indicates which <a href="callingconv">calling
3372 convention</a> the call should use. If none is specified, the call defaults
3373 to using C calling conventions.
3376 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
3377 being invoked. The argument types must match the types implied by this
3378 signature. This type can be omitted if the function is not varargs and
3379 if the function type does not return a pointer to a function.</p>
3382 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3383 be invoked. In most cases, this is a direct function invocation, but
3384 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3385 to function value.</p>
3388 <p>'<tt>function args</tt>': argument list whose types match the
3389 function signature argument types. All arguments must be of
3390 <a href="#t_firstclass">first class</a> type. If the function signature
3391 indicates the function accepts a variable number of arguments, the extra
3392 arguments can be specified.</p>
3398 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3399 transfer to a specified function, with its incoming arguments bound to
3400 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3401 instruction in the called function, control flow continues with the
3402 instruction after the function call, and the return value of the
3403 function is bound to the result argument. This is a simpler case of
3404 the <a href="#i_invoke">invoke</a> instruction.</p>
3409 %retval = call i32 %test(i32 %argc)
3410 call i32(i8 *, ...) *%printf(i8 * %msg, i32 12, i8 42);
3411 %X = tail call i32 %foo()
3412 %Y = tail call <a href="#callingconv">fastcc</a> i32 %foo()
3417 <!-- _______________________________________________________________________ -->
3418 <div class="doc_subsubsection">
3419 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3422 <div class="doc_text">
3427 <resultval> = va_arg <va_list*> <arglist>, <argty>
3432 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3433 the "variable argument" area of a function call. It is used to implement the
3434 <tt>va_arg</tt> macro in C.</p>
3438 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3439 the argument. It returns a value of the specified argument type and
3440 increments the <tt>va_list</tt> to point to the next argument. Again, the
3441 actual type of <tt>va_list</tt> is target specific.</p>
3445 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3446 type from the specified <tt>va_list</tt> and causes the
3447 <tt>va_list</tt> to point to the next argument. For more information,
3448 see the variable argument handling <a href="#int_varargs">Intrinsic
3451 <p>It is legal for this instruction to be called in a function which does not
3452 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3455 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3456 href="#intrinsics">intrinsic function</a> because it takes a type as an
3461 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3465 <!-- *********************************************************************** -->
3466 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3467 <!-- *********************************************************************** -->
3469 <div class="doc_text">
3471 <p>LLVM supports the notion of an "intrinsic function". These functions have
3472 well known names and semantics and are required to follow certain
3473 restrictions. Overall, these instructions represent an extension mechanism for
3474 the LLVM language that does not require changing all of the transformations in
3475 LLVM to add to the language (or the bytecode reader/writer, the parser,
3478 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3479 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
3480 this. Intrinsic functions must always be external functions: you cannot define
3481 the body of intrinsic functions. Intrinsic functions may only be used in call
3482 or invoke instructions: it is illegal to take the address of an intrinsic
3483 function. Additionally, because intrinsic functions are part of the LLVM
3484 language, it is required that they all be documented here if any are added.</p>
3487 <p>To learn how to add an intrinsic function, please see the <a
3488 href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3493 <!-- ======================================================================= -->
3494 <div class="doc_subsection">
3495 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3498 <div class="doc_text">
3500 <p>Variable argument support is defined in LLVM with the <a
3501 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3502 intrinsic functions. These functions are related to the similarly
3503 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3505 <p>All of these functions operate on arguments that use a
3506 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3507 language reference manual does not define what this type is, so all
3508 transformations should be prepared to handle intrinsics with any type
3511 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3512 instruction and the variable argument handling intrinsic functions are
3516 define i32 %test(i32 %X, ...) {
3517 ; Initialize variable argument processing
3519 call void %<a href="#i_va_start">llvm.va_start</a>(i8 ** %ap)
3521 ; Read a single integer argument
3522 %tmp = va_arg i8 ** %ap, i32
3524 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3526 call void %<a href="#i_va_copy">llvm.va_copy</a>(i8 ** %aq, i8 ** %ap)
3527 call void %<a href="#i_va_end">llvm.va_end</a>(i8 ** %aq)
3529 ; Stop processing of arguments.
3530 call void %<a href="#i_va_end">llvm.va_end</a>(i8 ** %ap)
3536 <!-- _______________________________________________________________________ -->
3537 <div class="doc_subsubsection">
3538 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3542 <div class="doc_text">
3544 <pre> declare void %llvm.va_start(<va_list>* <arglist>)<br></pre>
3546 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3547 <tt>*<arglist></tt> for subsequent use by <tt><a
3548 href="#i_va_arg">va_arg</a></tt>.</p>
3552 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3556 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3557 macro available in C. In a target-dependent way, it initializes the
3558 <tt>va_list</tt> element the argument points to, so that the next call to
3559 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3560 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3561 last argument of the function, the compiler can figure that out.</p>
3565 <!-- _______________________________________________________________________ -->
3566 <div class="doc_subsubsection">
3567 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
3570 <div class="doc_text">
3572 <pre> declare void %llvm.va_end(<va_list*> <arglist>)<br></pre>
3574 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
3575 which has been initialized previously with <tt><a href="#i_va_start">llvm.va_start</a></tt>
3576 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
3578 <p>The argument is a <tt>va_list</tt> to destroy.</p>
3580 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
3581 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
3582 Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and <a
3583 href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
3584 with calls to <tt>llvm.va_end</tt>.</p>
3587 <!-- _______________________________________________________________________ -->
3588 <div class="doc_subsubsection">
3589 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
3592 <div class="doc_text">
3597 declare void %llvm.va_copy(<va_list>* <destarglist>,
3598 <va_list>* <srcarglist>)
3603 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
3604 the source argument list to the destination argument list.</p>
3608 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
3609 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
3614 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
3615 available in C. In a target-dependent way, it copies the source
3616 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
3617 because the <tt><a href="i_va_begin">llvm.va_begin</a></tt> intrinsic may be
3618 arbitrarily complex and require memory allocation, for example.</p>
3622 <!-- ======================================================================= -->
3623 <div class="doc_subsection">
3624 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
3627 <div class="doc_text">
3630 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
3631 Collection</a> requires the implementation and generation of these intrinsics.
3632 These intrinsics allow identification of <a href="#i_gcroot">GC roots on the
3633 stack</a>, as well as garbage collector implementations that require <a
3634 href="#i_gcread">read</a> and <a href="#i_gcwrite">write</a> barriers.
3635 Front-ends for type-safe garbage collected languages should generate these
3636 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3637 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
3641 <!-- _______________________________________________________________________ -->
3642 <div class="doc_subsubsection">
3643 <a name="i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
3646 <div class="doc_text">
3651 declare void %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
3656 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
3657 the code generator, and allows some metadata to be associated with it.</p>
3661 <p>The first argument specifies the address of a stack object that contains the
3662 root pointer. The second pointer (which must be either a constant or a global
3663 value address) contains the meta-data to be associated with the root.</p>
3667 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
3668 location. At compile-time, the code generator generates information to allow
3669 the runtime to find the pointer at GC safe points.
3675 <!-- _______________________________________________________________________ -->
3676 <div class="doc_subsubsection">
3677 <a name="i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
3680 <div class="doc_text">
3685 declare i8 * %llvm.gcread(i8 * %ObjPtr, i8 ** %Ptr)
3690 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
3691 locations, allowing garbage collector implementations that require read
3696 <p>The second argument is the address to read from, which should be an address
3697 allocated from the garbage collector. The first object is a pointer to the
3698 start of the referenced object, if needed by the language runtime (otherwise
3703 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
3704 instruction, but may be replaced with substantially more complex code by the
3705 garbage collector runtime, as needed.</p>
3710 <!-- _______________________________________________________________________ -->
3711 <div class="doc_subsubsection">
3712 <a name="i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
3715 <div class="doc_text">
3720 declare void %llvm.gcwrite(i8 * %P1, i8 * %Obj, i8 ** %P2)
3725 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
3726 locations, allowing garbage collector implementations that require write
3727 barriers (such as generational or reference counting collectors).</p>
3731 <p>The first argument is the reference to store, the second is the start of the
3732 object to store it to, and the third is the address of the field of Obj to
3733 store to. If the runtime does not require a pointer to the object, Obj may be
3738 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
3739 instruction, but may be replaced with substantially more complex code by the
3740 garbage collector runtime, as needed.</p>
3746 <!-- ======================================================================= -->
3747 <div class="doc_subsection">
3748 <a name="int_codegen">Code Generator Intrinsics</a>
3751 <div class="doc_text">
3753 These intrinsics are provided by LLVM to expose special features that may only
3754 be implemented with code generator support.
3759 <!-- _______________________________________________________________________ -->
3760 <div class="doc_subsubsection">
3761 <a name="i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
3764 <div class="doc_text">
3768 declare i8 *%llvm.returnaddress(i32 <level>)
3774 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
3775 target-specific value indicating the return address of the current function
3776 or one of its callers.
3782 The argument to this intrinsic indicates which function to return the address
3783 for. Zero indicates the calling function, one indicates its caller, etc. The
3784 argument is <b>required</b> to be a constant integer value.
3790 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
3791 the return address of the specified call frame, or zero if it cannot be
3792 identified. The value returned by this intrinsic is likely to be incorrect or 0
3793 for arguments other than zero, so it should only be used for debugging purposes.
3797 Note that calling this intrinsic does not prevent function inlining or other
3798 aggressive transformations, so the value returned may not be that of the obvious
3799 source-language caller.
3804 <!-- _______________________________________________________________________ -->
3805 <div class="doc_subsubsection">
3806 <a name="i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
3809 <div class="doc_text">
3813 declare i8 *%llvm.frameaddress(i32 <level>)
3819 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
3820 target-specific frame pointer value for the specified stack frame.
3826 The argument to this intrinsic indicates which function to return the frame
3827 pointer for. Zero indicates the calling function, one indicates its caller,
3828 etc. The argument is <b>required</b> to be a constant integer value.
3834 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
3835 the frame address of the specified call frame, or zero if it cannot be
3836 identified. The value returned by this intrinsic is likely to be incorrect or 0
3837 for arguments other than zero, so it should only be used for debugging purposes.
3841 Note that calling this intrinsic does not prevent function inlining or other
3842 aggressive transformations, so the value returned may not be that of the obvious
3843 source-language caller.
3847 <!-- _______________________________________________________________________ -->
3848 <div class="doc_subsubsection">
3849 <a name="i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
3852 <div class="doc_text">
3856 declare i8 *%llvm.stacksave()
3862 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
3863 the function stack, for use with <a href="#i_stackrestore">
3864 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
3865 features like scoped automatic variable sized arrays in C99.
3871 This intrinsic returns a opaque pointer value that can be passed to <a
3872 href="#i_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
3873 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
3874 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
3875 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
3876 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
3877 that were allocated after the <tt>llvm.stacksave</tt> was executed.
3882 <!-- _______________________________________________________________________ -->
3883 <div class="doc_subsubsection">
3884 <a name="i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
3887 <div class="doc_text">
3891 declare void %llvm.stackrestore(i8 * %ptr)
3897 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
3898 the function stack to the state it was in when the corresponding <a
3899 href="#llvm.stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
3900 useful for implementing language features like scoped automatic variable sized
3907 See the description for <a href="#i_stacksave"><tt>llvm.stacksave</tt></a>.
3913 <!-- _______________________________________________________________________ -->
3914 <div class="doc_subsubsection">
3915 <a name="i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
3918 <div class="doc_text">
3922 declare void %llvm.prefetch(i8 * <address>,
3923 i32 <rw>, i32 <locality>)
3930 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
3931 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
3933 effect on the behavior of the program but can change its performance
3940 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
3941 determining if the fetch should be for a read (0) or write (1), and
3942 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
3943 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
3944 <tt>locality</tt> arguments must be constant integers.
3950 This intrinsic does not modify the behavior of the program. In particular,
3951 prefetches cannot trap and do not produce a value. On targets that support this
3952 intrinsic, the prefetch can provide hints to the processor cache for better
3958 <!-- _______________________________________________________________________ -->
3959 <div class="doc_subsubsection">
3960 <a name="i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
3963 <div class="doc_text">
3967 declare void %llvm.pcmarker( i32 <id> )
3974 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
3976 code to simulators and other tools. The method is target specific, but it is
3977 expected that the marker will use exported symbols to transmit the PC of the marker.
3978 The marker makes no guarantees that it will remain with any specific instruction
3979 after optimizations. It is possible that the presence of a marker will inhibit
3980 optimizations. The intended use is to be inserted after optimizations to allow
3981 correlations of simulation runs.
3987 <tt>id</tt> is a numerical id identifying the marker.
3993 This intrinsic does not modify the behavior of the program. Backends that do not
3994 support this intrinisic may ignore it.
3999 <!-- _______________________________________________________________________ -->
4000 <div class="doc_subsubsection">
4001 <a name="i_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4004 <div class="doc_text">
4008 declare i64 %llvm.readcyclecounter( )
4015 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4016 counter register (or similar low latency, high accuracy clocks) on those targets
4017 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4018 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4019 should only be used for small timings.
4025 When directly supported, reading the cycle counter should not modify any memory.
4026 Implementations are allowed to either return a application specific value or a
4027 system wide value. On backends without support, this is lowered to a constant 0.
4032 <!-- ======================================================================= -->
4033 <div class="doc_subsection">
4034 <a name="int_libc">Standard C Library Intrinsics</a>
4037 <div class="doc_text">
4039 LLVM provides intrinsics for a few important standard C library functions.
4040 These intrinsics allow source-language front-ends to pass information about the
4041 alignment of the pointer arguments to the code generator, providing opportunity
4042 for more efficient code generation.
4047 <!-- _______________________________________________________________________ -->
4048 <div class="doc_subsubsection">
4049 <a name="i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4052 <div class="doc_text">
4056 declare void %llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4057 i32 <len>, i32 <align>)
4058 declare void %llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4059 i64 <len>, i32 <align>)
4065 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4066 location to the destination location.
4070 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4071 intrinsics do not return a value, and takes an extra alignment argument.
4077 The first argument is a pointer to the destination, the second is a pointer to
4078 the source. The third argument is an integer argument
4079 specifying the number of bytes to copy, and the fourth argument is the alignment
4080 of the source and destination locations.
4084 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4085 the caller guarantees that both the source and destination pointers are aligned
4092 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4093 location to the destination location, which are not allowed to overlap. It
4094 copies "len" bytes of memory over. If the argument is known to be aligned to
4095 some boundary, this can be specified as the fourth argument, otherwise it should
4101 <!-- _______________________________________________________________________ -->
4102 <div class="doc_subsubsection">
4103 <a name="i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4106 <div class="doc_text">
4110 declare void %llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4111 i32 <len>, i32 <align>)
4112 declare void %llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4113 i64 <len>, i32 <align>)
4119 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4120 location to the destination location. It is similar to the
4121 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
4125 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4126 intrinsics do not return a value, and takes an extra alignment argument.
4132 The first argument is a pointer to the destination, the second is a pointer to
4133 the source. The third argument is an integer argument
4134 specifying the number of bytes to copy, and the fourth argument is the alignment
4135 of the source and destination locations.
4139 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4140 the caller guarantees that the source and destination pointers are aligned to
4147 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4148 location to the destination location, which may overlap. It
4149 copies "len" bytes of memory over. If the argument is known to be aligned to
4150 some boundary, this can be specified as the fourth argument, otherwise it should
4156 <!-- _______________________________________________________________________ -->
4157 <div class="doc_subsubsection">
4158 <a name="i_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4161 <div class="doc_text">
4165 declare void %llvm.memset.i32(i8 * <dest>, i8 <val>,
4166 i32 <len>, i32 <align>)
4167 declare void %llvm.memset.i64(i8 * <dest>, i8 <val>,
4168 i64 <len>, i32 <align>)
4174 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4179 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4180 does not return a value, and takes an extra alignment argument.
4186 The first argument is a pointer to the destination to fill, the second is the
4187 byte value to fill it with, the third argument is an integer
4188 argument specifying the number of bytes to fill, and the fourth argument is the
4189 known alignment of destination location.
4193 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4194 the caller guarantees that the destination pointer is aligned to that boundary.
4200 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4202 destination location. If the argument is known to be aligned to some boundary,
4203 this can be specified as the fourth argument, otherwise it should be set to 0 or
4209 <!-- _______________________________________________________________________ -->
4210 <div class="doc_subsubsection">
4211 <a name="i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a>
4214 <div class="doc_text">
4218 declare bool %llvm.isunordered.f32(float Val1, float Val2)
4219 declare bool %llvm.isunordered.f64(double Val1, double Val2)
4225 The '<tt>llvm.isunordered</tt>' intrinsics return true if either or both of the
4226 specified floating point values is a NAN.
4232 The arguments are floating point numbers of the same type.
4238 If either or both of the arguments is a SNAN or QNAN, it returns true, otherwise
4244 <!-- _______________________________________________________________________ -->
4245 <div class="doc_subsubsection">
4246 <a name="i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4249 <div class="doc_text">
4253 declare float %llvm.sqrt.f32(float %Val)
4254 declare double %llvm.sqrt.f64(double %Val)
4260 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4261 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
4262 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4263 negative numbers (which allows for better optimization).
4269 The argument and return value are floating point numbers of the same type.
4275 This function returns the sqrt of the specified operand if it is a positive
4276 floating point number.
4280 <!-- _______________________________________________________________________ -->
4281 <div class="doc_subsubsection">
4282 <a name="i_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4285 <div class="doc_text">
4289 declare float %llvm.powi.f32(float %Val, i32 %power)
4290 declare double %llvm.powi.f64(double %Val, i32 %power)
4296 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4297 specified (positive or negative) power. The order of evaluation of
4298 multiplications is not defined.
4304 The second argument is an integer power, and the first is a value to raise to
4311 This function returns the first value raised to the second power with an
4312 unspecified sequence of rounding operations.</p>
4316 <!-- ======================================================================= -->
4317 <div class="doc_subsection">
4318 <a name="int_manip">Bit Manipulation Intrinsics</a>
4321 <div class="doc_text">
4323 LLVM provides intrinsics for a few important bit manipulation operations.
4324 These allow efficient code generation for some algorithms.
4329 <!-- _______________________________________________________________________ -->
4330 <div class="doc_subsubsection">
4331 <a name="i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4334 <div class="doc_text">
4338 declare i16 %llvm.bswap.i16(i16 <id>)
4339 declare i32 %llvm.bswap.i32(i32 <id>)
4340 declare i64 %llvm.bswap.i64(i64 <id>)
4346 The '<tt>llvm.bwsap</tt>' family of intrinsics is used to byteswap a 16, 32 or
4347 64 bit quantity. These are useful for performing operations on data that is not
4348 in the target's native byte order.
4354 The <tt>llvm.bswap.16</tt> intrinsic returns an i16 value that has the high
4355 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4356 intrinsic returns an i32 value that has the four bytes of the input i32
4357 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
4358 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i64</tt>
4359 intrinsic extends this concept to 64 bits.
4364 <!-- _______________________________________________________________________ -->
4365 <div class="doc_subsubsection">
4366 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4369 <div class="doc_text">
4373 declare i8 %llvm.ctpop.i8 (i8 <src>)
4374 declare i16 %llvm.ctpop.i16(i16 <src>)
4375 declare i32 %llvm.ctpop.i32(i32 <src>)
4376 declare i64 %llvm.ctpop.i64(i64 <src>)
4382 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4389 The only argument is the value to be counted. The argument may be of any
4390 unsigned integer type. The return type must match the argument type.
4396 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4400 <!-- _______________________________________________________________________ -->
4401 <div class="doc_subsubsection">
4402 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4405 <div class="doc_text">
4409 declare i8 %llvm.ctlz.i8 (i8 <src>)
4410 declare i16 %llvm.ctlz.i16(i16 <src>)
4411 declare i32 %llvm.ctlz.i32(i32 <src>)
4412 declare i64 %llvm.ctlz.i64(i64 <src>)
4418 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4419 leading zeros in a variable.
4425 The only argument is the value to be counted. The argument may be of any
4426 unsigned integer type. The return type must match the argument type.
4432 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4433 in a variable. If the src == 0 then the result is the size in bits of the type
4434 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
4440 <!-- _______________________________________________________________________ -->
4441 <div class="doc_subsubsection">
4442 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
4445 <div class="doc_text">
4449 declare i8 %llvm.cttz.i8 (i8 <src>)
4450 declare i16 %llvm.cttz.i16(i16 <src>)
4451 declare i32 %llvm.cttz.i32(i32 <src>)
4452 declare i64 %llvm.cttz.i64(i64 <src>)
4458 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
4465 The only argument is the value to be counted. The argument may be of any
4466 unsigned integer type. The return type must match the argument type.
4472 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
4473 in a variable. If the src == 0 then the result is the size in bits of the type
4474 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
4478 <!-- ======================================================================= -->
4479 <div class="doc_subsection">
4480 <a name="int_debugger">Debugger Intrinsics</a>
4483 <div class="doc_text">
4485 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
4486 are described in the <a
4487 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
4488 Debugging</a> document.
4493 <!-- *********************************************************************** -->
4496 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
4497 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
4498 <a href="http://validator.w3.org/check/referer"><img
4499 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
4501 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
4502 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
4503 Last modified: $Date$