1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <div class="doc_title"> LLVM Language Reference Manual </div>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a></li>
24 <li><a href="#callingconv">Calling Conventions</a></li>
25 <li><a href="#globalvars">Global Variables</a></li>
26 <li><a href="#functionstructure">Functions</a></li>
27 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
30 <li><a href="#typesystem">Type System</a>
32 <li><a href="#t_primitive">Primitive Types</a>
34 <li><a href="#t_classifications">Type Classifications</a></li>
37 <li><a href="#t_derived">Derived Types</a>
39 <li><a href="#t_array">Array Type</a></li>
40 <li><a href="#t_function">Function Type</a></li>
41 <li><a href="#t_pointer">Pointer Type</a></li>
42 <li><a href="#t_struct">Structure Type</a></li>
43 <li><a href="#t_packed">Packed Type</a></li>
44 <li><a href="#t_opaque">Opaque Type</a></li>
49 <li><a href="#constants">Constants</a>
51 <li><a href="#simpleconstants">Simple Constants</a>
52 <li><a href="#aggregateconstants">Aggregate Constants</a>
53 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
54 <li><a href="#undefvalues">Undefined Values</a>
55 <li><a href="#constantexprs">Constant Expressions</a>
58 <li><a href="#othervalues">Other Values</a>
60 <li><a href="#inlineasm">Inline Assembler Expressions</a>
63 <li><a href="#instref">Instruction Reference</a>
65 <li><a href="#terminators">Terminator Instructions</a>
67 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
68 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
69 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
70 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
71 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
72 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
75 <li><a href="#binaryops">Binary Operations</a>
77 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
78 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
79 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
80 <li><a href="#i_div">'<tt>div</tt>' Instruction</a></li>
81 <li><a href="#i_rem">'<tt>rem</tt>' Instruction</a></li>
82 <li><a href="#i_setcc">'<tt>set<i>cc</i></tt>' Instructions</a></li>
85 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
87 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
88 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
89 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
90 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
91 <li><a href="#i_shr">'<tt>shr</tt>' Instruction</a></li>
94 <li><a href="#memoryops">Memory Access Operations</a>
96 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
97 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
98 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
99 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
100 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
101 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
104 <li><a href="#otherops">Other Operations</a>
106 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
107 <li><a href="#i_cast">'<tt>cast .. to</tt>' Instruction</a></li>
108 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
109 <li><a href="#i_vset">'<tt>vset</tt>' Instruction</a></li>
110 <li><a href="#i_vselect">'<tt>vselect</tt>' Instruction</a></li>
111 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
112 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
113 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
114 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
119 <li><a href="#intrinsics">Intrinsic Functions</a>
121 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
123 <li><a href="#i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
124 <li><a href="#i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
125 <li><a href="#i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
128 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
130 <li><a href="#i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
131 <li><a href="#i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
132 <li><a href="#i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
135 <li><a href="#int_codegen">Code Generator Intrinsics</a>
137 <li><a href="#i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
138 <li><a href="#i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
139 <li><a href="#i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
140 <li><a href="#i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
141 <li><a href="#i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
142 <li><a href="#i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
143 <li><a href="#i_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
146 <li><a href="#int_os">Operating System Intrinsics</a>
148 <li><a href="#i_readport">'<tt>llvm.readport</tt>' Intrinsic</a></li>
149 <li><a href="#i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a></li>
150 <li><a href="#i_readio">'<tt>llvm.readio</tt>' Intrinsic</a></li>
151 <li><a href="#i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a></li>
153 <li><a href="#int_libc">Standard C Library Intrinsics</a>
155 <li><a href="#i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a></li>
156 <li><a href="#i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a></li>
157 <li><a href="#i_memset">'<tt>llvm.memset</tt>' Intrinsic</a></li>
158 <li><a href="#i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a></li>
159 <li><a href="#i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
163 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
165 <li><a href="#i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
166 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
167 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
168 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
171 <li><a href="#int_debugger">Debugger intrinsics</a></li>
176 <div class="doc_author">
177 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
178 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
181 <!-- *********************************************************************** -->
182 <div class="doc_section"> <a name="abstract">Abstract </a></div>
183 <!-- *********************************************************************** -->
185 <div class="doc_text">
186 <p>This document is a reference manual for the LLVM assembly language.
187 LLVM is an SSA based representation that provides type safety,
188 low-level operations, flexibility, and the capability of representing
189 'all' high-level languages cleanly. It is the common code
190 representation used throughout all phases of the LLVM compilation
194 <!-- *********************************************************************** -->
195 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
196 <!-- *********************************************************************** -->
198 <div class="doc_text">
200 <p>The LLVM code representation is designed to be used in three
201 different forms: as an in-memory compiler IR, as an on-disk bytecode
202 representation (suitable for fast loading by a Just-In-Time compiler),
203 and as a human readable assembly language representation. This allows
204 LLVM to provide a powerful intermediate representation for efficient
205 compiler transformations and analysis, while providing a natural means
206 to debug and visualize the transformations. The three different forms
207 of LLVM are all equivalent. This document describes the human readable
208 representation and notation.</p>
210 <p>The LLVM representation aims to be light-weight and low-level
211 while being expressive, typed, and extensible at the same time. It
212 aims to be a "universal IR" of sorts, by being at a low enough level
213 that high-level ideas may be cleanly mapped to it (similar to how
214 microprocessors are "universal IR's", allowing many source languages to
215 be mapped to them). By providing type information, LLVM can be used as
216 the target of optimizations: for example, through pointer analysis, it
217 can be proven that a C automatic variable is never accessed outside of
218 the current function... allowing it to be promoted to a simple SSA
219 value instead of a memory location.</p>
223 <!-- _______________________________________________________________________ -->
224 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
226 <div class="doc_text">
228 <p>It is important to note that this document describes 'well formed'
229 LLVM assembly language. There is a difference between what the parser
230 accepts and what is considered 'well formed'. For example, the
231 following instruction is syntactically okay, but not well formed:</p>
234 %x = <a href="#i_add">add</a> int 1, %x
237 <p>...because the definition of <tt>%x</tt> does not dominate all of
238 its uses. The LLVM infrastructure provides a verification pass that may
239 be used to verify that an LLVM module is well formed. This pass is
240 automatically run by the parser after parsing input assembly and by
241 the optimizer before it outputs bytecode. The violations pointed out
242 by the verifier pass indicate bugs in transformation passes or input to
245 <!-- Describe the typesetting conventions here. --> </div>
247 <!-- *********************************************************************** -->
248 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
249 <!-- *********************************************************************** -->
251 <div class="doc_text">
253 <p>LLVM uses three different forms of identifiers, for different
257 <li>Named values are represented as a string of characters with a '%' prefix.
258 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
259 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
260 Identifiers which require other characters in their names can be surrounded
261 with quotes. In this way, anything except a <tt>"</tt> character can be used
264 <li>Unnamed values are represented as an unsigned numeric value with a '%'
265 prefix. For example, %12, %2, %44.</li>
267 <li>Constants, which are described in a <a href="#constants">section about
268 constants</a>, below.</li>
271 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
272 don't need to worry about name clashes with reserved words, and the set of
273 reserved words may be expanded in the future without penalty. Additionally,
274 unnamed identifiers allow a compiler to quickly come up with a temporary
275 variable without having to avoid symbol table conflicts.</p>
277 <p>Reserved words in LLVM are very similar to reserved words in other
278 languages. There are keywords for different opcodes ('<tt><a
279 href="#i_add">add</a></tt>', '<tt><a href="#i_cast">cast</a></tt>', '<tt><a
280 href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
281 href="#t_void">void</a></tt>', '<tt><a href="#t_uint">uint</a></tt>', etc...),
282 and others. These reserved words cannot conflict with variable names, because
283 none of them start with a '%' character.</p>
285 <p>Here is an example of LLVM code to multiply the integer variable
286 '<tt>%X</tt>' by 8:</p>
291 %result = <a href="#i_mul">mul</a> uint %X, 8
294 <p>After strength reduction:</p>
297 %result = <a href="#i_shl">shl</a> uint %X, ubyte 3
300 <p>And the hard way:</p>
303 <a href="#i_add">add</a> uint %X, %X <i>; yields {uint}:%0</i>
304 <a href="#i_add">add</a> uint %0, %0 <i>; yields {uint}:%1</i>
305 %result = <a href="#i_add">add</a> uint %1, %1
308 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
309 important lexical features of LLVM:</p>
313 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
316 <li>Unnamed temporaries are created when the result of a computation is not
317 assigned to a named value.</li>
319 <li>Unnamed temporaries are numbered sequentially</li>
323 <p>...and it also shows a convention that we follow in this document. When
324 demonstrating instructions, we will follow an instruction with a comment that
325 defines the type and name of value produced. Comments are shown in italic
330 <!-- *********************************************************************** -->
331 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
332 <!-- *********************************************************************** -->
334 <!-- ======================================================================= -->
335 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
338 <div class="doc_text">
340 <p>LLVM programs are composed of "Module"s, each of which is a
341 translation unit of the input programs. Each module consists of
342 functions, global variables, and symbol table entries. Modules may be
343 combined together with the LLVM linker, which merges function (and
344 global variable) definitions, resolves forward declarations, and merges
345 symbol table entries. Here is an example of the "hello world" module:</p>
347 <pre><i>; Declare the string constant as a global constant...</i>
348 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
349 href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
351 <i>; External declaration of the puts function</i>
352 <a href="#functionstructure">declare</a> int %puts(sbyte*) <i>; int(sbyte*)* </i>
354 <i>; Definition of main function</i>
355 int %main() { <i>; int()* </i>
356 <i>; Convert [13x sbyte]* to sbyte *...</i>
358 href="#i_getelementptr">getelementptr</a> [13 x sbyte]* %.LC0, long 0, long 0 <i>; sbyte*</i>
360 <i>; Call puts function to write out the string to stdout...</i>
362 href="#i_call">call</a> int %puts(sbyte* %cast210) <i>; int</i>
364 href="#i_ret">ret</a> int 0<br>}<br></pre>
366 <p>This example is made up of a <a href="#globalvars">global variable</a>
367 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
368 function, and a <a href="#functionstructure">function definition</a>
369 for "<tt>main</tt>".</p>
371 <p>In general, a module is made up of a list of global values,
372 where both functions and global variables are global values. Global values are
373 represented by a pointer to a memory location (in this case, a pointer to an
374 array of char, and a pointer to a function), and have one of the following <a
375 href="#linkage">linkage types</a>.</p>
379 <!-- ======================================================================= -->
380 <div class="doc_subsection">
381 <a name="linkage">Linkage Types</a>
384 <div class="doc_text">
387 All Global Variables and Functions have one of the following types of linkage:
392 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
394 <dd>Global values with internal linkage are only directly accessible by
395 objects in the current module. In particular, linking code into a module with
396 an internal global value may cause the internal to be renamed as necessary to
397 avoid collisions. Because the symbol is internal to the module, all
398 references can be updated. This corresponds to the notion of the
399 '<tt>static</tt>' keyword in C, or the idea of "anonymous namespaces" in C++.
402 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
404 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt> linkage, with
405 the twist that linking together two modules defining the same
406 <tt>linkonce</tt> globals will cause one of the globals to be discarded. This
407 is typically used to implement inline functions. Unreferenced
408 <tt>linkonce</tt> globals are allowed to be discarded.
411 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
413 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
414 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
415 used to implement constructs in C such as "<tt>int X;</tt>" at global scope.
418 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
420 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
421 pointer to array type. When two global variables with appending linkage are
422 linked together, the two global arrays are appended together. This is the
423 LLVM, typesafe, equivalent of having the system linker append together
424 "sections" with identical names when .o files are linked.
427 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
429 <dd>If none of the above identifiers are used, the global is externally
430 visible, meaning that it participates in linkage and can be used to resolve
431 external symbol references.
435 <p><a name="linkage_external">For example, since the "<tt>.LC0</tt>"
436 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
437 variable and was linked with this one, one of the two would be renamed,
438 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
439 external (i.e., lacking any linkage declarations), they are accessible
440 outside of the current module. It is illegal for a function <i>declaration</i>
441 to have any linkage type other than "externally visible".</a></p>
445 <!-- ======================================================================= -->
446 <div class="doc_subsection">
447 <a name="callingconv">Calling Conventions</a>
450 <div class="doc_text">
452 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
453 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
454 specified for the call. The calling convention of any pair of dynamic
455 caller/callee must match, or the behavior of the program is undefined. The
456 following calling conventions are supported by LLVM, and more may be added in
460 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
462 <dd>This calling convention (the default if no other calling convention is
463 specified) matches the target C calling conventions. This calling convention
464 supports varargs function calls and tolerates some mismatch in the declared
465 prototype and implemented declaration of the function (as does normal C).
468 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
470 <dd>This calling convention attempts to make calls as fast as possible
471 (e.g. by passing things in registers). This calling convention allows the
472 target to use whatever tricks it wants to produce fast code for the target,
473 without having to conform to an externally specified ABI. Implementations of
474 this convention should allow arbitrary tail call optimization to be supported.
475 This calling convention does not support varargs and requires the prototype of
476 all callees to exactly match the prototype of the function definition.
479 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
481 <dd>This calling convention attempts to make code in the caller as efficient
482 as possible under the assumption that the call is not commonly executed. As
483 such, these calls often preserve all registers so that the call does not break
484 any live ranges in the caller side. This calling convention does not support
485 varargs and requires the prototype of all callees to exactly match the
486 prototype of the function definition.
489 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
491 <dd>Any calling convention may be specified by number, allowing
492 target-specific calling conventions to be used. Target specific calling
493 conventions start at 64.
497 <p>More calling conventions can be added/defined on an as-needed basis, to
498 support pascal conventions or any other well-known target-independent
503 <!-- ======================================================================= -->
504 <div class="doc_subsection">
505 <a name="globalvars">Global Variables</a>
508 <div class="doc_text">
510 <p>Global variables define regions of memory allocated at compilation time
511 instead of run-time. Global variables may optionally be initialized, may have
512 an explicit section to be placed in, and may
513 have an optional explicit alignment specified. A
514 variable may be defined as a global "constant," which indicates that the
515 contents of the variable will <b>never</b> be modified (enabling better
516 optimization, allowing the global data to be placed in the read-only section of
517 an executable, etc). Note that variables that need runtime initialization
518 cannot be marked "constant" as there is a store to the variable.</p>
521 LLVM explicitly allows <em>declarations</em> of global variables to be marked
522 constant, even if the final definition of the global is not. This capability
523 can be used to enable slightly better optimization of the program, but requires
524 the language definition to guarantee that optimizations based on the
525 'constantness' are valid for the translation units that do not include the
529 <p>As SSA values, global variables define pointer values that are in
530 scope (i.e. they dominate) all basic blocks in the program. Global
531 variables always define a pointer to their "content" type because they
532 describe a region of memory, and all memory objects in LLVM are
533 accessed through pointers.</p>
535 <p>LLVM allows an explicit section to be specified for globals. If the target
536 supports it, it will emit globals to the section specified.</p>
538 <p>An explicit alignment may be specified for a global. If not present, or if
539 the alignment is set to zero, the alignment of the global is set by the target
540 to whatever it feels convenient. If an explicit alignment is specified, the
541 global is forced to have at least that much alignment. All alignments must be
547 <!-- ======================================================================= -->
548 <div class="doc_subsection">
549 <a name="functionstructure">Functions</a>
552 <div class="doc_text">
554 <p>LLVM function definitions consist of an optional <a href="#linkage">linkage
555 type</a>, an optional <a href="#callingconv">calling convention</a>, a return
556 type, a function name, a (possibly empty) argument list, an optional section,
557 an optional alignment, an opening curly brace,
558 a list of basic blocks, and a closing curly brace. LLVM function declarations
559 are defined with the "<tt>declare</tt>" keyword, an optional <a
560 href="#callingconv">calling convention</a>, a return type, a function name,
561 a possibly empty list of arguments, and an optional alignment.</p>
563 <p>A function definition contains a list of basic blocks, forming the CFG for
564 the function. Each basic block may optionally start with a label (giving the
565 basic block a symbol table entry), contains a list of instructions, and ends
566 with a <a href="#terminators">terminator</a> instruction (such as a branch or
567 function return).</p>
569 <p>The first basic block in a program is special in two ways: it is immediately
570 executed on entrance to the function, and it is not allowed to have predecessor
571 basic blocks (i.e. there can not be any branches to the entry block of a
572 function). Because the block can have no predecessors, it also cannot have any
573 <a href="#i_phi">PHI nodes</a>.</p>
575 <p>LLVM functions are identified by their name and type signature. Hence, two
576 functions with the same name but different parameter lists or return values are
577 considered different functions, and LLVM will resolve references to each
580 <p>LLVM allows an explicit section to be specified for functions. If the target
581 supports it, it will emit functions to the section specified.</p>
583 <p>An explicit alignment may be specified for a function. If not present, or if
584 the alignment is set to zero, the alignment of the function is set by the target
585 to whatever it feels convenient. If an explicit alignment is specified, the
586 function is forced to have at least that much alignment. All alignments must be
591 <!-- ======================================================================= -->
592 <div class="doc_subsection">
593 <a name="moduleasm">Module-Level Inline Assembly</a></li>
596 <div class="doc_text">
598 Modules may contain "module-level inline asm" blocks, which corresponds to the
599 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
600 LLVM and treated as a single unit, but may be separated in the .ll file if
601 desired. The syntax is very simple:
604 <div class="doc_code"><pre>
605 module asm "inline asm code goes here"
606 module asm "more can go here"
609 <p>The strings can contain any character by escaping non-printable characters.
610 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
615 The inline asm code is simply printed to the machine code .s file when
616 assembly code is generated.
621 <!-- *********************************************************************** -->
622 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
623 <!-- *********************************************************************** -->
625 <div class="doc_text">
627 <p>The LLVM type system is one of the most important features of the
628 intermediate representation. Being typed enables a number of
629 optimizations to be performed on the IR directly, without having to do
630 extra analyses on the side before the transformation. A strong type
631 system makes it easier to read the generated code and enables novel
632 analyses and transformations that are not feasible to perform on normal
633 three address code representations.</p>
637 <!-- ======================================================================= -->
638 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
639 <div class="doc_text">
640 <p>The primitive types are the fundamental building blocks of the LLVM
641 system. The current set of primitive types is as follows:</p>
643 <table class="layout">
648 <tr><th>Type</th><th>Description</th></tr>
649 <tr><td><tt>void</tt></td><td>No value</td></tr>
650 <tr><td><tt>ubyte</tt></td><td>Unsigned 8-bit value</td></tr>
651 <tr><td><tt>ushort</tt></td><td>Unsigned 16-bit value</td></tr>
652 <tr><td><tt>uint</tt></td><td>Unsigned 32-bit value</td></tr>
653 <tr><td><tt>ulong</tt></td><td>Unsigned 64-bit value</td></tr>
654 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
655 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
662 <tr><th>Type</th><th>Description</th></tr>
663 <tr><td><tt>bool</tt></td><td>True or False value</td></tr>
664 <tr><td><tt>sbyte</tt></td><td>Signed 8-bit value</td></tr>
665 <tr><td><tt>short</tt></td><td>Signed 16-bit value</td></tr>
666 <tr><td><tt>int</tt></td><td>Signed 32-bit value</td></tr>
667 <tr><td><tt>long</tt></td><td>Signed 64-bit value</td></tr>
668 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
676 <!-- _______________________________________________________________________ -->
677 <div class="doc_subsubsection"> <a name="t_classifications">Type
678 Classifications</a> </div>
679 <div class="doc_text">
680 <p>These different primitive types fall into a few useful
683 <table border="1" cellspacing="0" cellpadding="4">
685 <tr><th>Classification</th><th>Types</th></tr>
687 <td><a name="t_signed">signed</a></td>
688 <td><tt>sbyte, short, int, long, float, double</tt></td>
691 <td><a name="t_unsigned">unsigned</a></td>
692 <td><tt>ubyte, ushort, uint, ulong</tt></td>
695 <td><a name="t_integer">integer</a></td>
696 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
699 <td><a name="t_integral">integral</a></td>
700 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt>
704 <td><a name="t_floating">floating point</a></td>
705 <td><tt>float, double</tt></td>
708 <td><a name="t_firstclass">first class</a></td>
709 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long,<br>
710 float, double, <a href="#t_pointer">pointer</a>,
711 <a href="#t_packed">packed</a></tt></td>
716 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
717 most important. Values of these types are the only ones which can be
718 produced by instructions, passed as arguments, or used as operands to
719 instructions. This means that all structures and arrays must be
720 manipulated either by pointer or by component.</p>
723 <!-- ======================================================================= -->
724 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
726 <div class="doc_text">
728 <p>The real power in LLVM comes from the derived types in the system.
729 This is what allows a programmer to represent arrays, functions,
730 pointers, and other useful types. Note that these derived types may be
731 recursive: For example, it is possible to have a two dimensional array.</p>
735 <!-- _______________________________________________________________________ -->
736 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
738 <div class="doc_text">
742 <p>The array type is a very simple derived type that arranges elements
743 sequentially in memory. The array type requires a size (number of
744 elements) and an underlying data type.</p>
749 [<# elements> x <elementtype>]
752 <p>The number of elements is a constant integer value; elementtype may
753 be any type with a size.</p>
756 <table class="layout">
759 <tt>[40 x int ]</tt><br/>
760 <tt>[41 x int ]</tt><br/>
761 <tt>[40 x uint]</tt><br/>
764 Array of 40 integer values.<br/>
765 Array of 41 integer values.<br/>
766 Array of 40 unsigned integer values.<br/>
770 <p>Here are some examples of multidimensional arrays:</p>
771 <table class="layout">
774 <tt>[3 x [4 x int]]</tt><br/>
775 <tt>[12 x [10 x float]]</tt><br/>
776 <tt>[2 x [3 x [4 x uint]]]</tt><br/>
779 3x4 array of integer values.<br/>
780 12x10 array of single precision floating point values.<br/>
781 2x3x4 array of unsigned integer values.<br/>
786 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
787 length array. Normally, accesses past the end of an array are undefined in
788 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
789 As a special case, however, zero length arrays are recognized to be variable
790 length. This allows implementation of 'pascal style arrays' with the LLVM
791 type "{ int, [0 x float]}", for example.</p>
795 <!-- _______________________________________________________________________ -->
796 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
797 <div class="doc_text">
799 <p>The function type can be thought of as a function signature. It
800 consists of a return type and a list of formal parameter types.
801 Function types are usually used to build virtual function tables
802 (which are structures of pointers to functions), for indirect function
803 calls, and when defining a function.</p>
805 The return type of a function type cannot be an aggregate type.
808 <pre> <returntype> (<parameter list>)<br></pre>
809 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
810 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
811 which indicates that the function takes a variable number of arguments.
812 Variable argument functions can access their arguments with the <a
813 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
815 <table class="layout">
818 <tt>int (int)</tt> <br/>
819 <tt>float (int, int *) *</tt><br/>
820 <tt>int (sbyte *, ...)</tt><br/>
823 function taking an <tt>int</tt>, returning an <tt>int</tt><br/>
824 <a href="#t_pointer">Pointer</a> to a function that takes an
825 <tt>int</tt> and a <a href="#t_pointer">pointer</a> to <tt>int</tt>,
826 returning <tt>float</tt>.<br/>
827 A vararg function that takes at least one <a href="#t_pointer">pointer</a>
828 to <tt>sbyte</tt> (signed char in C), which returns an integer. This is
829 the signature for <tt>printf</tt> in LLVM.<br/>
835 <!-- _______________________________________________________________________ -->
836 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
837 <div class="doc_text">
839 <p>The structure type is used to represent a collection of data members
840 together in memory. The packing of the field types is defined to match
841 the ABI of the underlying processor. The elements of a structure may
842 be any type that has a size.</p>
843 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
844 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
845 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
848 <pre> { <type list> }<br></pre>
850 <table class="layout">
853 <tt>{ int, int, int }</tt><br/>
854 <tt>{ float, int (int) * }</tt><br/>
857 a triple of three <tt>int</tt> values<br/>
858 A pair, where the first element is a <tt>float</tt> and the second element
859 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
860 that takes an <tt>int</tt>, returning an <tt>int</tt>.<br/>
866 <!-- _______________________________________________________________________ -->
867 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
868 <div class="doc_text">
870 <p>As in many languages, the pointer type represents a pointer or
871 reference to another object, which must live in memory.</p>
873 <pre> <type> *<br></pre>
875 <table class="layout">
878 <tt>[4x int]*</tt><br/>
879 <tt>int (int *) *</tt><br/>
882 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
883 four <tt>int</tt> values<br/>
884 A <a href="#t_pointer">pointer</a> to a <a
885 href="#t_function">function</a> that takes an <tt>int*</tt>, returning an
892 <!-- _______________________________________________________________________ -->
893 <div class="doc_subsubsection"> <a name="t_packed">Packed Type</a> </div>
894 <div class="doc_text">
898 <p>A packed type is a simple derived type that represents a vector
899 of elements. Packed types are used when multiple primitive data
900 are operated in parallel using a single instruction (SIMD).
901 A packed type requires a size (number of
902 elements) and an underlying primitive data type. Vectors must have a power
903 of two length (1, 2, 4, 8, 16 ...). Packed types are
904 considered <a href="#t_firstclass">first class</a>.</p>
909 < <# elements> x <elementtype> >
912 <p>The number of elements is a constant integer value; elementtype may
913 be any integral or floating point type.</p>
917 <table class="layout">
920 <tt><4 x int></tt><br/>
921 <tt><8 x float></tt><br/>
922 <tt><2 x uint></tt><br/>
925 Packed vector of 4 integer values.<br/>
926 Packed vector of 8 floating-point values.<br/>
927 Packed vector of 2 unsigned integer values.<br/>
933 <!-- _______________________________________________________________________ -->
934 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
935 <div class="doc_text">
939 <p>Opaque types are used to represent unknown types in the system. This
940 corresponds (for example) to the C notion of a foward declared structure type.
941 In LLVM, opaque types can eventually be resolved to any type (not just a
952 <table class="layout">
965 <!-- *********************************************************************** -->
966 <div class="doc_section"> <a name="constants">Constants</a> </div>
967 <!-- *********************************************************************** -->
969 <div class="doc_text">
971 <p>LLVM has several different basic types of constants. This section describes
972 them all and their syntax.</p>
976 <!-- ======================================================================= -->
977 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
979 <div class="doc_text">
982 <dt><b>Boolean constants</b></dt>
984 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
985 constants of the <tt><a href="#t_primitive">bool</a></tt> type.
988 <dt><b>Integer constants</b></dt>
990 <dd>Standard integers (such as '4') are constants of the <a
991 href="#t_integer">integer</a> type. Negative numbers may be used with signed
995 <dt><b>Floating point constants</b></dt>
997 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
998 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
999 notation (see below). Floating point constants must have a <a
1000 href="#t_floating">floating point</a> type. </dd>
1002 <dt><b>Null pointer constants</b></dt>
1004 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1005 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1009 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1010 of floating point constants. For example, the form '<tt>double
1011 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1012 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1013 (and the only time that they are generated by the disassembler) is when a
1014 floating point constant must be emitted but it cannot be represented as a
1015 decimal floating point number. For example, NaN's, infinities, and other
1016 special values are represented in their IEEE hexadecimal format so that
1017 assembly and disassembly do not cause any bits to change in the constants.</p>
1021 <!-- ======================================================================= -->
1022 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1025 <div class="doc_text">
1026 <p>Aggregate constants arise from aggregation of simple constants
1027 and smaller aggregate constants.</p>
1030 <dt><b>Structure constants</b></dt>
1032 <dd>Structure constants are represented with notation similar to structure
1033 type definitions (a comma separated list of elements, surrounded by braces
1034 (<tt>{}</tt>)). For example: "<tt>{ int 4, float 17.0, int* %G }</tt>",
1035 where "<tt>%G</tt>" is declared as "<tt>%G = external global int</tt>". Structure constants
1036 must have <a href="#t_struct">structure type</a>, and the number and
1037 types of elements must match those specified by the type.
1040 <dt><b>Array constants</b></dt>
1042 <dd>Array constants are represented with notation similar to array type
1043 definitions (a comma separated list of elements, surrounded by square brackets
1044 (<tt>[]</tt>)). For example: "<tt>[ int 42, int 11, int 74 ]</tt>". Array
1045 constants must have <a href="#t_array">array type</a>, and the number and
1046 types of elements must match those specified by the type.
1049 <dt><b>Packed constants</b></dt>
1051 <dd>Packed constants are represented with notation similar to packed type
1052 definitions (a comma separated list of elements, surrounded by
1053 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< int 42,
1054 int 11, int 74, int 100 ></tt>". Packed constants must have <a
1055 href="#t_packed">packed type</a>, and the number and types of elements must
1056 match those specified by the type.
1059 <dt><b>Zero initialization</b></dt>
1061 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1062 value to zero of <em>any</em> type, including scalar and aggregate types.
1063 This is often used to avoid having to print large zero initializers (e.g. for
1064 large arrays) and is always exactly equivalent to using explicit zero
1071 <!-- ======================================================================= -->
1072 <div class="doc_subsection">
1073 <a name="globalconstants">Global Variable and Function Addresses</a>
1076 <div class="doc_text">
1078 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1079 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1080 constants. These constants are explicitly referenced when the <a
1081 href="#identifiers">identifier for the global</a> is used and always have <a
1082 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1088 %Z = global [2 x int*] [ int* %X, int* %Y ]
1093 <!-- ======================================================================= -->
1094 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1095 <div class="doc_text">
1096 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1097 no specific value. Undefined values may be of any type and be used anywhere
1098 a constant is permitted.</p>
1100 <p>Undefined values indicate to the compiler that the program is well defined
1101 no matter what value is used, giving the compiler more freedom to optimize.
1105 <!-- ======================================================================= -->
1106 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1109 <div class="doc_text">
1111 <p>Constant expressions are used to allow expressions involving other constants
1112 to be used as constants. Constant expressions may be of any <a
1113 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1114 that does not have side effects (e.g. load and call are not supported). The
1115 following is the syntax for constant expressions:</p>
1118 <dt><b><tt>cast ( CST to TYPE )</tt></b></dt>
1120 <dd>Cast a constant to another type.</dd>
1122 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1124 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1125 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1126 instruction, the index list may have zero or more indexes, which are required
1127 to make sense for the type of "CSTPTR".</dd>
1129 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1131 <dd>Perform the <a href="#i_select">select operation</a> on
1134 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1136 <dd>Perform the <a href="#i_extractelement">extractelement
1137 operation</a> on constants.
1139 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1141 <dd>Perform the <a href="#i_insertelement">insertelement
1142 operation</a> on constants.
1144 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1146 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1147 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1148 binary</a> operations. The constraints on operands are the same as those for
1149 the corresponding instruction (e.g. no bitwise operations on floating point
1150 values are allowed).</dd>
1154 <!-- *********************************************************************** -->
1155 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1156 <!-- *********************************************************************** -->
1158 <!-- ======================================================================= -->
1159 <div class="doc_subsection">
1160 <a name="inlineasm">Inline Assembler Expressions</a>
1163 <div class="doc_text">
1166 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1167 Module-Level Inline Assembly</a>) through the use of a special value. This
1168 value represents the inline assembler as a string (containing the instructions
1169 to emit), a list of operand constraints (stored as a string), and a flag that
1170 indicates whether or not the inline asm expression has side effects. An example
1171 inline assembler expression is:
1175 int(int) asm "bswap $0", "=r,r"
1179 Inline assembler expressions may <b>only</b> be used as the callee operand of
1180 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1184 %X = call int asm "<a href="#i_bswap">bswap</a> $0", "=r,r"(int %Y)
1188 Inline asms with side effects not visible in the constraint list must be marked
1189 as having side effects. This is done through the use of the
1190 '<tt>sideeffect</tt>' keyword, like so:
1194 call void asm sideeffect "eieio", ""()
1197 <p>TODO: The format of the asm and constraints string still need to be
1198 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1199 need to be documented).
1204 <!-- *********************************************************************** -->
1205 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1206 <!-- *********************************************************************** -->
1208 <div class="doc_text">
1210 <p>The LLVM instruction set consists of several different
1211 classifications of instructions: <a href="#terminators">terminator
1212 instructions</a>, <a href="#binaryops">binary instructions</a>,
1213 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1214 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1215 instructions</a>.</p>
1219 <!-- ======================================================================= -->
1220 <div class="doc_subsection"> <a name="terminators">Terminator
1221 Instructions</a> </div>
1223 <div class="doc_text">
1225 <p>As mentioned <a href="#functionstructure">previously</a>, every
1226 basic block in a program ends with a "Terminator" instruction, which
1227 indicates which block should be executed after the current block is
1228 finished. These terminator instructions typically yield a '<tt>void</tt>'
1229 value: they produce control flow, not values (the one exception being
1230 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1231 <p>There are six different terminator instructions: the '<a
1232 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1233 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1234 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1235 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1236 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1240 <!-- _______________________________________________________________________ -->
1241 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1242 Instruction</a> </div>
1243 <div class="doc_text">
1245 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1246 ret void <i>; Return from void function</i>
1249 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1250 value) from a function back to the caller.</p>
1251 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1252 returns a value and then causes control flow, and one that just causes
1253 control flow to occur.</p>
1255 <p>The '<tt>ret</tt>' instruction may return any '<a
1256 href="#t_firstclass">first class</a>' type. Notice that a function is
1257 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1258 instruction inside of the function that returns a value that does not
1259 match the return type of the function.</p>
1261 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1262 returns back to the calling function's context. If the caller is a "<a
1263 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1264 the instruction after the call. If the caller was an "<a
1265 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1266 at the beginning of the "normal" destination block. If the instruction
1267 returns a value, that value shall set the call or invoke instruction's
1270 <pre> ret int 5 <i>; Return an integer value of 5</i>
1271 ret void <i>; Return from a void function</i>
1274 <!-- _______________________________________________________________________ -->
1275 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1276 <div class="doc_text">
1278 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1281 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1282 transfer to a different basic block in the current function. There are
1283 two forms of this instruction, corresponding to a conditional branch
1284 and an unconditional branch.</p>
1286 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1287 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
1288 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
1289 value as a target.</p>
1291 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
1292 argument is evaluated. If the value is <tt>true</tt>, control flows
1293 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1294 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1296 <pre>Test:<br> %cond = <a href="#i_setcc">seteq</a> int %a, %b<br> br bool %cond, label %IfEqual, label %IfUnequal<br>IfEqual:<br> <a
1297 href="#i_ret">ret</a> int 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> int 0<br></pre>
1299 <!-- _______________________________________________________________________ -->
1300 <div class="doc_subsubsection">
1301 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1304 <div class="doc_text">
1308 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1313 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1314 several different places. It is a generalization of the '<tt>br</tt>'
1315 instruction, allowing a branch to occur to one of many possible
1321 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1322 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1323 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1324 table is not allowed to contain duplicate constant entries.</p>
1328 <p>The <tt>switch</tt> instruction specifies a table of values and
1329 destinations. When the '<tt>switch</tt>' instruction is executed, this
1330 table is searched for the given value. If the value is found, control flow is
1331 transfered to the corresponding destination; otherwise, control flow is
1332 transfered to the default destination.</p>
1334 <h5>Implementation:</h5>
1336 <p>Depending on properties of the target machine and the particular
1337 <tt>switch</tt> instruction, this instruction may be code generated in different
1338 ways. For example, it could be generated as a series of chained conditional
1339 branches or with a lookup table.</p>
1344 <i>; Emulate a conditional br instruction</i>
1345 %Val = <a href="#i_cast">cast</a> bool %value to int
1346 switch int %Val, label %truedest [int 0, label %falsedest ]
1348 <i>; Emulate an unconditional br instruction</i>
1349 switch uint 0, label %dest [ ]
1351 <i>; Implement a jump table:</i>
1352 switch uint %val, label %otherwise [ uint 0, label %onzero
1353 uint 1, label %onone
1354 uint 2, label %ontwo ]
1358 <!-- _______________________________________________________________________ -->
1359 <div class="doc_subsubsection">
1360 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1363 <div class="doc_text">
1368 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1369 to label <normal label> except label <exception label>
1374 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1375 function, with the possibility of control flow transfer to either the
1376 '<tt>normal</tt>' label or the
1377 '<tt>exception</tt>' label. If the callee function returns with the
1378 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1379 "normal" label. If the callee (or any indirect callees) returns with the "<a
1380 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1381 continued at the dynamically nearest "exception" label.</p>
1385 <p>This instruction requires several arguments:</p>
1389 The optional "cconv" marker indicates which <a href="callingconv">calling
1390 convention</a> the call should use. If none is specified, the call defaults
1391 to using C calling conventions.
1393 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1394 function value being invoked. In most cases, this is a direct function
1395 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1396 an arbitrary pointer to function value.
1399 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1400 function to be invoked. </li>
1402 <li>'<tt>function args</tt>': argument list whose types match the function
1403 signature argument types. If the function signature indicates the function
1404 accepts a variable number of arguments, the extra arguments can be
1407 <li>'<tt>normal label</tt>': the label reached when the called function
1408 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1410 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1411 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1417 <p>This instruction is designed to operate as a standard '<tt><a
1418 href="#i_call">call</a></tt>' instruction in most regards. The primary
1419 difference is that it establishes an association with a label, which is used by
1420 the runtime library to unwind the stack.</p>
1422 <p>This instruction is used in languages with destructors to ensure that proper
1423 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1424 exception. Additionally, this is important for implementation of
1425 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1429 %retval = invoke int %Test(int 15) to label %Continue
1430 except label %TestCleanup <i>; {int}:retval set</i>
1431 %retval = invoke <a href="#callingconv">coldcc</a> int %Test(int 15) to label %Continue
1432 except label %TestCleanup <i>; {int}:retval set</i>
1437 <!-- _______________________________________________________________________ -->
1439 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1440 Instruction</a> </div>
1442 <div class="doc_text">
1451 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1452 at the first callee in the dynamic call stack which used an <a
1453 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1454 primarily used to implement exception handling.</p>
1458 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1459 immediately halt. The dynamic call stack is then searched for the first <a
1460 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1461 execution continues at the "exceptional" destination block specified by the
1462 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1463 dynamic call chain, undefined behavior results.</p>
1466 <!-- _______________________________________________________________________ -->
1468 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1469 Instruction</a> </div>
1471 <div class="doc_text">
1480 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1481 instruction is used to inform the optimizer that a particular portion of the
1482 code is not reachable. This can be used to indicate that the code after a
1483 no-return function cannot be reached, and other facts.</p>
1487 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1492 <!-- ======================================================================= -->
1493 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1494 <div class="doc_text">
1495 <p>Binary operators are used to do most of the computation in a
1496 program. They require two operands, execute an operation on them, and
1497 produce a single value. The operands might represent
1498 multiple data, as is the case with the <a href="#t_packed">packed</a> data type.
1499 The result value of a binary operator is not
1500 necessarily the same type as its operands.</p>
1501 <p>There are several different binary operators:</p>
1503 <!-- _______________________________________________________________________ -->
1504 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1505 Instruction</a> </div>
1506 <div class="doc_text">
1508 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1511 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1513 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1514 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1515 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1516 Both arguments must have identical types.</p>
1518 <p>The value produced is the integer or floating point sum of the two
1521 <pre> <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
1524 <!-- _______________________________________________________________________ -->
1525 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1526 Instruction</a> </div>
1527 <div class="doc_text">
1529 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1532 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1534 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1535 instruction present in most other intermediate representations.</p>
1537 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1538 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1540 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1541 Both arguments must have identical types.</p>
1543 <p>The value produced is the integer or floating point difference of
1544 the two operands.</p>
1546 <pre> <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
1547 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
1550 <!-- _______________________________________________________________________ -->
1551 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1552 Instruction</a> </div>
1553 <div class="doc_text">
1555 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1558 <p>The '<tt>mul</tt>' instruction returns the product of its two
1561 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1562 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1564 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1565 Both arguments must have identical types.</p>
1567 <p>The value produced is the integer or floating point product of the
1569 <p>There is no signed vs unsigned multiplication. The appropriate
1570 action is taken based on the type of the operand.</p>
1572 <pre> <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1575 <!-- _______________________________________________________________________ -->
1576 <div class="doc_subsubsection"> <a name="i_div">'<tt>div</tt>'
1577 Instruction</a> </div>
1578 <div class="doc_text">
1580 <pre> <result> = div <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1583 <p>The '<tt>div</tt>' instruction returns the quotient of its two
1586 <p>The two arguments to the '<tt>div</tt>' instruction must be either <a
1587 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1589 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1590 Both arguments must have identical types.</p>
1592 <p>The value produced is the integer or floating point quotient of the
1595 <pre> <result> = div int 4, %var <i>; yields {int}:result = 4 / %var</i>
1598 <!-- _______________________________________________________________________ -->
1599 <div class="doc_subsubsection"> <a name="i_rem">'<tt>rem</tt>'
1600 Instruction</a> </div>
1601 <div class="doc_text">
1603 <pre> <result> = rem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1606 <p>The '<tt>rem</tt>' instruction returns the remainder from the
1607 division of its two operands.</p>
1609 <p>The two arguments to the '<tt>rem</tt>' instruction must be either <a
1610 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1612 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1613 Both arguments must have identical types.</p>
1615 <p>This returns the <i>remainder</i> of a division (where the result
1616 has the same sign as the divisor), not the <i>modulus</i> (where the
1617 result has the same sign as the dividend) of a value. For more
1618 information about the difference, see <a
1619 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
1622 <pre> <result> = rem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1626 <!-- _______________________________________________________________________ -->
1627 <div class="doc_subsubsection"> <a name="i_setcc">'<tt>set<i>cc</i></tt>'
1628 Instructions</a> </div>
1629 <div class="doc_text">
1631 <pre> <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1632 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1633 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1634 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1635 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1636 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1639 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean
1640 value based on a comparison of their two operands.</p>
1642 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must
1643 be of <a href="#t_firstclass">first class</a> type (it is not possible
1644 to compare '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>'
1645 or '<tt>void</tt>' values, etc...). Both arguments must have identical
1648 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1649 value if both operands are equal.<br>
1650 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1651 value if both operands are unequal.<br>
1652 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1653 value if the first operand is less than the second operand.<br>
1654 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1655 value if the first operand is greater than the second operand.<br>
1656 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1657 value if the first operand is less than or equal to the second operand.<br>
1658 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1659 value if the first operand is greater than or equal to the second
1662 <pre> <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1663 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1664 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1665 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1666 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1667 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1671 <!-- ======================================================================= -->
1672 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
1673 Operations</a> </div>
1674 <div class="doc_text">
1675 <p>Bitwise binary operators are used to do various forms of
1676 bit-twiddling in a program. They are generally very efficient
1677 instructions and can commonly be strength reduced from other
1678 instructions. They require two operands, execute an operation on them,
1679 and produce a single value. The resulting value of the bitwise binary
1680 operators is always the same type as its first operand.</p>
1682 <!-- _______________________________________________________________________ -->
1683 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
1684 Instruction</a> </div>
1685 <div class="doc_text">
1687 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1690 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1691 its two operands.</p>
1693 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1694 href="#t_integral">integral</a> values. Both arguments must have
1695 identical types.</p>
1697 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1699 <div style="align: center">
1700 <table border="1" cellspacing="0" cellpadding="4">
1731 <pre> <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1732 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1733 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1736 <!-- _______________________________________________________________________ -->
1737 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
1738 <div class="doc_text">
1740 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1743 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1744 or of its two operands.</p>
1746 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1747 href="#t_integral">integral</a> values. Both arguments must have
1748 identical types.</p>
1750 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1752 <div style="align: center">
1753 <table border="1" cellspacing="0" cellpadding="4">
1784 <pre> <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1785 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1786 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1789 <!-- _______________________________________________________________________ -->
1790 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
1791 Instruction</a> </div>
1792 <div class="doc_text">
1794 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1797 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
1798 or of its two operands. The <tt>xor</tt> is used to implement the
1799 "one's complement" operation, which is the "~" operator in C.</p>
1801 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
1802 href="#t_integral">integral</a> values. Both arguments must have
1803 identical types.</p>
1805 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
1807 <div style="align: center">
1808 <table border="1" cellspacing="0" cellpadding="4">
1840 <pre> <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
1841 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
1842 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
1843 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
1846 <!-- _______________________________________________________________________ -->
1847 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
1848 Instruction</a> </div>
1849 <div class="doc_text">
1851 <pre> <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1854 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
1855 the left a specified number of bits.</p>
1857 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
1858 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1861 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
1863 <pre> <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
1864 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
1865 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
1868 <!-- _______________________________________________________________________ -->
1869 <div class="doc_subsubsection"> <a name="i_shr">'<tt>shr</tt>'
1870 Instruction</a> </div>
1871 <div class="doc_text">
1873 <pre> <result> = shr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1876 <p>The '<tt>shr</tt>' instruction returns the first operand shifted to
1877 the right a specified number of bits.</p>
1879 <p>The first argument to the '<tt>shr</tt>' instruction must be an <a
1880 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1883 <p>If the first argument is a <a href="#t_signed">signed</a> type, the
1884 most significant bit is duplicated in the newly free'd bit positions.
1885 If the first argument is unsigned, zero bits shall fill the empty
1888 <pre> <result> = shr int 4, ubyte %var <i>; yields {int}:result = 4 >> %var</i>
1889 <result> = shr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
1890 <result> = shr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
1891 <result> = shr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
1892 <result> = shr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
1896 <!-- ======================================================================= -->
1897 <div class="doc_subsection">
1898 <a name="memoryops">Memory Access Operations</a>
1901 <div class="doc_text">
1903 <p>A key design point of an SSA-based representation is how it
1904 represents memory. In LLVM, no memory locations are in SSA form, which
1905 makes things very simple. This section describes how to read, write,
1906 allocate, and free memory in LLVM.</p>
1910 <!-- _______________________________________________________________________ -->
1911 <div class="doc_subsubsection">
1912 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
1915 <div class="doc_text">
1920 <result> = malloc <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
1925 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
1926 heap and returns a pointer to it.</p>
1930 <p>The '<tt>malloc</tt>' instruction allocates
1931 <tt>sizeof(<type>)*NumElements</tt>
1932 bytes of memory from the operating system and returns a pointer of the
1933 appropriate type to the program. If "NumElements" is specified, it is the
1934 number of elements allocated. If an alignment is specified, the value result
1935 of the allocation is guaranteed to be aligned to at least that boundary. If
1936 not specified, or if zero, the target can choose to align the allocation on any
1937 convenient boundary.</p>
1939 <p>'<tt>type</tt>' must be a sized type.</p>
1943 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
1944 a pointer is returned.</p>
1949 %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
1951 %size = <a href="#i_add">add</a> uint 2, 2 <i>; yields {uint}:size = uint 4</i>
1952 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
1953 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
1954 %array3 = malloc int, uint 4, align 1024 <i>; yields {int*}:array3</i>
1955 %array4 = malloc int, align 1024 <i>; yields {int*}:array4</i>
1959 <!-- _______________________________________________________________________ -->
1960 <div class="doc_subsubsection">
1961 <a name="i_free">'<tt>free</tt>' Instruction</a>
1964 <div class="doc_text">
1969 free <type> <value> <i>; yields {void}</i>
1974 <p>The '<tt>free</tt>' instruction returns memory back to the unused
1975 memory heap to be reallocated in the future.</p>
1979 <p>'<tt>value</tt>' shall be a pointer value that points to a value
1980 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
1985 <p>Access to the memory pointed to by the pointer is no longer defined
1986 after this instruction executes.</p>
1991 %array = <a href="#i_malloc">malloc</a> [4 x ubyte] <i>; yields {[4 x ubyte]*}:array</i>
1992 free [4 x ubyte]* %array
1996 <!-- _______________________________________________________________________ -->
1997 <div class="doc_subsubsection">
1998 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2001 <div class="doc_text">
2006 <result> = alloca <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2011 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
2012 stack frame of the procedure that is live until the current function
2013 returns to its caller.</p>
2017 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2018 bytes of memory on the runtime stack, returning a pointer of the
2019 appropriate type to the program. If "NumElements" is specified, it is the
2020 number of elements allocated. If an alignment is specified, the value result
2021 of the allocation is guaranteed to be aligned to at least that boundary. If
2022 not specified, or if zero, the target can choose to align the allocation on any
2023 convenient boundary.</p>
2025 <p>'<tt>type</tt>' may be any sized type.</p>
2029 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2030 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2031 instruction is commonly used to represent automatic variables that must
2032 have an address available. When the function returns (either with the <tt><a
2033 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2034 instructions), the memory is reclaimed.</p>
2039 %ptr = alloca int <i>; yields {int*}:ptr</i>
2040 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
2041 %ptr = alloca int, uint 4, align 1024 <i>; yields {int*}:ptr</i>
2042 %ptr = alloca int, align 1024 <i>; yields {int*}:ptr</i>
2046 <!-- _______________________________________________________________________ -->
2047 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2048 Instruction</a> </div>
2049 <div class="doc_text">
2051 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
2053 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2055 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2056 address from which to load. The pointer must point to a <a
2057 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2058 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2059 the number or order of execution of this <tt>load</tt> with other
2060 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2063 <p>The location of memory pointed to is loaded.</p>
2065 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2067 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2068 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2071 <!-- _______________________________________________________________________ -->
2072 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2073 Instruction</a> </div>
2075 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2076 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2079 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2081 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2082 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
2083 operand must be a pointer to the type of the '<tt><value></tt>'
2084 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2085 optimizer is not allowed to modify the number or order of execution of
2086 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2087 href="#i_store">store</a></tt> instructions.</p>
2089 <p>The contents of memory are updated to contain '<tt><value></tt>'
2090 at the location specified by the '<tt><pointer></tt>' operand.</p>
2092 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2094 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2095 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2097 <!-- _______________________________________________________________________ -->
2098 <div class="doc_subsubsection">
2099 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2102 <div class="doc_text">
2105 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2111 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2112 subelement of an aggregate data structure.</p>
2116 <p>This instruction takes a list of integer constants that indicate what
2117 elements of the aggregate object to index to. The actual types of the arguments
2118 provided depend on the type of the first pointer argument. The
2119 '<tt>getelementptr</tt>' instruction is used to index down through the type
2120 levels of a structure or to a specific index in an array. When indexing into a
2121 structure, only <tt>uint</tt>
2122 integer constants are allowed. When indexing into an array or pointer,
2123 <tt>int</tt> and <tt>long</tt> indexes are allowed of any sign.</p>
2125 <p>For example, let's consider a C code fragment and how it gets
2126 compiled to LLVM:</p>
2140 int *foo(struct ST *s) {
2141 return &s[1].Z.B[5][13];
2145 <p>The LLVM code generated by the GCC frontend is:</p>
2148 %RT = type { sbyte, [10 x [20 x int]], sbyte }
2149 %ST = type { int, double, %RT }
2153 int* %foo(%ST* %s) {
2155 %reg = getelementptr %ST* %s, int 1, uint 2, uint 1, int 5, int 13
2162 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2163 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2164 and <a href="#t_array">array</a> types require <tt>uint</tt>, <tt>int</tt>,
2165 <tt>ulong</tt>, or <tt>long</tt> values, and <a href="#t_struct">structure</a>
2166 types require <tt>uint</tt> <b>constants</b>.</p>
2168 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2169 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
2170 }</tt>' type, a structure. The second index indexes into the third element of
2171 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
2172 sbyte }</tt>' type, another structure. The third index indexes into the second
2173 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
2174 array. The two dimensions of the array are subscripted into, yielding an
2175 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2176 to this element, thus computing a value of '<tt>int*</tt>' type.</p>
2178 <p>Note that it is perfectly legal to index partially through a
2179 structure, returning a pointer to an inner element. Because of this,
2180 the LLVM code for the given testcase is equivalent to:</p>
2183 int* %foo(%ST* %s) {
2184 %t1 = getelementptr %ST* %s, int 1 <i>; yields %ST*:%t1</i>
2185 %t2 = getelementptr %ST* %t1, int 0, uint 2 <i>; yields %RT*:%t2</i>
2186 %t3 = getelementptr %RT* %t2, int 0, uint 1 <i>; yields [10 x [20 x int]]*:%t3</i>
2187 %t4 = getelementptr [10 x [20 x int]]* %t3, int 0, int 5 <i>; yields [20 x int]*:%t4</i>
2188 %t5 = getelementptr [20 x int]* %t4, int 0, int 13 <i>; yields int*:%t5</i>
2193 <p>Note that it is undefined to access an array out of bounds: array and
2194 pointer indexes must always be within the defined bounds of the array type.
2195 The one exception for this rules is zero length arrays. These arrays are
2196 defined to be accessible as variable length arrays, which requires access
2197 beyond the zero'th element.</p>
2202 <i>; yields [12 x ubyte]*:aptr</i>
2203 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, uint 1
2207 <!-- ======================================================================= -->
2208 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
2209 <div class="doc_text">
2210 <p>The instructions in this category are the "miscellaneous"
2211 instructions, which defy better classification.</p>
2213 <!-- _______________________________________________________________________ -->
2214 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
2215 Instruction</a> </div>
2216 <div class="doc_text">
2218 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
2220 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
2221 the SSA graph representing the function.</p>
2223 <p>The type of the incoming values are specified with the first type
2224 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
2225 as arguments, with one pair for each predecessor basic block of the
2226 current block. Only values of <a href="#t_firstclass">first class</a>
2227 type may be used as the value arguments to the PHI node. Only labels
2228 may be used as the label arguments.</p>
2229 <p>There must be no non-phi instructions between the start of a basic
2230 block and the PHI instructions: i.e. PHI instructions must be first in
2233 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
2234 value specified by the parameter, depending on which basic block we
2235 came from in the last <a href="#terminators">terminator</a> instruction.</p>
2237 <pre>Loop: ; Infinite loop that counts from 0 on up...<br> %indvar = phi uint [ 0, %LoopHeader ], [ %nextindvar, %Loop ]<br> %nextindvar = add uint %indvar, 1<br> br label %Loop<br></pre>
2240 <!-- _______________________________________________________________________ -->
2241 <div class="doc_subsubsection">
2242 <a name="i_cast">'<tt>cast .. to</tt>' Instruction</a>
2245 <div class="doc_text">
2250 <result> = cast <ty> <value> to <ty2> <i>; yields ty2</i>
2256 The '<tt>cast</tt>' instruction is used as the primitive means to convert
2257 integers to floating point, change data type sizes, and break type safety (by
2265 The '<tt>cast</tt>' instruction takes a value to cast, which must be a first
2266 class value, and a type to cast it to, which must also be a <a
2267 href="#t_firstclass">first class</a> type.
2273 This instruction follows the C rules for explicit casts when determining how the
2274 data being cast must change to fit in its new container.
2278 When casting to bool, any value that would be considered true in the context of
2279 a C '<tt>if</tt>' condition is converted to the boolean '<tt>true</tt>' values,
2280 all else are '<tt>false</tt>'.
2284 When extending an integral value from a type of one signness to another (for
2285 example '<tt>sbyte</tt>' to '<tt>ulong</tt>'), the value is sign-extended if the
2286 <b>source</b> value is signed, and zero-extended if the source value is
2287 unsigned. <tt>bool</tt> values are always zero extended into either zero or
2294 %X = cast int 257 to ubyte <i>; yields ubyte:1</i>
2295 %Y = cast int 123 to bool <i>; yields bool:true</i>
2299 <!-- _______________________________________________________________________ -->
2300 <div class="doc_subsubsection">
2301 <a name="i_select">'<tt>select</tt>' Instruction</a>
2304 <div class="doc_text">
2309 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
2315 The '<tt>select</tt>' instruction is used to choose one value based on a
2316 condition, without branching.
2323 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.
2329 If the boolean condition evaluates to true, the instruction returns the first
2330 value argument; otherwise, it returns the second value argument.
2336 %X = select bool true, ubyte 17, ubyte 42 <i>; yields ubyte:17</i>
2340 <!-- _______________________________________________________________________ -->
2341 <div class="doc_subsubsection"> <a name="i_vset">'<tt>vset</tt>'
2342 Instruction</a> </div>
2343 <div class="doc_text">
2345 <pre><result> = vset <op>, <n x <ty>> <var1>, <var2> <i>; yields <n x bool></i>
2350 <p>The '<tt>vset</tt>' instruction returns a vector of boolean
2351 values representing, at each position, the result of the comparison
2352 between the values at that position in the two operands.</p>
2356 <p>The arguments to a '<tt>vset</tt>' instruction are a comparison
2357 operation and two value arguments. The value arguments must be of <a
2358 href="#t_packed">packed</a> type, and they must have identical types.
2359 For value arguments of integral element type, the operation argument
2360 must be one of <tt>eq</tt>, <tt>ne</tt>, <tt>lt</tt>, <tt>gt</tt>,
2361 <tt>le</tt>, <tt>ge</tt>, <tt>ult</tt>, <tt>ugt</tt>, <tt>ule</tt>,
2362 <tt>uge</tt>, <tt>true</tt>, and <tt>false</tt>. For value arguments
2363 of floating point element type, the operation argument must be one of
2364 <tt>eq</tt>, <tt>ne</tt>, <tt>lt</tt>, <tt>gt</tt>, <tt>le</tt>,
2365 <tt>ge</tt>, <tt>oeq</tt>, <tt>one</tt>, <tt>olt</tt>, <tt>ogt</tt>,
2366 <tt>ole</tt>, <tt>oge</tt>, <tt>ueq</tt>, <tt>une</tt>, <tt>ult</tt>,
2367 <tt>ugt</tt>, <tt>ule</tt>, <tt>uge</tt>, <tt>o</tt>, <tt>u</tt>,
2368 <tt>true</tt>, and <tt>false</tt>. The result is a packed
2369 <tt>bool</tt> value with the same length as each operand.</p>
2373 <p>The following table shows the semantics of '<tt>vset</tt>' for
2374 integral value arguments. For each position of the result, the
2375 comparison is done on the corresponding positions of the two value
2376 arguments. Note that the signedness of the comparison depends on the
2377 comparison opcode and <i>not</i> on the signedness of the value
2378 operands. E.g., <tt>vset lt <4 x unsigned> %x, %y</tt> does an
2379 elementwise <i>signed</i> comparison of <tt>%x</tt> and
2382 <table border="1" cellspacing="0" cellpadding="4">
2384 <tr><th>Operation</th><th>Result is true iff</th><th>Comparison is</th></tr>
2385 <tr><td><tt>eq</tt></td><td>var1 == var2</td><td>--</td></tr>
2386 <tr><td><tt>ne</tt></td><td>var1 != var2</td><td>--</td></tr>
2387 <tr><td><tt>lt</tt></td><td>var1 < var2</td><td>signed</td></tr>
2388 <tr><td><tt>gt</tt></td><td>var1 > var2</td><td>signed</td></tr>
2389 <tr><td><tt>le</tt></td><td>var1 <= var2</td><td>signed</td></tr>
2390 <tr><td><tt>ge</tt></td><td>var1 >= var2</td><td>signed</td></tr>
2391 <tr><td><tt>ult</tt></td><td>var1 < var2</td><td>unsigned</td></tr>
2392 <tr><td><tt>ugt</tt></td><td>var1 > var2</td><td>unsigned</td></tr>
2393 <tr><td><tt>ule</tt></td><td>var1 <= var2</td><td>unsigned</td></tr>
2394 <tr><td><tt>uge</tt></td><td>var1 >= var2</td><td>unsigned</td></tr>
2395 <tr><td><tt>true</tt></td><td>always</td><td>--</td></tr>
2396 <tr><td><tt>false</tt></td><td>never</td><td>--</td></tr>
2400 <p>The following table shows the semantics of '<tt>vset</tt>' for
2401 floating point types. If either operand is a floating point Not a
2402 Number (NaN) value, the operation is unordered, and the value in the
2403 first column below is produced at that position. Otherwise, the
2404 operation is ordered, and the value in the second column is
2407 <table border="1" cellspacing="0" cellpadding="4">
2409 <tr><th>Operation</th><th>If unordered<th>Otherwise true iff</th></tr>
2410 <tr><td><tt>eq</tt></td><td>undefined</td><td>var1 == var2</td></tr>
2411 <tr><td><tt>ne</tt></td><td>undefined</td><td>var1 != var2</td></tr>
2412 <tr><td><tt>lt</tt></td><td>undefined</td><td>var1 < var2</td></tr>
2413 <tr><td><tt>gt</tt></td><td>undefined</td><td>var1 > var2</td></tr>
2414 <tr><td><tt>le</tt></td><td>undefined</td><td>var1 <= var2</td></tr>
2415 <tr><td><tt>ge</tt></td><td>undefined</td><td>var1 >= var2</td></tr>
2416 <tr><td><tt>oeq</tt></td><td>false</td><td>var1 == var2</td></tr>
2417 <tr><td><tt>one</tt></td><td>false</td><td>var1 != var2</td></tr>
2418 <tr><td><tt>olt</tt></td><td>false</td><td>var1 < var2</td></tr>
2419 <tr><td><tt>ogt</tt></td><td>false</td><td>var1 > var2</td></tr>
2420 <tr><td><tt>ole</tt></td><td>false</td><td>var1 <= var2</td></tr>
2421 <tr><td><tt>oge</tt></td><td>false</td><td>var1 >= var2</td></tr>
2422 <tr><td><tt>ueq</tt></td><td>true</td><td>var1 == var2</td></tr>
2423 <tr><td><tt>une</tt></td><td>true</td><td>var1 != var2</td></tr>
2424 <tr><td><tt>ult</tt></td><td>true</td><td>var1 < var2</td></tr>
2425 <tr><td><tt>ugt</tt></td><td>true</td><td>var1 > var2</td></tr>
2426 <tr><td><tt>ule</tt></td><td>true</td><td>var1 <= var2</td></tr>
2427 <tr><td><tt>uge</tt></td><td>true</td><td>var1 >= var2</td></tr>
2428 <tr><td><tt>o</tt></td><td>false</td><td>always</td></tr>
2429 <tr><td><tt>u</tt></td><td>true</td><td>never</td></tr>
2430 <tr><td><tt>true</tt></td><td>true</td><td>always</td></tr>
2431 <tr><td><tt>false</tt></td><td>false</td><td>never</td></tr>
2436 <pre> <result> = vset eq <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = false, false</i>
2437 <result> = vset ne <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = true, true</i>
2438 <result> = vset lt <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = true, false</i>
2439 <result> = vset gt <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = false, true</i>
2440 <result> = vset le <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = true, false</i>
2441 <result> = vset ge <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = false, true</i>
2445 <!-- _______________________________________________________________________ -->
2446 <div class="doc_subsubsection">
2447 <a name="i_vselect">'<tt>vselect</tt>' Instruction</a>
2450 <div class="doc_text">
2455 <result> = vselect <n x bool> <cond>, <n x <ty>> <val1>, <n x <ty>> <val2> <i>; yields <n x <ty>></i>
2461 The '<tt>vselect</tt>' instruction chooses one value at each position
2462 of a vector based on a condition.
2469 The '<tt>vselect</tt>' instruction requires a <a
2470 href="#t_packed">packed</a> <tt>bool</tt> value indicating the
2471 condition at each vector position, and two values of the same packed
2472 type. All three operands must have the same length. The type of the
2473 result is the same as the type of the two value operands.</p>
2478 At each position where the <tt>bool</tt> vector is true, that position
2479 of the result gets its value from the first value argument; otherwise,
2480 it gets its value from the second value argument.
2486 %X = vselect bool <2 x bool> <bool true, bool false>, <2 x ubyte> <ubyte 17, ubyte 17>,
2487 <2 x ubyte> <ubyte 42, ubyte 42> <i>; yields <2 x ubyte>:17, 42</i>
2491 <!-- _______________________________________________________________________ -->
2492 <div class="doc_subsubsection">
2493 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2496 <div class="doc_text">
2501 <result> = extractelement <n x <ty>> <val>, uint <idx> <i>; yields <ty></i>
2507 The '<tt>extractelement</tt>' instruction extracts a single scalar
2508 element from a packed vector at a specified index.
2515 The first operand of an '<tt>extractelement</tt>' instruction is a
2516 value of <a href="#t_packed">packed</a> type. The second operand is
2517 an index indicating the position from which to extract the element.
2518 The index may be a variable.</p>
2523 The result is a scalar of the same type as the element type of
2524 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2525 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2526 results are undefined.
2532 %result = extractelement <4 x int> %vec, uint 0 <i>; yields int</i>
2537 <!-- _______________________________________________________________________ -->
2538 <div class="doc_subsubsection">
2539 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2542 <div class="doc_text">
2547 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, uint <idx> <i>; yields <n x <ty>></i>
2553 The '<tt>insertelement</tt>' instruction inserts a scalar
2554 element into a packed vector at a specified index.
2561 The first operand of an '<tt>insertelement</tt>' instruction is a
2562 value of <a href="#t_packed">packed</a> type. The second operand is a
2563 scalar value whose type must equal the element type of the first
2564 operand. The third operand is an index indicating the position at
2565 which to insert the value. The index may be a variable.</p>
2570 The result is a packed vector of the same type as <tt>val</tt>. Its
2571 element values are those of <tt>val</tt> except at position
2572 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2573 exceeds the length of <tt>val</tt>, the results are undefined.
2579 %result = insertelement <4 x int> %vec, int 1, uint 0 <i>; yields <4 x int></i>
2584 <!-- _______________________________________________________________________ -->
2585 <div class="doc_subsubsection">
2586 <a name="i_call">'<tt>call</tt>' Instruction</a>
2589 <div class="doc_text">
2593 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
2598 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
2602 <p>This instruction requires several arguments:</p>
2606 <p>The optional "tail" marker indicates whether the callee function accesses
2607 any allocas or varargs in the caller. If the "tail" marker is present, the
2608 function call is eligible for tail call optimization. Note that calls may
2609 be marked "tail" even if they do not occur before a <a
2610 href="#i_ret"><tt>ret</tt></a> instruction.
2613 <p>The optional "cconv" marker indicates which <a href="callingconv">calling
2614 convention</a> the call should use. If none is specified, the call defaults
2615 to using C calling conventions.
2618 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
2619 being invoked. The argument types must match the types implied by this
2620 signature. This type can be omitted if the function is not varargs and
2621 if the function type does not return a pointer to a function.</p>
2624 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
2625 be invoked. In most cases, this is a direct function invocation, but
2626 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
2627 to function value.</p>
2630 <p>'<tt>function args</tt>': argument list whose types match the
2631 function signature argument types. All arguments must be of
2632 <a href="#t_firstclass">first class</a> type. If the function signature
2633 indicates the function accepts a variable number of arguments, the extra
2634 arguments can be specified.</p>
2640 <p>The '<tt>call</tt>' instruction is used to cause control flow to
2641 transfer to a specified function, with its incoming arguments bound to
2642 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
2643 instruction in the called function, control flow continues with the
2644 instruction after the function call, and the return value of the
2645 function is bound to the result argument. This is a simpler case of
2646 the <a href="#i_invoke">invoke</a> instruction.</p>
2651 %retval = call int %test(int %argc)
2652 call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);
2653 %X = tail call int %foo()
2654 %Y = tail call <a href="#callingconv">fastcc</a> int %foo()
2659 <!-- _______________________________________________________________________ -->
2660 <div class="doc_subsubsection">
2661 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
2664 <div class="doc_text">
2669 <resultval> = va_arg <va_list*> <arglist>, <argty>
2674 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
2675 the "variable argument" area of a function call. It is used to implement the
2676 <tt>va_arg</tt> macro in C.</p>
2680 <p>This instruction takes a <tt>va_list*</tt> value and the type of
2681 the argument. It returns a value of the specified argument type and
2682 increments the <tt>va_list</tt> to point to the next argument. Again, the
2683 actual type of <tt>va_list</tt> is target specific.</p>
2687 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
2688 type from the specified <tt>va_list</tt> and causes the
2689 <tt>va_list</tt> to point to the next argument. For more information,
2690 see the variable argument handling <a href="#int_varargs">Intrinsic
2693 <p>It is legal for this instruction to be called in a function which does not
2694 take a variable number of arguments, for example, the <tt>vfprintf</tt>
2697 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
2698 href="#intrinsics">intrinsic function</a> because it takes a type as an
2703 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
2707 <!-- *********************************************************************** -->
2708 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
2709 <!-- *********************************************************************** -->
2711 <div class="doc_text">
2713 <p>LLVM supports the notion of an "intrinsic function". These functions have
2714 well known names and semantics and are required to follow certain
2715 restrictions. Overall, these instructions represent an extension mechanism for
2716 the LLVM language that does not require changing all of the transformations in
2717 LLVM to add to the language (or the bytecode reader/writer, the parser,
2720 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
2721 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
2722 this. Intrinsic functions must always be external functions: you cannot define
2723 the body of intrinsic functions. Intrinsic functions may only be used in call
2724 or invoke instructions: it is illegal to take the address of an intrinsic
2725 function. Additionally, because intrinsic functions are part of the LLVM
2726 language, it is required that they all be documented here if any are added.</p>
2729 <p>To learn how to add an intrinsic function, please see the <a
2730 href="ExtendingLLVM.html">Extending LLVM Guide</a>.
2735 <!-- ======================================================================= -->
2736 <div class="doc_subsection">
2737 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
2740 <div class="doc_text">
2742 <p>Variable argument support is defined in LLVM with the <a
2743 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
2744 intrinsic functions. These functions are related to the similarly
2745 named macros defined in the <tt><stdarg.h></tt> header file.</p>
2747 <p>All of these functions operate on arguments that use a
2748 target-specific value type "<tt>va_list</tt>". The LLVM assembly
2749 language reference manual does not define what this type is, so all
2750 transformations should be prepared to handle intrinsics with any type
2753 <p>This example shows how the <a href="#i_vanext"><tt>vanext</tt></a>
2754 instruction and the variable argument handling intrinsic functions are
2758 int %test(int %X, ...) {
2759 ; Initialize variable argument processing
2761 call void %<a href="#i_va_start">llvm.va_start</a>(sbyte** %ap)
2763 ; Read a single integer argument
2764 %tmp = va_arg sbyte** %ap, int
2766 ; Demonstrate usage of llvm.va_copy and llvm.va_end
2768 call void %<a href="#i_va_copy">llvm.va_copy</a>(sbyte** %aq, sbyte** %ap)
2769 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %aq)
2771 ; Stop processing of arguments.
2772 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %ap)
2778 <!-- _______________________________________________________________________ -->
2779 <div class="doc_subsubsection">
2780 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
2784 <div class="doc_text">
2786 <pre> declare void %llvm.va_start(<va_list>* <arglist>)<br></pre>
2788 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
2789 <tt>*<arglist></tt> for subsequent use by <tt><a
2790 href="#i_va_arg">va_arg</a></tt>.</p>
2794 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
2798 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
2799 macro available in C. In a target-dependent way, it initializes the
2800 <tt>va_list</tt> element the argument points to, so that the next call to
2801 <tt>va_arg</tt> will produce the first variable argument passed to the function.
2802 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
2803 last argument of the function, the compiler can figure that out.</p>
2807 <!-- _______________________________________________________________________ -->
2808 <div class="doc_subsubsection">
2809 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
2812 <div class="doc_text">
2814 <pre> declare void %llvm.va_end(<va_list*> <arglist>)<br></pre>
2816 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
2817 which has been initialized previously with <tt><a href="#i_va_start">llvm.va_start</a></tt>
2818 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
2820 <p>The argument is a <tt>va_list</tt> to destroy.</p>
2822 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
2823 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
2824 Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and <a
2825 href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
2826 with calls to <tt>llvm.va_end</tt>.</p>
2829 <!-- _______________________________________________________________________ -->
2830 <div class="doc_subsubsection">
2831 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
2834 <div class="doc_text">
2839 declare void %llvm.va_copy(<va_list>* <destarglist>,
2840 <va_list>* <srcarglist>)
2845 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
2846 the source argument list to the destination argument list.</p>
2850 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
2851 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
2856 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
2857 available in C. In a target-dependent way, it copies the source
2858 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
2859 because the <tt><a href="i_va_begin">llvm.va_begin</a></tt> intrinsic may be
2860 arbitrarily complex and require memory allocation, for example.</p>
2864 <!-- ======================================================================= -->
2865 <div class="doc_subsection">
2866 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
2869 <div class="doc_text">
2872 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
2873 Collection</a> requires the implementation and generation of these intrinsics.
2874 These intrinsics allow identification of <a href="#i_gcroot">GC roots on the
2875 stack</a>, as well as garbage collector implementations that require <a
2876 href="#i_gcread">read</a> and <a href="#i_gcwrite">write</a> barriers.
2877 Front-ends for type-safe garbage collected languages should generate these
2878 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
2879 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
2883 <!-- _______________________________________________________________________ -->
2884 <div class="doc_subsubsection">
2885 <a name="i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
2888 <div class="doc_text">
2893 declare void %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
2898 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
2899 the code generator, and allows some metadata to be associated with it.</p>
2903 <p>The first argument specifies the address of a stack object that contains the
2904 root pointer. The second pointer (which must be either a constant or a global
2905 value address) contains the meta-data to be associated with the root.</p>
2909 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
2910 location. At compile-time, the code generator generates information to allow
2911 the runtime to find the pointer at GC safe points.
2917 <!-- _______________________________________________________________________ -->
2918 <div class="doc_subsubsection">
2919 <a name="i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
2922 <div class="doc_text">
2927 declare sbyte* %llvm.gcread(sbyte** %Ptr)
2932 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
2933 locations, allowing garbage collector implementations that require read
2938 <p>The argument is the address to read from, which should be an address
2939 allocated from the garbage collector.</p>
2943 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
2944 instruction, but may be replaced with substantially more complex code by the
2945 garbage collector runtime, as needed.</p>
2950 <!-- _______________________________________________________________________ -->
2951 <div class="doc_subsubsection">
2952 <a name="i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
2955 <div class="doc_text">
2960 declare void %llvm.gcwrite(sbyte* %P1, sbyte** %P2)
2965 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
2966 locations, allowing garbage collector implementations that require write
2967 barriers (such as generational or reference counting collectors).</p>
2971 <p>The first argument is the reference to store, and the second is the heap
2972 location to store to.</p>
2976 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
2977 instruction, but may be replaced with substantially more complex code by the
2978 garbage collector runtime, as needed.</p>
2984 <!-- ======================================================================= -->
2985 <div class="doc_subsection">
2986 <a name="int_codegen">Code Generator Intrinsics</a>
2989 <div class="doc_text">
2991 These intrinsics are provided by LLVM to expose special features that may only
2992 be implemented with code generator support.
2997 <!-- _______________________________________________________________________ -->
2998 <div class="doc_subsubsection">
2999 <a name="i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
3002 <div class="doc_text">
3006 declare sbyte *%llvm.returnaddress(uint <level>)
3012 The '<tt>llvm.returnaddress</tt>' intrinsic returns a target-specific value
3013 indicating the return address of the current function or one of its callers.
3019 The argument to this intrinsic indicates which function to return the address
3020 for. Zero indicates the calling function, one indicates its caller, etc. The
3021 argument is <b>required</b> to be a constant integer value.
3027 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
3028 the return address of the specified call frame, or zero if it cannot be
3029 identified. The value returned by this intrinsic is likely to be incorrect or 0
3030 for arguments other than zero, so it should only be used for debugging purposes.
3034 Note that calling this intrinsic does not prevent function inlining or other
3035 aggressive transformations, so the value returned may not be that of the obvious
3036 source-language caller.
3041 <!-- _______________________________________________________________________ -->
3042 <div class="doc_subsubsection">
3043 <a name="i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
3046 <div class="doc_text">
3050 declare sbyte *%llvm.frameaddress(uint <level>)
3056 The '<tt>llvm.frameaddress</tt>' intrinsic returns the target-specific frame
3057 pointer value for the specified stack frame.
3063 The argument to this intrinsic indicates which function to return the frame
3064 pointer for. Zero indicates the calling function, one indicates its caller,
3065 etc. The argument is <b>required</b> to be a constant integer value.
3071 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
3072 the frame address of the specified call frame, or zero if it cannot be
3073 identified. The value returned by this intrinsic is likely to be incorrect or 0
3074 for arguments other than zero, so it should only be used for debugging purposes.
3078 Note that calling this intrinsic does not prevent function inlining or other
3079 aggressive transformations, so the value returned may not be that of the obvious
3080 source-language caller.
3084 <!-- _______________________________________________________________________ -->
3085 <div class="doc_subsubsection">
3086 <a name="i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
3089 <div class="doc_text">
3093 declare sbyte *%llvm.stacksave()
3099 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
3100 the function stack, for use with <a href="#i_stackrestore">
3101 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
3102 features like scoped automatic variable sized arrays in C99.
3108 This intrinsic returns a opaque pointer value that can be passed to <a
3109 href="#i_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
3110 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
3111 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
3112 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
3113 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
3114 that were allocated after the <tt>llvm.stacksave</tt> was executed.
3119 <!-- _______________________________________________________________________ -->
3120 <div class="doc_subsubsection">
3121 <a name="i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
3124 <div class="doc_text">
3128 declare void %llvm.stackrestore(sbyte* %ptr)
3134 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
3135 the function stack to the state it was in when the corresponding <a
3136 href="#llvm.stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
3137 useful for implementing language features like scoped automatic variable sized
3144 See the description for <a href="#i_stacksave"><tt>llvm.stacksave</tt></a>.
3150 <!-- _______________________________________________________________________ -->
3151 <div class="doc_subsubsection">
3152 <a name="i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
3155 <div class="doc_text">
3159 declare void %llvm.prefetch(sbyte * <address>,
3160 uint <rw>, uint <locality>)
3167 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
3168 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
3170 effect on the behavior of the program but can change its performance
3177 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
3178 determining if the fetch should be for a read (0) or write (1), and
3179 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
3180 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
3181 <tt>locality</tt> arguments must be constant integers.
3187 This intrinsic does not modify the behavior of the program. In particular,
3188 prefetches cannot trap and do not produce a value. On targets that support this
3189 intrinsic, the prefetch can provide hints to the processor cache for better
3195 <!-- _______________________________________________________________________ -->
3196 <div class="doc_subsubsection">
3197 <a name="i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
3200 <div class="doc_text">
3204 declare void %llvm.pcmarker( uint <id> )
3211 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
3213 code to simulators and other tools. The method is target specific, but it is
3214 expected that the marker will use exported symbols to transmit the PC of the marker.
3215 The marker makes no guarantees that it will remain with any specific instruction
3216 after optimizations. It is possible that the presence of a marker will inhibit
3217 optimizations. The intended use is to be inserted after optmizations to allow
3218 correlations of simulation runs.
3224 <tt>id</tt> is a numerical id identifying the marker.
3230 This intrinsic does not modify the behavior of the program. Backends that do not
3231 support this intrinisic may ignore it.
3236 <!-- _______________________________________________________________________ -->
3237 <div class="doc_subsubsection">
3238 <a name="i_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
3241 <div class="doc_text">
3245 declare ulong %llvm.readcyclecounter( )
3252 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
3253 counter register (or similar low latency, high accuracy clocks) on those targets
3254 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
3255 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
3256 should only be used for small timings.
3262 When directly supported, reading the cycle counter should not modify any memory.
3263 Implementations are allowed to either return a application specific value or a
3264 system wide value. On backends without support, this is lowered to a constant 0.
3270 <!-- ======================================================================= -->
3271 <div class="doc_subsection">
3272 <a name="int_os">Operating System Intrinsics</a>
3275 <div class="doc_text">
3277 These intrinsics are provided by LLVM to support the implementation of
3278 operating system level code.
3283 <!-- _______________________________________________________________________ -->
3284 <div class="doc_subsubsection">
3285 <a name="i_readport">'<tt>llvm.readport</tt>' Intrinsic</a>
3288 <div class="doc_text">
3292 declare <integer type> %llvm.readport (<integer type> <address>)
3298 The '<tt>llvm.readport</tt>' intrinsic reads data from the specified hardware
3305 The argument to this intrinsic indicates the hardware I/O address from which
3306 to read the data. The address is in the hardware I/O address namespace (as
3307 opposed to being a memory location for memory mapped I/O).
3313 The '<tt>llvm.readport</tt>' intrinsic reads data from the hardware I/O port
3314 specified by <i>address</i> and returns the value. The address and return
3315 value must be integers, but the size is dependent upon the platform upon which
3316 the program is code generated. For example, on x86, the address must be an
3317 unsigned 16-bit value, and the return value must be 8, 16, or 32 bits.
3322 <!-- _______________________________________________________________________ -->
3323 <div class="doc_subsubsection">
3324 <a name="i_writeport">'<tt>llvm.writeport</tt>' Intrinsic</a>
3327 <div class="doc_text">
3331 call void (<integer type>, <integer type>)*
3332 %llvm.writeport (<integer type> <value>,
3333 <integer type> <address>)
3339 The '<tt>llvm.writeport</tt>' intrinsic writes data to the specified hardware
3346 The first argument is the value to write to the I/O port.
3350 The second argument indicates the hardware I/O address to which data should be
3351 written. The address is in the hardware I/O address namespace (as opposed to
3352 being a memory location for memory mapped I/O).
3358 The '<tt>llvm.writeport</tt>' intrinsic writes <i>value</i> to the I/O port
3359 specified by <i>address</i>. The address and value must be integers, but the
3360 size is dependent upon the platform upon which the program is code generated.
3361 For example, on x86, the address must be an unsigned 16-bit value, and the
3362 value written must be 8, 16, or 32 bits in length.
3367 <!-- _______________________________________________________________________ -->
3368 <div class="doc_subsubsection">
3369 <a name="i_readio">'<tt>llvm.readio</tt>' Intrinsic</a>
3372 <div class="doc_text">
3376 declare <result> %llvm.readio (<ty> * <pointer>)
3382 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
3389 The argument to this intrinsic is a pointer indicating the memory address from
3390 which to read the data. The data must be a
3391 <a href="#t_firstclass">first class</a> type.
3397 The '<tt>llvm.readio</tt>' intrinsic reads data from a memory mapped I/O
3398 location specified by <i>pointer</i> and returns the value. The argument must
3399 be a pointer, and the return value must be a
3400 <a href="#t_firstclass">first class</a> type. However, certain architectures
3401 may not support I/O on all first class types. For example, 32-bit processors
3402 may only support I/O on data types that are 32 bits or less.
3406 This intrinsic enforces an in-order memory model for llvm.readio and
3407 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
3408 scheduled processors may execute loads and stores out of order, re-ordering at
3409 run time accesses to memory mapped I/O registers. Using these intrinsics
3410 ensures that accesses to memory mapped I/O registers occur in program order.
3415 <!-- _______________________________________________________________________ -->
3416 <div class="doc_subsubsection">
3417 <a name="i_writeio">'<tt>llvm.writeio</tt>' Intrinsic</a>
3420 <div class="doc_text">
3424 declare void %llvm.writeio (<ty1> <value>, <ty2> * <pointer>)
3430 The '<tt>llvm.writeio</tt>' intrinsic writes data to the specified memory
3437 The first argument is the value to write to the memory mapped I/O location.
3438 The second argument is a pointer indicating the memory address to which the
3439 data should be written.
3445 The '<tt>llvm.writeio</tt>' intrinsic writes <i>value</i> to the memory mapped
3446 I/O address specified by <i>pointer</i>. The value must be a
3447 <a href="#t_firstclass">first class</a> type. However, certain architectures
3448 may not support I/O on all first class types. For example, 32-bit processors
3449 may only support I/O on data types that are 32 bits or less.
3453 This intrinsic enforces an in-order memory model for llvm.readio and
3454 llvm.writeio calls on machines that use dynamic scheduling. Dynamically
3455 scheduled processors may execute loads and stores out of order, re-ordering at
3456 run time accesses to memory mapped I/O registers. Using these intrinsics
3457 ensures that accesses to memory mapped I/O registers occur in program order.
3462 <!-- ======================================================================= -->
3463 <div class="doc_subsection">
3464 <a name="int_libc">Standard C Library Intrinsics</a>
3467 <div class="doc_text">
3469 LLVM provides intrinsics for a few important standard C library functions.
3470 These intrinsics allow source-language front-ends to pass information about the
3471 alignment of the pointer arguments to the code generator, providing opportunity
3472 for more efficient code generation.
3477 <!-- _______________________________________________________________________ -->
3478 <div class="doc_subsubsection">
3479 <a name="i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
3482 <div class="doc_text">
3486 declare void %llvm.memcpy(sbyte* <dest>, sbyte* <src>,
3487 uint <len>, uint <align>)
3493 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
3494 location to the destination location.
3498 Note that, unlike the standard libc function, the <tt>llvm.memcpy</tt> intrinsic
3499 does not return a value, and takes an extra alignment argument.
3505 The first argument is a pointer to the destination, the second is a pointer to
3506 the source. The third argument is an (arbitrarily sized) integer argument
3507 specifying the number of bytes to copy, and the fourth argument is the alignment
3508 of the source and destination locations.
3512 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3513 the caller guarantees that the size of the copy is a multiple of the alignment
3514 and that both the source and destination pointers are aligned to that boundary.
3520 The '<tt>llvm.memcpy</tt>' intrinsic copies a block of memory from the source
3521 location to the destination location, which are not allowed to overlap. It
3522 copies "len" bytes of memory over. If the argument is known to be aligned to
3523 some boundary, this can be specified as the fourth argument, otherwise it should
3529 <!-- _______________________________________________________________________ -->
3530 <div class="doc_subsubsection">
3531 <a name="i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
3534 <div class="doc_text">
3538 declare void %llvm.memmove(sbyte* <dest>, sbyte* <src>,
3539 uint <len>, uint <align>)
3545 The '<tt>llvm.memmove</tt>' intrinsic moves a block of memory from the source
3546 location to the destination location. It is similar to the '<tt>llvm.memcpy</tt>'
3547 intrinsic but allows the two memory locations to overlap.
3551 Note that, unlike the standard libc function, the <tt>llvm.memmove</tt> intrinsic
3552 does not return a value, and takes an extra alignment argument.
3558 The first argument is a pointer to the destination, the second is a pointer to
3559 the source. The third argument is an (arbitrarily sized) integer argument
3560 specifying the number of bytes to copy, and the fourth argument is the alignment
3561 of the source and destination locations.
3565 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3566 the caller guarantees that the size of the copy is a multiple of the alignment
3567 and that both the source and destination pointers are aligned to that boundary.
3573 The '<tt>llvm.memmove</tt>' intrinsic copies a block of memory from the source
3574 location to the destination location, which may overlap. It
3575 copies "len" bytes of memory over. If the argument is known to be aligned to
3576 some boundary, this can be specified as the fourth argument, otherwise it should
3582 <!-- _______________________________________________________________________ -->
3583 <div class="doc_subsubsection">
3584 <a name="i_memset">'<tt>llvm.memset</tt>' Intrinsic</a>
3587 <div class="doc_text">
3591 declare void %llvm.memset(sbyte* <dest>, ubyte <val>,
3592 uint <len>, uint <align>)
3598 The '<tt>llvm.memset</tt>' intrinsic fills a block of memory with a particular
3603 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
3604 does not return a value, and takes an extra alignment argument.
3610 The first argument is a pointer to the destination to fill, the second is the
3611 byte value to fill it with, the third argument is an (arbitrarily sized) integer
3612 argument specifying the number of bytes to fill, and the fourth argument is the
3613 known alignment of destination location.
3617 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3618 the caller guarantees that the size of the copy is a multiple of the alignment
3619 and that the destination pointer is aligned to that boundary.
3625 The '<tt>llvm.memset</tt>' intrinsic fills "len" bytes of memory starting at the
3626 destination location. If the argument is known to be aligned to some boundary,
3627 this can be specified as the fourth argument, otherwise it should be set to 0 or
3633 <!-- _______________________________________________________________________ -->
3634 <div class="doc_subsubsection">
3635 <a name="i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a>
3638 <div class="doc_text">
3642 declare bool %llvm.isunordered.f32(float Val1, float Val2)
3643 declare bool %llvm.isunordered.f64(double Val1, double Val2)
3649 The '<tt>llvm.isunordered</tt>' intrinsics return true if either or both of the
3650 specified floating point values is a NAN.
3656 The arguments are floating point numbers of the same type.
3662 If either or both of the arguments is a SNAN or QNAN, it returns true, otherwise
3668 <!-- _______________________________________________________________________ -->
3669 <div class="doc_subsubsection">
3670 <a name="i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
3673 <div class="doc_text">
3677 declare double %llvm.sqrt.f32(float Val)
3678 declare double %llvm.sqrt.f64(double Val)
3684 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
3685 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
3686 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
3687 negative numbers (which allows for better optimization).
3693 The argument and return value are floating point numbers of the same type.
3699 This function returns the sqrt of the specified operand if it is a positive
3700 floating point number.
3704 <!-- ======================================================================= -->
3705 <div class="doc_subsection">
3706 <a name="int_manip">Bit Manipulation Intrinsics</a>
3709 <div class="doc_text">
3711 LLVM provides intrinsics for a few important bit manipulation operations.
3712 These allow efficient code generation for some algorithms.
3717 <!-- _______________________________________________________________________ -->
3718 <div class="doc_subsubsection">
3719 <a name="i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
3722 <div class="doc_text">
3726 declare ushort %llvm.bswap.i16(ushort <id>)
3727 declare uint %llvm.bswap.i32(uint <id>)
3728 declare ulong %llvm.bswap.i64(ulong <id>)
3734 The '<tt>llvm.bwsap</tt>' family of intrinsics is used to byteswap a 16, 32 or
3735 64 bit quantity. These are useful for performing operations on data that is not
3736 in the target's native byte order.
3742 The <tt>llvm.bswap.16</tt> intrinsic returns a ushort value that has the high and low
3743 byte of the input ushort swapped. Similarly, the <tt>llvm.bswap.i32</tt> intrinsic
3744 returns a uint value that has the four bytes of the input uint swapped, so that
3745 if the input bytes are numbered 0, 1, 2, 3 then the returned uint will have its
3746 bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i64</tt> intrinsic extends this concept
3752 <!-- _______________________________________________________________________ -->
3753 <div class="doc_subsubsection">
3754 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
3757 <div class="doc_text">
3761 declare ubyte %llvm.ctpop.i8 (ubyte <src>)
3762 declare ushort %llvm.ctpop.i16(ushort <src>)
3763 declare uint %llvm.ctpop.i32(uint <src>)
3764 declare ulong %llvm.ctpop.i64(ulong <src>)
3770 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
3777 The only argument is the value to be counted. The argument may be of any
3778 unsigned integer type. The return type must match the argument type.
3784 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
3788 <!-- _______________________________________________________________________ -->
3789 <div class="doc_subsubsection">
3790 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
3793 <div class="doc_text">
3797 declare ubyte %llvm.ctlz.i8 (ubyte <src>)
3798 declare ushort %llvm.ctlz.i16(ushort <src>)
3799 declare uint %llvm.ctlz.i32(uint <src>)
3800 declare ulong %llvm.ctlz.i64(ulong <src>)
3806 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
3807 leading zeros in a variable.
3813 The only argument is the value to be counted. The argument may be of any
3814 unsigned integer type. The return type must match the argument type.
3820 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
3821 in a variable. If the src == 0 then the result is the size in bits of the type
3822 of src. For example, <tt>llvm.cttz(int 2) = 30</tt>.
3828 <!-- _______________________________________________________________________ -->
3829 <div class="doc_subsubsection">
3830 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
3833 <div class="doc_text">
3837 declare ubyte %llvm.cttz.i8 (ubyte <src>)
3838 declare ushort %llvm.cttz.i16(ushort <src>)
3839 declare uint %llvm.cttz.i32(uint <src>)
3840 declare ulong %llvm.cttz.i64(ulong <src>)
3846 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
3853 The only argument is the value to be counted. The argument may be of any
3854 unsigned integer type. The return type must match the argument type.
3860 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
3861 in a variable. If the src == 0 then the result is the size in bits of the type
3862 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
3866 <!-- ======================================================================= -->
3867 <div class="doc_subsection">
3868 <a name="int_debugger">Debugger Intrinsics</a>
3871 <div class="doc_text">
3873 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
3874 are described in the <a
3875 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
3876 Debugging</a> document.
3881 <!-- *********************************************************************** -->
3884 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3885 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
3886 <a href="http://validator.w3.org/check/referer"><img
3887 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
3889 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3890 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br>
3891 Last modified: $Date$