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="#aliasstructure">Aliases</a>
28 <li><a href="#paramattrs">Parameter Attributes</a></li>
29 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
30 <li><a href="#datalayout">Data Layout</a></li>
33 <li><a href="#typesystem">Type System</a>
35 <li><a href="#t_primitive">Primitive Types</a>
37 <li><a href="#t_classifications">Type Classifications</a></li>
40 <li><a href="#t_derived">Derived Types</a>
42 <li><a href="#t_array">Array Type</a></li>
43 <li><a href="#t_function">Function Type</a></li>
44 <li><a href="#t_pointer">Pointer Type</a></li>
45 <li><a href="#t_struct">Structure Type</a></li>
46 <li><a href="#t_pstruct">Packed Structure Type</a></li>
47 <li><a href="#t_vector">Vector Type</a></li>
48 <li><a href="#t_opaque">Opaque Type</a></li>
53 <li><a href="#constants">Constants</a>
55 <li><a href="#simpleconstants">Simple Constants</a>
56 <li><a href="#aggregateconstants">Aggregate Constants</a>
57 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
58 <li><a href="#undefvalues">Undefined Values</a>
59 <li><a href="#constantexprs">Constant Expressions</a>
62 <li><a href="#othervalues">Other Values</a>
64 <li><a href="#inlineasm">Inline Assembler Expressions</a>
67 <li><a href="#instref">Instruction Reference</a>
69 <li><a href="#terminators">Terminator Instructions</a>
71 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
72 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
73 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
74 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
75 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
76 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
79 <li><a href="#binaryops">Binary Operations</a>
81 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
82 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
83 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
84 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
85 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
86 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
87 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
88 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
89 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
92 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
94 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
95 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
96 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
97 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
98 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
99 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
102 <li><a href="#vectorops">Vector Operations</a>
104 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
105 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
106 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
109 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
111 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
112 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
113 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
114 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
115 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
116 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
119 <li><a href="#convertops">Conversion Operations</a>
121 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
122 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
123 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
124 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
125 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
126 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
127 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
128 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
129 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
130 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
131 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
132 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
134 <li><a href="#otherops">Other Operations</a>
136 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
137 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
138 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
139 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
140 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
141 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
146 <li><a href="#intrinsics">Intrinsic Functions</a>
148 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
150 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
151 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
152 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
155 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
157 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
158 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
159 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
162 <li><a href="#int_codegen">Code Generator Intrinsics</a>
164 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
165 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
166 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
167 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
168 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
169 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
170 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
173 <li><a href="#int_libc">Standard C Library Intrinsics</a>
175 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
176 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
177 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
178 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
179 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
182 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
184 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
185 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
186 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
187 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
188 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
189 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
192 <li><a href="#int_debugger">Debugger intrinsics</a></li>
193 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
194 <li><a href="#int_atomics">Atomic Operations and Synchronization Intrinsics</a>
196 <li><a href="#int_lcs">'<tt>llvm.atomic.lcs.*</tt>' Intrinsic</a></li>
197 <li><a href="#int_ls">'<tt>llvm.atomic.ls.*</tt>' Intrinsic</a></li>
198 <li><a href="#int_las">'<tt>llvm.atomic.las.*</tt>' Intrinsic</a></li>
199 <li><a href="#int_lss">'<tt>llvm.atomic.lss.*</tt>' Intrinsic</a></li>
200 <li><a href="#int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a></li>
203 <li><a href="#int_trampoline">Trampoline Intrinsics</a>
205 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
206 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
209 <li><a href="#int_general">General intrinsics</a>
211 <li><a href="#int_var_annotation">
212 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
219 <div class="doc_author">
220 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
221 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
224 <!-- *********************************************************************** -->
225 <div class="doc_section"> <a name="abstract">Abstract </a></div>
226 <!-- *********************************************************************** -->
228 <div class="doc_text">
229 <p>This document is a reference manual for the LLVM assembly language.
230 LLVM is an SSA based representation that provides type safety,
231 low-level operations, flexibility, and the capability of representing
232 'all' high-level languages cleanly. It is the common code
233 representation used throughout all phases of the LLVM compilation
237 <!-- *********************************************************************** -->
238 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
239 <!-- *********************************************************************** -->
241 <div class="doc_text">
243 <p>The LLVM code representation is designed to be used in three
244 different forms: as an in-memory compiler IR, as an on-disk bitcode
245 representation (suitable for fast loading by a Just-In-Time compiler),
246 and as a human readable assembly language representation. This allows
247 LLVM to provide a powerful intermediate representation for efficient
248 compiler transformations and analysis, while providing a natural means
249 to debug and visualize the transformations. The three different forms
250 of LLVM are all equivalent. This document describes the human readable
251 representation and notation.</p>
253 <p>The LLVM representation aims to be light-weight and low-level
254 while being expressive, typed, and extensible at the same time. It
255 aims to be a "universal IR" of sorts, by being at a low enough level
256 that high-level ideas may be cleanly mapped to it (similar to how
257 microprocessors are "universal IR's", allowing many source languages to
258 be mapped to them). By providing type information, LLVM can be used as
259 the target of optimizations: for example, through pointer analysis, it
260 can be proven that a C automatic variable is never accessed outside of
261 the current function... allowing it to be promoted to a simple SSA
262 value instead of a memory location.</p>
266 <!-- _______________________________________________________________________ -->
267 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
269 <div class="doc_text">
271 <p>It is important to note that this document describes 'well formed'
272 LLVM assembly language. There is a difference between what the parser
273 accepts and what is considered 'well formed'. For example, the
274 following instruction is syntactically okay, but not well formed:</p>
276 <div class="doc_code">
278 %x = <a href="#i_add">add</a> i32 1, %x
282 <p>...because the definition of <tt>%x</tt> does not dominate all of
283 its uses. The LLVM infrastructure provides a verification pass that may
284 be used to verify that an LLVM module is well formed. This pass is
285 automatically run by the parser after parsing input assembly and by
286 the optimizer before it outputs bitcode. The violations pointed out
287 by the verifier pass indicate bugs in transformation passes or input to
291 <!-- Describe the typesetting conventions here. -->
293 <!-- *********************************************************************** -->
294 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
295 <!-- *********************************************************************** -->
297 <div class="doc_text">
299 <p>LLVM uses three different forms of identifiers, for different
303 <li>Named values are represented as a string of characters with a '%' prefix.
304 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
305 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
306 Identifiers which require other characters in their names can be surrounded
307 with quotes. In this way, anything except a <tt>"</tt> character can be used
310 <li>Unnamed values are represented as an unsigned numeric value with a '%'
311 prefix. For example, %12, %2, %44.</li>
313 <li>Constants, which are described in a <a href="#constants">section about
314 constants</a>, below.</li>
317 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
318 don't need to worry about name clashes with reserved words, and the set of
319 reserved words may be expanded in the future without penalty. Additionally,
320 unnamed identifiers allow a compiler to quickly come up with a temporary
321 variable without having to avoid symbol table conflicts.</p>
323 <p>Reserved words in LLVM are very similar to reserved words in other
324 languages. There are keywords for different opcodes
325 ('<tt><a href="#i_add">add</a></tt>',
326 '<tt><a href="#i_bitcast">bitcast</a></tt>',
327 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
328 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
329 and others. These reserved words cannot conflict with variable names, because
330 none of them start with a '%' character.</p>
332 <p>Here is an example of LLVM code to multiply the integer variable
333 '<tt>%X</tt>' by 8:</p>
337 <div class="doc_code">
339 %result = <a href="#i_mul">mul</a> i32 %X, 8
343 <p>After strength reduction:</p>
345 <div class="doc_code">
347 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
351 <p>And the hard way:</p>
353 <div class="doc_code">
355 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
356 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
357 %result = <a href="#i_add">add</a> i32 %1, %1
361 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
362 important lexical features of LLVM:</p>
366 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
369 <li>Unnamed temporaries are created when the result of a computation is not
370 assigned to a named value.</li>
372 <li>Unnamed temporaries are numbered sequentially</li>
376 <p>...and it also shows a convention that we follow in this document. When
377 demonstrating instructions, we will follow an instruction with a comment that
378 defines the type and name of value produced. Comments are shown in italic
383 <!-- *********************************************************************** -->
384 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
385 <!-- *********************************************************************** -->
387 <!-- ======================================================================= -->
388 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
391 <div class="doc_text">
393 <p>LLVM programs are composed of "Module"s, each of which is a
394 translation unit of the input programs. Each module consists of
395 functions, global variables, and symbol table entries. Modules may be
396 combined together with the LLVM linker, which merges function (and
397 global variable) definitions, resolves forward declarations, and merges
398 symbol table entries. Here is an example of the "hello world" module:</p>
400 <div class="doc_code">
401 <pre><i>; Declare the string constant as a global constant...</i>
402 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
403 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
405 <i>; External declaration of the puts function</i>
406 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
408 <i>; Definition of main function</i>
409 define i32 @main() { <i>; i32()* </i>
410 <i>; Convert [13x i8 ]* to i8 *...</i>
412 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
414 <i>; Call puts function to write out the string to stdout...</i>
416 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
418 href="#i_ret">ret</a> i32 0<br>}<br>
422 <p>This example is made up of a <a href="#globalvars">global variable</a>
423 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
424 function, and a <a href="#functionstructure">function definition</a>
425 for "<tt>main</tt>".</p>
427 <p>In general, a module is made up of a list of global values,
428 where both functions and global variables are global values. Global values are
429 represented by a pointer to a memory location (in this case, a pointer to an
430 array of char, and a pointer to a function), and have one of the following <a
431 href="#linkage">linkage types</a>.</p>
435 <!-- ======================================================================= -->
436 <div class="doc_subsection">
437 <a name="linkage">Linkage Types</a>
440 <div class="doc_text">
443 All Global Variables and Functions have one of the following types of linkage:
448 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
450 <dd>Global values with internal linkage are only directly accessible by
451 objects in the current module. In particular, linking code into a module with
452 an internal global value may cause the internal to be renamed as necessary to
453 avoid collisions. Because the symbol is internal to the module, all
454 references can be updated. This corresponds to the notion of the
455 '<tt>static</tt>' keyword in C.
458 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
460 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
461 the same name when linkage occurs. This is typically used to implement
462 inline functions, templates, or other code which must be generated in each
463 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
464 allowed to be discarded.
467 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
469 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
470 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
471 used for globals that may be emitted in multiple translation units, but that
472 are not guaranteed to be emitted into every translation unit that uses them.
473 One example of this are common globals in C, such as "<tt>int X;</tt>" at
477 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
479 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
480 pointer to array type. When two global variables with appending linkage are
481 linked together, the two global arrays are appended together. This is the
482 LLVM, typesafe, equivalent of having the system linker append together
483 "sections" with identical names when .o files are linked.
486 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
487 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
488 until linked, if not linked, the symbol becomes null instead of being an
492 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
494 <dd>If none of the above identifiers are used, the global is externally
495 visible, meaning that it participates in linkage and can be used to resolve
496 external symbol references.
501 The next two types of linkage are targeted for Microsoft Windows platform
502 only. They are designed to support importing (exporting) symbols from (to)
507 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
509 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
510 or variable via a global pointer to a pointer that is set up by the DLL
511 exporting the symbol. On Microsoft Windows targets, the pointer name is
512 formed by combining <code>_imp__</code> and the function or variable name.
515 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
517 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
518 pointer to a pointer in a DLL, so that it can be referenced with the
519 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
520 name is formed by combining <code>_imp__</code> and the function or variable
526 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
527 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
528 variable and was linked with this one, one of the two would be renamed,
529 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
530 external (i.e., lacking any linkage declarations), they are accessible
531 outside of the current module.</p>
532 <p>It is illegal for a function <i>declaration</i>
533 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
534 or <tt>extern_weak</tt>.</p>
535 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
539 <!-- ======================================================================= -->
540 <div class="doc_subsection">
541 <a name="callingconv">Calling Conventions</a>
544 <div class="doc_text">
546 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
547 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
548 specified for the call. The calling convention of any pair of dynamic
549 caller/callee must match, or the behavior of the program is undefined. The
550 following calling conventions are supported by LLVM, and more may be added in
554 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
556 <dd>This calling convention (the default if no other calling convention is
557 specified) matches the target C calling conventions. This calling convention
558 supports varargs function calls and tolerates some mismatch in the declared
559 prototype and implemented declaration of the function (as does normal C).
562 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
564 <dd>This calling convention attempts to make calls as fast as possible
565 (e.g. by passing things in registers). This calling convention allows the
566 target to use whatever tricks it wants to produce fast code for the target,
567 without having to conform to an externally specified ABI. Implementations of
568 this convention should allow arbitrary tail call optimization to be supported.
569 This calling convention does not support varargs and requires the prototype of
570 all callees to exactly match the prototype of the function definition.
573 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
575 <dd>This calling convention attempts to make code in the caller as efficient
576 as possible under the assumption that the call is not commonly executed. As
577 such, these calls often preserve all registers so that the call does not break
578 any live ranges in the caller side. This calling convention does not support
579 varargs and requires the prototype of all callees to exactly match the
580 prototype of the function definition.
583 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
585 <dd>Any calling convention may be specified by number, allowing
586 target-specific calling conventions to be used. Target specific calling
587 conventions start at 64.
591 <p>More calling conventions can be added/defined on an as-needed basis, to
592 support pascal conventions or any other well-known target-independent
597 <!-- ======================================================================= -->
598 <div class="doc_subsection">
599 <a name="visibility">Visibility Styles</a>
602 <div class="doc_text">
605 All Global Variables and Functions have one of the following visibility styles:
609 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
611 <dd>On ELF, default visibility means that the declaration is visible to other
612 modules and, in shared libraries, means that the declared entity may be
613 overridden. On Darwin, default visibility means that the declaration is
614 visible to other modules. Default visibility corresponds to "external
615 linkage" in the language.
618 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
620 <dd>Two declarations of an object with hidden visibility refer to the same
621 object if they are in the same shared object. Usually, hidden visibility
622 indicates that the symbol will not be placed into the dynamic symbol table,
623 so no other module (executable or shared library) can reference it
627 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
629 <dd>On ELF, protected visibility indicates that the symbol will be placed in
630 the dynamic symbol table, but that references within the defining module will
631 bind to the local symbol. That is, the symbol cannot be overridden by another
638 <!-- ======================================================================= -->
639 <div class="doc_subsection">
640 <a name="globalvars">Global Variables</a>
643 <div class="doc_text">
645 <p>Global variables define regions of memory allocated at compilation time
646 instead of run-time. Global variables may optionally be initialized, may have
647 an explicit section to be placed in, and may have an optional explicit alignment
648 specified. A variable may be defined as "thread_local", which means that it
649 will not be shared by threads (each thread will have a separated copy of the
650 variable). A variable may be defined as a global "constant," which indicates
651 that the contents of the variable will <b>never</b> be modified (enabling better
652 optimization, allowing the global data to be placed in the read-only section of
653 an executable, etc). Note that variables that need runtime initialization
654 cannot be marked "constant" as there is a store to the variable.</p>
657 LLVM explicitly allows <em>declarations</em> of global variables to be marked
658 constant, even if the final definition of the global is not. This capability
659 can be used to enable slightly better optimization of the program, but requires
660 the language definition to guarantee that optimizations based on the
661 'constantness' are valid for the translation units that do not include the
665 <p>As SSA values, global variables define pointer values that are in
666 scope (i.e. they dominate) all basic blocks in the program. Global
667 variables always define a pointer to their "content" type because they
668 describe a region of memory, and all memory objects in LLVM are
669 accessed through pointers.</p>
671 <p>LLVM allows an explicit section to be specified for globals. If the target
672 supports it, it will emit globals to the section specified.</p>
674 <p>An explicit alignment may be specified for a global. If not present, or if
675 the alignment is set to zero, the alignment of the global is set by the target
676 to whatever it feels convenient. If an explicit alignment is specified, the
677 global is forced to have at least that much alignment. All alignments must be
680 <p>For example, the following defines a global with an initializer, section,
683 <div class="doc_code">
685 @G = constant float 1.0, section "foo", align 4
692 <!-- ======================================================================= -->
693 <div class="doc_subsection">
694 <a name="functionstructure">Functions</a>
697 <div class="doc_text">
699 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
700 an optional <a href="#linkage">linkage type</a>, an optional
701 <a href="#visibility">visibility style</a>, an optional
702 <a href="#callingconv">calling convention</a>, a return type, an optional
703 <a href="#paramattrs">parameter attribute</a> for the return type, a function
704 name, a (possibly empty) argument list (each with optional
705 <a href="#paramattrs">parameter attributes</a>), an optional section, an
706 optional alignment, an opening curly brace, a list of basic blocks, and a
709 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
710 optional <a href="#linkage">linkage type</a>, an optional
711 <a href="#visibility">visibility style</a>, an optional
712 <a href="#callingconv">calling convention</a>, a return type, an optional
713 <a href="#paramattrs">parameter attribute</a> for the return type, a function
714 name, a possibly empty list of arguments, and an optional alignment.</p>
716 <p>A function definition contains a list of basic blocks, forming the CFG for
717 the function. Each basic block may optionally start with a label (giving the
718 basic block a symbol table entry), contains a list of instructions, and ends
719 with a <a href="#terminators">terminator</a> instruction (such as a branch or
720 function return).</p>
722 <p>The first basic block in a function is special in two ways: it is immediately
723 executed on entrance to the function, and it is not allowed to have predecessor
724 basic blocks (i.e. there can not be any branches to the entry block of a
725 function). Because the block can have no predecessors, it also cannot have any
726 <a href="#i_phi">PHI nodes</a>.</p>
728 <p>LLVM allows an explicit section to be specified for functions. If the target
729 supports it, it will emit functions to the section specified.</p>
731 <p>An explicit alignment may be specified for a function. If not present, or if
732 the alignment is set to zero, the alignment of the function is set by the target
733 to whatever it feels convenient. If an explicit alignment is specified, the
734 function is forced to have at least that much alignment. All alignments must be
740 <!-- ======================================================================= -->
741 <div class="doc_subsection">
742 <a name="aliasstructure">Aliases</a>
744 <div class="doc_text">
745 <p>Aliases act as "second name" for the aliasee value (which can be either
746 function or global variable or bitcast of global value). Aliases may have an
747 optional <a href="#linkage">linkage type</a>, and an
748 optional <a href="#visibility">visibility style</a>.</p>
752 <div class="doc_code">
754 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
762 <!-- ======================================================================= -->
763 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
764 <div class="doc_text">
765 <p>The return type and each parameter of a function type may have a set of
766 <i>parameter attributes</i> associated with them. Parameter attributes are
767 used to communicate additional information about the result or parameters of
768 a function. Parameter attributes are considered to be part of the function
769 type so two functions types that differ only by the parameter attributes
770 are different function types.</p>
772 <p>Parameter attributes are simple keywords that follow the type specified. If
773 multiple parameter attributes are needed, they are space separated. For
776 <div class="doc_code">
778 %someFunc = i16 (i8 signext %someParam) zeroext
779 %someFunc = i16 (i8 zeroext %someParam) zeroext
783 <p>Note that the two function types above are unique because the parameter has
784 a different attribute (<tt>signext</tt> in the first one, <tt>zeroext</tt> in
785 the second). Also note that the attribute for the function result
786 (<tt>zeroext</tt>) comes immediately after the argument list.</p>
788 <p>Currently, only the following parameter attributes are defined:</p>
790 <dt><tt>zeroext</tt></dt>
791 <dd>This indicates that the parameter should be zero extended just before
792 a call to this function.</dd>
793 <dt><tt>signext</tt></dt>
794 <dd>This indicates that the parameter should be sign extended just before
795 a call to this function.</dd>
796 <dt><tt>inreg</tt></dt>
797 <dd>This indicates that the parameter should be placed in register (if
798 possible) during assembling function call. Support for this attribute is
800 <dt><tt>sret</tt></dt>
801 <dd>This indicates that the parameter specifies the address of a structure
802 that is the return value of the function in the source program.</dd>
803 <dt><tt>noalias</tt></dt>
804 <dd>This indicates that the parameter not alias any other object or any
805 other "noalias" objects during the function call.
806 <dt><tt>noreturn</tt></dt>
807 <dd>This function attribute indicates that the function never returns. This
808 indicates to LLVM that every call to this function should be treated as if
809 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
810 <dt><tt>nounwind</tt></dt>
811 <dd>This function attribute indicates that the function type does not use
812 the unwind instruction and does not allow stack unwinding to propagate
814 <dt><tt>nest</tt></dt>
815 <dd>This indicates that the parameter can be excised using the
816 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
821 <!-- ======================================================================= -->
822 <div class="doc_subsection">
823 <a name="moduleasm">Module-Level Inline Assembly</a>
826 <div class="doc_text">
828 Modules may contain "module-level inline asm" blocks, which corresponds to the
829 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
830 LLVM and treated as a single unit, but may be separated in the .ll file if
831 desired. The syntax is very simple:
834 <div class="doc_code">
836 module asm "inline asm code goes here"
837 module asm "more can go here"
841 <p>The strings can contain any character by escaping non-printable characters.
842 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
847 The inline asm code is simply printed to the machine code .s file when
848 assembly code is generated.
852 <!-- ======================================================================= -->
853 <div class="doc_subsection">
854 <a name="datalayout">Data Layout</a>
857 <div class="doc_text">
858 <p>A module may specify a target specific data layout string that specifies how
859 data is to be laid out in memory. The syntax for the data layout is simply:</p>
860 <pre> target datalayout = "<i>layout specification</i>"</pre>
861 <p>The <i>layout specification</i> consists of a list of specifications
862 separated by the minus sign character ('-'). Each specification starts with a
863 letter and may include other information after the letter to define some
864 aspect of the data layout. The specifications accepted are as follows: </p>
867 <dd>Specifies that the target lays out data in big-endian form. That is, the
868 bits with the most significance have the lowest address location.</dd>
870 <dd>Specifies that hte target lays out data in little-endian form. That is,
871 the bits with the least significance have the lowest address location.</dd>
872 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
873 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
874 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
875 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
877 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
878 <dd>This specifies the alignment for an integer type of a given bit
879 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
880 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
881 <dd>This specifies the alignment for a vector type of a given bit
883 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
884 <dd>This specifies the alignment for a floating point type of a given bit
885 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
887 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
888 <dd>This specifies the alignment for an aggregate type of a given bit
891 <p>When constructing the data layout for a given target, LLVM starts with a
892 default set of specifications which are then (possibly) overriden by the
893 specifications in the <tt>datalayout</tt> keyword. The default specifications
894 are given in this list:</p>
896 <li><tt>E</tt> - big endian</li>
897 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
898 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
899 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
900 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
901 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
902 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
903 alignment of 64-bits</li>
904 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
905 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
906 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
907 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
908 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
910 <p>When llvm is determining the alignment for a given type, it uses the
913 <li>If the type sought is an exact match for one of the specifications, that
914 specification is used.</li>
915 <li>If no match is found, and the type sought is an integer type, then the
916 smallest integer type that is larger than the bitwidth of the sought type is
917 used. If none of the specifications are larger than the bitwidth then the the
918 largest integer type is used. For example, given the default specifications
919 above, the i7 type will use the alignment of i8 (next largest) while both
920 i65 and i256 will use the alignment of i64 (largest specified).</li>
921 <li>If no match is found, and the type sought is a vector type, then the
922 largest vector type that is smaller than the sought vector type will be used
923 as a fall back. This happens because <128 x double> can be implemented in
924 terms of 64 <2 x double>, for example.</li>
928 <!-- *********************************************************************** -->
929 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
930 <!-- *********************************************************************** -->
932 <div class="doc_text">
934 <p>The LLVM type system is one of the most important features of the
935 intermediate representation. Being typed enables a number of
936 optimizations to be performed on the IR directly, without having to do
937 extra analyses on the side before the transformation. A strong type
938 system makes it easier to read the generated code and enables novel
939 analyses and transformations that are not feasible to perform on normal
940 three address code representations.</p>
944 <!-- ======================================================================= -->
945 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
946 <div class="doc_text">
947 <p>The primitive types are the fundamental building blocks of the LLVM
948 system. The current set of primitive types is as follows:</p>
950 <table class="layout">
955 <tr><th>Type</th><th>Description</th></tr>
956 <tr><td><tt><a name="t_void">void</a></tt></td><td>No value</td></tr>
957 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
964 <tr><th>Type</th><th>Description</th></tr>
965 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
966 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
974 <!-- _______________________________________________________________________ -->
975 <div class="doc_subsubsection"> <a name="t_classifications">Type
976 Classifications</a> </div>
977 <div class="doc_text">
978 <p>These different primitive types fall into a few useful
981 <table border="1" cellspacing="0" cellpadding="4">
983 <tr><th>Classification</th><th>Types</th></tr>
985 <td><a name="t_integer">integer</a></td>
986 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
989 <td><a name="t_floating">floating point</a></td>
990 <td><tt>float, double</tt></td>
993 <td><a name="t_firstclass">first class</a></td>
994 <td><tt>i1, ..., float, double, <br/>
995 <a href="#t_pointer">pointer</a>,<a href="#t_vector">vector</a></tt>
1001 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1002 most important. Values of these types are the only ones which can be
1003 produced by instructions, passed as arguments, or used as operands to
1004 instructions. This means that all structures and arrays must be
1005 manipulated either by pointer or by component.</p>
1008 <!-- ======================================================================= -->
1009 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1011 <div class="doc_text">
1013 <p>The real power in LLVM comes from the derived types in the system.
1014 This is what allows a programmer to represent arrays, functions,
1015 pointers, and other useful types. Note that these derived types may be
1016 recursive: For example, it is possible to have a two dimensional array.</p>
1020 <!-- _______________________________________________________________________ -->
1021 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1023 <div class="doc_text">
1026 <p>The integer type is a very simple derived type that simply specifies an
1027 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1028 2^23-1 (about 8 million) can be specified.</p>
1036 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1040 <table class="layout">
1050 <tt>i1942652</tt><br/>
1053 A boolean integer of 1 bit<br/>
1054 A nibble sized integer of 4 bits.<br/>
1055 A byte sized integer of 8 bits.<br/>
1056 A half word sized integer of 16 bits.<br/>
1057 A word sized integer of 32 bits.<br/>
1058 An integer whose bit width is the answer. <br/>
1059 A double word sized integer of 64 bits.<br/>
1060 A really big integer of over 1 million bits.<br/>
1066 <!-- _______________________________________________________________________ -->
1067 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1069 <div class="doc_text">
1073 <p>The array type is a very simple derived type that arranges elements
1074 sequentially in memory. The array type requires a size (number of
1075 elements) and an underlying data type.</p>
1080 [<# elements> x <elementtype>]
1083 <p>The number of elements is a constant integer value; elementtype may
1084 be any type with a size.</p>
1087 <table class="layout">
1090 <tt>[40 x i32 ]</tt><br/>
1091 <tt>[41 x i32 ]</tt><br/>
1092 <tt>[40 x i8]</tt><br/>
1095 Array of 40 32-bit integer values.<br/>
1096 Array of 41 32-bit integer values.<br/>
1097 Array of 40 8-bit integer values.<br/>
1101 <p>Here are some examples of multidimensional arrays:</p>
1102 <table class="layout">
1105 <tt>[3 x [4 x i32]]</tt><br/>
1106 <tt>[12 x [10 x float]]</tt><br/>
1107 <tt>[2 x [3 x [4 x i16]]]</tt><br/>
1110 3x4 array of 32-bit integer values.<br/>
1111 12x10 array of single precision floating point values.<br/>
1112 2x3x4 array of 16-bit integer values.<br/>
1117 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1118 length array. Normally, accesses past the end of an array are undefined in
1119 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1120 As a special case, however, zero length arrays are recognized to be variable
1121 length. This allows implementation of 'pascal style arrays' with the LLVM
1122 type "{ i32, [0 x float]}", for example.</p>
1126 <!-- _______________________________________________________________________ -->
1127 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1128 <div class="doc_text">
1130 <p>The function type can be thought of as a function signature. It
1131 consists of a return type and a list of formal parameter types.
1132 Function types are usually used to build virtual function tables
1133 (which are structures of pointers to functions), for indirect function
1134 calls, and when defining a function.</p>
1136 The return type of a function type cannot be an aggregate type.
1139 <pre> <returntype> (<parameter list>)<br></pre>
1140 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1141 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1142 which indicates that the function takes a variable number of arguments.
1143 Variable argument functions can access their arguments with the <a
1144 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
1146 <table class="layout">
1148 <td class="left"><tt>i32 (i32)</tt></td>
1149 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1151 </tr><tr class="layout">
1152 <td class="left"><tt>float (i16 signext, i32 *) *
1154 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1155 an <tt>i16</tt> that should be sign extended and a
1156 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1159 </tr><tr class="layout">
1160 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1161 <td class="left">A vararg function that takes at least one
1162 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1163 which returns an integer. This is the signature for <tt>printf</tt> in
1170 <!-- _______________________________________________________________________ -->
1171 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1172 <div class="doc_text">
1174 <p>The structure type is used to represent a collection of data members
1175 together in memory. The packing of the field types is defined to match
1176 the ABI of the underlying processor. The elements of a structure may
1177 be any type that has a size.</p>
1178 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1179 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1180 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1183 <pre> { <type list> }<br></pre>
1185 <table class="layout">
1187 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1188 <td class="left">A triple of three <tt>i32</tt> values</td>
1189 </tr><tr class="layout">
1190 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1191 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1192 second element is a <a href="#t_pointer">pointer</a> to a
1193 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1194 an <tt>i32</tt>.</td>
1199 <!-- _______________________________________________________________________ -->
1200 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1202 <div class="doc_text">
1204 <p>The packed structure type is used to represent a collection of data members
1205 together in memory. There is no padding between fields. Further, the alignment
1206 of a packed structure is 1 byte. The elements of a packed structure may
1207 be any type that has a size.</p>
1208 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1209 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1210 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1213 <pre> < { <type list> } > <br></pre>
1215 <table class="layout">
1217 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1218 <td class="left">A triple of three <tt>i32</tt> values</td>
1219 </tr><tr class="layout">
1220 <td class="left"><tt>< { float, i32 (i32) * } ></tt></td>
1221 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1222 second element is a <a href="#t_pointer">pointer</a> to a
1223 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1224 an <tt>i32</tt>.</td>
1229 <!-- _______________________________________________________________________ -->
1230 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1231 <div class="doc_text">
1233 <p>As in many languages, the pointer type represents a pointer or
1234 reference to another object, which must live in memory.</p>
1236 <pre> <type> *<br></pre>
1238 <table class="layout">
1241 <tt>[4x i32]*</tt><br/>
1242 <tt>i32 (i32 *) *</tt><br/>
1245 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
1246 four <tt>i32</tt> values<br/>
1247 A <a href="#t_pointer">pointer</a> to a <a
1248 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1255 <!-- _______________________________________________________________________ -->
1256 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1257 <div class="doc_text">
1261 <p>A vector type is a simple derived type that represents a vector
1262 of elements. Vector types are used when multiple primitive data
1263 are operated in parallel using a single instruction (SIMD).
1264 A vector type requires a size (number of
1265 elements) and an underlying primitive data type. Vectors must have a power
1266 of two length (1, 2, 4, 8, 16 ...). Vector types are
1267 considered <a href="#t_firstclass">first class</a>.</p>
1272 < <# elements> x <elementtype> >
1275 <p>The number of elements is a constant integer value; elementtype may
1276 be any integer or floating point type.</p>
1280 <table class="layout">
1283 <tt><4 x i32></tt><br/>
1284 <tt><8 x float></tt><br/>
1285 <tt><2 x i64></tt><br/>
1288 Vector of 4 32-bit integer values.<br/>
1289 Vector of 8 floating-point values.<br/>
1290 Vector of 2 64-bit integer values.<br/>
1296 <!-- _______________________________________________________________________ -->
1297 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1298 <div class="doc_text">
1302 <p>Opaque types are used to represent unknown types in the system. This
1303 corresponds (for example) to the C notion of a foward declared structure type.
1304 In LLVM, opaque types can eventually be resolved to any type (not just a
1305 structure type).</p>
1315 <table class="layout">
1321 An opaque type.<br/>
1328 <!-- *********************************************************************** -->
1329 <div class="doc_section"> <a name="constants">Constants</a> </div>
1330 <!-- *********************************************************************** -->
1332 <div class="doc_text">
1334 <p>LLVM has several different basic types of constants. This section describes
1335 them all and their syntax.</p>
1339 <!-- ======================================================================= -->
1340 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1342 <div class="doc_text">
1345 <dt><b>Boolean constants</b></dt>
1347 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1348 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1351 <dt><b>Integer constants</b></dt>
1353 <dd>Standard integers (such as '4') are constants of the <a
1354 href="#t_integer">integer</a> type. Negative numbers may be used with
1358 <dt><b>Floating point constants</b></dt>
1360 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1361 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1362 notation (see below). Floating point constants must have a <a
1363 href="#t_floating">floating point</a> type. </dd>
1365 <dt><b>Null pointer constants</b></dt>
1367 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1368 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1372 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1373 of floating point constants. For example, the form '<tt>double
1374 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1375 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1376 (and the only time that they are generated by the disassembler) is when a
1377 floating point constant must be emitted but it cannot be represented as a
1378 decimal floating point number. For example, NaN's, infinities, and other
1379 special values are represented in their IEEE hexadecimal format so that
1380 assembly and disassembly do not cause any bits to change in the constants.</p>
1384 <!-- ======================================================================= -->
1385 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1388 <div class="doc_text">
1389 <p>Aggregate constants arise from aggregation of simple constants
1390 and smaller aggregate constants.</p>
1393 <dt><b>Structure constants</b></dt>
1395 <dd>Structure constants are represented with notation similar to structure
1396 type definitions (a comma separated list of elements, surrounded by braces
1397 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* %G }</tt>",
1398 where "<tt>%G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1399 must have <a href="#t_struct">structure type</a>, and the number and
1400 types of elements must match those specified by the type.
1403 <dt><b>Array constants</b></dt>
1405 <dd>Array constants are represented with notation similar to array type
1406 definitions (a comma separated list of elements, surrounded by square brackets
1407 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1408 constants must have <a href="#t_array">array type</a>, and the number and
1409 types of elements must match those specified by the type.
1412 <dt><b>Vector constants</b></dt>
1414 <dd>Vector constants are represented with notation similar to vector type
1415 definitions (a comma separated list of elements, surrounded by
1416 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1417 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1418 href="#t_vector">vector type</a>, and the number and types of elements must
1419 match those specified by the type.
1422 <dt><b>Zero initialization</b></dt>
1424 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1425 value to zero of <em>any</em> type, including scalar and aggregate types.
1426 This is often used to avoid having to print large zero initializers (e.g. for
1427 large arrays) and is always exactly equivalent to using explicit zero
1434 <!-- ======================================================================= -->
1435 <div class="doc_subsection">
1436 <a name="globalconstants">Global Variable and Function Addresses</a>
1439 <div class="doc_text">
1441 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1442 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1443 constants. These constants are explicitly referenced when the <a
1444 href="#identifiers">identifier for the global</a> is used and always have <a
1445 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1448 <div class="doc_code">
1452 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1458 <!-- ======================================================================= -->
1459 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1460 <div class="doc_text">
1461 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1462 no specific value. Undefined values may be of any type and be used anywhere
1463 a constant is permitted.</p>
1465 <p>Undefined values indicate to the compiler that the program is well defined
1466 no matter what value is used, giving the compiler more freedom to optimize.
1470 <!-- ======================================================================= -->
1471 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1474 <div class="doc_text">
1476 <p>Constant expressions are used to allow expressions involving other constants
1477 to be used as constants. Constant expressions may be of any <a
1478 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1479 that does not have side effects (e.g. load and call are not supported). The
1480 following is the syntax for constant expressions:</p>
1483 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1484 <dd>Truncate a constant to another type. The bit size of CST must be larger
1485 than the bit size of TYPE. Both types must be integers.</dd>
1487 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1488 <dd>Zero extend a constant to another type. The bit size of CST must be
1489 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1491 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1492 <dd>Sign extend a constant to another type. The bit size of CST must be
1493 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1495 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1496 <dd>Truncate a floating point constant to another floating point type. The
1497 size of CST must be larger than the size of TYPE. Both types must be
1498 floating point.</dd>
1500 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1501 <dd>Floating point extend a constant to another type. The size of CST must be
1502 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1504 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1505 <dd>Convert a floating point constant to the corresponding unsigned integer
1506 constant. TYPE must be an integer type. CST must be floating point. If the
1507 value won't fit in the integer type, the results are undefined.</dd>
1509 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1510 <dd>Convert a floating point constant to the corresponding signed integer
1511 constant. TYPE must be an integer type. CST must be floating point. If the
1512 value won't fit in the integer type, the results are undefined.</dd>
1514 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1515 <dd>Convert an unsigned integer constant to the corresponding floating point
1516 constant. TYPE must be floating point. CST must be of integer type. If the
1517 value won't fit in the floating point type, the results are undefined.</dd>
1519 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1520 <dd>Convert a signed integer constant to the corresponding floating point
1521 constant. TYPE must be floating point. CST must be of integer type. If the
1522 value won't fit in the floating point type, the results are undefined.</dd>
1524 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1525 <dd>Convert a pointer typed constant to the corresponding integer constant
1526 TYPE must be an integer type. CST must be of pointer type. The CST value is
1527 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1529 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1530 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1531 pointer type. CST must be of integer type. The CST value is zero extended,
1532 truncated, or unchanged to make it fit in a pointer size. This one is
1533 <i>really</i> dangerous!</dd>
1535 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1536 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1537 identical (same number of bits). The conversion is done as if the CST value
1538 was stored to memory and read back as TYPE. In other words, no bits change
1539 with this operator, just the type. This can be used for conversion of
1540 vector types to any other type, as long as they have the same bit width. For
1541 pointers it is only valid to cast to another pointer type.
1544 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1546 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1547 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1548 instruction, the index list may have zero or more indexes, which are required
1549 to make sense for the type of "CSTPTR".</dd>
1551 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1553 <dd>Perform the <a href="#i_select">select operation</a> on
1556 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1557 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1559 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1560 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1562 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1564 <dd>Perform the <a href="#i_extractelement">extractelement
1565 operation</a> on constants.
1567 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1569 <dd>Perform the <a href="#i_insertelement">insertelement
1570 operation</a> on constants.</dd>
1573 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1575 <dd>Perform the <a href="#i_shufflevector">shufflevector
1576 operation</a> on constants.</dd>
1578 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1580 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1581 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1582 binary</a> operations. The constraints on operands are the same as those for
1583 the corresponding instruction (e.g. no bitwise operations on floating point
1584 values are allowed).</dd>
1588 <!-- *********************************************************************** -->
1589 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1590 <!-- *********************************************************************** -->
1592 <!-- ======================================================================= -->
1593 <div class="doc_subsection">
1594 <a name="inlineasm">Inline Assembler Expressions</a>
1597 <div class="doc_text">
1600 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1601 Module-Level Inline Assembly</a>) through the use of a special value. This
1602 value represents the inline assembler as a string (containing the instructions
1603 to emit), a list of operand constraints (stored as a string), and a flag that
1604 indicates whether or not the inline asm expression has side effects. An example
1605 inline assembler expression is:
1608 <div class="doc_code">
1610 i32 (i32) asm "bswap $0", "=r,r"
1615 Inline assembler expressions may <b>only</b> be used as the callee operand of
1616 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1619 <div class="doc_code">
1621 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1626 Inline asms with side effects not visible in the constraint list must be marked
1627 as having side effects. This is done through the use of the
1628 '<tt>sideeffect</tt>' keyword, like so:
1631 <div class="doc_code">
1633 call void asm sideeffect "eieio", ""()
1637 <p>TODO: The format of the asm and constraints string still need to be
1638 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1639 need to be documented).
1644 <!-- *********************************************************************** -->
1645 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1646 <!-- *********************************************************************** -->
1648 <div class="doc_text">
1650 <p>The LLVM instruction set consists of several different
1651 classifications of instructions: <a href="#terminators">terminator
1652 instructions</a>, <a href="#binaryops">binary instructions</a>,
1653 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1654 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1655 instructions</a>.</p>
1659 <!-- ======================================================================= -->
1660 <div class="doc_subsection"> <a name="terminators">Terminator
1661 Instructions</a> </div>
1663 <div class="doc_text">
1665 <p>As mentioned <a href="#functionstructure">previously</a>, every
1666 basic block in a program ends with a "Terminator" instruction, which
1667 indicates which block should be executed after the current block is
1668 finished. These terminator instructions typically yield a '<tt>void</tt>'
1669 value: they produce control flow, not values (the one exception being
1670 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1671 <p>There are six different terminator instructions: the '<a
1672 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1673 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1674 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1675 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1676 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1680 <!-- _______________________________________________________________________ -->
1681 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1682 Instruction</a> </div>
1683 <div class="doc_text">
1685 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1686 ret void <i>; Return from void function</i>
1689 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1690 value) from a function back to the caller.</p>
1691 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1692 returns a value and then causes control flow, and one that just causes
1693 control flow to occur.</p>
1695 <p>The '<tt>ret</tt>' instruction may return any '<a
1696 href="#t_firstclass">first class</a>' type. Notice that a function is
1697 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1698 instruction inside of the function that returns a value that does not
1699 match the return type of the function.</p>
1701 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1702 returns back to the calling function's context. If the caller is a "<a
1703 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1704 the instruction after the call. If the caller was an "<a
1705 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1706 at the beginning of the "normal" destination block. If the instruction
1707 returns a value, that value shall set the call or invoke instruction's
1710 <pre> ret i32 5 <i>; Return an integer value of 5</i>
1711 ret void <i>; Return from a void function</i>
1714 <!-- _______________________________________________________________________ -->
1715 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1716 <div class="doc_text">
1718 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1721 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1722 transfer to a different basic block in the current function. There are
1723 two forms of this instruction, corresponding to a conditional branch
1724 and an unconditional branch.</p>
1726 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1727 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1728 unconditional form of the '<tt>br</tt>' instruction takes a single
1729 '<tt>label</tt>' value as a target.</p>
1731 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1732 argument is evaluated. If the value is <tt>true</tt>, control flows
1733 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1734 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1736 <pre>Test:<br> %cond = <a href="#i_icmp">icmp</a> eq, i32 %a, %b<br> br i1 %cond, label %IfEqual, label %IfUnequal<br>IfEqual:<br> <a
1737 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1739 <!-- _______________________________________________________________________ -->
1740 <div class="doc_subsubsection">
1741 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1744 <div class="doc_text">
1748 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1753 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1754 several different places. It is a generalization of the '<tt>br</tt>'
1755 instruction, allowing a branch to occur to one of many possible
1761 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1762 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1763 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1764 table is not allowed to contain duplicate constant entries.</p>
1768 <p>The <tt>switch</tt> instruction specifies a table of values and
1769 destinations. When the '<tt>switch</tt>' instruction is executed, this
1770 table is searched for the given value. If the value is found, control flow is
1771 transfered to the corresponding destination; otherwise, control flow is
1772 transfered to the default destination.</p>
1774 <h5>Implementation:</h5>
1776 <p>Depending on properties of the target machine and the particular
1777 <tt>switch</tt> instruction, this instruction may be code generated in different
1778 ways. For example, it could be generated as a series of chained conditional
1779 branches or with a lookup table.</p>
1784 <i>; Emulate a conditional br instruction</i>
1785 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1786 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1788 <i>; Emulate an unconditional br instruction</i>
1789 switch i32 0, label %dest [ ]
1791 <i>; Implement a jump table:</i>
1792 switch i32 %val, label %otherwise [ i32 0, label %onzero
1794 i32 2, label %ontwo ]
1798 <!-- _______________________________________________________________________ -->
1799 <div class="doc_subsubsection">
1800 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1803 <div class="doc_text">
1808 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1809 to label <normal label> unwind label <exception label>
1814 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1815 function, with the possibility of control flow transfer to either the
1816 '<tt>normal</tt>' label or the
1817 '<tt>exception</tt>' label. If the callee function returns with the
1818 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1819 "normal" label. If the callee (or any indirect callees) returns with the "<a
1820 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1821 continued at the dynamically nearest "exception" label.</p>
1825 <p>This instruction requires several arguments:</p>
1829 The optional "cconv" marker indicates which <a href="#callingconv">calling
1830 convention</a> the call should use. If none is specified, the call defaults
1831 to using C calling conventions.
1833 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1834 function value being invoked. In most cases, this is a direct function
1835 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1836 an arbitrary pointer to function value.
1839 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1840 function to be invoked. </li>
1842 <li>'<tt>function args</tt>': argument list whose types match the function
1843 signature argument types. If the function signature indicates the function
1844 accepts a variable number of arguments, the extra arguments can be
1847 <li>'<tt>normal label</tt>': the label reached when the called function
1848 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1850 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1851 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1857 <p>This instruction is designed to operate as a standard '<tt><a
1858 href="#i_call">call</a></tt>' instruction in most regards. The primary
1859 difference is that it establishes an association with a label, which is used by
1860 the runtime library to unwind the stack.</p>
1862 <p>This instruction is used in languages with destructors to ensure that proper
1863 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1864 exception. Additionally, this is important for implementation of
1865 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1869 %retval = invoke i32 %Test(i32 15) to label %Continue
1870 unwind label %TestCleanup <i>; {i32}:retval set</i>
1871 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Test(i32 15) to label %Continue
1872 unwind label %TestCleanup <i>; {i32}:retval set</i>
1877 <!-- _______________________________________________________________________ -->
1879 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1880 Instruction</a> </div>
1882 <div class="doc_text">
1891 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1892 at the first callee in the dynamic call stack which used an <a
1893 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1894 primarily used to implement exception handling.</p>
1898 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1899 immediately halt. The dynamic call stack is then searched for the first <a
1900 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1901 execution continues at the "exceptional" destination block specified by the
1902 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1903 dynamic call chain, undefined behavior results.</p>
1906 <!-- _______________________________________________________________________ -->
1908 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1909 Instruction</a> </div>
1911 <div class="doc_text">
1920 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1921 instruction is used to inform the optimizer that a particular portion of the
1922 code is not reachable. This can be used to indicate that the code after a
1923 no-return function cannot be reached, and other facts.</p>
1927 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1932 <!-- ======================================================================= -->
1933 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1934 <div class="doc_text">
1935 <p>Binary operators are used to do most of the computation in a
1936 program. They require two operands, execute an operation on them, and
1937 produce a single value. The operands might represent
1938 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
1939 The result value of a binary operator is not
1940 necessarily the same type as its operands.</p>
1941 <p>There are several different binary operators:</p>
1943 <!-- _______________________________________________________________________ -->
1944 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1945 Instruction</a> </div>
1946 <div class="doc_text">
1948 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1951 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1953 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1954 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1955 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1956 Both arguments must have identical types.</p>
1958 <p>The value produced is the integer or floating point sum of the two
1961 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
1964 <!-- _______________________________________________________________________ -->
1965 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1966 Instruction</a> </div>
1967 <div class="doc_text">
1969 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1972 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1974 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1975 instruction present in most other intermediate representations.</p>
1977 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1978 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1980 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1981 Both arguments must have identical types.</p>
1983 <p>The value produced is the integer or floating point difference of
1984 the two operands.</p>
1987 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
1988 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
1991 <!-- _______________________________________________________________________ -->
1992 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1993 Instruction</a> </div>
1994 <div class="doc_text">
1996 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1999 <p>The '<tt>mul</tt>' instruction returns the product of its two
2002 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2003 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2005 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2006 Both arguments must have identical types.</p>
2008 <p>The value produced is the integer or floating point product of the
2010 <p>Because the operands are the same width, the result of an integer
2011 multiplication is the same whether the operands should be deemed unsigned or
2014 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2017 <!-- _______________________________________________________________________ -->
2018 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2020 <div class="doc_text">
2022 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2025 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2028 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2029 <a href="#t_integer">integer</a> values. Both arguments must have identical
2030 types. This instruction can also take <a href="#t_vector">vector</a> versions
2031 of the values in which case the elements must be integers.</p>
2033 <p>The value produced is the unsigned integer quotient of the two operands. This
2034 instruction always performs an unsigned division operation, regardless of
2035 whether the arguments are unsigned or not.</p>
2037 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2040 <!-- _______________________________________________________________________ -->
2041 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2043 <div class="doc_text">
2045 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2048 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2051 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2052 <a href="#t_integer">integer</a> values. Both arguments must have identical
2053 types. This instruction can also take <a href="#t_vector">vector</a> versions
2054 of the values in which case the elements must be integers.</p>
2056 <p>The value produced is the signed integer quotient of the two operands. This
2057 instruction always performs a signed division operation, regardless of whether
2058 the arguments are signed or not.</p>
2060 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2063 <!-- _______________________________________________________________________ -->
2064 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2065 Instruction</a> </div>
2066 <div class="doc_text">
2068 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2071 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2074 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2075 <a href="#t_floating">floating point</a> values. Both arguments must have
2076 identical types. This instruction can also take <a href="#t_vector">vector</a>
2077 versions of floating point values.</p>
2079 <p>The value produced is the floating point quotient of the two operands.</p>
2081 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2084 <!-- _______________________________________________________________________ -->
2085 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2087 <div class="doc_text">
2089 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2092 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2093 unsigned division of its two arguments.</p>
2095 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2096 <a href="#t_integer">integer</a> values. Both arguments must have identical
2099 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2100 This instruction always performs an unsigned division to get the remainder,
2101 regardless of whether the arguments are unsigned or not.</p>
2103 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2107 <!-- _______________________________________________________________________ -->
2108 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2109 Instruction</a> </div>
2110 <div class="doc_text">
2112 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2115 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2116 signed division of its two operands.</p>
2118 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2119 <a href="#t_integer">integer</a> values. Both arguments must have identical
2122 <p>This instruction returns the <i>remainder</i> of a division (where the result
2123 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2124 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2125 a value. For more information about the difference, see <a
2126 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2127 Math Forum</a>. For a table of how this is implemented in various languages,
2128 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2129 Wikipedia: modulo operation</a>.</p>
2131 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2135 <!-- _______________________________________________________________________ -->
2136 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2137 Instruction</a> </div>
2138 <div class="doc_text">
2140 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2143 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2144 division of its two operands.</p>
2146 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2147 <a href="#t_floating">floating point</a> values. Both arguments must have
2148 identical types.</p>
2150 <p>This instruction returns the <i>remainder</i> of a division.</p>
2152 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2156 <!-- ======================================================================= -->
2157 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2158 Operations</a> </div>
2159 <div class="doc_text">
2160 <p>Bitwise binary operators are used to do various forms of
2161 bit-twiddling in a program. They are generally very efficient
2162 instructions and can commonly be strength reduced from other
2163 instructions. They require two operands, execute an operation on them,
2164 and produce a single value. The resulting value of the bitwise binary
2165 operators is always the same type as its first operand.</p>
2168 <!-- _______________________________________________________________________ -->
2169 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2170 Instruction</a> </div>
2171 <div class="doc_text">
2173 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2176 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2177 the left a specified number of bits.</p>
2179 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2180 href="#t_integer">integer</a> type.</p>
2182 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
2183 <h5>Example:</h5><pre>
2184 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2185 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2186 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2189 <!-- _______________________________________________________________________ -->
2190 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2191 Instruction</a> </div>
2192 <div class="doc_text">
2194 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2198 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2199 operand shifted to the right a specified number of bits with zero fill.</p>
2202 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2203 <a href="#t_integer">integer</a> type.</p>
2206 <p>This instruction always performs a logical shift right operation. The most
2207 significant bits of the result will be filled with zero bits after the
2212 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2213 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2214 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2215 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2219 <!-- _______________________________________________________________________ -->
2220 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2221 Instruction</a> </div>
2222 <div class="doc_text">
2225 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2229 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2230 operand shifted to the right a specified number of bits with sign extension.</p>
2233 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2234 <a href="#t_integer">integer</a> type.</p>
2237 <p>This instruction always performs an arithmetic shift right operation,
2238 The most significant bits of the result will be filled with the sign bit
2239 of <tt>var1</tt>.</p>
2243 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2244 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2245 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2246 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2250 <!-- _______________________________________________________________________ -->
2251 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2252 Instruction</a> </div>
2253 <div class="doc_text">
2255 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2258 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2259 its two operands.</p>
2261 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2262 href="#t_integer">integer</a> values. Both arguments must have
2263 identical types.</p>
2265 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2267 <div style="align: center">
2268 <table border="1" cellspacing="0" cellpadding="4">
2299 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2300 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2301 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2304 <!-- _______________________________________________________________________ -->
2305 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2306 <div class="doc_text">
2308 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2311 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2312 or of its two operands.</p>
2314 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2315 href="#t_integer">integer</a> values. Both arguments must have
2316 identical types.</p>
2318 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2320 <div style="align: center">
2321 <table border="1" cellspacing="0" cellpadding="4">
2352 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2353 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2354 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2357 <!-- _______________________________________________________________________ -->
2358 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2359 Instruction</a> </div>
2360 <div class="doc_text">
2362 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2365 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2366 or of its two operands. The <tt>xor</tt> is used to implement the
2367 "one's complement" operation, which is the "~" operator in C.</p>
2369 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2370 href="#t_integer">integer</a> values. Both arguments must have
2371 identical types.</p>
2373 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2375 <div style="align: center">
2376 <table border="1" cellspacing="0" cellpadding="4">
2408 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2409 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2410 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2411 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2415 <!-- ======================================================================= -->
2416 <div class="doc_subsection">
2417 <a name="vectorops">Vector Operations</a>
2420 <div class="doc_text">
2422 <p>LLVM supports several instructions to represent vector operations in a
2423 target-independent manner. These instructions cover the element-access and
2424 vector-specific operations needed to process vectors effectively. While LLVM
2425 does directly support these vector operations, many sophisticated algorithms
2426 will want to use target-specific intrinsics to take full advantage of a specific
2431 <!-- _______________________________________________________________________ -->
2432 <div class="doc_subsubsection">
2433 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2436 <div class="doc_text">
2441 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2447 The '<tt>extractelement</tt>' instruction extracts a single scalar
2448 element from a vector at a specified index.
2455 The first operand of an '<tt>extractelement</tt>' instruction is a
2456 value of <a href="#t_vector">vector</a> type. The second operand is
2457 an index indicating the position from which to extract the element.
2458 The index may be a variable.</p>
2463 The result is a scalar of the same type as the element type of
2464 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2465 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2466 results are undefined.
2472 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2477 <!-- _______________________________________________________________________ -->
2478 <div class="doc_subsubsection">
2479 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2482 <div class="doc_text">
2487 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2493 The '<tt>insertelement</tt>' instruction inserts a scalar
2494 element into a vector at a specified index.
2501 The first operand of an '<tt>insertelement</tt>' instruction is a
2502 value of <a href="#t_vector">vector</a> type. The second operand is a
2503 scalar value whose type must equal the element type of the first
2504 operand. The third operand is an index indicating the position at
2505 which to insert the value. The index may be a variable.</p>
2510 The result is a vector of the same type as <tt>val</tt>. Its
2511 element values are those of <tt>val</tt> except at position
2512 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2513 exceeds the length of <tt>val</tt>, the results are undefined.
2519 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2523 <!-- _______________________________________________________________________ -->
2524 <div class="doc_subsubsection">
2525 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2528 <div class="doc_text">
2533 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2539 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2540 from two input vectors, returning a vector of the same type.
2546 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2547 with types that match each other and types that match the result of the
2548 instruction. The third argument is a shuffle mask, which has the same number
2549 of elements as the other vector type, but whose element type is always 'i32'.
2553 The shuffle mask operand is required to be a constant vector with either
2554 constant integer or undef values.
2560 The elements of the two input vectors are numbered from left to right across
2561 both of the vectors. The shuffle mask operand specifies, for each element of
2562 the result vector, which element of the two input registers the result element
2563 gets. The element selector may be undef (meaning "don't care") and the second
2564 operand may be undef if performing a shuffle from only one vector.
2570 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2571 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2572 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2573 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2578 <!-- ======================================================================= -->
2579 <div class="doc_subsection">
2580 <a name="memoryops">Memory Access and Addressing Operations</a>
2583 <div class="doc_text">
2585 <p>A key design point of an SSA-based representation is how it
2586 represents memory. In LLVM, no memory locations are in SSA form, which
2587 makes things very simple. This section describes how to read, write,
2588 allocate, and free memory in LLVM.</p>
2592 <!-- _______________________________________________________________________ -->
2593 <div class="doc_subsubsection">
2594 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2597 <div class="doc_text">
2602 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2607 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2608 heap and returns a pointer to it.</p>
2612 <p>The '<tt>malloc</tt>' instruction allocates
2613 <tt>sizeof(<type>)*NumElements</tt>
2614 bytes of memory from the operating system and returns a pointer of the
2615 appropriate type to the program. If "NumElements" is specified, it is the
2616 number of elements allocated. If an alignment is specified, the value result
2617 of the allocation is guaranteed to be aligned to at least that boundary. If
2618 not specified, or if zero, the target can choose to align the allocation on any
2619 convenient boundary.</p>
2621 <p>'<tt>type</tt>' must be a sized type.</p>
2625 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2626 a pointer is returned.</p>
2631 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2633 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2634 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2635 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2636 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2637 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2641 <!-- _______________________________________________________________________ -->
2642 <div class="doc_subsubsection">
2643 <a name="i_free">'<tt>free</tt>' Instruction</a>
2646 <div class="doc_text">
2651 free <type> <value> <i>; yields {void}</i>
2656 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2657 memory heap to be reallocated in the future.</p>
2661 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2662 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2667 <p>Access to the memory pointed to by the pointer is no longer defined
2668 after this instruction executes.</p>
2673 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2674 free [4 x i8]* %array
2678 <!-- _______________________________________________________________________ -->
2679 <div class="doc_subsubsection">
2680 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2683 <div class="doc_text">
2688 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2693 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
2694 currently executing function, to be automatically released when this function
2695 returns to its caller.</p>
2699 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2700 bytes of memory on the runtime stack, returning a pointer of the
2701 appropriate type to the program. If "NumElements" is specified, it is the
2702 number of elements allocated. If an alignment is specified, the value result
2703 of the allocation is guaranteed to be aligned to at least that boundary. If
2704 not specified, or if zero, the target can choose to align the allocation on any
2705 convenient boundary.</p>
2707 <p>'<tt>type</tt>' may be any sized type.</p>
2711 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2712 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2713 instruction is commonly used to represent automatic variables that must
2714 have an address available. When the function returns (either with the <tt><a
2715 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2716 instructions), the memory is reclaimed.</p>
2721 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2722 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2723 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2724 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2728 <!-- _______________________________________________________________________ -->
2729 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2730 Instruction</a> </div>
2731 <div class="doc_text">
2733 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2735 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2737 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2738 address from which to load. The pointer must point to a <a
2739 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2740 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2741 the number or order of execution of this <tt>load</tt> with other
2742 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2745 <p>The location of memory pointed to is loaded.</p>
2747 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2749 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2750 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2753 <!-- _______________________________________________________________________ -->
2754 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2755 Instruction</a> </div>
2756 <div class="doc_text">
2758 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2759 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2762 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2764 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2765 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
2766 operand must be a pointer to the type of the '<tt><value></tt>'
2767 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2768 optimizer is not allowed to modify the number or order of execution of
2769 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2770 href="#i_store">store</a></tt> instructions.</p>
2772 <p>The contents of memory are updated to contain '<tt><value></tt>'
2773 at the location specified by the '<tt><pointer></tt>' operand.</p>
2775 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2777 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2778 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2782 <!-- _______________________________________________________________________ -->
2783 <div class="doc_subsubsection">
2784 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2787 <div class="doc_text">
2790 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2796 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2797 subelement of an aggregate data structure.</p>
2801 <p>This instruction takes a list of integer operands that indicate what
2802 elements of the aggregate object to index to. The actual types of the arguments
2803 provided depend on the type of the first pointer argument. The
2804 '<tt>getelementptr</tt>' instruction is used to index down through the type
2805 levels of a structure or to a specific index in an array. When indexing into a
2806 structure, only <tt>i32</tt> integer constants are allowed. When indexing
2807 into an array or pointer, only integers of 32 or 64 bits are allowed, and will
2808 be sign extended to 64-bit values.</p>
2810 <p>For example, let's consider a C code fragment and how it gets
2811 compiled to LLVM:</p>
2813 <div class="doc_code">
2826 int *foo(struct ST *s) {
2827 return &s[1].Z.B[5][13];
2832 <p>The LLVM code generated by the GCC frontend is:</p>
2834 <div class="doc_code">
2836 %RT = type { i8 , [10 x [20 x i32]], i8 }
2837 %ST = type { i32, double, %RT }
2839 define i32* %foo(%ST* %s) {
2841 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
2849 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2850 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2851 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
2852 <a href="#t_integer">integer</a> type but the value will always be sign extended
2853 to 64-bits. <a href="#t_struct">Structure</a> types require <tt>i32</tt>
2854 <b>constants</b>.</p>
2856 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2857 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
2858 }</tt>' type, a structure. The second index indexes into the third element of
2859 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
2860 i8 }</tt>' type, another structure. The third index indexes into the second
2861 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
2862 array. The two dimensions of the array are subscripted into, yielding an
2863 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2864 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
2866 <p>Note that it is perfectly legal to index partially through a
2867 structure, returning a pointer to an inner element. Because of this,
2868 the LLVM code for the given testcase is equivalent to:</p>
2871 define i32* %foo(%ST* %s) {
2872 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
2873 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
2874 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
2875 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
2876 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
2881 <p>Note that it is undefined to access an array out of bounds: array and
2882 pointer indexes must always be within the defined bounds of the array type.
2883 The one exception for this rules is zero length arrays. These arrays are
2884 defined to be accessible as variable length arrays, which requires access
2885 beyond the zero'th element.</p>
2887 <p>The getelementptr instruction is often confusing. For some more insight
2888 into how it works, see <a href="GetElementPtr.html">the getelementptr
2894 <i>; yields [12 x i8]*:aptr</i>
2895 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
2899 <!-- ======================================================================= -->
2900 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
2902 <div class="doc_text">
2903 <p>The instructions in this category are the conversion instructions (casting)
2904 which all take a single operand and a type. They perform various bit conversions
2908 <!-- _______________________________________________________________________ -->
2909 <div class="doc_subsubsection">
2910 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
2912 <div class="doc_text">
2916 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
2921 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
2926 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
2927 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
2928 and type of the result, which must be an <a href="#t_integer">integer</a>
2929 type. The bit size of <tt>value</tt> must be larger than the bit size of
2930 <tt>ty2</tt>. Equal sized types are not allowed.</p>
2934 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
2935 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
2936 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
2937 It will always truncate bits.</p>
2941 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
2942 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
2943 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
2947 <!-- _______________________________________________________________________ -->
2948 <div class="doc_subsubsection">
2949 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
2951 <div class="doc_text">
2955 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
2959 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
2964 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
2965 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
2966 also be of <a href="#t_integer">integer</a> type. The bit size of the
2967 <tt>value</tt> must be smaller than the bit size of the destination type,
2971 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
2972 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
2974 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
2978 %X = zext i32 257 to i64 <i>; yields i64:257</i>
2979 %Y = zext i1 true to i32 <i>; yields i32:1</i>
2983 <!-- _______________________________________________________________________ -->
2984 <div class="doc_subsubsection">
2985 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
2987 <div class="doc_text">
2991 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
2995 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
2999 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3000 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3001 also be of <a href="#t_integer">integer</a> type. The bit size of the
3002 <tt>value</tt> must be smaller than the bit size of the destination type,
3007 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3008 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3009 the type <tt>ty2</tt>.</p>
3011 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3015 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3016 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3020 <!-- _______________________________________________________________________ -->
3021 <div class="doc_subsubsection">
3022 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3025 <div class="doc_text">
3030 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3034 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3039 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3040 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3041 cast it to. The size of <tt>value</tt> must be larger than the size of
3042 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3043 <i>no-op cast</i>.</p>
3046 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3047 <a href="#t_floating">floating point</a> type to a smaller
3048 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3049 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3053 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3054 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3058 <!-- _______________________________________________________________________ -->
3059 <div class="doc_subsubsection">
3060 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3062 <div class="doc_text">
3066 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3070 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3071 floating point value.</p>
3074 <p>The '<tt>fpext</tt>' instruction takes a
3075 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3076 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3077 type must be smaller than the destination type.</p>
3080 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3081 <a href="#t_floating">floating point</a> type to a larger
3082 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3083 used to make a <i>no-op cast</i> because it always changes bits. Use
3084 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3088 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3089 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3093 <!-- _______________________________________________________________________ -->
3094 <div class="doc_subsubsection">
3095 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3097 <div class="doc_text">
3101 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3105 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3106 unsigned integer equivalent of type <tt>ty2</tt>.
3110 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3111 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
3112 must be an <a href="#t_integer">integer</a> type.</p>
3115 <p> The '<tt>fptoui</tt>' instruction converts its
3116 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3117 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3118 the results are undefined.</p>
3120 <p>When converting to i1, the conversion is done as a comparison against
3121 zero. If the <tt>value</tt> was zero, the i1 result will be <tt>false</tt>.
3122 If the <tt>value</tt> was non-zero, the i1 result will be <tt>true</tt>.</p>
3126 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3127 %Y = fptoui float 1.0E+300 to i1 <i>; yields i1:true</i>
3128 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3132 <!-- _______________________________________________________________________ -->
3133 <div class="doc_subsubsection">
3134 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3136 <div class="doc_text">
3140 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3144 <p>The '<tt>fptosi</tt>' instruction converts
3145 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3150 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3151 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
3152 must also be an <a href="#t_integer">integer</a> type.</p>
3155 <p>The '<tt>fptosi</tt>' instruction converts its
3156 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3157 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3158 the results are undefined.</p>
3160 <p>When converting to i1, the conversion is done as a comparison against
3161 zero. If the <tt>value</tt> was zero, the i1 result will be <tt>false</tt>.
3162 If the <tt>value</tt> was non-zero, the i1 result will be <tt>true</tt>.</p>
3166 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3167 %Y = fptosi float 1.0E-247 to i1 <i>; yields i1:true</i>
3168 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3172 <!-- _______________________________________________________________________ -->
3173 <div class="doc_subsubsection">
3174 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3176 <div class="doc_text">
3180 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3184 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3185 integer and converts that value to the <tt>ty2</tt> type.</p>
3189 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be an
3190 <a href="#t_integer">integer</a> value, and a type to cast it to, which must
3191 be a <a href="#t_floating">floating point</a> type.</p>
3194 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3195 integer quantity and converts it to the corresponding floating point value. If
3196 the value cannot fit in the floating point value, the results are undefined.</p>
3201 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3202 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3206 <!-- _______________________________________________________________________ -->
3207 <div class="doc_subsubsection">
3208 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3210 <div class="doc_text">
3214 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3218 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3219 integer and converts that value to the <tt>ty2</tt> type.</p>
3222 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be an
3223 <a href="#t_integer">integer</a> value, and a type to cast it to, which must be
3224 a <a href="#t_floating">floating point</a> type.</p>
3227 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3228 integer quantity and converts it to the corresponding floating point value. If
3229 the value cannot fit in the floating point value, the results are undefined.</p>
3233 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3234 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3238 <!-- _______________________________________________________________________ -->
3239 <div class="doc_subsubsection">
3240 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3242 <div class="doc_text">
3246 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3250 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3251 the integer type <tt>ty2</tt>.</p>
3254 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3255 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3256 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3259 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3260 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3261 truncating or zero extending that value to the size of the integer type. If
3262 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3263 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3264 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3269 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3270 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3274 <!-- _______________________________________________________________________ -->
3275 <div class="doc_subsubsection">
3276 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3278 <div class="doc_text">
3282 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3286 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3287 a pointer type, <tt>ty2</tt>.</p>
3290 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3291 value to cast, and a type to cast it to, which must be a
3292 <a href="#t_pointer">pointer</a> type.
3295 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3296 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3297 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3298 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3299 the size of a pointer then a zero extension is done. If they are the same size,
3300 nothing is done (<i>no-op cast</i>).</p>
3304 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3305 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3306 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3310 <!-- _______________________________________________________________________ -->
3311 <div class="doc_subsubsection">
3312 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3314 <div class="doc_text">
3318 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3322 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3323 <tt>ty2</tt> without changing any bits.</p>
3326 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3327 a first class value, and a type to cast it to, which must also be a <a
3328 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3329 and the destination type, <tt>ty2</tt>, must be identical. If the source
3330 type is a pointer, the destination type must also be a pointer.</p>
3333 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3334 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3335 this conversion. The conversion is done as if the <tt>value</tt> had been
3336 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3337 converted to other pointer types with this instruction. To convert pointers to
3338 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3339 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3343 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3344 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3345 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3349 <!-- ======================================================================= -->
3350 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3351 <div class="doc_text">
3352 <p>The instructions in this category are the "miscellaneous"
3353 instructions, which defy better classification.</p>
3356 <!-- _______________________________________________________________________ -->
3357 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3359 <div class="doc_text">
3361 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3364 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3365 of its two integer operands.</p>
3367 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3368 the condition code indicating the kind of comparison to perform. It is not
3369 a value, just a keyword. The possible condition code are:
3371 <li><tt>eq</tt>: equal</li>
3372 <li><tt>ne</tt>: not equal </li>
3373 <li><tt>ugt</tt>: unsigned greater than</li>
3374 <li><tt>uge</tt>: unsigned greater or equal</li>
3375 <li><tt>ult</tt>: unsigned less than</li>
3376 <li><tt>ule</tt>: unsigned less or equal</li>
3377 <li><tt>sgt</tt>: signed greater than</li>
3378 <li><tt>sge</tt>: signed greater or equal</li>
3379 <li><tt>slt</tt>: signed less than</li>
3380 <li><tt>sle</tt>: signed less or equal</li>
3382 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3383 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3385 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3386 the condition code given as <tt>cond</tt>. The comparison performed always
3387 yields a <a href="#t_primitive">i1</a> result, as follows:
3389 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3390 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3392 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3393 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3394 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3395 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3396 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3397 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3398 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3399 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3400 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3401 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3402 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3403 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3404 <li><tt>sge</tt>: interprets the operands as signed values and yields
3405 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3406 <li><tt>slt</tt>: interprets the operands as signed values and yields
3407 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3408 <li><tt>sle</tt>: interprets the operands as signed values and yields
3409 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3411 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3412 values are compared as if they were integers.</p>
3415 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3416 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3417 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3418 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3419 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3420 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3424 <!-- _______________________________________________________________________ -->
3425 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3427 <div class="doc_text">
3429 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3432 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3433 of its floating point operands.</p>
3435 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3436 the condition code indicating the kind of comparison to perform. It is not
3437 a value, just a keyword. The possible condition code are:
3439 <li><tt>false</tt>: no comparison, always returns false</li>
3440 <li><tt>oeq</tt>: ordered and equal</li>
3441 <li><tt>ogt</tt>: ordered and greater than </li>
3442 <li><tt>oge</tt>: ordered and greater than or equal</li>
3443 <li><tt>olt</tt>: ordered and less than </li>
3444 <li><tt>ole</tt>: ordered and less than or equal</li>
3445 <li><tt>one</tt>: ordered and not equal</li>
3446 <li><tt>ord</tt>: ordered (no nans)</li>
3447 <li><tt>ueq</tt>: unordered or equal</li>
3448 <li><tt>ugt</tt>: unordered or greater than </li>
3449 <li><tt>uge</tt>: unordered or greater than or equal</li>
3450 <li><tt>ult</tt>: unordered or less than </li>
3451 <li><tt>ule</tt>: unordered or less than or equal</li>
3452 <li><tt>une</tt>: unordered or not equal</li>
3453 <li><tt>uno</tt>: unordered (either nans)</li>
3454 <li><tt>true</tt>: no comparison, always returns true</li>
3456 <p><i>Ordered</i> means that neither operand is a QNAN while
3457 <i>unordered</i> means that either operand may be a QNAN.</p>
3458 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3459 <a href="#t_floating">floating point</a> typed. They must have identical
3462 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3463 the condition code given as <tt>cond</tt>. The comparison performed always
3464 yields a <a href="#t_primitive">i1</a> result, as follows:
3466 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3467 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3468 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3469 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3470 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3471 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3472 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3473 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3474 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3475 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3476 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3477 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3478 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3479 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3480 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3481 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3482 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3483 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3484 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3485 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3486 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3487 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3488 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3489 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3490 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3491 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3492 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3493 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3497 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3498 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3499 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3500 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3504 <!-- _______________________________________________________________________ -->
3505 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3506 Instruction</a> </div>
3507 <div class="doc_text">
3509 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3511 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3512 the SSA graph representing the function.</p>
3514 <p>The type of the incoming values is specified with the first type
3515 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3516 as arguments, with one pair for each predecessor basic block of the
3517 current block. Only values of <a href="#t_firstclass">first class</a>
3518 type may be used as the value arguments to the PHI node. Only labels
3519 may be used as the label arguments.</p>
3520 <p>There must be no non-phi instructions between the start of a basic
3521 block and the PHI instructions: i.e. PHI instructions must be first in
3524 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3525 specified by the pair corresponding to the predecessor basic block that executed
3526 just prior to the current block.</p>
3528 <pre>Loop: ; Infinite loop that counts from 0 on up...<br> %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]<br> %nextindvar = add i32 %indvar, 1<br> br label %Loop<br></pre>
3531 <!-- _______________________________________________________________________ -->
3532 <div class="doc_subsubsection">
3533 <a name="i_select">'<tt>select</tt>' Instruction</a>
3536 <div class="doc_text">
3541 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3547 The '<tt>select</tt>' instruction is used to choose one value based on a
3548 condition, without branching.
3555 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.
3561 If the boolean condition evaluates to true, the instruction returns the first
3562 value argument; otherwise, it returns the second value argument.
3568 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3573 <!-- _______________________________________________________________________ -->
3574 <div class="doc_subsubsection">
3575 <a name="i_call">'<tt>call</tt>' Instruction</a>
3578 <div class="doc_text">
3582 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
3587 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3591 <p>This instruction requires several arguments:</p>
3595 <p>The optional "tail" marker indicates whether the callee function accesses
3596 any allocas or varargs in the caller. If the "tail" marker is present, the
3597 function call is eligible for tail call optimization. Note that calls may
3598 be marked "tail" even if they do not occur before a <a
3599 href="#i_ret"><tt>ret</tt></a> instruction.
3602 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3603 convention</a> the call should use. If none is specified, the call defaults
3604 to using C calling conventions.
3607 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
3608 being invoked. The argument types must match the types implied by this
3609 signature. This type can be omitted if the function is not varargs and
3610 if the function type does not return a pointer to a function.</p>
3613 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3614 be invoked. In most cases, this is a direct function invocation, but
3615 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3616 to function value.</p>
3619 <p>'<tt>function args</tt>': argument list whose types match the
3620 function signature argument types. All arguments must be of
3621 <a href="#t_firstclass">first class</a> type. If the function signature
3622 indicates the function accepts a variable number of arguments, the extra
3623 arguments can be specified.</p>
3629 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3630 transfer to a specified function, with its incoming arguments bound to
3631 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3632 instruction in the called function, control flow continues with the
3633 instruction after the function call, and the return value of the
3634 function is bound to the result argument. This is a simpler case of
3635 the <a href="#i_invoke">invoke</a> instruction.</p>
3640 %retval = call i32 %test(i32 %argc)
3641 call i32(i8 *, ...) *%printf(i8 * %msg, i32 12, i8 42);
3642 %X = tail call i32 %foo()
3643 %Y = tail call <a href="#callingconv">fastcc</a> i32 %foo()
3648 <!-- _______________________________________________________________________ -->
3649 <div class="doc_subsubsection">
3650 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3653 <div class="doc_text">
3658 <resultval> = va_arg <va_list*> <arglist>, <argty>
3663 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3664 the "variable argument" area of a function call. It is used to implement the
3665 <tt>va_arg</tt> macro in C.</p>
3669 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3670 the argument. It returns a value of the specified argument type and
3671 increments the <tt>va_list</tt> to point to the next argument. The
3672 actual type of <tt>va_list</tt> is target specific.</p>
3676 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3677 type from the specified <tt>va_list</tt> and causes the
3678 <tt>va_list</tt> to point to the next argument. For more information,
3679 see the variable argument handling <a href="#int_varargs">Intrinsic
3682 <p>It is legal for this instruction to be called in a function which does not
3683 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3686 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3687 href="#intrinsics">intrinsic function</a> because it takes a type as an
3692 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3696 <!-- *********************************************************************** -->
3697 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3698 <!-- *********************************************************************** -->
3700 <div class="doc_text">
3702 <p>LLVM supports the notion of an "intrinsic function". These functions have
3703 well known names and semantics and are required to follow certain restrictions.
3704 Overall, these intrinsics represent an extension mechanism for the LLVM
3705 language that does not require changing all of the transformations in LLVM when
3706 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3708 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3709 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3710 begin with this prefix. Intrinsic functions must always be external functions:
3711 you cannot define the body of intrinsic functions. Intrinsic functions may
3712 only be used in call or invoke instructions: it is illegal to take the address
3713 of an intrinsic function. Additionally, because intrinsic functions are part
3714 of the LLVM language, it is required if any are added that they be documented
3717 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3718 a family of functions that perform the same operation but on different data
3719 types. Because LLVM can represent over 8 million different integer types,
3720 overloading is used commonly to allow an intrinsic function to operate on any
3721 integer type. One or more of the argument types or the result type can be
3722 overloaded to accept any integer type. Argument types may also be defined as
3723 exactly matching a previous argument's type or the result type. This allows an
3724 intrinsic function which accepts multiple arguments, but needs all of them to
3725 be of the same type, to only be overloaded with respect to a single argument or
3728 <p>Overloaded intrinsics will have the names of its overloaded argument types
3729 encoded into its function name, each preceded by a period. Only those types
3730 which are overloaded result in a name suffix. Arguments whose type is matched
3731 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
3732 take an integer of any width and returns an integer of exactly the same integer
3733 width. This leads to a family of functions such as
3734 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
3735 Only one type, the return type, is overloaded, and only one type suffix is
3736 required. Because the argument's type is matched against the return type, it
3737 does not require its own name suffix.</p>
3739 <p>To learn how to add an intrinsic function, please see the
3740 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3745 <!-- ======================================================================= -->
3746 <div class="doc_subsection">
3747 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3750 <div class="doc_text">
3752 <p>Variable argument support is defined in LLVM with the <a
3753 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3754 intrinsic functions. These functions are related to the similarly
3755 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3757 <p>All of these functions operate on arguments that use a
3758 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3759 language reference manual does not define what this type is, so all
3760 transformations should be prepared to handle these functions regardless of
3763 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3764 instruction and the variable argument handling intrinsic functions are
3767 <div class="doc_code">
3769 define i32 @test(i32 %X, ...) {
3770 ; Initialize variable argument processing
3772 %ap2 = bitcast i8** %ap to i8*
3773 call void @llvm.va_start(i8* %ap2)
3775 ; Read a single integer argument
3776 %tmp = va_arg i8** %ap, i32
3778 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3780 %aq2 = bitcast i8** %aq to i8*
3781 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3782 call void @llvm.va_end(i8* %aq2)
3784 ; Stop processing of arguments.
3785 call void @llvm.va_end(i8* %ap2)
3789 declare void @llvm.va_start(i8*)
3790 declare void @llvm.va_copy(i8*, i8*)
3791 declare void @llvm.va_end(i8*)
3797 <!-- _______________________________________________________________________ -->
3798 <div class="doc_subsubsection">
3799 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3803 <div class="doc_text">
3805 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
3807 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3808 <tt>*<arglist></tt> for subsequent use by <tt><a
3809 href="#i_va_arg">va_arg</a></tt>.</p>
3813 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3817 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3818 macro available in C. In a target-dependent way, it initializes the
3819 <tt>va_list</tt> element to which the argument points, so that the next call to
3820 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3821 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3822 last argument of the function as the compiler can figure that out.</p>
3826 <!-- _______________________________________________________________________ -->
3827 <div class="doc_subsubsection">
3828 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
3831 <div class="doc_text">
3833 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
3836 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
3837 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
3838 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
3842 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
3846 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
3847 macro available in C. In a target-dependent way, it destroys the
3848 <tt>va_list</tt> element to which the argument points. Calls to <a
3849 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
3850 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
3851 <tt>llvm.va_end</tt>.</p>
3855 <!-- _______________________________________________________________________ -->
3856 <div class="doc_subsubsection">
3857 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
3860 <div class="doc_text">
3865 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
3870 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
3871 from the source argument list to the destination argument list.</p>
3875 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
3876 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
3881 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
3882 macro available in C. In a target-dependent way, it copies the source
3883 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
3884 intrinsic is necessary because the <tt><a href="#int_va_start">
3885 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
3886 example, memory allocation.</p>
3890 <!-- ======================================================================= -->
3891 <div class="doc_subsection">
3892 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
3895 <div class="doc_text">
3898 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
3899 Collection</a> requires the implementation and generation of these intrinsics.
3900 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
3901 stack</a>, as well as garbage collector implementations that require <a
3902 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
3903 Front-ends for type-safe garbage collected languages should generate these
3904 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3905 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
3909 <!-- _______________________________________________________________________ -->
3910 <div class="doc_subsubsection">
3911 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
3914 <div class="doc_text">
3919 declare void @llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
3924 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
3925 the code generator, and allows some metadata to be associated with it.</p>
3929 <p>The first argument specifies the address of a stack object that contains the
3930 root pointer. The second pointer (which must be either a constant or a global
3931 value address) contains the meta-data to be associated with the root.</p>
3935 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
3936 location. At compile-time, the code generator generates information to allow
3937 the runtime to find the pointer at GC safe points.
3943 <!-- _______________________________________________________________________ -->
3944 <div class="doc_subsubsection">
3945 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
3948 <div class="doc_text">
3953 declare i8 * @llvm.gcread(i8 * %ObjPtr, i8 ** %Ptr)
3958 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
3959 locations, allowing garbage collector implementations that require read
3964 <p>The second argument is the address to read from, which should be an address
3965 allocated from the garbage collector. The first object is a pointer to the
3966 start of the referenced object, if needed by the language runtime (otherwise
3971 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
3972 instruction, but may be replaced with substantially more complex code by the
3973 garbage collector runtime, as needed.</p>
3978 <!-- _______________________________________________________________________ -->
3979 <div class="doc_subsubsection">
3980 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
3983 <div class="doc_text">
3988 declare void @llvm.gcwrite(i8 * %P1, i8 * %Obj, i8 ** %P2)
3993 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
3994 locations, allowing garbage collector implementations that require write
3995 barriers (such as generational or reference counting collectors).</p>
3999 <p>The first argument is the reference to store, the second is the start of the
4000 object to store it to, and the third is the address of the field of Obj to
4001 store to. If the runtime does not require a pointer to the object, Obj may be
4006 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4007 instruction, but may be replaced with substantially more complex code by the
4008 garbage collector runtime, as needed.</p>
4014 <!-- ======================================================================= -->
4015 <div class="doc_subsection">
4016 <a name="int_codegen">Code Generator Intrinsics</a>
4019 <div class="doc_text">
4021 These intrinsics are provided by LLVM to expose special features that may only
4022 be implemented with code generator support.
4027 <!-- _______________________________________________________________________ -->
4028 <div class="doc_subsubsection">
4029 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4032 <div class="doc_text">
4036 declare i8 *@llvm.returnaddress(i32 <level>)
4042 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4043 target-specific value indicating the return address of the current function
4044 or one of its callers.
4050 The argument to this intrinsic indicates which function to return the address
4051 for. Zero indicates the calling function, one indicates its caller, etc. The
4052 argument is <b>required</b> to be a constant integer value.
4058 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4059 the return address of the specified call frame, or zero if it cannot be
4060 identified. The value returned by this intrinsic is likely to be incorrect or 0
4061 for arguments other than zero, so it should only be used for debugging purposes.
4065 Note that calling this intrinsic does not prevent function inlining or other
4066 aggressive transformations, so the value returned may not be that of the obvious
4067 source-language caller.
4072 <!-- _______________________________________________________________________ -->
4073 <div class="doc_subsubsection">
4074 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4077 <div class="doc_text">
4081 declare i8 *@llvm.frameaddress(i32 <level>)
4087 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4088 target-specific frame pointer value for the specified stack frame.
4094 The argument to this intrinsic indicates which function to return the frame
4095 pointer for. Zero indicates the calling function, one indicates its caller,
4096 etc. The argument is <b>required</b> to be a constant integer value.
4102 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4103 the frame address of the specified call frame, or zero if it cannot be
4104 identified. The value returned by this intrinsic is likely to be incorrect or 0
4105 for arguments other than zero, so it should only be used for debugging purposes.
4109 Note that calling this intrinsic does not prevent function inlining or other
4110 aggressive transformations, so the value returned may not be that of the obvious
4111 source-language caller.
4115 <!-- _______________________________________________________________________ -->
4116 <div class="doc_subsubsection">
4117 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4120 <div class="doc_text">
4124 declare i8 *@llvm.stacksave()
4130 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4131 the function stack, for use with <a href="#int_stackrestore">
4132 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4133 features like scoped automatic variable sized arrays in C99.
4139 This intrinsic returns a opaque pointer value that can be passed to <a
4140 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4141 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4142 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4143 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4144 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4145 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4150 <!-- _______________________________________________________________________ -->
4151 <div class="doc_subsubsection">
4152 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4155 <div class="doc_text">
4159 declare void @llvm.stackrestore(i8 * %ptr)
4165 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4166 the function stack to the state it was in when the corresponding <a
4167 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4168 useful for implementing language features like scoped automatic variable sized
4175 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4181 <!-- _______________________________________________________________________ -->
4182 <div class="doc_subsubsection">
4183 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4186 <div class="doc_text">
4190 declare void @llvm.prefetch(i8 * <address>,
4191 i32 <rw>, i32 <locality>)
4198 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4199 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4201 effect on the behavior of the program but can change its performance
4208 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4209 determining if the fetch should be for a read (0) or write (1), and
4210 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4211 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4212 <tt>locality</tt> arguments must be constant integers.
4218 This intrinsic does not modify the behavior of the program. In particular,
4219 prefetches cannot trap and do not produce a value. On targets that support this
4220 intrinsic, the prefetch can provide hints to the processor cache for better
4226 <!-- _______________________________________________________________________ -->
4227 <div class="doc_subsubsection">
4228 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4231 <div class="doc_text">
4235 declare void @llvm.pcmarker( i32 <id> )
4242 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4244 code to simulators and other tools. The method is target specific, but it is
4245 expected that the marker will use exported symbols to transmit the PC of the marker.
4246 The marker makes no guarantees that it will remain with any specific instruction
4247 after optimizations. It is possible that the presence of a marker will inhibit
4248 optimizations. The intended use is to be inserted after optimizations to allow
4249 correlations of simulation runs.
4255 <tt>id</tt> is a numerical id identifying the marker.
4261 This intrinsic does not modify the behavior of the program. Backends that do not
4262 support this intrinisic may ignore it.
4267 <!-- _______________________________________________________________________ -->
4268 <div class="doc_subsubsection">
4269 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4272 <div class="doc_text">
4276 declare i64 @llvm.readcyclecounter( )
4283 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4284 counter register (or similar low latency, high accuracy clocks) on those targets
4285 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4286 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4287 should only be used for small timings.
4293 When directly supported, reading the cycle counter should not modify any memory.
4294 Implementations are allowed to either return a application specific value or a
4295 system wide value. On backends without support, this is lowered to a constant 0.
4300 <!-- ======================================================================= -->
4301 <div class="doc_subsection">
4302 <a name="int_libc">Standard C Library Intrinsics</a>
4305 <div class="doc_text">
4307 LLVM provides intrinsics for a few important standard C library functions.
4308 These intrinsics allow source-language front-ends to pass information about the
4309 alignment of the pointer arguments to the code generator, providing opportunity
4310 for more efficient code generation.
4315 <!-- _______________________________________________________________________ -->
4316 <div class="doc_subsubsection">
4317 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4320 <div class="doc_text">
4324 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4325 i32 <len>, i32 <align>)
4326 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4327 i64 <len>, i32 <align>)
4333 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4334 location to the destination location.
4338 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4339 intrinsics do not return a value, and takes an extra alignment argument.
4345 The first argument is a pointer to the destination, the second is a pointer to
4346 the source. The third argument is an integer argument
4347 specifying the number of bytes to copy, and the fourth argument is the alignment
4348 of the source and destination locations.
4352 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4353 the caller guarantees that both the source and destination pointers are aligned
4360 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4361 location to the destination location, which are not allowed to overlap. It
4362 copies "len" bytes of memory over. If the argument is known to be aligned to
4363 some boundary, this can be specified as the fourth argument, otherwise it should
4369 <!-- _______________________________________________________________________ -->
4370 <div class="doc_subsubsection">
4371 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4374 <div class="doc_text">
4378 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4379 i32 <len>, i32 <align>)
4380 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4381 i64 <len>, i32 <align>)
4387 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4388 location to the destination location. It is similar to the
4389 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
4393 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4394 intrinsics do not return a value, and takes an extra alignment argument.
4400 The first argument is a pointer to the destination, the second is a pointer to
4401 the source. The third argument is an integer argument
4402 specifying the number of bytes to copy, and the fourth argument is the alignment
4403 of the source and destination locations.
4407 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4408 the caller guarantees that the source and destination pointers are aligned to
4415 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4416 location to the destination location, which may overlap. It
4417 copies "len" bytes of memory over. If the argument is known to be aligned to
4418 some boundary, this can be specified as the fourth argument, otherwise it should
4424 <!-- _______________________________________________________________________ -->
4425 <div class="doc_subsubsection">
4426 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4429 <div class="doc_text">
4433 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4434 i32 <len>, i32 <align>)
4435 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4436 i64 <len>, i32 <align>)
4442 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4447 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4448 does not return a value, and takes an extra alignment argument.
4454 The first argument is a pointer to the destination to fill, the second is the
4455 byte value to fill it with, the third argument is an integer
4456 argument specifying the number of bytes to fill, and the fourth argument is the
4457 known alignment of destination location.
4461 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4462 the caller guarantees that the destination pointer is aligned to that boundary.
4468 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4470 destination location. If the argument is known to be aligned to some boundary,
4471 this can be specified as the fourth argument, otherwise it should be set to 0 or
4477 <!-- _______________________________________________________________________ -->
4478 <div class="doc_subsubsection">
4479 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4482 <div class="doc_text">
4486 declare float @llvm.sqrt.f32(float %Val)
4487 declare double @llvm.sqrt.f64(double %Val)
4493 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4494 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
4495 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4496 negative numbers (which allows for better optimization).
4502 The argument and return value are floating point numbers of the same type.
4508 This function returns the sqrt of the specified operand if it is a nonnegative
4509 floating point number.
4513 <!-- _______________________________________________________________________ -->
4514 <div class="doc_subsubsection">
4515 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4518 <div class="doc_text">
4522 declare float @llvm.powi.f32(float %Val, i32 %power)
4523 declare double @llvm.powi.f64(double %Val, i32 %power)
4529 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4530 specified (positive or negative) power. The order of evaluation of
4531 multiplications is not defined.
4537 The second argument is an integer power, and the first is a value to raise to
4544 This function returns the first value raised to the second power with an
4545 unspecified sequence of rounding operations.</p>
4549 <!-- ======================================================================= -->
4550 <div class="doc_subsection">
4551 <a name="int_manip">Bit Manipulation Intrinsics</a>
4554 <div class="doc_text">
4556 LLVM provides intrinsics for a few important bit manipulation operations.
4557 These allow efficient code generation for some algorithms.
4562 <!-- _______________________________________________________________________ -->
4563 <div class="doc_subsubsection">
4564 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4567 <div class="doc_text">
4570 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4571 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
4573 declare i16 @llvm.bswap.i16(i16 <id>)
4574 declare i32 @llvm.bswap.i32(i32 <id>)
4575 declare i64 @llvm.bswap.i64(i64 <id>)
4581 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4582 values with an even number of bytes (positive multiple of 16 bits). These are
4583 useful for performing operations on data that is not in the target's native
4590 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
4591 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4592 intrinsic returns an i32 value that has the four bytes of the input i32
4593 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
4594 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
4595 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
4596 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4601 <!-- _______________________________________________________________________ -->
4602 <div class="doc_subsubsection">
4603 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4606 <div class="doc_text">
4609 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4610 width. Not all targets support all bit widths however.
4612 declare i8 @llvm.ctpop.i8 (i8 <src>)
4613 declare i16 @llvm.ctpop.i16(i16 <src>)
4614 declare i32 @llvm.ctpop.i32(i32 <src>)
4615 declare i64 @llvm.ctpop.i64(i64 <src>)
4616 declare i256 @llvm.ctpop.i256(i256 <src>)
4622 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4629 The only argument is the value to be counted. The argument may be of any
4630 integer type. The return type must match the argument type.
4636 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4640 <!-- _______________________________________________________________________ -->
4641 <div class="doc_subsubsection">
4642 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4645 <div class="doc_text">
4648 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
4649 integer bit width. Not all targets support all bit widths however.
4651 declare i8 @llvm.ctlz.i8 (i8 <src>)
4652 declare i16 @llvm.ctlz.i16(i16 <src>)
4653 declare i32 @llvm.ctlz.i32(i32 <src>)
4654 declare i64 @llvm.ctlz.i64(i64 <src>)
4655 declare i256 @llvm.ctlz.i256(i256 <src>)
4661 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4662 leading zeros in a variable.
4668 The only argument is the value to be counted. The argument may be of any
4669 integer type. The return type must match the argument type.
4675 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4676 in a variable. If the src == 0 then the result is the size in bits of the type
4677 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
4683 <!-- _______________________________________________________________________ -->
4684 <div class="doc_subsubsection">
4685 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
4688 <div class="doc_text">
4691 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
4692 integer bit width. Not all targets support all bit widths however.
4694 declare i8 @llvm.cttz.i8 (i8 <src>)
4695 declare i16 @llvm.cttz.i16(i16 <src>)
4696 declare i32 @llvm.cttz.i32(i32 <src>)
4697 declare i64 @llvm.cttz.i64(i64 <src>)
4698 declare i256 @llvm.cttz.i256(i256 <src>)
4704 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
4711 The only argument is the value to be counted. The argument may be of any
4712 integer type. The return type must match the argument type.
4718 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
4719 in a variable. If the src == 0 then the result is the size in bits of the type
4720 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
4724 <!-- _______________________________________________________________________ -->
4725 <div class="doc_subsubsection">
4726 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
4729 <div class="doc_text">
4732 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
4733 on any integer bit width.
4735 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
4736 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
4740 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
4741 range of bits from an integer value and returns them in the same bit width as
4742 the original value.</p>
4745 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4746 any bit width but they must have the same bit width. The second and third
4747 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
4750 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
4751 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
4752 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
4753 operates in forward mode.</p>
4754 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
4755 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
4756 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
4758 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
4759 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
4760 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
4761 to determine the number of bits to retain.</li>
4762 <li>A mask of the retained bits is created by shifting a -1 value.</li>
4763 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
4765 <p>In reverse mode, a similar computation is made except that the bits are
4766 returned in the reverse order. So, for example, if <tt>X</tt> has the value
4767 <tt>i16 0x0ACF (101011001111)</tt> and we apply
4768 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
4769 <tt>i16 0x0026 (000000100110)</tt>.</p>
4772 <div class="doc_subsubsection">
4773 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
4776 <div class="doc_text">
4779 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
4780 on any integer bit width.
4782 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
4783 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
4787 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
4788 of bits in an integer value with another integer value. It returns the integer
4789 with the replaced bits.</p>
4792 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4793 any bit width but they must have the same bit width. <tt>%val</tt> is the value
4794 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
4795 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
4796 type since they specify only a bit index.</p>
4799 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
4800 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
4801 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
4802 operates in forward mode.</p>
4803 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
4804 truncating it down to the size of the replacement area or zero extending it
4805 up to that size.</p>
4806 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
4807 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
4808 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
4809 to the <tt>%hi</tt>th bit.
4810 <p>In reverse mode, a similar computation is made except that the bits are
4811 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
4812 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
4815 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
4816 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
4817 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
4818 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
4819 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
4823 <!-- ======================================================================= -->
4824 <div class="doc_subsection">
4825 <a name="int_debugger">Debugger Intrinsics</a>
4828 <div class="doc_text">
4830 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
4831 are described in the <a
4832 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
4833 Debugging</a> document.
4838 <!-- ======================================================================= -->
4839 <div class="doc_subsection">
4840 <a name="int_eh">Exception Handling Intrinsics</a>
4843 <div class="doc_text">
4844 <p> The LLVM exception handling intrinsics (which all start with
4845 <tt>llvm.eh.</tt> prefix), are described in the <a
4846 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
4847 Handling</a> document. </p>
4850 <!-- ======================================================================= -->
4851 <div class="doc_subsection">
4852 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
4855 <div class="doc_text">
4857 These intrinsic functions expand the "universal IR" of LLVM to represent
4858 hardware constructs for atomic operations and memory synchronization. This
4859 provides an interface to the hardware, not an interface to the programmer. It
4860 is aimed at a low enough level to allow any programming models or APIs which
4861 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
4862 hardware behavior. Just as hardware provides a "universal IR" for source
4863 languages, it also provides a starting point for developing a "universal"
4864 atomic operation and synchronization IR.
4867 These do <em>not</em> form an API such as high-level threading libraries,
4868 software transaction memory systems, atomic primitives, and intrinsic
4869 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
4870 application libraries. The hardware interface provided by LLVM should allow
4871 a clean implementation of all of these APIs and parallel programming models.
4872 No one model or paradigm should be selected above others unless the hardware
4873 itself ubiquitously does so.
4877 <!-- _______________________________________________________________________ -->
4878 <div class="doc_subsubsection">
4879 <a name="int_lcs">'<tt>llvm.atomic.lcs.*</tt>' Intrinsic</a>
4881 <div class="doc_text">
4884 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lcs</tt> on any
4885 integer bit width. Not all targets support all bit widths however.</p>
4887 declare i8 @llvm.atomic.lcs.i8.i8p.i8.i8( i8* <ptr>, i8 <cmp>, i8 <val> )
4888 declare i16 @llvm.atomic.lcs.i16.i16p.i16.i16( i16* <ptr>, i16 <cmp>, i16 <val> )
4889 declare i32 @llvm.atomic.lcs.i32.i32p.i32.i32( i32* <ptr>, i32 <cmp>, i32 <val> )
4890 declare i64 @llvm.atomic.lcs.i64.i64p.i64.i64( i64* <ptr>, i64 <cmp>, i64 <val> )
4894 This loads a value in memory and compares it to a given value. If they are
4895 equal, it stores a new value into the memory.
4899 The <tt>llvm.atomic.lcs</tt> intrinsic takes three arguments. The result as
4900 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
4901 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
4902 this integer type. While any bit width integer may be used, targets may only
4903 lower representations they support in hardware.
4907 This entire intrinsic must be executed atomically. It first loads the value
4908 in memory pointed to by <tt>ptr</tt> and compares it with the value
4909 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
4910 loaded value is yielded in all cases. This provides the equivalent of an
4911 atomic compare-and-swap operation within the SSA framework.
4918 %val1 = add i32 4, 4
4919 %result1 = call i32 @llvm.atomic.lcs( i32* %ptr, i32 4, %val1 )
4920 <i>; yields {i32}:result1 = 4</i>
4921 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
4922 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
4924 %val2 = add i32 1, 1
4925 %result2 = call i32 @llvm.atomic.lcs( i32* %ptr, i32 5, %val2 )
4926 <i>; yields {i32}:result2 = 8</i>
4927 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
4928 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
4932 <!-- _______________________________________________________________________ -->
4933 <div class="doc_subsubsection">
4934 <a name="int_ls">'<tt>llvm.atomic.ls.*</tt>' Intrinsic</a>
4936 <div class="doc_text">
4939 This is an overloaded intrinsic. You can use <tt>llvm.atomic.ls</tt> on any
4940 integer bit width. Not all targets support all bit widths however.</p>
4942 declare i8 @llvm.atomic.ls.i8.i8p.i8( i8* <ptr>, i8 <val> )
4943 declare i16 @llvm.atomic.ls.i16.i16p.i16( i16* <ptr>, i16 <val> )
4944 declare i32 @llvm.atomic.ls.i32.i32p.i32( i32* <ptr>, i32 <val> )
4945 declare i64 @llvm.atomic.ls.i64.i64p.i64( i64* <ptr>, i64 <val> )
4949 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
4950 the value from memory. It then stores the value in <tt>val</tt> in the memory
4955 The <tt>llvm.atomic.ls</tt> intrinsic takes two arguments. Both the
4956 <tt>val</tt> argument and the result must be integers of the same bit width.
4957 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
4958 integer type. The targets may only lower integer representations they
4963 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
4964 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
4965 equivalent of an atomic swap operation within the SSA framework.
4972 %val1 = add i32 4, 4
4973 %result1 = call i32 @llvm.atomic.ls( i32* %ptr, i32 %val1 )
4974 <i>; yields {i32}:result1 = 4</i>
4975 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
4976 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
4978 %val2 = add i32 1, 1
4979 %result2 = call i32 @llvm.atomic.ls( i32* %ptr, i32 %val2 )
4980 <i>; yields {i32}:result2 = 8</i>
4981 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
4982 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
4986 <!-- _______________________________________________________________________ -->
4987 <div class="doc_subsubsection">
4988 <a name="int_las">'<tt>llvm.atomic.las.*</tt>' Intrinsic</a>
4990 <div class="doc_text">
4993 This is an overloaded intrinsic. You can use <tt>llvm.atomic.las</tt> on any
4994 integer bit width. Not all targets support all bit widths however.</p>
4996 declare i8 @llvm.atomic.las.i8.i8p.i8( i8* <ptr>, i8 <delta> )
4997 declare i16 @llvm.atomic.las.i16.i16p.i16( i16* <ptr>, i16 <delta> )
4998 declare i32 @llvm.atomic.las.i32.i32p.i32( i32* <ptr>, i32 <delta> )
4999 declare i64 @llvm.atomic.las.i64.i64p.i64( i64* <ptr>, i64 <delta> )
5003 This intrinsic adds <tt>delta</tt> to the value stored in memory at
5004 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5008 The intrinsic takes two arguments, the first a pointer to an integer value
5009 and the second an integer value. The result is also an integer value. These
5010 integer types can have any bit width, but they must all have the same bit
5011 width. The targets may only lower integer representations they support.
5015 This intrinsic does a series of operations atomically. It first loads the
5016 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
5017 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5023 %result1 = call i32 @llvm.atomic.las( i32* %ptr, i32 4 )
5024 <i>; yields {i32}:result1 = 4</i>
5025 %result2 = call i32 @llvm.atomic.las( i32* %ptr, i32 2 )
5026 <i>; yields {i32}:result2 = 8</i>
5027 %result3 = call i32 @llvm.atomic.las( i32* %ptr, i32 5 )
5028 <i>; yields {i32}:result3 = 10</i>
5029 %memval = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
5033 <!-- _______________________________________________________________________ -->
5034 <div class="doc_subsubsection">
5035 <a name="int_lss">'<tt>llvm.atomic.lss.*</tt>' Intrinsic</a>
5037 <div class="doc_text">
5040 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lss</tt> on any
5041 integer bit width. Not all targets support all bit widths however.</p>
5043 declare i8 @llvm.atomic.lss.i8.i8.i8( i8* <ptr>, i8 <delta> )
5044 declare i16 @llvm.atomic.lss.i16.i16.i16( i16* <ptr>, i16 <delta> )
5045 declare i32 @llvm.atomic.lss.i32.i32.i32( i32* <ptr>, i32 <delta> )
5046 declare i64 @llvm.atomic.lss.i64.i64.i64( i64* <ptr>, i64 <delta> )
5050 This intrinsic subtracts <tt>delta</tt> from the value stored in memory at
5051 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5055 The intrinsic takes two arguments, the first a pointer to an integer value
5056 and the second an integer value. The result is also an integer value. These
5057 integer types can have any bit width, but they must all have the same bit
5058 width. The targets may only lower integer representations they support.
5062 This intrinsic does a series of operations atomically. It first loads the
5063 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>,
5064 stores the result to <tt>ptr</tt>. It yields the original value stored
5071 %result1 = call i32 @llvm.atomic.lss( i32* %ptr, i32 4 )
5072 <i>; yields {i32}:result1 = 32</i>
5073 %result2 = call i32 @llvm.atomic.lss( i32* %ptr, i32 2 )
5074 <i>; yields {i32}:result2 = 28</i>
5075 %result3 = call i32 @llvm.atomic.lss( i32* %ptr, i32 5 )
5076 <i>; yields {i32}:result3 = 26</i>
5077 %memval = load i32* %ptr <i>; yields {i32}:memval1 = 21</i>
5081 <!-- _______________________________________________________________________ -->
5082 <div class="doc_subsubsection">
5083 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5085 <div class="doc_text">
5088 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss> )
5092 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5093 specific pairs of memory access types.
5097 The <tt>llvm.memory.barrier</tt> intrinsic requires four boolean arguments.
5098 Each argument enables a specific barrier as listed below.
5101 <li><tt>ll</tt>: load-load barrier</li>
5102 <li><tt>ls</tt>: load-store barrier</li>
5103 <li><tt>sl</tt>: store-load barrier</li>
5104 <li><tt>ss</tt>: store-store barrier</li>
5108 This intrinsic causes the system to enforce some ordering constraints upon
5109 the loads and stores of the program. This barrier does not indicate
5110 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5111 which they occur. For any of the specified pairs of load and store operations
5112 (f.ex. load-load, or store-load), all of the first operations preceding the
5113 barrier will complete before any of the second operations succeeding the
5114 barrier begin. Specifically the semantics for each pairing is as follows:
5117 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5118 after the barrier begins.</li>
5119 <li><tt>ls</tt>: All loads before the barrier must complete before any
5120 store after the barrier begins.</li>
5121 <li><tt>ss</tt>: All stores before the barrier must complete before any
5122 store after the barrier begins.</li>
5123 <li><tt>sl</tt>: All stores before the barrier must complete before any
5124 load after the barrier begins.</li>
5127 These semantics are applied with a logical "and" behavior when more than one
5128 is enabled in a single memory barrier intrinsic.
5135 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5136 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5137 <i>; guarantee the above finishes</i>
5138 store i32 8, %ptr <i>; before this begins</i>
5142 <!-- ======================================================================= -->
5143 <div class="doc_subsection">
5144 <a name="int_trampoline">Trampoline Intrinsics</a>
5147 <div class="doc_text">
5149 These intrinsics make it possible to excise one parameter, marked with
5150 the <tt>nest</tt> attribute, from a function. The result is a callable
5151 function pointer lacking the nest parameter - the caller does not need
5152 to provide a value for it. Instead, the value to use is stored in
5153 advance in a "trampoline", a block of memory usually allocated
5154 on the stack, which also contains code to splice the nest value into the
5155 argument list. This is used to implement the GCC nested function address
5159 For example, if the function is
5160 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5161 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:
5163 %tramp1 = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5164 %tramp = getelementptr [10 x i8]* %tramp1, i32 0, i32 0
5165 call void @llvm.init.trampoline( i8* %tramp, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5166 %adj = call i8* @llvm.adjust.trampoline( i8* %tramp )
5167 %fp = bitcast i8* %adj to i32 (i32, i32)*
5169 The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent to
5170 <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.
5173 Trampolines are currently only supported on the X86 architecture.
5177 <!-- _______________________________________________________________________ -->
5178 <div class="doc_subsubsection">
5179 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5181 <div class="doc_text">
5184 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5188 This initializes the memory pointed to by <tt>tramp</tt> as a trampoline.
5192 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5193 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5194 and sufficiently aligned block of memory; this memory is written to by the
5195 intrinsic. Currently LLVM provides no help in determining just how big and
5196 aligned the memory needs to be. The <tt>func</tt> argument must hold a
5197 function bitcast to an <tt>i8*</tt>.
5201 The block of memory pointed to by <tt>tramp</tt> is filled with target
5202 dependent code, turning it into a function.
5203 The new function's signature is the same as that of <tt>func</tt> with
5204 any arguments marked with the <tt>nest</tt> attribute removed. At most
5205 one such <tt>nest</tt> argument is allowed, and it must be of pointer
5206 type. Calling the new function is equivalent to calling <tt>func</tt>
5207 with the same argument list, but with <tt>nval</tt> used for the missing
5208 <tt>nest</tt> argument.
5212 <!-- _______________________________________________________________________ -->
5213 <div class="doc_subsubsection">
5214 <a name="int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a>
5216 <div class="doc_text">
5219 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
5223 This intrinsic returns a function pointer suitable for executing
5224 the trampoline code pointed to by <tt>tramp</tt>.
5228 The <tt>llvm.adjust.trampoline</tt> takes one argument, a pointer to a
5229 trampoline initialized by the
5230 <a href="#int_it">'<tt>llvm.init.trampoline</tt>' intrinsic</a>.
5234 A function pointer that can be used to execute the trampoline code in
5235 <tt>tramp</tt> is returned. The returned value should be bitcast to an
5236 <a href="#int_trampoline">appropriate function pointer type</a>
5237 before being called.
5241 <!-- ======================================================================= -->
5242 <div class="doc_subsection">
5243 <a name="int_general">General Intrinsics</a>
5246 <div class="doc_text">
5247 <p> This class of intrinsics is designed to be generic and has
5248 no specific purpose. </p>
5251 <!-- _______________________________________________________________________ -->
5252 <div class="doc_subsubsection">
5253 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5256 <div class="doc_text">
5260 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5266 The '<tt>llvm.var.annotation</tt>' intrinsic
5272 The first argument is a pointer to a value, the second is a pointer to a
5273 global string, the third is a pointer to a global string which is the source
5274 file name, and the last argument is the line number.
5280 This intrinsic allows annotation of local variables with arbitrary strings.
5281 This can be useful for special purpose optimizations that want to look for these
5282 annotations. These have no other defined use, they are ignored by code
5283 generation and optimization.
5287 <!-- *********************************************************************** -->
5290 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5291 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5292 <a href="http://validator.w3.org/check/referer"><img
5293 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
5295 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5296 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5297 Last modified: $Date$