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 Intrinsic</a>
205 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
208 <li><a href="#int_general">General intrinsics</a>
210 <li><a href="#int_var_annotation">
211 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
218 <div class="doc_author">
219 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
220 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
223 <!-- *********************************************************************** -->
224 <div class="doc_section"> <a name="abstract">Abstract </a></div>
225 <!-- *********************************************************************** -->
227 <div class="doc_text">
228 <p>This document is a reference manual for the LLVM assembly language.
229 LLVM is an SSA based representation that provides type safety,
230 low-level operations, flexibility, and the capability of representing
231 'all' high-level languages cleanly. It is the common code
232 representation used throughout all phases of the LLVM compilation
236 <!-- *********************************************************************** -->
237 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
238 <!-- *********************************************************************** -->
240 <div class="doc_text">
242 <p>The LLVM code representation is designed to be used in three
243 different forms: as an in-memory compiler IR, as an on-disk bitcode
244 representation (suitable for fast loading by a Just-In-Time compiler),
245 and as a human readable assembly language representation. This allows
246 LLVM to provide a powerful intermediate representation for efficient
247 compiler transformations and analysis, while providing a natural means
248 to debug and visualize the transformations. The three different forms
249 of LLVM are all equivalent. This document describes the human readable
250 representation and notation.</p>
252 <p>The LLVM representation aims to be light-weight and low-level
253 while being expressive, typed, and extensible at the same time. It
254 aims to be a "universal IR" of sorts, by being at a low enough level
255 that high-level ideas may be cleanly mapped to it (similar to how
256 microprocessors are "universal IR's", allowing many source languages to
257 be mapped to them). By providing type information, LLVM can be used as
258 the target of optimizations: for example, through pointer analysis, it
259 can be proven that a C automatic variable is never accessed outside of
260 the current function... allowing it to be promoted to a simple SSA
261 value instead of a memory location.</p>
265 <!-- _______________________________________________________________________ -->
266 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
268 <div class="doc_text">
270 <p>It is important to note that this document describes 'well formed'
271 LLVM assembly language. There is a difference between what the parser
272 accepts and what is considered 'well formed'. For example, the
273 following instruction is syntactically okay, but not well formed:</p>
275 <div class="doc_code">
277 %x = <a href="#i_add">add</a> i32 1, %x
281 <p>...because the definition of <tt>%x</tt> does not dominate all of
282 its uses. The LLVM infrastructure provides a verification pass that may
283 be used to verify that an LLVM module is well formed. This pass is
284 automatically run by the parser after parsing input assembly and by
285 the optimizer before it outputs bitcode. The violations pointed out
286 by the verifier pass indicate bugs in transformation passes or input to
290 <!-- Describe the typesetting conventions here. -->
292 <!-- *********************************************************************** -->
293 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
294 <!-- *********************************************************************** -->
296 <div class="doc_text">
298 <p>LLVM identifiers come in two basic types: global and local. Global
299 identifiers (functions, global variables) begin with the @ character. Local
300 identifiers (register names, types) begin with the % character. Additionally,
301 there are three different formats for identifiers, for different purposes:
304 <li>Named values are represented as a string of characters with their prefix.
305 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
306 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
307 Identifiers which require other characters in their names can be surrounded
308 with quotes. In this way, anything except a <tt>"</tt> character can
309 be used in a named value.</li>
311 <li>Unnamed values are represented as an unsigned numeric value with their
312 prefix. For example, %12, @2, %44.</li>
314 <li>Constants, which are described in a <a href="#constants">section about
315 constants</a>, below.</li>
318 <p>LLVM requires that values start with a prefix for two reasons: Compilers
319 don't need to worry about name clashes with reserved words, and the set of
320 reserved words may be expanded in the future without penalty. Additionally,
321 unnamed identifiers allow a compiler to quickly come up with a temporary
322 variable without having to avoid symbol table conflicts.</p>
324 <p>Reserved words in LLVM are very similar to reserved words in other
325 languages. There are keywords for different opcodes
326 ('<tt><a href="#i_add">add</a></tt>',
327 '<tt><a href="#i_bitcast">bitcast</a></tt>',
328 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
329 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
330 and others. These reserved words cannot conflict with variable names, because
331 none of them start with a prefix character ('%' or '@').</p>
333 <p>Here is an example of LLVM code to multiply the integer variable
334 '<tt>%X</tt>' by 8:</p>
338 <div class="doc_code">
340 %result = <a href="#i_mul">mul</a> i32 %X, 8
344 <p>After strength reduction:</p>
346 <div class="doc_code">
348 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
352 <p>And the hard way:</p>
354 <div class="doc_code">
356 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
357 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
358 %result = <a href="#i_add">add</a> i32 %1, %1
362 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
363 important lexical features of LLVM:</p>
367 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
370 <li>Unnamed temporaries are created when the result of a computation is not
371 assigned to a named value.</li>
373 <li>Unnamed temporaries are numbered sequentially</li>
377 <p>...and it also shows a convention that we follow in this document. When
378 demonstrating instructions, we will follow an instruction with a comment that
379 defines the type and name of value produced. Comments are shown in italic
384 <!-- *********************************************************************** -->
385 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
386 <!-- *********************************************************************** -->
388 <!-- ======================================================================= -->
389 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
392 <div class="doc_text">
394 <p>LLVM programs are composed of "Module"s, each of which is a
395 translation unit of the input programs. Each module consists of
396 functions, global variables, and symbol table entries. Modules may be
397 combined together with the LLVM linker, which merges function (and
398 global variable) definitions, resolves forward declarations, and merges
399 symbol table entries. Here is an example of the "hello world" module:</p>
401 <div class="doc_code">
402 <pre><i>; Declare the string constant as a global constant...</i>
403 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
404 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
406 <i>; External declaration of the puts function</i>
407 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
409 <i>; Definition of main function</i>
410 define i32 @main() { <i>; i32()* </i>
411 <i>; Convert [13x i8 ]* to i8 *...</i>
413 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
415 <i>; Call puts function to write out the string to stdout...</i>
417 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
419 href="#i_ret">ret</a> i32 0<br>}<br>
423 <p>This example is made up of a <a href="#globalvars">global variable</a>
424 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
425 function, and a <a href="#functionstructure">function definition</a>
426 for "<tt>main</tt>".</p>
428 <p>In general, a module is made up of a list of global values,
429 where both functions and global variables are global values. Global values are
430 represented by a pointer to a memory location (in this case, a pointer to an
431 array of char, and a pointer to a function), and have one of the following <a
432 href="#linkage">linkage types</a>.</p>
436 <!-- ======================================================================= -->
437 <div class="doc_subsection">
438 <a name="linkage">Linkage Types</a>
441 <div class="doc_text">
444 All Global Variables and Functions have one of the following types of linkage:
449 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
451 <dd>Global values with internal linkage are only directly accessible by
452 objects in the current module. In particular, linking code into a module with
453 an internal global value may cause the internal to be renamed as necessary to
454 avoid collisions. Because the symbol is internal to the module, all
455 references can be updated. This corresponds to the notion of the
456 '<tt>static</tt>' keyword in C.
459 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
461 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
462 the same name when linkage occurs. This is typically used to implement
463 inline functions, templates, or other code which must be generated in each
464 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
465 allowed to be discarded.
468 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
470 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
471 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
472 used for globals that may be emitted in multiple translation units, but that
473 are not guaranteed to be emitted into every translation unit that uses them.
474 One example of this are common globals in C, such as "<tt>int X;</tt>" at
478 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
480 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
481 pointer to array type. When two global variables with appending linkage are
482 linked together, the two global arrays are appended together. This is the
483 LLVM, typesafe, equivalent of having the system linker append together
484 "sections" with identical names when .o files are linked.
487 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
488 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
489 until linked, if not linked, the symbol becomes null instead of being an
493 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
495 <dd>If none of the above identifiers are used, the global is externally
496 visible, meaning that it participates in linkage and can be used to resolve
497 external symbol references.
502 The next two types of linkage are targeted for Microsoft Windows platform
503 only. They are designed to support importing (exporting) symbols from (to)
508 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
510 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
511 or variable via a global pointer to a pointer that is set up by the DLL
512 exporting the symbol. On Microsoft Windows targets, the pointer name is
513 formed by combining <code>_imp__</code> and the function or variable name.
516 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
518 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
519 pointer to a pointer in a DLL, so that it can be referenced with the
520 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
521 name is formed by combining <code>_imp__</code> and the function or variable
527 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
528 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
529 variable and was linked with this one, one of the two would be renamed,
530 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
531 external (i.e., lacking any linkage declarations), they are accessible
532 outside of the current module.</p>
533 <p>It is illegal for a function <i>declaration</i>
534 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
535 or <tt>extern_weak</tt>.</p>
536 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
540 <!-- ======================================================================= -->
541 <div class="doc_subsection">
542 <a name="callingconv">Calling Conventions</a>
545 <div class="doc_text">
547 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
548 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
549 specified for the call. The calling convention of any pair of dynamic
550 caller/callee must match, or the behavior of the program is undefined. The
551 following calling conventions are supported by LLVM, and more may be added in
555 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
557 <dd>This calling convention (the default if no other calling convention is
558 specified) matches the target C calling conventions. This calling convention
559 supports varargs function calls and tolerates some mismatch in the declared
560 prototype and implemented declaration of the function (as does normal C).
563 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
565 <dd>This calling convention attempts to make calls as fast as possible
566 (e.g. by passing things in registers). This calling convention allows the
567 target to use whatever tricks it wants to produce fast code for the target,
568 without having to conform to an externally specified ABI. Implementations of
569 this convention should allow arbitrary tail call optimization to be supported.
570 This calling convention does not support varargs and requires the prototype of
571 all callees to exactly match the prototype of the function definition.
574 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
576 <dd>This calling convention attempts to make code in the caller as efficient
577 as possible under the assumption that the call is not commonly executed. As
578 such, these calls often preserve all registers so that the call does not break
579 any live ranges in the caller side. This calling convention does not support
580 varargs and requires the prototype of all callees to exactly match the
581 prototype of the function definition.
584 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
586 <dd>Any calling convention may be specified by number, allowing
587 target-specific calling conventions to be used. Target specific calling
588 conventions start at 64.
592 <p>More calling conventions can be added/defined on an as-needed basis, to
593 support pascal conventions or any other well-known target-independent
598 <!-- ======================================================================= -->
599 <div class="doc_subsection">
600 <a name="visibility">Visibility Styles</a>
603 <div class="doc_text">
606 All Global Variables and Functions have one of the following visibility styles:
610 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
612 <dd>On ELF, default visibility means that the declaration is visible to other
613 modules and, in shared libraries, means that the declared entity may be
614 overridden. On Darwin, default visibility means that the declaration is
615 visible to other modules. Default visibility corresponds to "external
616 linkage" in the language.
619 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
621 <dd>Two declarations of an object with hidden visibility refer to the same
622 object if they are in the same shared object. Usually, hidden visibility
623 indicates that the symbol will not be placed into the dynamic symbol table,
624 so no other module (executable or shared library) can reference it
628 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
630 <dd>On ELF, protected visibility indicates that the symbol will be placed in
631 the dynamic symbol table, but that references within the defining module will
632 bind to the local symbol. That is, the symbol cannot be overridden by another
639 <!-- ======================================================================= -->
640 <div class="doc_subsection">
641 <a name="globalvars">Global Variables</a>
644 <div class="doc_text">
646 <p>Global variables define regions of memory allocated at compilation time
647 instead of run-time. Global variables may optionally be initialized, may have
648 an explicit section to be placed in, and may have an optional explicit alignment
649 specified. A variable may be defined as "thread_local", which means that it
650 will not be shared by threads (each thread will have a separated copy of the
651 variable). A variable may be defined as a global "constant," which indicates
652 that the contents of the variable will <b>never</b> be modified (enabling better
653 optimization, allowing the global data to be placed in the read-only section of
654 an executable, etc). Note that variables that need runtime initialization
655 cannot be marked "constant" as there is a store to the variable.</p>
658 LLVM explicitly allows <em>declarations</em> of global variables to be marked
659 constant, even if the final definition of the global is not. This capability
660 can be used to enable slightly better optimization of the program, but requires
661 the language definition to guarantee that optimizations based on the
662 'constantness' are valid for the translation units that do not include the
666 <p>As SSA values, global variables define pointer values that are in
667 scope (i.e. they dominate) all basic blocks in the program. Global
668 variables always define a pointer to their "content" type because they
669 describe a region of memory, and all memory objects in LLVM are
670 accessed through pointers.</p>
672 <p>LLVM allows an explicit section to be specified for globals. If the target
673 supports it, it will emit globals to the section specified.</p>
675 <p>An explicit alignment may be specified for a global. If not present, or if
676 the alignment is set to zero, the alignment of the global is set by the target
677 to whatever it feels convenient. If an explicit alignment is specified, the
678 global is forced to have at least that much alignment. All alignments must be
681 <p>For example, the following defines a global with an initializer, section,
684 <div class="doc_code">
686 @G = constant float 1.0, section "foo", align 4
693 <!-- ======================================================================= -->
694 <div class="doc_subsection">
695 <a name="functionstructure">Functions</a>
698 <div class="doc_text">
700 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
701 an optional <a href="#linkage">linkage type</a>, an optional
702 <a href="#visibility">visibility style</a>, an optional
703 <a href="#callingconv">calling convention</a>, a return type, an optional
704 <a href="#paramattrs">parameter attribute</a> for the return type, a function
705 name, a (possibly empty) argument list (each with optional
706 <a href="#paramattrs">parameter attributes</a>), an optional section, an
707 optional alignment, an opening curly brace, a list of basic blocks, and a
710 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
711 optional <a href="#linkage">linkage type</a>, an optional
712 <a href="#visibility">visibility style</a>, an optional
713 <a href="#callingconv">calling convention</a>, a return type, an optional
714 <a href="#paramattrs">parameter attribute</a> for the return type, a function
715 name, a possibly empty list of arguments, and an optional alignment.</p>
717 <p>A function definition contains a list of basic blocks, forming the CFG for
718 the function. Each basic block may optionally start with a label (giving the
719 basic block a symbol table entry), contains a list of instructions, and ends
720 with a <a href="#terminators">terminator</a> instruction (such as a branch or
721 function return).</p>
723 <p>The first basic block in a function is special in two ways: it is immediately
724 executed on entrance to the function, and it is not allowed to have predecessor
725 basic blocks (i.e. there can not be any branches to the entry block of a
726 function). Because the block can have no predecessors, it also cannot have any
727 <a href="#i_phi">PHI nodes</a>.</p>
729 <p>LLVM allows an explicit section to be specified for functions. If the target
730 supports it, it will emit functions to the section specified.</p>
732 <p>An explicit alignment may be specified for a function. If not present, or if
733 the alignment is set to zero, the alignment of the function is set by the target
734 to whatever it feels convenient. If an explicit alignment is specified, the
735 function is forced to have at least that much alignment. All alignments must be
741 <!-- ======================================================================= -->
742 <div class="doc_subsection">
743 <a name="aliasstructure">Aliases</a>
745 <div class="doc_text">
746 <p>Aliases act as "second name" for the aliasee value (which can be either
747 function or global variable or bitcast of global value). Aliases may have an
748 optional <a href="#linkage">linkage type</a>, and an
749 optional <a href="#visibility">visibility style</a>.</p>
753 <div class="doc_code">
755 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
763 <!-- ======================================================================= -->
764 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
765 <div class="doc_text">
766 <p>The return type and each parameter of a function type may have a set of
767 <i>parameter attributes</i> associated with them. Parameter attributes are
768 used to communicate additional information about the result or parameters of
769 a function. Parameter attributes are considered to be part of the function
770 type so two functions types that differ only by the parameter attributes
771 are different function types.</p>
773 <p>Parameter attributes are simple keywords that follow the type specified. If
774 multiple parameter attributes are needed, they are space separated. For
777 <div class="doc_code">
779 %someFunc = i16 (i8 signext %someParam) zeroext
780 %someFunc = i16 (i8 zeroext %someParam) zeroext
784 <p>Note that the two function types above are unique because the parameter has
785 a different attribute (<tt>signext</tt> in the first one, <tt>zeroext</tt> in
786 the second). Also note that the attribute for the function result
787 (<tt>zeroext</tt>) comes immediately after the argument list.</p>
789 <p>Currently, only the following parameter attributes are defined:</p>
791 <dt><tt>zeroext</tt></dt>
792 <dd>This indicates that the parameter should be zero extended just before
793 a call to this function.</dd>
794 <dt><tt>signext</tt></dt>
795 <dd>This indicates that the parameter should be sign extended just before
796 a call to this function.</dd>
797 <dt><tt>inreg</tt></dt>
798 <dd>This indicates that the parameter should be placed in register (if
799 possible) during assembling function call. Support for this attribute is
801 <dt><tt>sret</tt></dt>
802 <dd>This indicates that the parameter specifies the address of a structure
803 that is the return value of the function in the source program.</dd>
804 <dt><tt>noalias</tt></dt>
805 <dd>This indicates that the parameter not alias any other object or any
806 other "noalias" objects during the function call.
807 <dt><tt>noreturn</tt></dt>
808 <dd>This function attribute indicates that the function never returns. This
809 indicates to LLVM that every call to this function should be treated as if
810 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
811 <dt><tt>nounwind</tt></dt>
812 <dd>This function attribute indicates that the function type does not use
813 the unwind instruction and does not allow stack unwinding to propagate
815 <dt><tt>nest</tt></dt>
816 <dd>This indicates that the parameter can be excised using the
817 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
822 <!-- ======================================================================= -->
823 <div class="doc_subsection">
824 <a name="moduleasm">Module-Level Inline Assembly</a>
827 <div class="doc_text">
829 Modules may contain "module-level inline asm" blocks, which corresponds to the
830 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
831 LLVM and treated as a single unit, but may be separated in the .ll file if
832 desired. The syntax is very simple:
835 <div class="doc_code">
837 module asm "inline asm code goes here"
838 module asm "more can go here"
842 <p>The strings can contain any character by escaping non-printable characters.
843 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
848 The inline asm code is simply printed to the machine code .s file when
849 assembly code is generated.
853 <!-- ======================================================================= -->
854 <div class="doc_subsection">
855 <a name="datalayout">Data Layout</a>
858 <div class="doc_text">
859 <p>A module may specify a target specific data layout string that specifies how
860 data is to be laid out in memory. The syntax for the data layout is simply:</p>
861 <pre> target datalayout = "<i>layout specification</i>"</pre>
862 <p>The <i>layout specification</i> consists of a list of specifications
863 separated by the minus sign character ('-'). Each specification starts with a
864 letter and may include other information after the letter to define some
865 aspect of the data layout. The specifications accepted are as follows: </p>
868 <dd>Specifies that the target lays out data in big-endian form. That is, the
869 bits with the most significance have the lowest address location.</dd>
871 <dd>Specifies that hte target lays out data in little-endian form. That is,
872 the bits with the least significance have the lowest address location.</dd>
873 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
874 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
875 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
876 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
878 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
879 <dd>This specifies the alignment for an integer type of a given bit
880 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
881 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
882 <dd>This specifies the alignment for a vector type of a given bit
884 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
885 <dd>This specifies the alignment for a floating point type of a given bit
886 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
888 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
889 <dd>This specifies the alignment for an aggregate type of a given bit
892 <p>When constructing the data layout for a given target, LLVM starts with a
893 default set of specifications which are then (possibly) overriden by the
894 specifications in the <tt>datalayout</tt> keyword. The default specifications
895 are given in this list:</p>
897 <li><tt>E</tt> - big endian</li>
898 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
899 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
900 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
901 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
902 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
903 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
904 alignment of 64-bits</li>
905 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
906 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
907 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
908 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
909 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
911 <p>When llvm is determining the alignment for a given type, it uses the
914 <li>If the type sought is an exact match for one of the specifications, that
915 specification is used.</li>
916 <li>If no match is found, and the type sought is an integer type, then the
917 smallest integer type that is larger than the bitwidth of the sought type is
918 used. If none of the specifications are larger than the bitwidth then the the
919 largest integer type is used. For example, given the default specifications
920 above, the i7 type will use the alignment of i8 (next largest) while both
921 i65 and i256 will use the alignment of i64 (largest specified).</li>
922 <li>If no match is found, and the type sought is a vector type, then the
923 largest vector type that is smaller than the sought vector type will be used
924 as a fall back. This happens because <128 x double> can be implemented in
925 terms of 64 <2 x double>, for example.</li>
929 <!-- *********************************************************************** -->
930 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
931 <!-- *********************************************************************** -->
933 <div class="doc_text">
935 <p>The LLVM type system is one of the most important features of the
936 intermediate representation. Being typed enables a number of
937 optimizations to be performed on the IR directly, without having to do
938 extra analyses on the side before the transformation. A strong type
939 system makes it easier to read the generated code and enables novel
940 analyses and transformations that are not feasible to perform on normal
941 three address code representations.</p>
945 <!-- ======================================================================= -->
946 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
947 <div class="doc_text">
948 <p>The primitive types are the fundamental building blocks of the LLVM
949 system. The current set of primitive types is as follows:</p>
951 <table class="layout">
956 <tr><th>Type</th><th>Description</th></tr>
957 <tr><td><tt><a name="t_void">void</a></tt></td><td>No value</td></tr>
958 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
965 <tr><th>Type</th><th>Description</th></tr>
966 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
967 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
975 <!-- _______________________________________________________________________ -->
976 <div class="doc_subsubsection"> <a name="t_classifications">Type
977 Classifications</a> </div>
978 <div class="doc_text">
979 <p>These different primitive types fall into a few useful
982 <table border="1" cellspacing="0" cellpadding="4">
984 <tr><th>Classification</th><th>Types</th></tr>
986 <td><a name="t_integer">integer</a></td>
987 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
990 <td><a name="t_floating">floating point</a></td>
991 <td><tt>float, double</tt></td>
994 <td><a name="t_firstclass">first class</a></td>
995 <td><tt>i1, ..., float, double, <br/>
996 <a href="#t_pointer">pointer</a>,<a href="#t_vector">vector</a></tt>
1002 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1003 most important. Values of these types are the only ones which can be
1004 produced by instructions, passed as arguments, or used as operands to
1005 instructions. This means that all structures and arrays must be
1006 manipulated either by pointer or by component.</p>
1009 <!-- ======================================================================= -->
1010 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1012 <div class="doc_text">
1014 <p>The real power in LLVM comes from the derived types in the system.
1015 This is what allows a programmer to represent arrays, functions,
1016 pointers, and other useful types. Note that these derived types may be
1017 recursive: For example, it is possible to have a two dimensional array.</p>
1021 <!-- _______________________________________________________________________ -->
1022 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1024 <div class="doc_text">
1027 <p>The integer type is a very simple derived type that simply specifies an
1028 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1029 2^23-1 (about 8 million) can be specified.</p>
1037 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1041 <table class="layout">
1051 <tt>i1942652</tt><br/>
1054 A boolean integer of 1 bit<br/>
1055 A nibble sized integer of 4 bits.<br/>
1056 A byte sized integer of 8 bits.<br/>
1057 A half word sized integer of 16 bits.<br/>
1058 A word sized integer of 32 bits.<br/>
1059 An integer whose bit width is the answer. <br/>
1060 A double word sized integer of 64 bits.<br/>
1061 A really big integer of over 1 million bits.<br/>
1067 <!-- _______________________________________________________________________ -->
1068 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1070 <div class="doc_text">
1074 <p>The array type is a very simple derived type that arranges elements
1075 sequentially in memory. The array type requires a size (number of
1076 elements) and an underlying data type.</p>
1081 [<# elements> x <elementtype>]
1084 <p>The number of elements is a constant integer value; elementtype may
1085 be any type with a size.</p>
1088 <table class="layout">
1091 <tt>[40 x i32 ]</tt><br/>
1092 <tt>[41 x i32 ]</tt><br/>
1093 <tt>[40 x i8]</tt><br/>
1096 Array of 40 32-bit integer values.<br/>
1097 Array of 41 32-bit integer values.<br/>
1098 Array of 40 8-bit integer values.<br/>
1102 <p>Here are some examples of multidimensional arrays:</p>
1103 <table class="layout">
1106 <tt>[3 x [4 x i32]]</tt><br/>
1107 <tt>[12 x [10 x float]]</tt><br/>
1108 <tt>[2 x [3 x [4 x i16]]]</tt><br/>
1111 3x4 array of 32-bit integer values.<br/>
1112 12x10 array of single precision floating point values.<br/>
1113 2x3x4 array of 16-bit integer values.<br/>
1118 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1119 length array. Normally, accesses past the end of an array are undefined in
1120 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1121 As a special case, however, zero length arrays are recognized to be variable
1122 length. This allows implementation of 'pascal style arrays' with the LLVM
1123 type "{ i32, [0 x float]}", for example.</p>
1127 <!-- _______________________________________________________________________ -->
1128 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1129 <div class="doc_text">
1131 <p>The function type can be thought of as a function signature. It
1132 consists of a return type and a list of formal parameter types.
1133 Function types are usually used to build virtual function tables
1134 (which are structures of pointers to functions), for indirect function
1135 calls, and when defining a function.</p>
1137 The return type of a function type cannot be an aggregate type.
1140 <pre> <returntype> (<parameter list>)<br></pre>
1141 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1142 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1143 which indicates that the function takes a variable number of arguments.
1144 Variable argument functions can access their arguments with the <a
1145 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
1147 <table class="layout">
1149 <td class="left"><tt>i32 (i32)</tt></td>
1150 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1152 </tr><tr class="layout">
1153 <td class="left"><tt>float (i16 signext, i32 *) *
1155 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1156 an <tt>i16</tt> that should be sign extended and a
1157 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1160 </tr><tr class="layout">
1161 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1162 <td class="left">A vararg function that takes at least one
1163 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1164 which returns an integer. This is the signature for <tt>printf</tt> in
1171 <!-- _______________________________________________________________________ -->
1172 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1173 <div class="doc_text">
1175 <p>The structure type is used to represent a collection of data members
1176 together in memory. The packing of the field types is defined to match
1177 the ABI of the underlying processor. The elements of a structure may
1178 be any type that has a size.</p>
1179 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1180 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1181 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1184 <pre> { <type list> }<br></pre>
1186 <table class="layout">
1188 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1189 <td class="left">A triple of three <tt>i32</tt> values</td>
1190 </tr><tr class="layout">
1191 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1192 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1193 second element is a <a href="#t_pointer">pointer</a> to a
1194 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1195 an <tt>i32</tt>.</td>
1200 <!-- _______________________________________________________________________ -->
1201 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1203 <div class="doc_text">
1205 <p>The packed structure type is used to represent a collection of data members
1206 together in memory. There is no padding between fields. Further, the alignment
1207 of a packed structure is 1 byte. The elements of a packed structure may
1208 be any type that has a size.</p>
1209 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1210 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1211 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1214 <pre> < { <type list> } > <br></pre>
1216 <table class="layout">
1218 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1219 <td class="left">A triple of three <tt>i32</tt> values</td>
1220 </tr><tr class="layout">
1221 <td class="left"><tt>< { float, i32 (i32) * } ></tt></td>
1222 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1223 second element is a <a href="#t_pointer">pointer</a> to a
1224 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1225 an <tt>i32</tt>.</td>
1230 <!-- _______________________________________________________________________ -->
1231 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1232 <div class="doc_text">
1234 <p>As in many languages, the pointer type represents a pointer or
1235 reference to another object, which must live in memory.</p>
1237 <pre> <type> *<br></pre>
1239 <table class="layout">
1242 <tt>[4x i32]*</tt><br/>
1243 <tt>i32 (i32 *) *</tt><br/>
1246 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
1247 four <tt>i32</tt> values<br/>
1248 A <a href="#t_pointer">pointer</a> to a <a
1249 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1256 <!-- _______________________________________________________________________ -->
1257 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1258 <div class="doc_text">
1262 <p>A vector type is a simple derived type that represents a vector
1263 of elements. Vector types are used when multiple primitive data
1264 are operated in parallel using a single instruction (SIMD).
1265 A vector type requires a size (number of
1266 elements) and an underlying primitive data type. Vectors must have a power
1267 of two length (1, 2, 4, 8, 16 ...). Vector types are
1268 considered <a href="#t_firstclass">first class</a>.</p>
1273 < <# elements> x <elementtype> >
1276 <p>The number of elements is a constant integer value; elementtype may
1277 be any integer or floating point type.</p>
1281 <table class="layout">
1284 <tt><4 x i32></tt><br/>
1285 <tt><8 x float></tt><br/>
1286 <tt><2 x i64></tt><br/>
1289 Vector of 4 32-bit integer values.<br/>
1290 Vector of 8 floating-point values.<br/>
1291 Vector of 2 64-bit integer values.<br/>
1297 <!-- _______________________________________________________________________ -->
1298 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1299 <div class="doc_text">
1303 <p>Opaque types are used to represent unknown types in the system. This
1304 corresponds (for example) to the C notion of a foward declared structure type.
1305 In LLVM, opaque types can eventually be resolved to any type (not just a
1306 structure type).</p>
1316 <table class="layout">
1322 An opaque type.<br/>
1329 <!-- *********************************************************************** -->
1330 <div class="doc_section"> <a name="constants">Constants</a> </div>
1331 <!-- *********************************************************************** -->
1333 <div class="doc_text">
1335 <p>LLVM has several different basic types of constants. This section describes
1336 them all and their syntax.</p>
1340 <!-- ======================================================================= -->
1341 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1343 <div class="doc_text">
1346 <dt><b>Boolean constants</b></dt>
1348 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1349 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1352 <dt><b>Integer constants</b></dt>
1354 <dd>Standard integers (such as '4') are constants of the <a
1355 href="#t_integer">integer</a> type. Negative numbers may be used with
1359 <dt><b>Floating point constants</b></dt>
1361 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1362 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1363 notation (see below). Floating point constants must have a <a
1364 href="#t_floating">floating point</a> type. </dd>
1366 <dt><b>Null pointer constants</b></dt>
1368 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1369 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1373 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1374 of floating point constants. For example, the form '<tt>double
1375 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1376 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1377 (and the only time that they are generated by the disassembler) is when a
1378 floating point constant must be emitted but it cannot be represented as a
1379 decimal floating point number. For example, NaN's, infinities, and other
1380 special values are represented in their IEEE hexadecimal format so that
1381 assembly and disassembly do not cause any bits to change in the constants.</p>
1385 <!-- ======================================================================= -->
1386 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1389 <div class="doc_text">
1390 <p>Aggregate constants arise from aggregation of simple constants
1391 and smaller aggregate constants.</p>
1394 <dt><b>Structure constants</b></dt>
1396 <dd>Structure constants are represented with notation similar to structure
1397 type definitions (a comma separated list of elements, surrounded by braces
1398 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* %G }</tt>",
1399 where "<tt>%G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1400 must have <a href="#t_struct">structure type</a>, and the number and
1401 types of elements must match those specified by the type.
1404 <dt><b>Array constants</b></dt>
1406 <dd>Array constants are represented with notation similar to array type
1407 definitions (a comma separated list of elements, surrounded by square brackets
1408 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1409 constants must have <a href="#t_array">array type</a>, and the number and
1410 types of elements must match those specified by the type.
1413 <dt><b>Vector constants</b></dt>
1415 <dd>Vector constants are represented with notation similar to vector type
1416 definitions (a comma separated list of elements, surrounded by
1417 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1418 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1419 href="#t_vector">vector type</a>, and the number and types of elements must
1420 match those specified by the type.
1423 <dt><b>Zero initialization</b></dt>
1425 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1426 value to zero of <em>any</em> type, including scalar and aggregate types.
1427 This is often used to avoid having to print large zero initializers (e.g. for
1428 large arrays) and is always exactly equivalent to using explicit zero
1435 <!-- ======================================================================= -->
1436 <div class="doc_subsection">
1437 <a name="globalconstants">Global Variable and Function Addresses</a>
1440 <div class="doc_text">
1442 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1443 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1444 constants. These constants are explicitly referenced when the <a
1445 href="#identifiers">identifier for the global</a> is used and always have <a
1446 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1449 <div class="doc_code">
1453 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1459 <!-- ======================================================================= -->
1460 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1461 <div class="doc_text">
1462 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1463 no specific value. Undefined values may be of any type and be used anywhere
1464 a constant is permitted.</p>
1466 <p>Undefined values indicate to the compiler that the program is well defined
1467 no matter what value is used, giving the compiler more freedom to optimize.
1471 <!-- ======================================================================= -->
1472 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1475 <div class="doc_text">
1477 <p>Constant expressions are used to allow expressions involving other constants
1478 to be used as constants. Constant expressions may be of any <a
1479 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1480 that does not have side effects (e.g. load and call are not supported). The
1481 following is the syntax for constant expressions:</p>
1484 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1485 <dd>Truncate a constant to another type. The bit size of CST must be larger
1486 than the bit size of TYPE. Both types must be integers.</dd>
1488 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1489 <dd>Zero extend a constant to another type. The bit size of CST must be
1490 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1492 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1493 <dd>Sign extend a constant to another type. The bit size of CST must be
1494 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1496 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1497 <dd>Truncate a floating point constant to another floating point type. The
1498 size of CST must be larger than the size of TYPE. Both types must be
1499 floating point.</dd>
1501 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1502 <dd>Floating point extend a constant to another type. The size of CST must be
1503 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1505 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1506 <dd>Convert a floating point constant to the corresponding unsigned integer
1507 constant. TYPE must be an integer type. CST must be floating point. If the
1508 value won't fit in the integer type, the results are undefined.</dd>
1510 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1511 <dd>Convert a floating point constant to the corresponding signed integer
1512 constant. TYPE must be an integer type. CST must be floating point. If the
1513 value won't fit in the integer type, the results are undefined.</dd>
1515 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1516 <dd>Convert an unsigned integer constant to the corresponding floating point
1517 constant. TYPE must be floating point. CST must be of integer type. If the
1518 value won't fit in the floating point type, the results are undefined.</dd>
1520 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1521 <dd>Convert a signed integer constant to the corresponding floating point
1522 constant. TYPE must be floating point. CST must be of integer type. If the
1523 value won't fit in the floating point type, the results are undefined.</dd>
1525 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1526 <dd>Convert a pointer typed constant to the corresponding integer constant
1527 TYPE must be an integer type. CST must be of pointer type. The CST value is
1528 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1530 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1531 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1532 pointer type. CST must be of integer type. The CST value is zero extended,
1533 truncated, or unchanged to make it fit in a pointer size. This one is
1534 <i>really</i> dangerous!</dd>
1536 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1537 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1538 identical (same number of bits). The conversion is done as if the CST value
1539 was stored to memory and read back as TYPE. In other words, no bits change
1540 with this operator, just the type. This can be used for conversion of
1541 vector types to any other type, as long as they have the same bit width. For
1542 pointers it is only valid to cast to another pointer type.
1545 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1547 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1548 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1549 instruction, the index list may have zero or more indexes, which are required
1550 to make sense for the type of "CSTPTR".</dd>
1552 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1554 <dd>Perform the <a href="#i_select">select operation</a> on
1557 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1558 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1560 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1561 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1563 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1565 <dd>Perform the <a href="#i_extractelement">extractelement
1566 operation</a> on constants.
1568 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1570 <dd>Perform the <a href="#i_insertelement">insertelement
1571 operation</a> on constants.</dd>
1574 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1576 <dd>Perform the <a href="#i_shufflevector">shufflevector
1577 operation</a> on constants.</dd>
1579 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1581 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1582 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1583 binary</a> operations. The constraints on operands are the same as those for
1584 the corresponding instruction (e.g. no bitwise operations on floating point
1585 values are allowed).</dd>
1589 <!-- *********************************************************************** -->
1590 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1591 <!-- *********************************************************************** -->
1593 <!-- ======================================================================= -->
1594 <div class="doc_subsection">
1595 <a name="inlineasm">Inline Assembler Expressions</a>
1598 <div class="doc_text">
1601 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1602 Module-Level Inline Assembly</a>) through the use of a special value. This
1603 value represents the inline assembler as a string (containing the instructions
1604 to emit), a list of operand constraints (stored as a string), and a flag that
1605 indicates whether or not the inline asm expression has side effects. An example
1606 inline assembler expression is:
1609 <div class="doc_code">
1611 i32 (i32) asm "bswap $0", "=r,r"
1616 Inline assembler expressions may <b>only</b> be used as the callee operand of
1617 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1620 <div class="doc_code">
1622 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1627 Inline asms with side effects not visible in the constraint list must be marked
1628 as having side effects. This is done through the use of the
1629 '<tt>sideeffect</tt>' keyword, like so:
1632 <div class="doc_code">
1634 call void asm sideeffect "eieio", ""()
1638 <p>TODO: The format of the asm and constraints string still need to be
1639 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1640 need to be documented).
1645 <!-- *********************************************************************** -->
1646 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1647 <!-- *********************************************************************** -->
1649 <div class="doc_text">
1651 <p>The LLVM instruction set consists of several different
1652 classifications of instructions: <a href="#terminators">terminator
1653 instructions</a>, <a href="#binaryops">binary instructions</a>,
1654 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1655 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1656 instructions</a>.</p>
1660 <!-- ======================================================================= -->
1661 <div class="doc_subsection"> <a name="terminators">Terminator
1662 Instructions</a> </div>
1664 <div class="doc_text">
1666 <p>As mentioned <a href="#functionstructure">previously</a>, every
1667 basic block in a program ends with a "Terminator" instruction, which
1668 indicates which block should be executed after the current block is
1669 finished. These terminator instructions typically yield a '<tt>void</tt>'
1670 value: they produce control flow, not values (the one exception being
1671 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1672 <p>There are six different terminator instructions: the '<a
1673 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1674 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1675 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1676 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1677 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1681 <!-- _______________________________________________________________________ -->
1682 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1683 Instruction</a> </div>
1684 <div class="doc_text">
1686 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1687 ret void <i>; Return from void function</i>
1690 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1691 value) from a function back to the caller.</p>
1692 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1693 returns a value and then causes control flow, and one that just causes
1694 control flow to occur.</p>
1696 <p>The '<tt>ret</tt>' instruction may return any '<a
1697 href="#t_firstclass">first class</a>' type. Notice that a function is
1698 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1699 instruction inside of the function that returns a value that does not
1700 match the return type of the function.</p>
1702 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1703 returns back to the calling function's context. If the caller is a "<a
1704 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1705 the instruction after the call. If the caller was an "<a
1706 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1707 at the beginning of the "normal" destination block. If the instruction
1708 returns a value, that value shall set the call or invoke instruction's
1711 <pre> ret i32 5 <i>; Return an integer value of 5</i>
1712 ret void <i>; Return from a void function</i>
1715 <!-- _______________________________________________________________________ -->
1716 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1717 <div class="doc_text">
1719 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1722 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1723 transfer to a different basic block in the current function. There are
1724 two forms of this instruction, corresponding to a conditional branch
1725 and an unconditional branch.</p>
1727 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1728 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1729 unconditional form of the '<tt>br</tt>' instruction takes a single
1730 '<tt>label</tt>' value as a target.</p>
1732 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1733 argument is evaluated. If the value is <tt>true</tt>, control flows
1734 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1735 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1737 <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
1738 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1740 <!-- _______________________________________________________________________ -->
1741 <div class="doc_subsubsection">
1742 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1745 <div class="doc_text">
1749 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1754 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1755 several different places. It is a generalization of the '<tt>br</tt>'
1756 instruction, allowing a branch to occur to one of many possible
1762 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1763 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1764 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1765 table is not allowed to contain duplicate constant entries.</p>
1769 <p>The <tt>switch</tt> instruction specifies a table of values and
1770 destinations. When the '<tt>switch</tt>' instruction is executed, this
1771 table is searched for the given value. If the value is found, control flow is
1772 transfered to the corresponding destination; otherwise, control flow is
1773 transfered to the default destination.</p>
1775 <h5>Implementation:</h5>
1777 <p>Depending on properties of the target machine and the particular
1778 <tt>switch</tt> instruction, this instruction may be code generated in different
1779 ways. For example, it could be generated as a series of chained conditional
1780 branches or with a lookup table.</p>
1785 <i>; Emulate a conditional br instruction</i>
1786 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1787 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1789 <i>; Emulate an unconditional br instruction</i>
1790 switch i32 0, label %dest [ ]
1792 <i>; Implement a jump table:</i>
1793 switch i32 %val, label %otherwise [ i32 0, label %onzero
1795 i32 2, label %ontwo ]
1799 <!-- _______________________________________________________________________ -->
1800 <div class="doc_subsubsection">
1801 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1804 <div class="doc_text">
1809 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1810 to label <normal label> unwind label <exception label>
1815 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1816 function, with the possibility of control flow transfer to either the
1817 '<tt>normal</tt>' label or the
1818 '<tt>exception</tt>' label. If the callee function returns with the
1819 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1820 "normal" label. If the callee (or any indirect callees) returns with the "<a
1821 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1822 continued at the dynamically nearest "exception" label.</p>
1826 <p>This instruction requires several arguments:</p>
1830 The optional "cconv" marker indicates which <a href="#callingconv">calling
1831 convention</a> the call should use. If none is specified, the call defaults
1832 to using C calling conventions.
1834 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1835 function value being invoked. In most cases, this is a direct function
1836 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1837 an arbitrary pointer to function value.
1840 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1841 function to be invoked. </li>
1843 <li>'<tt>function args</tt>': argument list whose types match the function
1844 signature argument types. If the function signature indicates the function
1845 accepts a variable number of arguments, the extra arguments can be
1848 <li>'<tt>normal label</tt>': the label reached when the called function
1849 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1851 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1852 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1858 <p>This instruction is designed to operate as a standard '<tt><a
1859 href="#i_call">call</a></tt>' instruction in most regards. The primary
1860 difference is that it establishes an association with a label, which is used by
1861 the runtime library to unwind the stack.</p>
1863 <p>This instruction is used in languages with destructors to ensure that proper
1864 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1865 exception. Additionally, this is important for implementation of
1866 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1870 %retval = invoke i32 %Test(i32 15) to label %Continue
1871 unwind label %TestCleanup <i>; {i32}:retval set</i>
1872 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Test(i32 15) to label %Continue
1873 unwind label %TestCleanup <i>; {i32}:retval set</i>
1878 <!-- _______________________________________________________________________ -->
1880 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1881 Instruction</a> </div>
1883 <div class="doc_text">
1892 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1893 at the first callee in the dynamic call stack which used an <a
1894 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1895 primarily used to implement exception handling.</p>
1899 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1900 immediately halt. The dynamic call stack is then searched for the first <a
1901 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1902 execution continues at the "exceptional" destination block specified by the
1903 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1904 dynamic call chain, undefined behavior results.</p>
1907 <!-- _______________________________________________________________________ -->
1909 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1910 Instruction</a> </div>
1912 <div class="doc_text">
1921 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1922 instruction is used to inform the optimizer that a particular portion of the
1923 code is not reachable. This can be used to indicate that the code after a
1924 no-return function cannot be reached, and other facts.</p>
1928 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1933 <!-- ======================================================================= -->
1934 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1935 <div class="doc_text">
1936 <p>Binary operators are used to do most of the computation in a
1937 program. They require two operands, execute an operation on them, and
1938 produce a single value. The operands might represent
1939 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
1940 The result value of a binary operator is not
1941 necessarily the same type as its operands.</p>
1942 <p>There are several different binary operators:</p>
1944 <!-- _______________________________________________________________________ -->
1945 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1946 Instruction</a> </div>
1947 <div class="doc_text">
1949 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1952 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1954 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1955 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1956 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1957 Both arguments must have identical types.</p>
1959 <p>The value produced is the integer or floating point sum of the two
1962 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
1965 <!-- _______________________________________________________________________ -->
1966 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1967 Instruction</a> </div>
1968 <div class="doc_text">
1970 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1973 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1975 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1976 instruction present in most other intermediate representations.</p>
1978 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1979 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1981 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1982 Both arguments must have identical types.</p>
1984 <p>The value produced is the integer or floating point difference of
1985 the two operands.</p>
1988 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
1989 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
1992 <!-- _______________________________________________________________________ -->
1993 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1994 Instruction</a> </div>
1995 <div class="doc_text">
1997 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2000 <p>The '<tt>mul</tt>' instruction returns the product of its two
2003 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2004 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2006 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2007 Both arguments must have identical types.</p>
2009 <p>The value produced is the integer or floating point product of the
2011 <p>Because the operands are the same width, the result of an integer
2012 multiplication is the same whether the operands should be deemed unsigned or
2015 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2018 <!-- _______________________________________________________________________ -->
2019 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2021 <div class="doc_text">
2023 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2026 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2029 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2030 <a href="#t_integer">integer</a> values. Both arguments must have identical
2031 types. This instruction can also take <a href="#t_vector">vector</a> versions
2032 of the values in which case the elements must be integers.</p>
2034 <p>The value produced is the unsigned integer quotient of the two operands. This
2035 instruction always performs an unsigned division operation, regardless of
2036 whether the arguments are unsigned or not.</p>
2038 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2041 <!-- _______________________________________________________________________ -->
2042 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2044 <div class="doc_text">
2046 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2049 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2052 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2053 <a href="#t_integer">integer</a> values. Both arguments must have identical
2054 types. This instruction can also take <a href="#t_vector">vector</a> versions
2055 of the values in which case the elements must be integers.</p>
2057 <p>The value produced is the signed integer quotient of the two operands. This
2058 instruction always performs a signed division operation, regardless of whether
2059 the arguments are signed or not.</p>
2061 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2064 <!-- _______________________________________________________________________ -->
2065 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2066 Instruction</a> </div>
2067 <div class="doc_text">
2069 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2072 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2075 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2076 <a href="#t_floating">floating point</a> values. Both arguments must have
2077 identical types. This instruction can also take <a href="#t_vector">vector</a>
2078 versions of floating point values.</p>
2080 <p>The value produced is the floating point quotient of the two operands.</p>
2082 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2085 <!-- _______________________________________________________________________ -->
2086 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2088 <div class="doc_text">
2090 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2093 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2094 unsigned division of its two arguments.</p>
2096 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2097 <a href="#t_integer">integer</a> values. Both arguments must have identical
2100 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2101 This instruction always performs an unsigned division to get the remainder,
2102 regardless of whether the arguments are unsigned or not.</p>
2104 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2108 <!-- _______________________________________________________________________ -->
2109 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2110 Instruction</a> </div>
2111 <div class="doc_text">
2113 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2116 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2117 signed division of its two operands.</p>
2119 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2120 <a href="#t_integer">integer</a> values. Both arguments must have identical
2123 <p>This instruction returns the <i>remainder</i> of a division (where the result
2124 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2125 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2126 a value. For more information about the difference, see <a
2127 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2128 Math Forum</a>. For a table of how this is implemented in various languages,
2129 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2130 Wikipedia: modulo operation</a>.</p>
2132 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2136 <!-- _______________________________________________________________________ -->
2137 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2138 Instruction</a> </div>
2139 <div class="doc_text">
2141 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2144 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2145 division of its two operands.</p>
2147 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2148 <a href="#t_floating">floating point</a> values. Both arguments must have
2149 identical types.</p>
2151 <p>This instruction returns the <i>remainder</i> of a division.</p>
2153 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2157 <!-- ======================================================================= -->
2158 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2159 Operations</a> </div>
2160 <div class="doc_text">
2161 <p>Bitwise binary operators are used to do various forms of
2162 bit-twiddling in a program. They are generally very efficient
2163 instructions and can commonly be strength reduced from other
2164 instructions. They require two operands, execute an operation on them,
2165 and produce a single value. The resulting value of the bitwise binary
2166 operators is always the same type as its first operand.</p>
2169 <!-- _______________________________________________________________________ -->
2170 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2171 Instruction</a> </div>
2172 <div class="doc_text">
2174 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2177 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2178 the left a specified number of bits.</p>
2180 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2181 href="#t_integer">integer</a> type.</p>
2183 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
2184 <h5>Example:</h5><pre>
2185 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2186 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2187 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2190 <!-- _______________________________________________________________________ -->
2191 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2192 Instruction</a> </div>
2193 <div class="doc_text">
2195 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2199 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2200 operand shifted to the right a specified number of bits with zero fill.</p>
2203 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2204 <a href="#t_integer">integer</a> type.</p>
2207 <p>This instruction always performs a logical shift right operation. The most
2208 significant bits of the result will be filled with zero bits after the
2213 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2214 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2215 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2216 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2220 <!-- _______________________________________________________________________ -->
2221 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2222 Instruction</a> </div>
2223 <div class="doc_text">
2226 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2230 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2231 operand shifted to the right a specified number of bits with sign extension.</p>
2234 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2235 <a href="#t_integer">integer</a> type.</p>
2238 <p>This instruction always performs an arithmetic shift right operation,
2239 The most significant bits of the result will be filled with the sign bit
2240 of <tt>var1</tt>.</p>
2244 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2245 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2246 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2247 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2251 <!-- _______________________________________________________________________ -->
2252 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2253 Instruction</a> </div>
2254 <div class="doc_text">
2256 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2259 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2260 its two operands.</p>
2262 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2263 href="#t_integer">integer</a> values. Both arguments must have
2264 identical types.</p>
2266 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2268 <div style="align: center">
2269 <table border="1" cellspacing="0" cellpadding="4">
2300 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2301 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2302 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2305 <!-- _______________________________________________________________________ -->
2306 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2307 <div class="doc_text">
2309 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2312 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2313 or of its two operands.</p>
2315 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2316 href="#t_integer">integer</a> values. Both arguments must have
2317 identical types.</p>
2319 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2321 <div style="align: center">
2322 <table border="1" cellspacing="0" cellpadding="4">
2353 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2354 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2355 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2358 <!-- _______________________________________________________________________ -->
2359 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2360 Instruction</a> </div>
2361 <div class="doc_text">
2363 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2366 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2367 or of its two operands. The <tt>xor</tt> is used to implement the
2368 "one's complement" operation, which is the "~" operator in C.</p>
2370 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2371 href="#t_integer">integer</a> values. Both arguments must have
2372 identical types.</p>
2374 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2376 <div style="align: center">
2377 <table border="1" cellspacing="0" cellpadding="4">
2409 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2410 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2411 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2412 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2416 <!-- ======================================================================= -->
2417 <div class="doc_subsection">
2418 <a name="vectorops">Vector Operations</a>
2421 <div class="doc_text">
2423 <p>LLVM supports several instructions to represent vector operations in a
2424 target-independent manner. These instructions cover the element-access and
2425 vector-specific operations needed to process vectors effectively. While LLVM
2426 does directly support these vector operations, many sophisticated algorithms
2427 will want to use target-specific intrinsics to take full advantage of a specific
2432 <!-- _______________________________________________________________________ -->
2433 <div class="doc_subsubsection">
2434 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2437 <div class="doc_text">
2442 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2448 The '<tt>extractelement</tt>' instruction extracts a single scalar
2449 element from a vector at a specified index.
2456 The first operand of an '<tt>extractelement</tt>' instruction is a
2457 value of <a href="#t_vector">vector</a> type. The second operand is
2458 an index indicating the position from which to extract the element.
2459 The index may be a variable.</p>
2464 The result is a scalar of the same type as the element type of
2465 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2466 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2467 results are undefined.
2473 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2478 <!-- _______________________________________________________________________ -->
2479 <div class="doc_subsubsection">
2480 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2483 <div class="doc_text">
2488 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2494 The '<tt>insertelement</tt>' instruction inserts a scalar
2495 element into a vector at a specified index.
2502 The first operand of an '<tt>insertelement</tt>' instruction is a
2503 value of <a href="#t_vector">vector</a> type. The second operand is a
2504 scalar value whose type must equal the element type of the first
2505 operand. The third operand is an index indicating the position at
2506 which to insert the value. The index may be a variable.</p>
2511 The result is a vector of the same type as <tt>val</tt>. Its
2512 element values are those of <tt>val</tt> except at position
2513 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2514 exceeds the length of <tt>val</tt>, the results are undefined.
2520 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2524 <!-- _______________________________________________________________________ -->
2525 <div class="doc_subsubsection">
2526 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2529 <div class="doc_text">
2534 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2540 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2541 from two input vectors, returning a vector of the same type.
2547 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2548 with types that match each other and types that match the result of the
2549 instruction. The third argument is a shuffle mask, which has the same number
2550 of elements as the other vector type, but whose element type is always 'i32'.
2554 The shuffle mask operand is required to be a constant vector with either
2555 constant integer or undef values.
2561 The elements of the two input vectors are numbered from left to right across
2562 both of the vectors. The shuffle mask operand specifies, for each element of
2563 the result vector, which element of the two input registers the result element
2564 gets. The element selector may be undef (meaning "don't care") and the second
2565 operand may be undef if performing a shuffle from only one vector.
2571 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2572 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2573 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2574 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2579 <!-- ======================================================================= -->
2580 <div class="doc_subsection">
2581 <a name="memoryops">Memory Access and Addressing Operations</a>
2584 <div class="doc_text">
2586 <p>A key design point of an SSA-based representation is how it
2587 represents memory. In LLVM, no memory locations are in SSA form, which
2588 makes things very simple. This section describes how to read, write,
2589 allocate, and free memory in LLVM.</p>
2593 <!-- _______________________________________________________________________ -->
2594 <div class="doc_subsubsection">
2595 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2598 <div class="doc_text">
2603 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2608 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2609 heap and returns a pointer to it.</p>
2613 <p>The '<tt>malloc</tt>' instruction allocates
2614 <tt>sizeof(<type>)*NumElements</tt>
2615 bytes of memory from the operating system and returns a pointer of the
2616 appropriate type to the program. If "NumElements" is specified, it is the
2617 number of elements allocated. If an alignment is specified, the value result
2618 of the allocation is guaranteed to be aligned to at least that boundary. If
2619 not specified, or if zero, the target can choose to align the allocation on any
2620 convenient boundary.</p>
2622 <p>'<tt>type</tt>' must be a sized type.</p>
2626 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2627 a pointer is returned.</p>
2632 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2634 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2635 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2636 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2637 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2638 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2642 <!-- _______________________________________________________________________ -->
2643 <div class="doc_subsubsection">
2644 <a name="i_free">'<tt>free</tt>' Instruction</a>
2647 <div class="doc_text">
2652 free <type> <value> <i>; yields {void}</i>
2657 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2658 memory heap to be reallocated in the future.</p>
2662 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2663 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2668 <p>Access to the memory pointed to by the pointer is no longer defined
2669 after this instruction executes.</p>
2674 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2675 free [4 x i8]* %array
2679 <!-- _______________________________________________________________________ -->
2680 <div class="doc_subsubsection">
2681 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2684 <div class="doc_text">
2689 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2694 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
2695 currently executing function, to be automatically released when this function
2696 returns to its caller.</p>
2700 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2701 bytes of memory on the runtime stack, returning a pointer of the
2702 appropriate type to the program. If "NumElements" is specified, it is the
2703 number of elements allocated. If an alignment is specified, the value result
2704 of the allocation is guaranteed to be aligned to at least that boundary. If
2705 not specified, or if zero, the target can choose to align the allocation on any
2706 convenient boundary.</p>
2708 <p>'<tt>type</tt>' may be any sized type.</p>
2712 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2713 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2714 instruction is commonly used to represent automatic variables that must
2715 have an address available. When the function returns (either with the <tt><a
2716 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2717 instructions), the memory is reclaimed.</p>
2722 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2723 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2724 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2725 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2729 <!-- _______________________________________________________________________ -->
2730 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2731 Instruction</a> </div>
2732 <div class="doc_text">
2734 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2736 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2738 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2739 address from which to load. The pointer must point to a <a
2740 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2741 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2742 the number or order of execution of this <tt>load</tt> with other
2743 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2746 <p>The location of memory pointed to is loaded.</p>
2748 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2750 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2751 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2754 <!-- _______________________________________________________________________ -->
2755 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2756 Instruction</a> </div>
2757 <div class="doc_text">
2759 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2760 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2763 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2765 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2766 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
2767 operand must be a pointer to the type of the '<tt><value></tt>'
2768 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2769 optimizer is not allowed to modify the number or order of execution of
2770 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2771 href="#i_store">store</a></tt> instructions.</p>
2773 <p>The contents of memory are updated to contain '<tt><value></tt>'
2774 at the location specified by the '<tt><pointer></tt>' operand.</p>
2776 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2778 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2779 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2783 <!-- _______________________________________________________________________ -->
2784 <div class="doc_subsubsection">
2785 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2788 <div class="doc_text">
2791 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2797 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2798 subelement of an aggregate data structure.</p>
2802 <p>This instruction takes a list of integer operands that indicate what
2803 elements of the aggregate object to index to. The actual types of the arguments
2804 provided depend on the type of the first pointer argument. The
2805 '<tt>getelementptr</tt>' instruction is used to index down through the type
2806 levels of a structure or to a specific index in an array. When indexing into a
2807 structure, only <tt>i32</tt> integer constants are allowed. When indexing
2808 into an array or pointer, only integers of 32 or 64 bits are allowed, and will
2809 be sign extended to 64-bit values.</p>
2811 <p>For example, let's consider a C code fragment and how it gets
2812 compiled to LLVM:</p>
2814 <div class="doc_code">
2827 int *foo(struct ST *s) {
2828 return &s[1].Z.B[5][13];
2833 <p>The LLVM code generated by the GCC frontend is:</p>
2835 <div class="doc_code">
2837 %RT = type { i8 , [10 x [20 x i32]], i8 }
2838 %ST = type { i32, double, %RT }
2840 define i32* %foo(%ST* %s) {
2842 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
2850 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2851 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2852 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
2853 <a href="#t_integer">integer</a> type but the value will always be sign extended
2854 to 64-bits. <a href="#t_struct">Structure</a> types require <tt>i32</tt>
2855 <b>constants</b>.</p>
2857 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2858 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
2859 }</tt>' type, a structure. The second index indexes into the third element of
2860 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
2861 i8 }</tt>' type, another structure. The third index indexes into the second
2862 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
2863 array. The two dimensions of the array are subscripted into, yielding an
2864 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2865 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
2867 <p>Note that it is perfectly legal to index partially through a
2868 structure, returning a pointer to an inner element. Because of this,
2869 the LLVM code for the given testcase is equivalent to:</p>
2872 define i32* %foo(%ST* %s) {
2873 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
2874 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
2875 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
2876 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
2877 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
2882 <p>Note that it is undefined to access an array out of bounds: array and
2883 pointer indexes must always be within the defined bounds of the array type.
2884 The one exception for this rules is zero length arrays. These arrays are
2885 defined to be accessible as variable length arrays, which requires access
2886 beyond the zero'th element.</p>
2888 <p>The getelementptr instruction is often confusing. For some more insight
2889 into how it works, see <a href="GetElementPtr.html">the getelementptr
2895 <i>; yields [12 x i8]*:aptr</i>
2896 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
2900 <!-- ======================================================================= -->
2901 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
2903 <div class="doc_text">
2904 <p>The instructions in this category are the conversion instructions (casting)
2905 which all take a single operand and a type. They perform various bit conversions
2909 <!-- _______________________________________________________________________ -->
2910 <div class="doc_subsubsection">
2911 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
2913 <div class="doc_text">
2917 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
2922 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
2927 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
2928 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
2929 and type of the result, which must be an <a href="#t_integer">integer</a>
2930 type. The bit size of <tt>value</tt> must be larger than the bit size of
2931 <tt>ty2</tt>. Equal sized types are not allowed.</p>
2935 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
2936 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
2937 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
2938 It will always truncate bits.</p>
2942 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
2943 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
2944 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
2948 <!-- _______________________________________________________________________ -->
2949 <div class="doc_subsubsection">
2950 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
2952 <div class="doc_text">
2956 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
2960 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
2965 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
2966 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
2967 also be of <a href="#t_integer">integer</a> type. The bit size of the
2968 <tt>value</tt> must be smaller than the bit size of the destination type,
2972 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
2973 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
2975 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
2979 %X = zext i32 257 to i64 <i>; yields i64:257</i>
2980 %Y = zext i1 true to i32 <i>; yields i32:1</i>
2984 <!-- _______________________________________________________________________ -->
2985 <div class="doc_subsubsection">
2986 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
2988 <div class="doc_text">
2992 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
2996 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3000 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3001 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3002 also be of <a href="#t_integer">integer</a> type. The bit size of the
3003 <tt>value</tt> must be smaller than the bit size of the destination type,
3008 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3009 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3010 the type <tt>ty2</tt>.</p>
3012 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3016 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3017 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3021 <!-- _______________________________________________________________________ -->
3022 <div class="doc_subsubsection">
3023 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3026 <div class="doc_text">
3031 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3035 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3040 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3041 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3042 cast it to. The size of <tt>value</tt> must be larger than the size of
3043 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3044 <i>no-op cast</i>.</p>
3047 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3048 <a href="#t_floating">floating point</a> type to a smaller
3049 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3050 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3054 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3055 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3059 <!-- _______________________________________________________________________ -->
3060 <div class="doc_subsubsection">
3061 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3063 <div class="doc_text">
3067 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3071 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3072 floating point value.</p>
3075 <p>The '<tt>fpext</tt>' instruction takes a
3076 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3077 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3078 type must be smaller than the destination type.</p>
3081 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3082 <a href="#t_floating">floating point</a> type to a larger
3083 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3084 used to make a <i>no-op cast</i> because it always changes bits. Use
3085 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3089 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3090 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3094 <!-- _______________________________________________________________________ -->
3095 <div class="doc_subsubsection">
3096 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3098 <div class="doc_text">
3102 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3106 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3107 unsigned integer equivalent of type <tt>ty2</tt>.
3111 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3112 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
3113 must be an <a href="#t_integer">integer</a> type.</p>
3116 <p> The '<tt>fptoui</tt>' instruction converts its
3117 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3118 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3119 the results are undefined.</p>
3121 <p>When converting to i1, the conversion is done as a comparison against
3122 zero. If the <tt>value</tt> was zero, the i1 result will be <tt>false</tt>.
3123 If the <tt>value</tt> was non-zero, the i1 result will be <tt>true</tt>.</p>
3127 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3128 %Y = fptoui float 1.0E+300 to i1 <i>; yields i1:true</i>
3129 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3133 <!-- _______________________________________________________________________ -->
3134 <div class="doc_subsubsection">
3135 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3137 <div class="doc_text">
3141 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3145 <p>The '<tt>fptosi</tt>' instruction converts
3146 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3151 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3152 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
3153 must also be an <a href="#t_integer">integer</a> type.</p>
3156 <p>The '<tt>fptosi</tt>' instruction converts its
3157 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3158 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3159 the results are undefined.</p>
3161 <p>When converting to i1, the conversion is done as a comparison against
3162 zero. If the <tt>value</tt> was zero, the i1 result will be <tt>false</tt>.
3163 If the <tt>value</tt> was non-zero, the i1 result will be <tt>true</tt>.</p>
3167 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3168 %Y = fptosi float 1.0E-247 to i1 <i>; yields i1:true</i>
3169 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3173 <!-- _______________________________________________________________________ -->
3174 <div class="doc_subsubsection">
3175 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3177 <div class="doc_text">
3181 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3185 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3186 integer and converts that value to the <tt>ty2</tt> type.</p>
3190 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be an
3191 <a href="#t_integer">integer</a> value, and a type to cast it to, which must
3192 be a <a href="#t_floating">floating point</a> type.</p>
3195 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3196 integer quantity and converts it to the corresponding floating point value. If
3197 the value cannot fit in the floating point value, the results are undefined.</p>
3202 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3203 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3207 <!-- _______________________________________________________________________ -->
3208 <div class="doc_subsubsection">
3209 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3211 <div class="doc_text">
3215 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3219 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3220 integer and converts that value to the <tt>ty2</tt> type.</p>
3223 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be an
3224 <a href="#t_integer">integer</a> value, and a type to cast it to, which must be
3225 a <a href="#t_floating">floating point</a> type.</p>
3228 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3229 integer quantity and converts it to the corresponding floating point value. If
3230 the value cannot fit in the floating point value, the results are undefined.</p>
3234 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3235 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3239 <!-- _______________________________________________________________________ -->
3240 <div class="doc_subsubsection">
3241 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3243 <div class="doc_text">
3247 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3251 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3252 the integer type <tt>ty2</tt>.</p>
3255 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3256 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3257 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3260 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3261 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3262 truncating or zero extending that value to the size of the integer type. If
3263 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3264 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3265 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3270 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3271 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3275 <!-- _______________________________________________________________________ -->
3276 <div class="doc_subsubsection">
3277 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3279 <div class="doc_text">
3283 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3287 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3288 a pointer type, <tt>ty2</tt>.</p>
3291 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3292 value to cast, and a type to cast it to, which must be a
3293 <a href="#t_pointer">pointer</a> type.
3296 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3297 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3298 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3299 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3300 the size of a pointer then a zero extension is done. If they are the same size,
3301 nothing is done (<i>no-op cast</i>).</p>
3305 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3306 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3307 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3311 <!-- _______________________________________________________________________ -->
3312 <div class="doc_subsubsection">
3313 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3315 <div class="doc_text">
3319 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3323 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3324 <tt>ty2</tt> without changing any bits.</p>
3327 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3328 a first class value, and a type to cast it to, which must also be a <a
3329 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3330 and the destination type, <tt>ty2</tt>, must be identical. If the source
3331 type is a pointer, the destination type must also be a pointer.</p>
3334 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3335 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3336 this conversion. The conversion is done as if the <tt>value</tt> had been
3337 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3338 converted to other pointer types with this instruction. To convert pointers to
3339 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3340 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3344 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3345 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3346 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3350 <!-- ======================================================================= -->
3351 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3352 <div class="doc_text">
3353 <p>The instructions in this category are the "miscellaneous"
3354 instructions, which defy better classification.</p>
3357 <!-- _______________________________________________________________________ -->
3358 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3360 <div class="doc_text">
3362 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3365 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3366 of its two integer operands.</p>
3368 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3369 the condition code indicating the kind of comparison to perform. It is not
3370 a value, just a keyword. The possible condition code are:
3372 <li><tt>eq</tt>: equal</li>
3373 <li><tt>ne</tt>: not equal </li>
3374 <li><tt>ugt</tt>: unsigned greater than</li>
3375 <li><tt>uge</tt>: unsigned greater or equal</li>
3376 <li><tt>ult</tt>: unsigned less than</li>
3377 <li><tt>ule</tt>: unsigned less or equal</li>
3378 <li><tt>sgt</tt>: signed greater than</li>
3379 <li><tt>sge</tt>: signed greater or equal</li>
3380 <li><tt>slt</tt>: signed less than</li>
3381 <li><tt>sle</tt>: signed less or equal</li>
3383 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3384 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3386 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3387 the condition code given as <tt>cond</tt>. The comparison performed always
3388 yields a <a href="#t_primitive">i1</a> result, as follows:
3390 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3391 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3393 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3394 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3395 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3396 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3397 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3398 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3399 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3400 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3401 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3402 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3403 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3404 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3405 <li><tt>sge</tt>: interprets the operands as signed values and yields
3406 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3407 <li><tt>slt</tt>: interprets the operands as signed values and yields
3408 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3409 <li><tt>sle</tt>: interprets the operands as signed values and yields
3410 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3412 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3413 values are compared as if they were integers.</p>
3416 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3417 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3418 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3419 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3420 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3421 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3425 <!-- _______________________________________________________________________ -->
3426 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3428 <div class="doc_text">
3430 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3433 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3434 of its floating point operands.</p>
3436 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3437 the condition code indicating the kind of comparison to perform. It is not
3438 a value, just a keyword. The possible condition code are:
3440 <li><tt>false</tt>: no comparison, always returns false</li>
3441 <li><tt>oeq</tt>: ordered and equal</li>
3442 <li><tt>ogt</tt>: ordered and greater than </li>
3443 <li><tt>oge</tt>: ordered and greater than or equal</li>
3444 <li><tt>olt</tt>: ordered and less than </li>
3445 <li><tt>ole</tt>: ordered and less than or equal</li>
3446 <li><tt>one</tt>: ordered and not equal</li>
3447 <li><tt>ord</tt>: ordered (no nans)</li>
3448 <li><tt>ueq</tt>: unordered or equal</li>
3449 <li><tt>ugt</tt>: unordered or greater than </li>
3450 <li><tt>uge</tt>: unordered or greater than or equal</li>
3451 <li><tt>ult</tt>: unordered or less than </li>
3452 <li><tt>ule</tt>: unordered or less than or equal</li>
3453 <li><tt>une</tt>: unordered or not equal</li>
3454 <li><tt>uno</tt>: unordered (either nans)</li>
3455 <li><tt>true</tt>: no comparison, always returns true</li>
3457 <p><i>Ordered</i> means that neither operand is a QNAN while
3458 <i>unordered</i> means that either operand may be a QNAN.</p>
3459 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3460 <a href="#t_floating">floating point</a> typed. They must have identical
3463 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3464 the condition code given as <tt>cond</tt>. The comparison performed always
3465 yields a <a href="#t_primitive">i1</a> result, as follows:
3467 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3468 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3469 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3470 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3471 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3472 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3473 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3474 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3475 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3476 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3477 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3478 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3479 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3480 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3481 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3482 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3483 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3484 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3485 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3486 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3487 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3488 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3489 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3490 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3491 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3492 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3493 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3494 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3498 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3499 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3500 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3501 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3505 <!-- _______________________________________________________________________ -->
3506 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3507 Instruction</a> </div>
3508 <div class="doc_text">
3510 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3512 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3513 the SSA graph representing the function.</p>
3515 <p>The type of the incoming values is specified with the first type
3516 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3517 as arguments, with one pair for each predecessor basic block of the
3518 current block. Only values of <a href="#t_firstclass">first class</a>
3519 type may be used as the value arguments to the PHI node. Only labels
3520 may be used as the label arguments.</p>
3521 <p>There must be no non-phi instructions between the start of a basic
3522 block and the PHI instructions: i.e. PHI instructions must be first in
3525 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3526 specified by the pair corresponding to the predecessor basic block that executed
3527 just prior to the current block.</p>
3529 <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>
3532 <!-- _______________________________________________________________________ -->
3533 <div class="doc_subsubsection">
3534 <a name="i_select">'<tt>select</tt>' Instruction</a>
3537 <div class="doc_text">
3542 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3548 The '<tt>select</tt>' instruction is used to choose one value based on a
3549 condition, without branching.
3556 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.
3562 If the boolean condition evaluates to true, the instruction returns the first
3563 value argument; otherwise, it returns the second value argument.
3569 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3574 <!-- _______________________________________________________________________ -->
3575 <div class="doc_subsubsection">
3576 <a name="i_call">'<tt>call</tt>' Instruction</a>
3579 <div class="doc_text">
3583 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
3588 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3592 <p>This instruction requires several arguments:</p>
3596 <p>The optional "tail" marker indicates whether the callee function accesses
3597 any allocas or varargs in the caller. If the "tail" marker is present, the
3598 function call is eligible for tail call optimization. Note that calls may
3599 be marked "tail" even if they do not occur before a <a
3600 href="#i_ret"><tt>ret</tt></a> instruction.
3603 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3604 convention</a> the call should use. If none is specified, the call defaults
3605 to using C calling conventions.
3608 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
3609 the type of the return value. Functions that return no value are marked
3610 <tt><a href="#t_void">void</a></tt>.</p>
3613 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
3614 value being invoked. The argument types must match the types implied by
3615 this signature. This type can be omitted if the function is not varargs
3616 and if the function type does not return a pointer to a function.</p>
3619 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3620 be invoked. In most cases, this is a direct function invocation, but
3621 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3622 to function value.</p>
3625 <p>'<tt>function args</tt>': argument list whose types match the
3626 function signature argument types. All arguments must be of
3627 <a href="#t_firstclass">first class</a> type. If the function signature
3628 indicates the function accepts a variable number of arguments, the extra
3629 arguments can be specified.</p>
3635 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3636 transfer to a specified function, with its incoming arguments bound to
3637 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3638 instruction in the called function, control flow continues with the
3639 instruction after the function call, and the return value of the
3640 function is bound to the result argument. This is a simpler case of
3641 the <a href="#i_invoke">invoke</a> instruction.</p>
3646 %retval = call i32 @test(i32 %argc)
3647 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42);
3648 %X = tail call i32 @foo()
3649 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo()
3650 %Z = call void %foo(i8 97 signext)
3655 <!-- _______________________________________________________________________ -->
3656 <div class="doc_subsubsection">
3657 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3660 <div class="doc_text">
3665 <resultval> = va_arg <va_list*> <arglist>, <argty>
3670 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3671 the "variable argument" area of a function call. It is used to implement the
3672 <tt>va_arg</tt> macro in C.</p>
3676 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3677 the argument. It returns a value of the specified argument type and
3678 increments the <tt>va_list</tt> to point to the next argument. The
3679 actual type of <tt>va_list</tt> is target specific.</p>
3683 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3684 type from the specified <tt>va_list</tt> and causes the
3685 <tt>va_list</tt> to point to the next argument. For more information,
3686 see the variable argument handling <a href="#int_varargs">Intrinsic
3689 <p>It is legal for this instruction to be called in a function which does not
3690 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3693 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3694 href="#intrinsics">intrinsic function</a> because it takes a type as an
3699 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3703 <!-- *********************************************************************** -->
3704 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3705 <!-- *********************************************************************** -->
3707 <div class="doc_text">
3709 <p>LLVM supports the notion of an "intrinsic function". These functions have
3710 well known names and semantics and are required to follow certain restrictions.
3711 Overall, these intrinsics represent an extension mechanism for the LLVM
3712 language that does not require changing all of the transformations in LLVM when
3713 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3715 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3716 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3717 begin with this prefix. Intrinsic functions must always be external functions:
3718 you cannot define the body of intrinsic functions. Intrinsic functions may
3719 only be used in call or invoke instructions: it is illegal to take the address
3720 of an intrinsic function. Additionally, because intrinsic functions are part
3721 of the LLVM language, it is required if any are added that they be documented
3724 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3725 a family of functions that perform the same operation but on different data
3726 types. Because LLVM can represent over 8 million different integer types,
3727 overloading is used commonly to allow an intrinsic function to operate on any
3728 integer type. One or more of the argument types or the result type can be
3729 overloaded to accept any integer type. Argument types may also be defined as
3730 exactly matching a previous argument's type or the result type. This allows an
3731 intrinsic function which accepts multiple arguments, but needs all of them to
3732 be of the same type, to only be overloaded with respect to a single argument or
3735 <p>Overloaded intrinsics will have the names of its overloaded argument types
3736 encoded into its function name, each preceded by a period. Only those types
3737 which are overloaded result in a name suffix. Arguments whose type is matched
3738 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
3739 take an integer of any width and returns an integer of exactly the same integer
3740 width. This leads to a family of functions such as
3741 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
3742 Only one type, the return type, is overloaded, and only one type suffix is
3743 required. Because the argument's type is matched against the return type, it
3744 does not require its own name suffix.</p>
3746 <p>To learn how to add an intrinsic function, please see the
3747 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3752 <!-- ======================================================================= -->
3753 <div class="doc_subsection">
3754 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3757 <div class="doc_text">
3759 <p>Variable argument support is defined in LLVM with the <a
3760 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3761 intrinsic functions. These functions are related to the similarly
3762 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3764 <p>All of these functions operate on arguments that use a
3765 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3766 language reference manual does not define what this type is, so all
3767 transformations should be prepared to handle these functions regardless of
3770 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3771 instruction and the variable argument handling intrinsic functions are
3774 <div class="doc_code">
3776 define i32 @test(i32 %X, ...) {
3777 ; Initialize variable argument processing
3779 %ap2 = bitcast i8** %ap to i8*
3780 call void @llvm.va_start(i8* %ap2)
3782 ; Read a single integer argument
3783 %tmp = va_arg i8** %ap, i32
3785 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3787 %aq2 = bitcast i8** %aq to i8*
3788 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3789 call void @llvm.va_end(i8* %aq2)
3791 ; Stop processing of arguments.
3792 call void @llvm.va_end(i8* %ap2)
3796 declare void @llvm.va_start(i8*)
3797 declare void @llvm.va_copy(i8*, i8*)
3798 declare void @llvm.va_end(i8*)
3804 <!-- _______________________________________________________________________ -->
3805 <div class="doc_subsubsection">
3806 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3810 <div class="doc_text">
3812 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
3814 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3815 <tt>*<arglist></tt> for subsequent use by <tt><a
3816 href="#i_va_arg">va_arg</a></tt>.</p>
3820 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3824 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3825 macro available in C. In a target-dependent way, it initializes the
3826 <tt>va_list</tt> element to which the argument points, so that the next call to
3827 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3828 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3829 last argument of the function as the compiler can figure that out.</p>
3833 <!-- _______________________________________________________________________ -->
3834 <div class="doc_subsubsection">
3835 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
3838 <div class="doc_text">
3840 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
3843 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
3844 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
3845 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
3849 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
3853 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
3854 macro available in C. In a target-dependent way, it destroys the
3855 <tt>va_list</tt> element to which the argument points. Calls to <a
3856 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
3857 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
3858 <tt>llvm.va_end</tt>.</p>
3862 <!-- _______________________________________________________________________ -->
3863 <div class="doc_subsubsection">
3864 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
3867 <div class="doc_text">
3872 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
3877 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
3878 from the source argument list to the destination argument list.</p>
3882 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
3883 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
3888 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
3889 macro available in C. In a target-dependent way, it copies the source
3890 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
3891 intrinsic is necessary because the <tt><a href="#int_va_start">
3892 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
3893 example, memory allocation.</p>
3897 <!-- ======================================================================= -->
3898 <div class="doc_subsection">
3899 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
3902 <div class="doc_text">
3905 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
3906 Collection</a> requires the implementation and generation of these intrinsics.
3907 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
3908 stack</a>, as well as garbage collector implementations that require <a
3909 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
3910 Front-ends for type-safe garbage collected languages should generate these
3911 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3912 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
3916 <!-- _______________________________________________________________________ -->
3917 <div class="doc_subsubsection">
3918 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
3921 <div class="doc_text">
3926 declare void @llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
3931 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
3932 the code generator, and allows some metadata to be associated with it.</p>
3936 <p>The first argument specifies the address of a stack object that contains the
3937 root pointer. The second pointer (which must be either a constant or a global
3938 value address) contains the meta-data to be associated with the root.</p>
3942 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
3943 location. At compile-time, the code generator generates information to allow
3944 the runtime to find the pointer at GC safe points.
3950 <!-- _______________________________________________________________________ -->
3951 <div class="doc_subsubsection">
3952 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
3955 <div class="doc_text">
3960 declare i8 * @llvm.gcread(i8 * %ObjPtr, i8 ** %Ptr)
3965 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
3966 locations, allowing garbage collector implementations that require read
3971 <p>The second argument is the address to read from, which should be an address
3972 allocated from the garbage collector. The first object is a pointer to the
3973 start of the referenced object, if needed by the language runtime (otherwise
3978 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
3979 instruction, but may be replaced with substantially more complex code by the
3980 garbage collector runtime, as needed.</p>
3985 <!-- _______________________________________________________________________ -->
3986 <div class="doc_subsubsection">
3987 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
3990 <div class="doc_text">
3995 declare void @llvm.gcwrite(i8 * %P1, i8 * %Obj, i8 ** %P2)
4000 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4001 locations, allowing garbage collector implementations that require write
4002 barriers (such as generational or reference counting collectors).</p>
4006 <p>The first argument is the reference to store, the second is the start of the
4007 object to store it to, and the third is the address of the field of Obj to
4008 store to. If the runtime does not require a pointer to the object, Obj may be
4013 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4014 instruction, but may be replaced with substantially more complex code by the
4015 garbage collector runtime, as needed.</p>
4021 <!-- ======================================================================= -->
4022 <div class="doc_subsection">
4023 <a name="int_codegen">Code Generator Intrinsics</a>
4026 <div class="doc_text">
4028 These intrinsics are provided by LLVM to expose special features that may only
4029 be implemented with code generator support.
4034 <!-- _______________________________________________________________________ -->
4035 <div class="doc_subsubsection">
4036 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4039 <div class="doc_text">
4043 declare i8 *@llvm.returnaddress(i32 <level>)
4049 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4050 target-specific value indicating the return address of the current function
4051 or one of its callers.
4057 The argument to this intrinsic indicates which function to return the address
4058 for. Zero indicates the calling function, one indicates its caller, etc. The
4059 argument is <b>required</b> to be a constant integer value.
4065 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4066 the return address of the specified call frame, or zero if it cannot be
4067 identified. The value returned by this intrinsic is likely to be incorrect or 0
4068 for arguments other than zero, so it should only be used for debugging purposes.
4072 Note that calling this intrinsic does not prevent function inlining or other
4073 aggressive transformations, so the value returned may not be that of the obvious
4074 source-language caller.
4079 <!-- _______________________________________________________________________ -->
4080 <div class="doc_subsubsection">
4081 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4084 <div class="doc_text">
4088 declare i8 *@llvm.frameaddress(i32 <level>)
4094 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4095 target-specific frame pointer value for the specified stack frame.
4101 The argument to this intrinsic indicates which function to return the frame
4102 pointer for. Zero indicates the calling function, one indicates its caller,
4103 etc. The argument is <b>required</b> to be a constant integer value.
4109 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4110 the frame address of the specified call frame, or zero if it cannot be
4111 identified. The value returned by this intrinsic is likely to be incorrect or 0
4112 for arguments other than zero, so it should only be used for debugging purposes.
4116 Note that calling this intrinsic does not prevent function inlining or other
4117 aggressive transformations, so the value returned may not be that of the obvious
4118 source-language caller.
4122 <!-- _______________________________________________________________________ -->
4123 <div class="doc_subsubsection">
4124 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4127 <div class="doc_text">
4131 declare i8 *@llvm.stacksave()
4137 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4138 the function stack, for use with <a href="#int_stackrestore">
4139 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4140 features like scoped automatic variable sized arrays in C99.
4146 This intrinsic returns a opaque pointer value that can be passed to <a
4147 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4148 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4149 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4150 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4151 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4152 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4157 <!-- _______________________________________________________________________ -->
4158 <div class="doc_subsubsection">
4159 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4162 <div class="doc_text">
4166 declare void @llvm.stackrestore(i8 * %ptr)
4172 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4173 the function stack to the state it was in when the corresponding <a
4174 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4175 useful for implementing language features like scoped automatic variable sized
4182 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4188 <!-- _______________________________________________________________________ -->
4189 <div class="doc_subsubsection">
4190 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4193 <div class="doc_text">
4197 declare void @llvm.prefetch(i8 * <address>,
4198 i32 <rw>, i32 <locality>)
4205 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4206 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4208 effect on the behavior of the program but can change its performance
4215 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4216 determining if the fetch should be for a read (0) or write (1), and
4217 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4218 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4219 <tt>locality</tt> arguments must be constant integers.
4225 This intrinsic does not modify the behavior of the program. In particular,
4226 prefetches cannot trap and do not produce a value. On targets that support this
4227 intrinsic, the prefetch can provide hints to the processor cache for better
4233 <!-- _______________________________________________________________________ -->
4234 <div class="doc_subsubsection">
4235 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4238 <div class="doc_text">
4242 declare void @llvm.pcmarker( i32 <id> )
4249 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4251 code to simulators and other tools. The method is target specific, but it is
4252 expected that the marker will use exported symbols to transmit the PC of the marker.
4253 The marker makes no guarantees that it will remain with any specific instruction
4254 after optimizations. It is possible that the presence of a marker will inhibit
4255 optimizations. The intended use is to be inserted after optimizations to allow
4256 correlations of simulation runs.
4262 <tt>id</tt> is a numerical id identifying the marker.
4268 This intrinsic does not modify the behavior of the program. Backends that do not
4269 support this intrinisic may ignore it.
4274 <!-- _______________________________________________________________________ -->
4275 <div class="doc_subsubsection">
4276 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4279 <div class="doc_text">
4283 declare i64 @llvm.readcyclecounter( )
4290 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4291 counter register (or similar low latency, high accuracy clocks) on those targets
4292 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4293 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4294 should only be used for small timings.
4300 When directly supported, reading the cycle counter should not modify any memory.
4301 Implementations are allowed to either return a application specific value or a
4302 system wide value. On backends without support, this is lowered to a constant 0.
4307 <!-- ======================================================================= -->
4308 <div class="doc_subsection">
4309 <a name="int_libc">Standard C Library Intrinsics</a>
4312 <div class="doc_text">
4314 LLVM provides intrinsics for a few important standard C library functions.
4315 These intrinsics allow source-language front-ends to pass information about the
4316 alignment of the pointer arguments to the code generator, providing opportunity
4317 for more efficient code generation.
4322 <!-- _______________________________________________________________________ -->
4323 <div class="doc_subsubsection">
4324 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4327 <div class="doc_text">
4331 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4332 i32 <len>, i32 <align>)
4333 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4334 i64 <len>, i32 <align>)
4340 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4341 location to the destination location.
4345 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4346 intrinsics do not return a value, and takes an extra alignment argument.
4352 The first argument is a pointer to the destination, the second is a pointer to
4353 the source. The third argument is an integer argument
4354 specifying the number of bytes to copy, and the fourth argument is the alignment
4355 of the source and destination locations.
4359 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4360 the caller guarantees that both the source and destination pointers are aligned
4367 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4368 location to the destination location, which are not allowed to overlap. It
4369 copies "len" bytes of memory over. If the argument is known to be aligned to
4370 some boundary, this can be specified as the fourth argument, otherwise it should
4376 <!-- _______________________________________________________________________ -->
4377 <div class="doc_subsubsection">
4378 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4381 <div class="doc_text">
4385 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4386 i32 <len>, i32 <align>)
4387 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4388 i64 <len>, i32 <align>)
4394 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4395 location to the destination location. It is similar to the
4396 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
4400 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4401 intrinsics do not return a value, and takes an extra alignment argument.
4407 The first argument is a pointer to the destination, the second is a pointer to
4408 the source. The third argument is an integer argument
4409 specifying the number of bytes to copy, and the fourth argument is the alignment
4410 of the source and destination locations.
4414 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4415 the caller guarantees that the source and destination pointers are aligned to
4422 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4423 location to the destination location, which may overlap. It
4424 copies "len" bytes of memory over. If the argument is known to be aligned to
4425 some boundary, this can be specified as the fourth argument, otherwise it should
4431 <!-- _______________________________________________________________________ -->
4432 <div class="doc_subsubsection">
4433 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4436 <div class="doc_text">
4440 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4441 i32 <len>, i32 <align>)
4442 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4443 i64 <len>, i32 <align>)
4449 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4454 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4455 does not return a value, and takes an extra alignment argument.
4461 The first argument is a pointer to the destination to fill, the second is the
4462 byte value to fill it with, the third argument is an integer
4463 argument specifying the number of bytes to fill, and the fourth argument is the
4464 known alignment of destination location.
4468 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4469 the caller guarantees that the destination pointer is aligned to that boundary.
4475 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4477 destination location. If the argument is known to be aligned to some boundary,
4478 this can be specified as the fourth argument, otherwise it should be set to 0 or
4484 <!-- _______________________________________________________________________ -->
4485 <div class="doc_subsubsection">
4486 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4489 <div class="doc_text">
4493 declare float @llvm.sqrt.f32(float %Val)
4494 declare double @llvm.sqrt.f64(double %Val)
4500 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4501 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
4502 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4503 negative numbers (which allows for better optimization).
4509 The argument and return value are floating point numbers of the same type.
4515 This function returns the sqrt of the specified operand if it is a nonnegative
4516 floating point number.
4520 <!-- _______________________________________________________________________ -->
4521 <div class="doc_subsubsection">
4522 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4525 <div class="doc_text">
4529 declare float @llvm.powi.f32(float %Val, i32 %power)
4530 declare double @llvm.powi.f64(double %Val, i32 %power)
4536 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4537 specified (positive or negative) power. The order of evaluation of
4538 multiplications is not defined.
4544 The second argument is an integer power, and the first is a value to raise to
4551 This function returns the first value raised to the second power with an
4552 unspecified sequence of rounding operations.</p>
4556 <!-- ======================================================================= -->
4557 <div class="doc_subsection">
4558 <a name="int_manip">Bit Manipulation Intrinsics</a>
4561 <div class="doc_text">
4563 LLVM provides intrinsics for a few important bit manipulation operations.
4564 These allow efficient code generation for some algorithms.
4569 <!-- _______________________________________________________________________ -->
4570 <div class="doc_subsubsection">
4571 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4574 <div class="doc_text">
4577 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4578 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
4580 declare i16 @llvm.bswap.i16(i16 <id>)
4581 declare i32 @llvm.bswap.i32(i32 <id>)
4582 declare i64 @llvm.bswap.i64(i64 <id>)
4588 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4589 values with an even number of bytes (positive multiple of 16 bits). These are
4590 useful for performing operations on data that is not in the target's native
4597 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
4598 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4599 intrinsic returns an i32 value that has the four bytes of the input i32
4600 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
4601 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
4602 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
4603 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4608 <!-- _______________________________________________________________________ -->
4609 <div class="doc_subsubsection">
4610 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4613 <div class="doc_text">
4616 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4617 width. Not all targets support all bit widths however.
4619 declare i8 @llvm.ctpop.i8 (i8 <src>)
4620 declare i16 @llvm.ctpop.i16(i16 <src>)
4621 declare i32 @llvm.ctpop.i32(i32 <src>)
4622 declare i64 @llvm.ctpop.i64(i64 <src>)
4623 declare i256 @llvm.ctpop.i256(i256 <src>)
4629 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4636 The only argument is the value to be counted. The argument may be of any
4637 integer type. The return type must match the argument type.
4643 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4647 <!-- _______________________________________________________________________ -->
4648 <div class="doc_subsubsection">
4649 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4652 <div class="doc_text">
4655 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
4656 integer bit width. Not all targets support all bit widths however.
4658 declare i8 @llvm.ctlz.i8 (i8 <src>)
4659 declare i16 @llvm.ctlz.i16(i16 <src>)
4660 declare i32 @llvm.ctlz.i32(i32 <src>)
4661 declare i64 @llvm.ctlz.i64(i64 <src>)
4662 declare i256 @llvm.ctlz.i256(i256 <src>)
4668 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4669 leading zeros in a variable.
4675 The only argument is the value to be counted. The argument may be of any
4676 integer type. The return type must match the argument type.
4682 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4683 in a variable. If the src == 0 then the result is the size in bits of the type
4684 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
4690 <!-- _______________________________________________________________________ -->
4691 <div class="doc_subsubsection">
4692 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
4695 <div class="doc_text">
4698 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
4699 integer bit width. Not all targets support all bit widths however.
4701 declare i8 @llvm.cttz.i8 (i8 <src>)
4702 declare i16 @llvm.cttz.i16(i16 <src>)
4703 declare i32 @llvm.cttz.i32(i32 <src>)
4704 declare i64 @llvm.cttz.i64(i64 <src>)
4705 declare i256 @llvm.cttz.i256(i256 <src>)
4711 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
4718 The only argument is the value to be counted. The argument may be of any
4719 integer type. The return type must match the argument type.
4725 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
4726 in a variable. If the src == 0 then the result is the size in bits of the type
4727 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
4731 <!-- _______________________________________________________________________ -->
4732 <div class="doc_subsubsection">
4733 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
4736 <div class="doc_text">
4739 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
4740 on any integer bit width.
4742 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
4743 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
4747 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
4748 range of bits from an integer value and returns them in the same bit width as
4749 the original value.</p>
4752 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4753 any bit width but they must have the same bit width. The second and third
4754 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
4757 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
4758 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
4759 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
4760 operates in forward mode.</p>
4761 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
4762 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
4763 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
4765 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
4766 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
4767 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
4768 to determine the number of bits to retain.</li>
4769 <li>A mask of the retained bits is created by shifting a -1 value.</li>
4770 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
4772 <p>In reverse mode, a similar computation is made except that the bits are
4773 returned in the reverse order. So, for example, if <tt>X</tt> has the value
4774 <tt>i16 0x0ACF (101011001111)</tt> and we apply
4775 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
4776 <tt>i16 0x0026 (000000100110)</tt>.</p>
4779 <div class="doc_subsubsection">
4780 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
4783 <div class="doc_text">
4786 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
4787 on any integer bit width.
4789 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
4790 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
4794 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
4795 of bits in an integer value with another integer value. It returns the integer
4796 with the replaced bits.</p>
4799 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4800 any bit width but they must have the same bit width. <tt>%val</tt> is the value
4801 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
4802 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
4803 type since they specify only a bit index.</p>
4806 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
4807 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
4808 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
4809 operates in forward mode.</p>
4810 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
4811 truncating it down to the size of the replacement area or zero extending it
4812 up to that size.</p>
4813 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
4814 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
4815 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
4816 to the <tt>%hi</tt>th bit.
4817 <p>In reverse mode, a similar computation is made except that the bits are
4818 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
4819 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
4822 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
4823 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
4824 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
4825 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
4826 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
4830 <!-- ======================================================================= -->
4831 <div class="doc_subsection">
4832 <a name="int_debugger">Debugger Intrinsics</a>
4835 <div class="doc_text">
4837 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
4838 are described in the <a
4839 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
4840 Debugging</a> document.
4845 <!-- ======================================================================= -->
4846 <div class="doc_subsection">
4847 <a name="int_eh">Exception Handling Intrinsics</a>
4850 <div class="doc_text">
4851 <p> The LLVM exception handling intrinsics (which all start with
4852 <tt>llvm.eh.</tt> prefix), are described in the <a
4853 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
4854 Handling</a> document. </p>
4857 <!-- ======================================================================= -->
4858 <div class="doc_subsection">
4859 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
4862 <div class="doc_text">
4864 These intrinsic functions expand the "universal IR" of LLVM to represent
4865 hardware constructs for atomic operations and memory synchronization. This
4866 provides an interface to the hardware, not an interface to the programmer. It
4867 is aimed at a low enough level to allow any programming models or APIs which
4868 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
4869 hardware behavior. Just as hardware provides a "universal IR" for source
4870 languages, it also provides a starting point for developing a "universal"
4871 atomic operation and synchronization IR.
4874 These do <em>not</em> form an API such as high-level threading libraries,
4875 software transaction memory systems, atomic primitives, and intrinsic
4876 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
4877 application libraries. The hardware interface provided by LLVM should allow
4878 a clean implementation of all of these APIs and parallel programming models.
4879 No one model or paradigm should be selected above others unless the hardware
4880 itself ubiquitously does so.
4884 <!-- _______________________________________________________________________ -->
4885 <div class="doc_subsubsection">
4886 <a name="int_lcs">'<tt>llvm.atomic.lcs.*</tt>' Intrinsic</a>
4888 <div class="doc_text">
4891 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lcs</tt> on any
4892 integer bit width. Not all targets support all bit widths however.</p>
4894 declare i8 @llvm.atomic.lcs.i8.i8p.i8.i8( i8* <ptr>, i8 <cmp>, i8 <val> )
4895 declare i16 @llvm.atomic.lcs.i16.i16p.i16.i16( i16* <ptr>, i16 <cmp>, i16 <val> )
4896 declare i32 @llvm.atomic.lcs.i32.i32p.i32.i32( i32* <ptr>, i32 <cmp>, i32 <val> )
4897 declare i64 @llvm.atomic.lcs.i64.i64p.i64.i64( i64* <ptr>, i64 <cmp>, i64 <val> )
4901 This loads a value in memory and compares it to a given value. If they are
4902 equal, it stores a new value into the memory.
4906 The <tt>llvm.atomic.lcs</tt> intrinsic takes three arguments. The result as
4907 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
4908 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
4909 this integer type. While any bit width integer may be used, targets may only
4910 lower representations they support in hardware.
4914 This entire intrinsic must be executed atomically. It first loads the value
4915 in memory pointed to by <tt>ptr</tt> and compares it with the value
4916 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
4917 loaded value is yielded in all cases. This provides the equivalent of an
4918 atomic compare-and-swap operation within the SSA framework.
4925 %val1 = add i32 4, 4
4926 %result1 = call i32 @llvm.atomic.lcs( i32* %ptr, i32 4, %val1 )
4927 <i>; yields {i32}:result1 = 4</i>
4928 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
4929 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
4931 %val2 = add i32 1, 1
4932 %result2 = call i32 @llvm.atomic.lcs( i32* %ptr, i32 5, %val2 )
4933 <i>; yields {i32}:result2 = 8</i>
4934 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
4935 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
4939 <!-- _______________________________________________________________________ -->
4940 <div class="doc_subsubsection">
4941 <a name="int_ls">'<tt>llvm.atomic.ls.*</tt>' Intrinsic</a>
4943 <div class="doc_text">
4946 This is an overloaded intrinsic. You can use <tt>llvm.atomic.ls</tt> on any
4947 integer bit width. Not all targets support all bit widths however.</p>
4949 declare i8 @llvm.atomic.ls.i8.i8p.i8( i8* <ptr>, i8 <val> )
4950 declare i16 @llvm.atomic.ls.i16.i16p.i16( i16* <ptr>, i16 <val> )
4951 declare i32 @llvm.atomic.ls.i32.i32p.i32( i32* <ptr>, i32 <val> )
4952 declare i64 @llvm.atomic.ls.i64.i64p.i64( i64* <ptr>, i64 <val> )
4956 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
4957 the value from memory. It then stores the value in <tt>val</tt> in the memory
4962 The <tt>llvm.atomic.ls</tt> intrinsic takes two arguments. Both the
4963 <tt>val</tt> argument and the result must be integers of the same bit width.
4964 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
4965 integer type. The targets may only lower integer representations they
4970 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
4971 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
4972 equivalent of an atomic swap operation within the SSA framework.
4979 %val1 = add i32 4, 4
4980 %result1 = call i32 @llvm.atomic.ls( i32* %ptr, i32 %val1 )
4981 <i>; yields {i32}:result1 = 4</i>
4982 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
4983 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
4985 %val2 = add i32 1, 1
4986 %result2 = call i32 @llvm.atomic.ls( i32* %ptr, i32 %val2 )
4987 <i>; yields {i32}:result2 = 8</i>
4988 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
4989 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
4993 <!-- _______________________________________________________________________ -->
4994 <div class="doc_subsubsection">
4995 <a name="int_las">'<tt>llvm.atomic.las.*</tt>' Intrinsic</a>
4997 <div class="doc_text">
5000 This is an overloaded intrinsic. You can use <tt>llvm.atomic.las</tt> on any
5001 integer bit width. Not all targets support all bit widths however.</p>
5003 declare i8 @llvm.atomic.las.i8.i8p.i8( i8* <ptr>, i8 <delta> )
5004 declare i16 @llvm.atomic.las.i16.i16p.i16( i16* <ptr>, i16 <delta> )
5005 declare i32 @llvm.atomic.las.i32.i32p.i32( i32* <ptr>, i32 <delta> )
5006 declare i64 @llvm.atomic.las.i64.i64p.i64( i64* <ptr>, i64 <delta> )
5010 This intrinsic adds <tt>delta</tt> to the value stored in memory at
5011 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5015 The intrinsic takes two arguments, the first a pointer to an integer value
5016 and the second an integer value. The result is also an integer value. These
5017 integer types can have any bit width, but they must all have the same bit
5018 width. The targets may only lower integer representations they support.
5022 This intrinsic does a series of operations atomically. It first loads the
5023 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
5024 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5030 %result1 = call i32 @llvm.atomic.las( i32* %ptr, i32 4 )
5031 <i>; yields {i32}:result1 = 4</i>
5032 %result2 = call i32 @llvm.atomic.las( i32* %ptr, i32 2 )
5033 <i>; yields {i32}:result2 = 8</i>
5034 %result3 = call i32 @llvm.atomic.las( i32* %ptr, i32 5 )
5035 <i>; yields {i32}:result3 = 10</i>
5036 %memval = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
5040 <!-- _______________________________________________________________________ -->
5041 <div class="doc_subsubsection">
5042 <a name="int_lss">'<tt>llvm.atomic.lss.*</tt>' Intrinsic</a>
5044 <div class="doc_text">
5047 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lss</tt> on any
5048 integer bit width. Not all targets support all bit widths however.</p>
5050 declare i8 @llvm.atomic.lss.i8.i8.i8( i8* <ptr>, i8 <delta> )
5051 declare i16 @llvm.atomic.lss.i16.i16.i16( i16* <ptr>, i16 <delta> )
5052 declare i32 @llvm.atomic.lss.i32.i32.i32( i32* <ptr>, i32 <delta> )
5053 declare i64 @llvm.atomic.lss.i64.i64.i64( i64* <ptr>, i64 <delta> )
5057 This intrinsic subtracts <tt>delta</tt> from the value stored in memory at
5058 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5062 The intrinsic takes two arguments, the first a pointer to an integer value
5063 and the second an integer value. The result is also an integer value. These
5064 integer types can have any bit width, but they must all have the same bit
5065 width. The targets may only lower integer representations they support.
5069 This intrinsic does a series of operations atomically. It first loads the
5070 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>,
5071 stores the result to <tt>ptr</tt>. It yields the original value stored
5078 %result1 = call i32 @llvm.atomic.lss( i32* %ptr, i32 4 )
5079 <i>; yields {i32}:result1 = 32</i>
5080 %result2 = call i32 @llvm.atomic.lss( i32* %ptr, i32 2 )
5081 <i>; yields {i32}:result2 = 28</i>
5082 %result3 = call i32 @llvm.atomic.lss( i32* %ptr, i32 5 )
5083 <i>; yields {i32}:result3 = 26</i>
5084 %memval = load i32* %ptr <i>; yields {i32}:memval1 = 21</i>
5088 <!-- _______________________________________________________________________ -->
5089 <div class="doc_subsubsection">
5090 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5092 <div class="doc_text">
5095 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss> )
5099 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5100 specific pairs of memory access types.
5104 The <tt>llvm.memory.barrier</tt> intrinsic requires four boolean arguments.
5105 Each argument enables a specific barrier as listed below.
5108 <li><tt>ll</tt>: load-load barrier</li>
5109 <li><tt>ls</tt>: load-store barrier</li>
5110 <li><tt>sl</tt>: store-load barrier</li>
5111 <li><tt>ss</tt>: store-store barrier</li>
5115 This intrinsic causes the system to enforce some ordering constraints upon
5116 the loads and stores of the program. This barrier does not indicate
5117 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5118 which they occur. For any of the specified pairs of load and store operations
5119 (f.ex. load-load, or store-load), all of the first operations preceding the
5120 barrier will complete before any of the second operations succeeding the
5121 barrier begin. Specifically the semantics for each pairing is as follows:
5124 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5125 after the barrier begins.</li>
5126 <li><tt>ls</tt>: All loads before the barrier must complete before any
5127 store after the barrier begins.</li>
5128 <li><tt>ss</tt>: All stores before the barrier must complete before any
5129 store after the barrier begins.</li>
5130 <li><tt>sl</tt>: All stores before the barrier must complete before any
5131 load after the barrier begins.</li>
5134 These semantics are applied with a logical "and" behavior when more than one
5135 is enabled in a single memory barrier intrinsic.
5142 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5143 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5144 <i>; guarantee the above finishes</i>
5145 store i32 8, %ptr <i>; before this begins</i>
5149 <!-- ======================================================================= -->
5150 <div class="doc_subsection">
5151 <a name="int_trampoline">Trampoline Intrinsic</a>
5154 <div class="doc_text">
5156 This intrinsic makes it possible to excise one parameter, marked with
5157 the <tt>nest</tt> attribute, from a function. The result is a callable
5158 function pointer lacking the nest parameter - the caller does not need
5159 to provide a value for it. Instead, the value to use is stored in
5160 advance in a "trampoline", a block of memory usually allocated
5161 on the stack, which also contains code to splice the nest value into the
5162 argument list. This is used to implement the GCC nested function address
5166 For example, if the function is
5167 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5168 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:
5170 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5171 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5172 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5173 %fp = bitcast i8* %p to i32 (i32, i32)*
5175 The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent to
5176 <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.
5180 <!-- _______________________________________________________________________ -->
5181 <div class="doc_subsubsection">
5182 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5184 <div class="doc_text">
5187 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5191 This fills the memory pointed to by <tt>tramp</tt> with code
5192 and returns a function pointer suitable for executing it.
5196 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5197 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5198 and sufficiently aligned block of memory; this memory is written to by the
5199 intrinsic. Note that the size and the alignment are target-specific - LLVM
5200 currently provides no portable way of determining them, so a front-end that
5201 generates this intrinsic needs to have some target-specific knowledge.
5202 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5206 The block of memory pointed to by <tt>tramp</tt> is filled with target
5207 dependent code, turning it into a function. A pointer to this function is
5208 returned, but needs to be bitcast to an
5209 <a href="#int_trampoline">appropriate function pointer type</a>
5210 before being called. The new function's signature is the same as that of
5211 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5212 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5213 of pointer type. Calling the new function is equivalent to calling
5214 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5215 missing <tt>nest</tt> argument. If, after calling
5216 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5217 modified, then the effect of any later call to the returned function pointer is
5222 <!-- ======================================================================= -->
5223 <div class="doc_subsection">
5224 <a name="int_general">General Intrinsics</a>
5227 <div class="doc_text">
5228 <p> This class of intrinsics is designed to be generic and has
5229 no specific purpose. </p>
5232 <!-- _______________________________________________________________________ -->
5233 <div class="doc_subsubsection">
5234 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5237 <div class="doc_text">
5241 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5247 The '<tt>llvm.var.annotation</tt>' intrinsic
5253 The first argument is a pointer to a value, the second is a pointer to a
5254 global string, the third is a pointer to a global string which is the source
5255 file name, and the last argument is the line number.
5261 This intrinsic allows annotation of local variables with arbitrary strings.
5262 This can be useful for special purpose optimizations that want to look for these
5263 annotations. These have no other defined use, they are ignored by code
5264 generation and optimization.
5268 <!-- *********************************************************************** -->
5271 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5272 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5273 <a href="http://validator.w3.org/check/referer"><img
5274 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
5276 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5277 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5278 Last modified: $Date$