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="#fnattrs">Function Attributes</a></li>
30 <li><a href="#gc">Garbage Collector Names</a></li>
31 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
32 <li><a href="#datalayout">Data Layout</a></li>
35 <li><a href="#typesystem">Type System</a>
37 <li><a href="#t_classifications">Type Classifications</a></li>
38 <li><a href="#t_primitive">Primitive Types</a>
40 <li><a href="#t_floating">Floating Point Types</a></li>
41 <li><a href="#t_void">Void Type</a></li>
42 <li><a href="#t_label">Label Type</a></li>
45 <li><a href="#t_derived">Derived Types</a>
47 <li><a href="#t_integer">Integer Type</a></li>
48 <li><a href="#t_array">Array Type</a></li>
49 <li><a href="#t_function">Function Type</a></li>
50 <li><a href="#t_pointer">Pointer Type</a></li>
51 <li><a href="#t_struct">Structure Type</a></li>
52 <li><a href="#t_pstruct">Packed Structure Type</a></li>
53 <li><a href="#t_vector">Vector Type</a></li>
54 <li><a href="#t_opaque">Opaque Type</a></li>
59 <li><a href="#constants">Constants</a>
61 <li><a href="#simpleconstants">Simple Constants</a>
62 <li><a href="#aggregateconstants">Aggregate Constants</a>
63 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
64 <li><a href="#undefvalues">Undefined Values</a>
65 <li><a href="#constantexprs">Constant Expressions</a>
68 <li><a href="#othervalues">Other Values</a>
70 <li><a href="#inlineasm">Inline Assembler Expressions</a>
73 <li><a href="#instref">Instruction Reference</a>
75 <li><a href="#terminators">Terminator Instructions</a>
77 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
78 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
79 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
80 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
81 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
82 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
85 <li><a href="#binaryops">Binary Operations</a>
87 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
88 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
89 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
90 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
91 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
92 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
93 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
94 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
95 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
98 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
100 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
101 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
102 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
103 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
104 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
105 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
108 <li><a href="#vectorops">Vector Operations</a>
110 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
111 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
112 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
115 <li><a href="#aggregateops">Aggregate Operations</a>
117 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
118 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
121 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
123 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
124 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
125 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
126 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
127 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
128 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
131 <li><a href="#convertops">Conversion Operations</a>
133 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
134 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
135 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
136 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
137 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
138 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
139 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
140 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
141 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
142 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
143 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
144 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
146 <li><a href="#otherops">Other Operations</a>
148 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
149 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
150 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
151 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
152 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
153 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
154 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
155 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
160 <li><a href="#intrinsics">Intrinsic Functions</a>
162 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
164 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
165 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
166 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
169 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
171 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
172 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
173 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
176 <li><a href="#int_codegen">Code Generator Intrinsics</a>
178 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
179 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
180 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
181 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
182 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
183 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
184 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
187 <li><a href="#int_libc">Standard C Library Intrinsics</a>
189 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
190 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
191 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
192 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
193 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
194 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
195 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
196 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
199 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
201 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
202 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
203 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
204 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
205 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
206 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
209 <li><a href="#int_debugger">Debugger intrinsics</a></li>
210 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
211 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
213 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
216 <li><a href="#int_atomics">Atomic intrinsics</a>
218 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
219 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
220 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
221 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
222 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
223 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
224 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
225 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
226 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
227 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
228 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
229 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
230 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
233 <li><a href="#int_general">General intrinsics</a>
235 <li><a href="#int_var_annotation">
236 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
237 <li><a href="#int_annotation">
238 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
239 <li><a href="#int_trap">
240 <tt>llvm.trap</tt>' Intrinsic</a></li>
247 <div class="doc_author">
248 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
249 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
252 <!-- *********************************************************************** -->
253 <div class="doc_section"> <a name="abstract">Abstract </a></div>
254 <!-- *********************************************************************** -->
256 <div class="doc_text">
257 <p>This document is a reference manual for the LLVM assembly language.
258 LLVM is a Static Single Assignment (SSA) based representation that provides
259 type safety, low-level operations, flexibility, and the capability of
260 representing 'all' high-level languages cleanly. It is the common code
261 representation used throughout all phases of the LLVM compilation
265 <!-- *********************************************************************** -->
266 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
267 <!-- *********************************************************************** -->
269 <div class="doc_text">
271 <p>The LLVM code representation is designed to be used in three
272 different forms: as an in-memory compiler IR, as an on-disk bitcode
273 representation (suitable for fast loading by a Just-In-Time compiler),
274 and as a human readable assembly language representation. This allows
275 LLVM to provide a powerful intermediate representation for efficient
276 compiler transformations and analysis, while providing a natural means
277 to debug and visualize the transformations. The three different forms
278 of LLVM are all equivalent. This document describes the human readable
279 representation and notation.</p>
281 <p>The LLVM representation aims to be light-weight and low-level
282 while being expressive, typed, and extensible at the same time. It
283 aims to be a "universal IR" of sorts, by being at a low enough level
284 that high-level ideas may be cleanly mapped to it (similar to how
285 microprocessors are "universal IR's", allowing many source languages to
286 be mapped to them). By providing type information, LLVM can be used as
287 the target of optimizations: for example, through pointer analysis, it
288 can be proven that a C automatic variable is never accessed outside of
289 the current function... allowing it to be promoted to a simple SSA
290 value instead of a memory location.</p>
294 <!-- _______________________________________________________________________ -->
295 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
297 <div class="doc_text">
299 <p>It is important to note that this document describes 'well formed'
300 LLVM assembly language. There is a difference between what the parser
301 accepts and what is considered 'well formed'. For example, the
302 following instruction is syntactically okay, but not well formed:</p>
304 <div class="doc_code">
306 %x = <a href="#i_add">add</a> i32 1, %x
310 <p>...because the definition of <tt>%x</tt> does not dominate all of
311 its uses. The LLVM infrastructure provides a verification pass that may
312 be used to verify that an LLVM module is well formed. This pass is
313 automatically run by the parser after parsing input assembly and by
314 the optimizer before it outputs bitcode. The violations pointed out
315 by the verifier pass indicate bugs in transformation passes or input to
319 <!-- Describe the typesetting conventions here. -->
321 <!-- *********************************************************************** -->
322 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
323 <!-- *********************************************************************** -->
325 <div class="doc_text">
327 <p>LLVM identifiers come in two basic types: global and local. Global
328 identifiers (functions, global variables) begin with the @ character. Local
329 identifiers (register names, types) begin with the % character. Additionally,
330 there are three different formats for identifiers, for different purposes:
333 <li>Named values are represented as a string of characters with their prefix.
334 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
335 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
336 Identifiers which require other characters in their names can be surrounded
337 with quotes. In this way, anything except a <tt>"</tt> character can
338 be used in a named value.</li>
340 <li>Unnamed values are represented as an unsigned numeric value with their
341 prefix. For example, %12, @2, %44.</li>
343 <li>Constants, which are described in a <a href="#constants">section about
344 constants</a>, below.</li>
347 <p>LLVM requires that values start with a prefix for two reasons: Compilers
348 don't need to worry about name clashes with reserved words, and the set of
349 reserved words may be expanded in the future without penalty. Additionally,
350 unnamed identifiers allow a compiler to quickly come up with a temporary
351 variable without having to avoid symbol table conflicts.</p>
353 <p>Reserved words in LLVM are very similar to reserved words in other
354 languages. There are keywords for different opcodes
355 ('<tt><a href="#i_add">add</a></tt>',
356 '<tt><a href="#i_bitcast">bitcast</a></tt>',
357 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
358 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
359 and others. These reserved words cannot conflict with variable names, because
360 none of them start with a prefix character ('%' or '@').</p>
362 <p>Here is an example of LLVM code to multiply the integer variable
363 '<tt>%X</tt>' by 8:</p>
367 <div class="doc_code">
369 %result = <a href="#i_mul">mul</a> i32 %X, 8
373 <p>After strength reduction:</p>
375 <div class="doc_code">
377 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
381 <p>And the hard way:</p>
383 <div class="doc_code">
385 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
386 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
387 %result = <a href="#i_add">add</a> i32 %1, %1
391 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
392 important lexical features of LLVM:</p>
396 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
399 <li>Unnamed temporaries are created when the result of a computation is not
400 assigned to a named value.</li>
402 <li>Unnamed temporaries are numbered sequentially</li>
406 <p>...and it also shows a convention that we follow in this document. When
407 demonstrating instructions, we will follow an instruction with a comment that
408 defines the type and name of value produced. Comments are shown in italic
413 <!-- *********************************************************************** -->
414 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
415 <!-- *********************************************************************** -->
417 <!-- ======================================================================= -->
418 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
421 <div class="doc_text">
423 <p>LLVM programs are composed of "Module"s, each of which is a
424 translation unit of the input programs. Each module consists of
425 functions, global variables, and symbol table entries. Modules may be
426 combined together with the LLVM linker, which merges function (and
427 global variable) definitions, resolves forward declarations, and merges
428 symbol table entries. Here is an example of the "hello world" module:</p>
430 <div class="doc_code">
431 <pre><i>; Declare the string constant as a global constant...</i>
432 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
433 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
435 <i>; External declaration of the puts function</i>
436 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
438 <i>; Definition of main function</i>
439 define i32 @main() { <i>; i32()* </i>
440 <i>; Convert [13x i8 ]* to i8 *...</i>
442 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
444 <i>; Call puts function to write out the string to stdout...</i>
446 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
448 href="#i_ret">ret</a> i32 0<br>}<br>
452 <p>This example is made up of a <a href="#globalvars">global variable</a>
453 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
454 function, and a <a href="#functionstructure">function definition</a>
455 for "<tt>main</tt>".</p>
457 <p>In general, a module is made up of a list of global values,
458 where both functions and global variables are global values. Global values are
459 represented by a pointer to a memory location (in this case, a pointer to an
460 array of char, and a pointer to a function), and have one of the following <a
461 href="#linkage">linkage types</a>.</p>
465 <!-- ======================================================================= -->
466 <div class="doc_subsection">
467 <a name="linkage">Linkage Types</a>
470 <div class="doc_text">
473 All Global Variables and Functions have one of the following types of linkage:
478 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
480 <dd>Global values with internal linkage are only directly accessible by
481 objects in the current module. In particular, linking code into a module with
482 an internal global value may cause the internal to be renamed as necessary to
483 avoid collisions. Because the symbol is internal to the module, all
484 references can be updated. This corresponds to the notion of the
485 '<tt>static</tt>' keyword in C.
488 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
490 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
491 the same name when linkage occurs. This is typically used to implement
492 inline functions, templates, or other code which must be generated in each
493 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
494 allowed to be discarded.
497 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
499 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
500 linkage, except that unreferenced <tt>common</tt> globals may not be
501 discarded. This is used for globals that may be emitted in multiple
502 translation units, but that are not guaranteed to be emitted into every
503 translation unit that uses them. One example of this is tentative
504 definitions in C, such as "<tt>int X;</tt>" at global scope.
507 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
509 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
510 that some targets may choose to emit different assembly sequences for them
511 for target-dependent reasons. This is used for globals that are declared
512 "weak" in C source code.
515 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
517 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
518 pointer to array type. When two global variables with appending linkage are
519 linked together, the two global arrays are appended together. This is the
520 LLVM, typesafe, equivalent of having the system linker append together
521 "sections" with identical names when .o files are linked.
524 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
525 <dd>The semantics of this linkage follow the ELF object file model: the
526 symbol is weak until linked, if not linked, the symbol becomes null instead
527 of being an undefined reference.
530 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
532 <dd>If none of the above identifiers are used, the global is externally
533 visible, meaning that it participates in linkage and can be used to resolve
534 external symbol references.
539 The next two types of linkage are targeted for Microsoft Windows platform
540 only. They are designed to support importing (exporting) symbols from (to)
541 DLLs (Dynamic Link Libraries).
545 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
547 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
548 or variable via a global pointer to a pointer that is set up by the DLL
549 exporting the symbol. On Microsoft Windows targets, the pointer name is
550 formed by combining <code>_imp__</code> and the function or variable name.
553 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
555 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
556 pointer to a pointer in a DLL, so that it can be referenced with the
557 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
558 name is formed by combining <code>_imp__</code> and the function or variable
564 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
565 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
566 variable and was linked with this one, one of the two would be renamed,
567 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
568 external (i.e., lacking any linkage declarations), they are accessible
569 outside of the current module.</p>
570 <p>It is illegal for a function <i>declaration</i>
571 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
572 or <tt>extern_weak</tt>.</p>
573 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
577 <!-- ======================================================================= -->
578 <div class="doc_subsection">
579 <a name="callingconv">Calling Conventions</a>
582 <div class="doc_text">
584 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
585 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
586 specified for the call. The calling convention of any pair of dynamic
587 caller/callee must match, or the behavior of the program is undefined. The
588 following calling conventions are supported by LLVM, and more may be added in
592 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
594 <dd>This calling convention (the default if no other calling convention is
595 specified) matches the target C calling conventions. This calling convention
596 supports varargs function calls and tolerates some mismatch in the declared
597 prototype and implemented declaration of the function (as does normal C).
600 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
602 <dd>This calling convention attempts to make calls as fast as possible
603 (e.g. by passing things in registers). This calling convention allows the
604 target to use whatever tricks it wants to produce fast code for the target,
605 without having to conform to an externally specified ABI (Application Binary
606 Interface). Implementations of this convention should allow arbitrary
607 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
608 supported. This calling convention does not support varargs and requires the
609 prototype of all callees to exactly match the prototype of the function
613 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
615 <dd>This calling convention attempts to make code in the caller as efficient
616 as possible under the assumption that the call is not commonly executed. As
617 such, these calls often preserve all registers so that the call does not break
618 any live ranges in the caller side. This calling convention does not support
619 varargs and requires the prototype of all callees to exactly match the
620 prototype of the function definition.
623 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
625 <dd>Any calling convention may be specified by number, allowing
626 target-specific calling conventions to be used. Target specific calling
627 conventions start at 64.
631 <p>More calling conventions can be added/defined on an as-needed basis, to
632 support pascal conventions or any other well-known target-independent
637 <!-- ======================================================================= -->
638 <div class="doc_subsection">
639 <a name="visibility">Visibility Styles</a>
642 <div class="doc_text">
645 All Global Variables and Functions have one of the following visibility styles:
649 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
651 <dd>On targets that use the ELF object file format, default visibility means
652 that the declaration is visible to other
653 modules and, in shared libraries, means that the declared entity may be
654 overridden. On Darwin, default visibility means that the declaration is
655 visible to other modules. Default visibility corresponds to "external
656 linkage" in the language.
659 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
661 <dd>Two declarations of an object with hidden visibility refer to the same
662 object if they are in the same shared object. Usually, hidden visibility
663 indicates that the symbol will not be placed into the dynamic symbol table,
664 so no other module (executable or shared library) can reference it
668 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
670 <dd>On ELF, protected visibility indicates that the symbol will be placed in
671 the dynamic symbol table, but that references within the defining module will
672 bind to the local symbol. That is, the symbol cannot be overridden by another
679 <!-- ======================================================================= -->
680 <div class="doc_subsection">
681 <a name="globalvars">Global Variables</a>
684 <div class="doc_text">
686 <p>Global variables define regions of memory allocated at compilation time
687 instead of run-time. Global variables may optionally be initialized, may have
688 an explicit section to be placed in, and may have an optional explicit alignment
689 specified. A variable may be defined as "thread_local", which means that it
690 will not be shared by threads (each thread will have a separated copy of the
691 variable). A variable may be defined as a global "constant," which indicates
692 that the contents of the variable will <b>never</b> be modified (enabling better
693 optimization, allowing the global data to be placed in the read-only section of
694 an executable, etc). Note that variables that need runtime initialization
695 cannot be marked "constant" as there is a store to the variable.</p>
698 LLVM explicitly allows <em>declarations</em> of global variables to be marked
699 constant, even if the final definition of the global is not. This capability
700 can be used to enable slightly better optimization of the program, but requires
701 the language definition to guarantee that optimizations based on the
702 'constantness' are valid for the translation units that do not include the
706 <p>As SSA values, global variables define pointer values that are in
707 scope (i.e. they dominate) all basic blocks in the program. Global
708 variables always define a pointer to their "content" type because they
709 describe a region of memory, and all memory objects in LLVM are
710 accessed through pointers.</p>
712 <p>A global variable may be declared to reside in a target-specifc numbered
713 address space. For targets that support them, address spaces may affect how
714 optimizations are performed and/or what target instructions are used to access
715 the variable. The default address space is zero. The address space qualifier
716 must precede any other attributes.</p>
718 <p>LLVM allows an explicit section to be specified for globals. If the target
719 supports it, it will emit globals to the section specified.</p>
721 <p>An explicit alignment may be specified for a global. If not present, or if
722 the alignment is set to zero, the alignment of the global is set by the target
723 to whatever it feels convenient. If an explicit alignment is specified, the
724 global is forced to have at least that much alignment. All alignments must be
727 <p>For example, the following defines a global in a numbered address space with
728 an initializer, section, and alignment:</p>
730 <div class="doc_code">
732 @G = constant float 1.0 addrspace(5), section "foo", align 4
739 <!-- ======================================================================= -->
740 <div class="doc_subsection">
741 <a name="functionstructure">Functions</a>
744 <div class="doc_text">
746 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
747 an optional <a href="#linkage">linkage type</a>, an optional
748 <a href="#visibility">visibility style</a>, an optional
749 <a href="#callingconv">calling convention</a>, a return type, an optional
750 <a href="#paramattrs">parameter attribute</a> for the return type, a function
751 name, a (possibly empty) argument list (each with optional
752 <a href="#paramattrs">parameter attributes</a>), optional
753 <a href="#fnattrs">function attributes</a>, an optional section,
754 an optional alignment, an optional <a href="#gc">garbage collector name</a>,
755 an opening curly brace, a list of basic blocks, and a closing curly brace.
757 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
758 optional <a href="#linkage">linkage type</a>, an optional
759 <a href="#visibility">visibility style</a>, an optional
760 <a href="#callingconv">calling convention</a>, a return type, an optional
761 <a href="#paramattrs">parameter attribute</a> for the return type, a function
762 name, a possibly empty list of arguments, an optional alignment, and an optional
763 <a href="#gc">garbage collector name</a>.</p>
765 <p>A function definition contains a list of basic blocks, forming the CFG
766 (Control Flow Graph) for
767 the function. Each basic block may optionally start with a label (giving the
768 basic block a symbol table entry), contains a list of instructions, and ends
769 with a <a href="#terminators">terminator</a> instruction (such as a branch or
770 function return).</p>
772 <p>The first basic block in a function is special in two ways: it is immediately
773 executed on entrance to the function, and it is not allowed to have predecessor
774 basic blocks (i.e. there can not be any branches to the entry block of a
775 function). Because the block can have no predecessors, it also cannot have any
776 <a href="#i_phi">PHI nodes</a>.</p>
778 <p>LLVM allows an explicit section to be specified for functions. If the target
779 supports it, it will emit functions to the section specified.</p>
781 <p>An explicit alignment may be specified for a function. If not present, or if
782 the alignment is set to zero, the alignment of the function is set by the target
783 to whatever it feels convenient. If an explicit alignment is specified, the
784 function is forced to have at least that much alignment. All alignments must be
790 <!-- ======================================================================= -->
791 <div class="doc_subsection">
792 <a name="aliasstructure">Aliases</a>
794 <div class="doc_text">
795 <p>Aliases act as "second name" for the aliasee value (which can be either
796 function, global variable, another alias or bitcast of global value). Aliases
797 may have an optional <a href="#linkage">linkage type</a>, and an
798 optional <a href="#visibility">visibility style</a>.</p>
802 <div class="doc_code">
804 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
812 <!-- ======================================================================= -->
813 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
814 <div class="doc_text">
815 <p>The return type and each parameter of a function type may have a set of
816 <i>parameter attributes</i> associated with them. Parameter attributes are
817 used to communicate additional information about the result or parameters of
818 a function. Parameter attributes are considered to be part of the function,
819 not of the function type, so functions with different parameter attributes
820 can have the same function type.</p>
822 <p>Parameter attributes are simple keywords that follow the type specified. If
823 multiple parameter attributes are needed, they are space separated. For
826 <div class="doc_code">
828 declare i32 @printf(i8* noalias , ...)
829 declare i32 @atoi(i8 zeroext)
830 declare signext i8 @returns_signed_char()
834 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
835 <tt>readonly</tt>) come immediately after the argument list.</p>
837 <p>Currently, only the following parameter attributes are defined:</p>
839 <dt><tt>zeroext</tt></dt>
840 <dd>This indicates to the code generator that the parameter or return value
841 should be zero-extended to a 32-bit value by the caller (for a parameter)
842 or the callee (for a return value).</dd>
844 <dt><tt>signext</tt></dt>
845 <dd>This indicates to the code generator that the parameter or return value
846 should be sign-extended to a 32-bit value by the caller (for a parameter)
847 or the callee (for a return value).</dd>
849 <dt><tt>inreg</tt></dt>
850 <dd>This indicates that this parameter or return value should be treated
851 in a special target-dependent fashion during while emitting code for a
852 function call or return (usually, by putting it in a register as opposed
853 to memory, though some targets use it to distinguish between two different
854 kinds of registers). Use of this attribute is target-specific.</dd>
856 <dt><tt><a name="byval">byval</a></tt></dt>
857 <dd>This indicates that the pointer parameter should really be passed by
858 value to the function. The attribute implies that a hidden copy of the
859 pointee is made between the caller and the callee, so the callee is unable
860 to modify the value in the callee. This attribute is only valid on LLVM
861 pointer arguments. It is generally used to pass structs and arrays by
862 value, but is also valid on pointers to scalars. The copy is considered to
863 belong to the caller not the callee (for example,
864 <tt><a href="#readonly">readonly</a></tt> functions should not write to
865 <tt>byval</tt> parameters). This is not a valid attribute for return
868 <dt><tt>sret</tt></dt>
869 <dd>This indicates that the pointer parameter specifies the address of a
870 structure that is the return value of the function in the source program.
871 This pointer must be guaranteed by the caller to be valid: loads and stores
872 to the structure may be assumed by the callee to not to trap. This may only
873 be applied to the first parameter. This is not a valid attribute for
876 <dt><tt>noalias</tt></dt>
877 <dd>This indicates that the parameter does not alias any global or any other
878 parameter. The caller is responsible for ensuring that this is the case,
879 usually by placing the value in a stack allocation. This is not a valid
880 attribute for return values.</dd>
882 <dt><tt>nest</tt></dt>
883 <dd>This indicates that the pointer parameter can be excised using the
884 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
885 attribute for return values.</dd>
890 <!-- ======================================================================= -->
891 <div class="doc_subsection">
892 <a name="gc">Garbage Collector Names</a>
895 <div class="doc_text">
896 <p>Each function may specify a garbage collector name, which is simply a
899 <div class="doc_code"><pre
900 >define void @f() gc "name" { ...</pre></div>
902 <p>The compiler declares the supported values of <i>name</i>. Specifying a
903 collector which will cause the compiler to alter its output in order to support
904 the named garbage collection algorithm.</p>
907 <!-- ======================================================================= -->
908 <div class="doc_subsection">
909 <a name="fnattrs">Function Attributes</a>
912 <div class="doc_text">
914 <p>Function attributes are set to communicate additional information about
915 a function. Function attributes are considered to be part of the function,
916 not of the function type, so functions with different parameter attributes
917 can have the same function type.</p>
919 <p>Function attributes are simple keywords that follow the type specified. If
920 multiple attributes are needed, they are space separated. For
923 <div class="doc_code">
925 define void @f() noinline { ... }
926 define void @f() alwaysinline { ... }
927 define void @f() alwaysinline optsize { ... }
928 define void @f() optsize
933 <dt><tt>alwaysinline</tt></dt>
934 <dd>This attribute indicates that the inliner should attempt to inline this
935 function into callers whenever possible, ignoring any active inlining size
936 threshold for this caller.</dd>
938 <dt><tt>noinline</tt></dt>
939 <dd>This attribute indicates that the inliner should never inline this function
940 in any situation. This attribute may not be used together with the
941 <tt>alwaysinline</tt> attribute.</dd>
943 <dt><tt>optsize</tt></dt>
944 <dd>This attribute suggests that optimization passes and code generator passes
945 make choices that keep the code size of this function low, and otherwise do
946 optimizations specifically to reduce code size.</dd>
948 <dt><tt>noreturn</tt></dt>
949 <dd>This function attribute indicates that the function never returns normally.
950 This produces undefined behavior at runtime if the function ever does
951 dynamically return.</dd>
953 <dt><tt>nounwind</tt></dt>
954 <dd>This function attribute indicates that the function never returns with an
955 unwind or exceptional control flow. If the function does unwind, its runtime
956 behavior is undefined.</dd>
958 <dt><tt>readnone</tt></dt>
959 <dd>This attribute indicates that the function computes its result (or the
960 exception it throws) based strictly on its arguments, without dereferencing any
961 pointer arguments or otherwise accessing any mutable state (e.g. memory, control
962 registers, etc) visible to caller functions. It does not write through any
963 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments) and
964 never changes any state visible to callers.</dd>
966 <dt><tt><a name="readonly">readonly</a></tt></dt>
967 <dd>This attribute indicates that the function does not write through any
968 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments)
969 or otherwise modify any state (e.g. memory, control registers, etc) visible to
970 caller functions. It may dereference pointer arguments and read state that may
971 be set in the caller. A readonly function always returns the same value (or
972 throws the same exception) when called with the same set of arguments and global
978 <!-- ======================================================================= -->
979 <div class="doc_subsection">
980 <a name="moduleasm">Module-Level Inline Assembly</a>
983 <div class="doc_text">
985 Modules may contain "module-level inline asm" blocks, which corresponds to the
986 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
987 LLVM and treated as a single unit, but may be separated in the .ll file if
988 desired. The syntax is very simple:
991 <div class="doc_code">
993 module asm "inline asm code goes here"
994 module asm "more can go here"
998 <p>The strings can contain any character by escaping non-printable characters.
999 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1004 The inline asm code is simply printed to the machine code .s file when
1005 assembly code is generated.
1009 <!-- ======================================================================= -->
1010 <div class="doc_subsection">
1011 <a name="datalayout">Data Layout</a>
1014 <div class="doc_text">
1015 <p>A module may specify a target specific data layout string that specifies how
1016 data is to be laid out in memory. The syntax for the data layout is simply:</p>
1017 <pre> target datalayout = "<i>layout specification</i>"</pre>
1018 <p>The <i>layout specification</i> consists of a list of specifications
1019 separated by the minus sign character ('-'). Each specification starts with a
1020 letter and may include other information after the letter to define some
1021 aspect of the data layout. The specifications accepted are as follows: </p>
1024 <dd>Specifies that the target lays out data in big-endian form. That is, the
1025 bits with the most significance have the lowest address location.</dd>
1027 <dd>Specifies that the target lays out data in little-endian form. That is,
1028 the bits with the least significance have the lowest address location.</dd>
1029 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1030 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1031 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
1032 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
1034 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1035 <dd>This specifies the alignment for an integer type of a given bit
1036 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1037 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1038 <dd>This specifies the alignment for a vector type of a given bit
1040 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1041 <dd>This specifies the alignment for a floating point type of a given bit
1042 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1044 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1045 <dd>This specifies the alignment for an aggregate type of a given bit
1048 <p>When constructing the data layout for a given target, LLVM starts with a
1049 default set of specifications which are then (possibly) overriden by the
1050 specifications in the <tt>datalayout</tt> keyword. The default specifications
1051 are given in this list:</p>
1053 <li><tt>E</tt> - big endian</li>
1054 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1055 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1056 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1057 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1058 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1059 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1060 alignment of 64-bits</li>
1061 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1062 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1063 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1064 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1065 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1067 <p>When LLVM is determining the alignment for a given type, it uses the
1070 <li>If the type sought is an exact match for one of the specifications, that
1071 specification is used.</li>
1072 <li>If no match is found, and the type sought is an integer type, then the
1073 smallest integer type that is larger than the bitwidth of the sought type is
1074 used. If none of the specifications are larger than the bitwidth then the the
1075 largest integer type is used. For example, given the default specifications
1076 above, the i7 type will use the alignment of i8 (next largest) while both
1077 i65 and i256 will use the alignment of i64 (largest specified).</li>
1078 <li>If no match is found, and the type sought is a vector type, then the
1079 largest vector type that is smaller than the sought vector type will be used
1080 as a fall back. This happens because <128 x double> can be implemented in
1081 terms of 64 <2 x double>, for example.</li>
1085 <!-- *********************************************************************** -->
1086 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1087 <!-- *********************************************************************** -->
1089 <div class="doc_text">
1091 <p>The LLVM type system is one of the most important features of the
1092 intermediate representation. Being typed enables a number of
1093 optimizations to be performed on the intermediate representation directly,
1094 without having to do
1095 extra analyses on the side before the transformation. A strong type
1096 system makes it easier to read the generated code and enables novel
1097 analyses and transformations that are not feasible to perform on normal
1098 three address code representations.</p>
1102 <!-- ======================================================================= -->
1103 <div class="doc_subsection"> <a name="t_classifications">Type
1104 Classifications</a> </div>
1105 <div class="doc_text">
1106 <p>The types fall into a few useful
1107 classifications:</p>
1109 <table border="1" cellspacing="0" cellpadding="4">
1111 <tr><th>Classification</th><th>Types</th></tr>
1113 <td><a href="#t_integer">integer</a></td>
1114 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1117 <td><a href="#t_floating">floating point</a></td>
1118 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1121 <td><a name="t_firstclass">first class</a></td>
1122 <td><a href="#t_integer">integer</a>,
1123 <a href="#t_floating">floating point</a>,
1124 <a href="#t_pointer">pointer</a>,
1125 <a href="#t_vector">vector</a>,
1126 <a href="#t_struct">structure</a>,
1127 <a href="#t_array">array</a>,
1128 <a href="#t_label">label</a>.
1132 <td><a href="#t_primitive">primitive</a></td>
1133 <td><a href="#t_label">label</a>,
1134 <a href="#t_void">void</a>,
1135 <a href="#t_floating">floating point</a>.</td>
1138 <td><a href="#t_derived">derived</a></td>
1139 <td><a href="#t_integer">integer</a>,
1140 <a href="#t_array">array</a>,
1141 <a href="#t_function">function</a>,
1142 <a href="#t_pointer">pointer</a>,
1143 <a href="#t_struct">structure</a>,
1144 <a href="#t_pstruct">packed structure</a>,
1145 <a href="#t_vector">vector</a>,
1146 <a href="#t_opaque">opaque</a>.
1151 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1152 most important. Values of these types are the only ones which can be
1153 produced by instructions, passed as arguments, or used as operands to
1157 <!-- ======================================================================= -->
1158 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1160 <div class="doc_text">
1161 <p>The primitive types are the fundamental building blocks of the LLVM
1166 <!-- _______________________________________________________________________ -->
1167 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1169 <div class="doc_text">
1172 <tr><th>Type</th><th>Description</th></tr>
1173 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1174 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1175 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1176 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1177 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1182 <!-- _______________________________________________________________________ -->
1183 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1185 <div class="doc_text">
1187 <p>The void type does not represent any value and has no size.</p>
1196 <!-- _______________________________________________________________________ -->
1197 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1199 <div class="doc_text">
1201 <p>The label type represents code labels.</p>
1211 <!-- ======================================================================= -->
1212 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1214 <div class="doc_text">
1216 <p>The real power in LLVM comes from the derived types in the system.
1217 This is what allows a programmer to represent arrays, functions,
1218 pointers, and other useful types. Note that these derived types may be
1219 recursive: For example, it is possible to have a two dimensional array.</p>
1223 <!-- _______________________________________________________________________ -->
1224 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1226 <div class="doc_text">
1229 <p>The integer type is a very simple derived type that simply specifies an
1230 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1231 2^23-1 (about 8 million) can be specified.</p>
1239 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1243 <table class="layout">
1246 <td><tt>i1</tt></td>
1247 <td>a single-bit integer.</td>
1249 <td><tt>i32</tt></td>
1250 <td>a 32-bit integer.</td>
1252 <td><tt>i1942652</tt></td>
1253 <td>a really big integer of over 1 million bits.</td>
1259 <!-- _______________________________________________________________________ -->
1260 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1262 <div class="doc_text">
1266 <p>The array type is a very simple derived type that arranges elements
1267 sequentially in memory. The array type requires a size (number of
1268 elements) and an underlying data type.</p>
1273 [<# elements> x <elementtype>]
1276 <p>The number of elements is a constant integer value; elementtype may
1277 be any type with a size.</p>
1280 <table class="layout">
1282 <td class="left"><tt>[40 x i32]</tt></td>
1283 <td class="left">Array of 40 32-bit integer values.</td>
1286 <td class="left"><tt>[41 x i32]</tt></td>
1287 <td class="left">Array of 41 32-bit integer values.</td>
1290 <td class="left"><tt>[4 x i8]</tt></td>
1291 <td class="left">Array of 4 8-bit integer values.</td>
1294 <p>Here are some examples of multidimensional arrays:</p>
1295 <table class="layout">
1297 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1298 <td class="left">3x4 array of 32-bit integer values.</td>
1301 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1302 <td class="left">12x10 array of single precision floating point values.</td>
1305 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1306 <td class="left">2x3x4 array of 16-bit integer values.</td>
1310 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1311 length array. Normally, accesses past the end of an array are undefined in
1312 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1313 As a special case, however, zero length arrays are recognized to be variable
1314 length. This allows implementation of 'pascal style arrays' with the LLVM
1315 type "{ i32, [0 x float]}", for example.</p>
1319 <!-- _______________________________________________________________________ -->
1320 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1321 <div class="doc_text">
1325 <p>The function type can be thought of as a function signature. It
1326 consists of a return type and a list of formal parameter types. The
1327 return type of a function type is a scalar type, a void type, or a struct type.
1328 If the return type is a struct type then all struct elements must be of first
1329 class types, and the struct must have at least one element.</p>
1334 <returntype list> (<parameter list>)
1337 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1338 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1339 which indicates that the function takes a variable number of arguments.
1340 Variable argument functions can access their arguments with the <a
1341 href="#int_varargs">variable argument handling intrinsic</a> functions.
1342 '<tt><returntype list></tt>' is a comma-separated list of
1343 <a href="#t_firstclass">first class</a> type specifiers.</p>
1346 <table class="layout">
1348 <td class="left"><tt>i32 (i32)</tt></td>
1349 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1351 </tr><tr class="layout">
1352 <td class="left"><tt>float (i16 signext, i32 *) *
1354 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1355 an <tt>i16</tt> that should be sign extended and a
1356 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1359 </tr><tr class="layout">
1360 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1361 <td class="left">A vararg function that takes at least one
1362 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1363 which returns an integer. This is the signature for <tt>printf</tt> in
1366 </tr><tr class="layout">
1367 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1368 <td class="left">A function taking an <tt>i32></tt>, returning two
1369 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1375 <!-- _______________________________________________________________________ -->
1376 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1377 <div class="doc_text">
1379 <p>The structure type is used to represent a collection of data members
1380 together in memory. The packing of the field types is defined to match
1381 the ABI of the underlying processor. The elements of a structure may
1382 be any type that has a size.</p>
1383 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1384 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1385 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1388 <pre> { <type list> }<br></pre>
1390 <table class="layout">
1392 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1393 <td class="left">A triple of three <tt>i32</tt> values</td>
1394 </tr><tr class="layout">
1395 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1396 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1397 second element is a <a href="#t_pointer">pointer</a> to a
1398 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1399 an <tt>i32</tt>.</td>
1404 <!-- _______________________________________________________________________ -->
1405 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1407 <div class="doc_text">
1409 <p>The packed structure type is used to represent a collection of data members
1410 together in memory. There is no padding between fields. Further, the alignment
1411 of a packed structure is 1 byte. The elements of a packed structure may
1412 be any type that has a size.</p>
1413 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1414 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1415 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1418 <pre> < { <type list> } > <br></pre>
1420 <table class="layout">
1422 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1423 <td class="left">A triple of three <tt>i32</tt> values</td>
1424 </tr><tr class="layout">
1426 <tt>< { float, i32 (i32)* } ></tt></td>
1427 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1428 second element is a <a href="#t_pointer">pointer</a> to a
1429 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1430 an <tt>i32</tt>.</td>
1435 <!-- _______________________________________________________________________ -->
1436 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1437 <div class="doc_text">
1439 <p>As in many languages, the pointer type represents a pointer or
1440 reference to another object, which must live in memory. Pointer types may have
1441 an optional address space attribute defining the target-specific numbered
1442 address space where the pointed-to object resides. The default address space is
1445 <pre> <type> *<br></pre>
1447 <table class="layout">
1449 <td class="left"><tt>[4x i32]*</tt></td>
1450 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1451 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1454 <td class="left"><tt>i32 (i32 *) *</tt></td>
1455 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1456 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1460 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1461 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1462 that resides in address space #5.</td>
1467 <!-- _______________________________________________________________________ -->
1468 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1469 <div class="doc_text">
1473 <p>A vector type is a simple derived type that represents a vector
1474 of elements. Vector types are used when multiple primitive data
1475 are operated in parallel using a single instruction (SIMD).
1476 A vector type requires a size (number of
1477 elements) and an underlying primitive data type. Vectors must have a power
1478 of two length (1, 2, 4, 8, 16 ...). Vector types are
1479 considered <a href="#t_firstclass">first class</a>.</p>
1484 < <# elements> x <elementtype> >
1487 <p>The number of elements is a constant integer value; elementtype may
1488 be any integer or floating point type.</p>
1492 <table class="layout">
1494 <td class="left"><tt><4 x i32></tt></td>
1495 <td class="left">Vector of 4 32-bit integer values.</td>
1498 <td class="left"><tt><8 x float></tt></td>
1499 <td class="left">Vector of 8 32-bit floating-point values.</td>
1502 <td class="left"><tt><2 x i64></tt></td>
1503 <td class="left">Vector of 2 64-bit integer values.</td>
1508 <!-- _______________________________________________________________________ -->
1509 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1510 <div class="doc_text">
1514 <p>Opaque types are used to represent unknown types in the system. This
1515 corresponds (for example) to the C notion of a forward declared structure type.
1516 In LLVM, opaque types can eventually be resolved to any type (not just a
1517 structure type).</p>
1527 <table class="layout">
1529 <td class="left"><tt>opaque</tt></td>
1530 <td class="left">An opaque type.</td>
1536 <!-- *********************************************************************** -->
1537 <div class="doc_section"> <a name="constants">Constants</a> </div>
1538 <!-- *********************************************************************** -->
1540 <div class="doc_text">
1542 <p>LLVM has several different basic types of constants. This section describes
1543 them all and their syntax.</p>
1547 <!-- ======================================================================= -->
1548 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1550 <div class="doc_text">
1553 <dt><b>Boolean constants</b></dt>
1555 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1556 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1559 <dt><b>Integer constants</b></dt>
1561 <dd>Standard integers (such as '4') are constants of the <a
1562 href="#t_integer">integer</a> type. Negative numbers may be used with
1566 <dt><b>Floating point constants</b></dt>
1568 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1569 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1570 notation (see below). The assembler requires the exact decimal value of
1571 a floating-point constant. For example, the assembler accepts 1.25 but
1572 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1573 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1575 <dt><b>Null pointer constants</b></dt>
1577 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1578 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1582 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1583 of floating point constants. For example, the form '<tt>double
1584 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1585 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1586 (and the only time that they are generated by the disassembler) is when a
1587 floating point constant must be emitted but it cannot be represented as a
1588 decimal floating point number. For example, NaN's, infinities, and other
1589 special values are represented in their IEEE hexadecimal format so that
1590 assembly and disassembly do not cause any bits to change in the constants.</p>
1594 <!-- ======================================================================= -->
1595 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1598 <div class="doc_text">
1599 <p>Aggregate constants arise from aggregation of simple constants
1600 and smaller aggregate constants.</p>
1603 <dt><b>Structure constants</b></dt>
1605 <dd>Structure constants are represented with notation similar to structure
1606 type definitions (a comma separated list of elements, surrounded by braces
1607 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1608 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1609 must have <a href="#t_struct">structure type</a>, and the number and
1610 types of elements must match those specified by the type.
1613 <dt><b>Array constants</b></dt>
1615 <dd>Array constants are represented with notation similar to array type
1616 definitions (a comma separated list of elements, surrounded by square brackets
1617 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1618 constants must have <a href="#t_array">array type</a>, and the number and
1619 types of elements must match those specified by the type.
1622 <dt><b>Vector constants</b></dt>
1624 <dd>Vector constants are represented with notation similar to vector type
1625 definitions (a comma separated list of elements, surrounded by
1626 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1627 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1628 href="#t_vector">vector type</a>, and the number and types of elements must
1629 match those specified by the type.
1632 <dt><b>Zero initialization</b></dt>
1634 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1635 value to zero of <em>any</em> type, including scalar and aggregate types.
1636 This is often used to avoid having to print large zero initializers (e.g. for
1637 large arrays) and is always exactly equivalent to using explicit zero
1644 <!-- ======================================================================= -->
1645 <div class="doc_subsection">
1646 <a name="globalconstants">Global Variable and Function Addresses</a>
1649 <div class="doc_text">
1651 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1652 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1653 constants. These constants are explicitly referenced when the <a
1654 href="#identifiers">identifier for the global</a> is used and always have <a
1655 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1658 <div class="doc_code">
1662 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1668 <!-- ======================================================================= -->
1669 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1670 <div class="doc_text">
1671 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1672 no specific value. Undefined values may be of any type and be used anywhere
1673 a constant is permitted.</p>
1675 <p>Undefined values indicate to the compiler that the program is well defined
1676 no matter what value is used, giving the compiler more freedom to optimize.
1680 <!-- ======================================================================= -->
1681 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1684 <div class="doc_text">
1686 <p>Constant expressions are used to allow expressions involving other constants
1687 to be used as constants. Constant expressions may be of any <a
1688 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1689 that does not have side effects (e.g. load and call are not supported). The
1690 following is the syntax for constant expressions:</p>
1693 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1694 <dd>Truncate a constant to another type. The bit size of CST must be larger
1695 than the bit size of TYPE. Both types must be integers.</dd>
1697 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1698 <dd>Zero extend a constant to another type. The bit size of CST must be
1699 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1701 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1702 <dd>Sign extend a constant to another type. The bit size of CST must be
1703 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1705 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1706 <dd>Truncate a floating point constant to another floating point type. The
1707 size of CST must be larger than the size of TYPE. Both types must be
1708 floating point.</dd>
1710 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1711 <dd>Floating point extend a constant to another type. The size of CST must be
1712 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1714 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1715 <dd>Convert a floating point constant to the corresponding unsigned integer
1716 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1717 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1718 of the same number of elements. If the value won't fit in the integer type,
1719 the results are undefined.</dd>
1721 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1722 <dd>Convert a floating point constant to the corresponding signed integer
1723 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1724 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1725 of the same number of elements. If the value won't fit in the integer type,
1726 the results are undefined.</dd>
1728 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1729 <dd>Convert an unsigned integer constant to the corresponding floating point
1730 constant. TYPE must be a scalar or vector floating point type. CST must be of
1731 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1732 of the same number of elements. If the value won't fit in the floating point
1733 type, the results are undefined.</dd>
1735 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1736 <dd>Convert a signed integer constant to the corresponding floating point
1737 constant. TYPE must be a scalar or vector floating point type. CST must be of
1738 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1739 of the same number of elements. If the value won't fit in the floating point
1740 type, the results are undefined.</dd>
1742 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1743 <dd>Convert a pointer typed constant to the corresponding integer constant
1744 TYPE must be an integer type. CST must be of pointer type. The CST value is
1745 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1747 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1748 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1749 pointer type. CST must be of integer type. The CST value is zero extended,
1750 truncated, or unchanged to make it fit in a pointer size. This one is
1751 <i>really</i> dangerous!</dd>
1753 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1754 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1755 identical (same number of bits). The conversion is done as if the CST value
1756 was stored to memory and read back as TYPE. In other words, no bits change
1757 with this operator, just the type. This can be used for conversion of
1758 vector types to any other type, as long as they have the same bit width. For
1759 pointers it is only valid to cast to another pointer type. It is not valid
1760 to bitcast to or from an aggregate type.
1763 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1765 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1766 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1767 instruction, the index list may have zero or more indexes, which are required
1768 to make sense for the type of "CSTPTR".</dd>
1770 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1772 <dd>Perform the <a href="#i_select">select operation</a> on
1775 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1776 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1778 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1779 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1781 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1782 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1784 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1785 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1787 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1789 <dd>Perform the <a href="#i_extractelement">extractelement
1790 operation</a> on constants.
1792 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1794 <dd>Perform the <a href="#i_insertelement">insertelement
1795 operation</a> on constants.</dd>
1798 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1800 <dd>Perform the <a href="#i_shufflevector">shufflevector
1801 operation</a> on constants.</dd>
1803 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1805 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1806 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1807 binary</a> operations. The constraints on operands are the same as those for
1808 the corresponding instruction (e.g. no bitwise operations on floating point
1809 values are allowed).</dd>
1813 <!-- *********************************************************************** -->
1814 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1815 <!-- *********************************************************************** -->
1817 <!-- ======================================================================= -->
1818 <div class="doc_subsection">
1819 <a name="inlineasm">Inline Assembler Expressions</a>
1822 <div class="doc_text">
1825 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1826 Module-Level Inline Assembly</a>) through the use of a special value. This
1827 value represents the inline assembler as a string (containing the instructions
1828 to emit), a list of operand constraints (stored as a string), and a flag that
1829 indicates whether or not the inline asm expression has side effects. An example
1830 inline assembler expression is:
1833 <div class="doc_code">
1835 i32 (i32) asm "bswap $0", "=r,r"
1840 Inline assembler expressions may <b>only</b> be used as the callee operand of
1841 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1844 <div class="doc_code">
1846 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1851 Inline asms with side effects not visible in the constraint list must be marked
1852 as having side effects. This is done through the use of the
1853 '<tt>sideeffect</tt>' keyword, like so:
1856 <div class="doc_code">
1858 call void asm sideeffect "eieio", ""()
1862 <p>TODO: The format of the asm and constraints string still need to be
1863 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1864 need to be documented). This is probably best done by reference to another
1865 document that covers inline asm from a holistic perspective.
1870 <!-- *********************************************************************** -->
1871 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1872 <!-- *********************************************************************** -->
1874 <div class="doc_text">
1876 <p>The LLVM instruction set consists of several different
1877 classifications of instructions: <a href="#terminators">terminator
1878 instructions</a>, <a href="#binaryops">binary instructions</a>,
1879 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1880 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1881 instructions</a>.</p>
1885 <!-- ======================================================================= -->
1886 <div class="doc_subsection"> <a name="terminators">Terminator
1887 Instructions</a> </div>
1889 <div class="doc_text">
1891 <p>As mentioned <a href="#functionstructure">previously</a>, every
1892 basic block in a program ends with a "Terminator" instruction, which
1893 indicates which block should be executed after the current block is
1894 finished. These terminator instructions typically yield a '<tt>void</tt>'
1895 value: they produce control flow, not values (the one exception being
1896 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1897 <p>There are six different terminator instructions: the '<a
1898 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1899 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1900 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1901 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1902 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1906 <!-- _______________________________________________________________________ -->
1907 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1908 Instruction</a> </div>
1909 <div class="doc_text">
1912 ret <type> <value> <i>; Return a value from a non-void function</i>
1913 ret void <i>; Return from void function</i>
1918 <p>The '<tt>ret</tt>' instruction is used to return control flow (and
1919 optionally a value) from a function back to the caller.</p>
1920 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1921 returns a value and then causes control flow, and one that just causes
1922 control flow to occur.</p>
1926 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument,
1927 the return value. The type of the return value must be a
1928 '<a href="#t_firstclass">first class</a>' type.</p>
1930 <p>A function is not <a href="#wellformed">well formed</a> if
1931 it it has a non-void return type and contains a '<tt>ret</tt>'
1932 instruction with no return value or a return value with a type that
1933 does not match its type, or if it has a void return type and contains
1934 a '<tt>ret</tt>' instruction with a return value.</p>
1938 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1939 returns back to the calling function's context. If the caller is a "<a
1940 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1941 the instruction after the call. If the caller was an "<a
1942 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1943 at the beginning of the "normal" destination block. If the instruction
1944 returns a value, that value shall set the call or invoke instruction's
1950 ret i32 5 <i>; Return an integer value of 5</i>
1951 ret void <i>; Return from a void function</i>
1952 ret { i32, i8 } { i32 4, i8 2 } <i>; Return an aggregate of values 4 and 2</i>
1955 <!-- _______________________________________________________________________ -->
1956 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1957 <div class="doc_text">
1959 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1962 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1963 transfer to a different basic block in the current function. There are
1964 two forms of this instruction, corresponding to a conditional branch
1965 and an unconditional branch.</p>
1967 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1968 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1969 unconditional form of the '<tt>br</tt>' instruction takes a single
1970 '<tt>label</tt>' value as a target.</p>
1972 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1973 argument is evaluated. If the value is <tt>true</tt>, control flows
1974 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1975 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1977 <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
1978 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1980 <!-- _______________________________________________________________________ -->
1981 <div class="doc_subsubsection">
1982 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1985 <div class="doc_text">
1989 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1994 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1995 several different places. It is a generalization of the '<tt>br</tt>'
1996 instruction, allowing a branch to occur to one of many possible
2002 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2003 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
2004 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
2005 table is not allowed to contain duplicate constant entries.</p>
2009 <p>The <tt>switch</tt> instruction specifies a table of values and
2010 destinations. When the '<tt>switch</tt>' instruction is executed, this
2011 table is searched for the given value. If the value is found, control flow is
2012 transfered to the corresponding destination; otherwise, control flow is
2013 transfered to the default destination.</p>
2015 <h5>Implementation:</h5>
2017 <p>Depending on properties of the target machine and the particular
2018 <tt>switch</tt> instruction, this instruction may be code generated in different
2019 ways. For example, it could be generated as a series of chained conditional
2020 branches or with a lookup table.</p>
2025 <i>; Emulate a conditional br instruction</i>
2026 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2027 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
2029 <i>; Emulate an unconditional br instruction</i>
2030 switch i32 0, label %dest [ ]
2032 <i>; Implement a jump table:</i>
2033 switch i32 %val, label %otherwise [ i32 0, label %onzero
2035 i32 2, label %ontwo ]
2039 <!-- _______________________________________________________________________ -->
2040 <div class="doc_subsubsection">
2041 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2044 <div class="doc_text">
2049 <result> = invoke [<a href="#callingconv">cconv</a>] [<a href="#pa\
2050 ramattrs">RetAttrs</a>] <ptr to function ty> <function ptr val>(<function args>)
2051 to label <normal label> unwind label <exception label>
2056 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2057 function, with the possibility of control flow transfer to either the
2058 '<tt>normal</tt>' label or the
2059 '<tt>exception</tt>' label. If the callee function returns with the
2060 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
2061 "normal" label. If the callee (or any indirect callees) returns with the "<a
2062 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
2063 continued at the dynamically nearest "exception" label.
2067 <p>This instruction requires several arguments:</p>
2071 The optional "cconv" marker indicates which <a href="#callingconv">calling
2072 convention</a> the call should use. If none is specified, the call defaults
2073 to using C calling conventions.
2076 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2077 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
2078 and '<tt>inreg</tt>' attributes are valid here.</li>
2080 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2081 function value being invoked. In most cases, this is a direct function
2082 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2083 an arbitrary pointer to function value.
2086 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2087 function to be invoked. </li>
2089 <li>'<tt>function args</tt>': argument list whose types match the function
2090 signature argument types. If the function signature indicates the function
2091 accepts a variable number of arguments, the extra arguments can be
2094 <li>'<tt>normal label</tt>': the label reached when the called function
2095 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2097 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2098 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2100 <li>The optional <a href="fnattrs">function attributes</a> list. Only
2101 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2102 '<tt>readnone</tt>' attributes are valid here.</li>
2107 <p>This instruction is designed to operate as a standard '<tt><a
2108 href="#i_call">call</a></tt>' instruction in most regards. The primary
2109 difference is that it establishes an association with a label, which is used by
2110 the runtime library to unwind the stack.</p>
2112 <p>This instruction is used in languages with destructors to ensure that proper
2113 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2114 exception. Additionally, this is important for implementation of
2115 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2119 %retval = invoke i32 @Test(i32 15) to label %Continue
2120 unwind label %TestCleanup <i>; {i32}:retval set</i>
2121 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2122 unwind label %TestCleanup <i>; {i32}:retval set</i>
2127 <!-- _______________________________________________________________________ -->
2129 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2130 Instruction</a> </div>
2132 <div class="doc_text">
2141 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2142 at the first callee in the dynamic call stack which used an <a
2143 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2144 primarily used to implement exception handling.</p>
2148 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2149 immediately halt. The dynamic call stack is then searched for the first <a
2150 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2151 execution continues at the "exceptional" destination block specified by the
2152 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2153 dynamic call chain, undefined behavior results.</p>
2156 <!-- _______________________________________________________________________ -->
2158 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2159 Instruction</a> </div>
2161 <div class="doc_text">
2170 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2171 instruction is used to inform the optimizer that a particular portion of the
2172 code is not reachable. This can be used to indicate that the code after a
2173 no-return function cannot be reached, and other facts.</p>
2177 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2182 <!-- ======================================================================= -->
2183 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2184 <div class="doc_text">
2185 <p>Binary operators are used to do most of the computation in a
2186 program. They require two operands of the same type, execute an operation on them, and
2187 produce a single value. The operands might represent
2188 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2189 The result value has the same type as its operands.</p>
2190 <p>There are several different binary operators:</p>
2192 <!-- _______________________________________________________________________ -->
2193 <div class="doc_subsubsection">
2194 <a name="i_add">'<tt>add</tt>' Instruction</a>
2197 <div class="doc_text">
2202 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2207 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2211 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2212 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2213 <a href="#t_vector">vector</a> values. Both arguments must have identical
2218 <p>The value produced is the integer or floating point sum of the two
2221 <p>If an integer sum has unsigned overflow, the result returned is the
2222 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2225 <p>Because LLVM integers use a two's complement representation, this
2226 instruction is appropriate for both signed and unsigned integers.</p>
2231 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2234 <!-- _______________________________________________________________________ -->
2235 <div class="doc_subsubsection">
2236 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2239 <div class="doc_text">
2244 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2249 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2252 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2253 '<tt>neg</tt>' instruction present in most other intermediate
2254 representations.</p>
2258 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2259 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2260 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2265 <p>The value produced is the integer or floating point difference of
2266 the two operands.</p>
2268 <p>If an integer difference has unsigned overflow, the result returned is the
2269 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2272 <p>Because LLVM integers use a two's complement representation, this
2273 instruction is appropriate for both signed and unsigned integers.</p>
2277 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2278 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2282 <!-- _______________________________________________________________________ -->
2283 <div class="doc_subsubsection">
2284 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2287 <div class="doc_text">
2290 <pre> <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2293 <p>The '<tt>mul</tt>' instruction returns the product of its two
2298 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2299 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2300 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2305 <p>The value produced is the integer or floating point product of the
2308 <p>If the result of an integer multiplication has unsigned overflow,
2309 the result returned is the mathematical result modulo
2310 2<sup>n</sup>, where n is the bit width of the result.</p>
2311 <p>Because LLVM integers use a two's complement representation, and the
2312 result is the same width as the operands, this instruction returns the
2313 correct result for both signed and unsigned integers. If a full product
2314 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2315 should be sign-extended or zero-extended as appropriate to the
2316 width of the full product.</p>
2318 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2322 <!-- _______________________________________________________________________ -->
2323 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2325 <div class="doc_text">
2327 <pre> <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2330 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2335 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2336 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2337 values. Both arguments must have identical types.</p>
2341 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2342 <p>Note that unsigned integer division and signed integer division are distinct
2343 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2344 <p>Division by zero leads to undefined behavior.</p>
2346 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2349 <!-- _______________________________________________________________________ -->
2350 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2352 <div class="doc_text">
2355 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2360 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2365 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2366 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2367 values. Both arguments must have identical types.</p>
2370 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2371 <p>Note that signed integer division and unsigned integer division are distinct
2372 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2373 <p>Division by zero leads to undefined behavior. Overflow also leads to
2374 undefined behavior; this is a rare case, but can occur, for example,
2375 by doing a 32-bit division of -2147483648 by -1.</p>
2377 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2380 <!-- _______________________________________________________________________ -->
2381 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2382 Instruction</a> </div>
2383 <div class="doc_text">
2386 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2390 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2395 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2396 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2397 of floating point values. Both arguments must have identical types.</p>
2401 <p>The value produced is the floating point quotient of the two operands.</p>
2406 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2410 <!-- _______________________________________________________________________ -->
2411 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2413 <div class="doc_text">
2415 <pre> <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2418 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2419 unsigned division of its two arguments.</p>
2421 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2422 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2423 values. Both arguments must have identical types.</p>
2425 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2426 This instruction always performs an unsigned division to get the remainder.</p>
2427 <p>Note that unsigned integer remainder and signed integer remainder are
2428 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2429 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2431 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2435 <!-- _______________________________________________________________________ -->
2436 <div class="doc_subsubsection">
2437 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2440 <div class="doc_text">
2445 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2450 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2451 signed division of its two operands. This instruction can also take
2452 <a href="#t_vector">vector</a> versions of the values in which case
2453 the elements must be integers.</p>
2457 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2458 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2459 values. Both arguments must have identical types.</p>
2463 <p>This instruction returns the <i>remainder</i> of a division (where the result
2464 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2465 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
2466 a value. For more information about the difference, see <a
2467 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2468 Math Forum</a>. For a table of how this is implemented in various languages,
2469 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2470 Wikipedia: modulo operation</a>.</p>
2471 <p>Note that signed integer remainder and unsigned integer remainder are
2472 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2473 <p>Taking the remainder of a division by zero leads to undefined behavior.
2474 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2475 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2476 (The remainder doesn't actually overflow, but this rule lets srem be
2477 implemented using instructions that return both the result of the division
2478 and the remainder.)</p>
2480 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2484 <!-- _______________________________________________________________________ -->
2485 <div class="doc_subsubsection">
2486 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2488 <div class="doc_text">
2491 <pre> <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2494 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2495 division of its two operands.</p>
2497 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2498 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2499 of floating point values. Both arguments must have identical types.</p>
2503 <p>This instruction returns the <i>remainder</i> of a division.
2504 The remainder has the same sign as the dividend.</p>
2509 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2513 <!-- ======================================================================= -->
2514 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2515 Operations</a> </div>
2516 <div class="doc_text">
2517 <p>Bitwise binary operators are used to do various forms of
2518 bit-twiddling in a program. They are generally very efficient
2519 instructions and can commonly be strength reduced from other
2520 instructions. They require two operands of the same type, execute an operation on them,
2521 and produce a single value. The resulting value is the same type as its operands.</p>
2524 <!-- _______________________________________________________________________ -->
2525 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2526 Instruction</a> </div>
2527 <div class="doc_text">
2529 <pre> <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2534 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2535 the left a specified number of bits.</p>
2539 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2540 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2541 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2545 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 2<sup>n</sup>,
2546 where n is the width of the result. If <tt>op2</tt> is (statically or dynamically) negative or
2547 equal to or larger than the number of bits in <tt>op1</tt>, the result is undefined.</p>
2549 <h5>Example:</h5><pre>
2550 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2551 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2552 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2553 <result> = shl i32 1, 32 <i>; undefined</i>
2556 <!-- _______________________________________________________________________ -->
2557 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2558 Instruction</a> </div>
2559 <div class="doc_text">
2561 <pre> <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2565 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2566 operand shifted to the right a specified number of bits with zero fill.</p>
2569 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2570 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2571 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2575 <p>This instruction always performs a logical shift right operation. The most
2576 significant bits of the result will be filled with zero bits after the
2577 shift. If <tt>op2</tt> is (statically or dynamically) equal to or larger than
2578 the number of bits in <tt>op1</tt>, the result is undefined.</p>
2582 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2583 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2584 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2585 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2586 <result> = lshr i32 1, 32 <i>; undefined</i>
2590 <!-- _______________________________________________________________________ -->
2591 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2592 Instruction</a> </div>
2593 <div class="doc_text">
2596 <pre> <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2600 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2601 operand shifted to the right a specified number of bits with sign extension.</p>
2604 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2605 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2606 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2609 <p>This instruction always performs an arithmetic shift right operation,
2610 The most significant bits of the result will be filled with the sign bit
2611 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
2612 larger than the number of bits in <tt>op1</tt>, the result is undefined.
2617 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2618 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2619 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2620 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2621 <result> = ashr i32 1, 32 <i>; undefined</i>
2625 <!-- _______________________________________________________________________ -->
2626 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2627 Instruction</a> </div>
2629 <div class="doc_text">
2634 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2639 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2640 its two operands.</p>
2644 <p>The two arguments to the '<tt>and</tt>' instruction must be
2645 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2646 values. Both arguments must have identical types.</p>
2649 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2652 <table border="1" cellspacing="0" cellpadding="4">
2684 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2685 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2686 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2689 <!-- _______________________________________________________________________ -->
2690 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2691 <div class="doc_text">
2693 <pre> <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2696 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2697 or of its two operands.</p>
2700 <p>The two arguments to the '<tt>or</tt>' instruction must be
2701 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2702 values. Both arguments must have identical types.</p>
2704 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2707 <table border="1" cellspacing="0" cellpadding="4">
2738 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2739 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2740 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2743 <!-- _______________________________________________________________________ -->
2744 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2745 Instruction</a> </div>
2746 <div class="doc_text">
2748 <pre> <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2751 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2752 or of its two operands. The <tt>xor</tt> is used to implement the
2753 "one's complement" operation, which is the "~" operator in C.</p>
2755 <p>The two arguments to the '<tt>xor</tt>' instruction must be
2756 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2757 values. Both arguments must have identical types.</p>
2761 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2764 <table border="1" cellspacing="0" cellpadding="4">
2796 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2797 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2798 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2799 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2803 <!-- ======================================================================= -->
2804 <div class="doc_subsection">
2805 <a name="vectorops">Vector Operations</a>
2808 <div class="doc_text">
2810 <p>LLVM supports several instructions to represent vector operations in a
2811 target-independent manner. These instructions cover the element-access and
2812 vector-specific operations needed to process vectors effectively. While LLVM
2813 does directly support these vector operations, many sophisticated algorithms
2814 will want to use target-specific intrinsics to take full advantage of a specific
2819 <!-- _______________________________________________________________________ -->
2820 <div class="doc_subsubsection">
2821 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2824 <div class="doc_text">
2829 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2835 The '<tt>extractelement</tt>' instruction extracts a single scalar
2836 element from a vector at a specified index.
2843 The first operand of an '<tt>extractelement</tt>' instruction is a
2844 value of <a href="#t_vector">vector</a> type. The second operand is
2845 an index indicating the position from which to extract the element.
2846 The index may be a variable.</p>
2851 The result is a scalar of the same type as the element type of
2852 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2853 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2854 results are undefined.
2860 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2865 <!-- _______________________________________________________________________ -->
2866 <div class="doc_subsubsection">
2867 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2870 <div class="doc_text">
2875 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2881 The '<tt>insertelement</tt>' instruction inserts a scalar
2882 element into a vector at a specified index.
2889 The first operand of an '<tt>insertelement</tt>' instruction is a
2890 value of <a href="#t_vector">vector</a> type. The second operand is a
2891 scalar value whose type must equal the element type of the first
2892 operand. The third operand is an index indicating the position at
2893 which to insert the value. The index may be a variable.</p>
2898 The result is a vector of the same type as <tt>val</tt>. Its
2899 element values are those of <tt>val</tt> except at position
2900 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2901 exceeds the length of <tt>val</tt>, the results are undefined.
2907 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2911 <!-- _______________________________________________________________________ -->
2912 <div class="doc_subsubsection">
2913 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2916 <div class="doc_text">
2921 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2927 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2928 from two input vectors, returning a vector of the same type.
2934 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2935 with types that match each other and types that match the result of the
2936 instruction. The third argument is a shuffle mask, which has the same number
2937 of elements as the other vector type, but whose element type is always 'i32'.
2941 The shuffle mask operand is required to be a constant vector with either
2942 constant integer or undef values.
2948 The elements of the two input vectors are numbered from left to right across
2949 both of the vectors. The shuffle mask operand specifies, for each element of
2950 the result vector, which element of the two input registers the result element
2951 gets. The element selector may be undef (meaning "don't care") and the second
2952 operand may be undef if performing a shuffle from only one vector.
2958 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2959 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2960 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2961 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2966 <!-- ======================================================================= -->
2967 <div class="doc_subsection">
2968 <a name="aggregateops">Aggregate Operations</a>
2971 <div class="doc_text">
2973 <p>LLVM supports several instructions for working with aggregate values.
2978 <!-- _______________________________________________________________________ -->
2979 <div class="doc_subsubsection">
2980 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
2983 <div class="doc_text">
2988 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
2994 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
2995 or array element from an aggregate value.
3002 The first operand of an '<tt>extractvalue</tt>' instruction is a
3003 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
3004 type. The operands are constant indices to specify which value to extract
3005 in a similar manner as indices in a
3006 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3012 The result is the value at the position in the aggregate specified by
3019 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3024 <!-- _______________________________________________________________________ -->
3025 <div class="doc_subsubsection">
3026 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3029 <div class="doc_text">
3034 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3040 The '<tt>insertvalue</tt>' instruction inserts a value
3041 into a struct field or array element in an aggregate.
3048 The first operand of an '<tt>insertvalue</tt>' instruction is a
3049 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
3050 The second operand is a first-class value to insert.
3051 The following operands are constant indices
3052 indicating the position at which to insert the value in a similar manner as
3054 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3055 The value to insert must have the same type as the value identified
3061 The result is an aggregate of the same type as <tt>val</tt>. Its
3062 value is that of <tt>val</tt> except that the value at the position
3063 specified by the indices is that of <tt>elt</tt>.
3069 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3074 <!-- ======================================================================= -->
3075 <div class="doc_subsection">
3076 <a name="memoryops">Memory Access and Addressing Operations</a>
3079 <div class="doc_text">
3081 <p>A key design point of an SSA-based representation is how it
3082 represents memory. In LLVM, no memory locations are in SSA form, which
3083 makes things very simple. This section describes how to read, write,
3084 allocate, and free memory in LLVM.</p>
3088 <!-- _______________________________________________________________________ -->
3089 <div class="doc_subsubsection">
3090 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3093 <div class="doc_text">
3098 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3103 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3104 heap and returns a pointer to it. The object is always allocated in the generic
3105 address space (address space zero).</p>
3109 <p>The '<tt>malloc</tt>' instruction allocates
3110 <tt>sizeof(<type>)*NumElements</tt>
3111 bytes of memory from the operating system and returns a pointer of the
3112 appropriate type to the program. If "NumElements" is specified, it is the
3113 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3114 If a constant alignment is specified, the value result of the allocation is guaranteed to
3115 be aligned to at least that boundary. If not specified, or if zero, the target can
3116 choose to align the allocation on any convenient boundary.</p>
3118 <p>'<tt>type</tt>' must be a sized type.</p>
3122 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3123 a pointer is returned. The result of a zero byte allocattion is undefined. The
3124 result is null if there is insufficient memory available.</p>
3129 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
3131 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3132 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3133 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3134 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3135 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3139 <!-- _______________________________________________________________________ -->
3140 <div class="doc_subsubsection">
3141 <a name="i_free">'<tt>free</tt>' Instruction</a>
3144 <div class="doc_text">
3149 free <type> <value> <i>; yields {void}</i>
3154 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3155 memory heap to be reallocated in the future.</p>
3159 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3160 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3165 <p>Access to the memory pointed to by the pointer is no longer defined
3166 after this instruction executes. If the pointer is null, the operation
3172 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3173 free [4 x i8]* %array
3177 <!-- _______________________________________________________________________ -->
3178 <div class="doc_subsubsection">
3179 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3182 <div class="doc_text">
3187 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3192 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3193 currently executing function, to be automatically released when this function
3194 returns to its caller. The object is always allocated in the generic address
3195 space (address space zero).</p>
3199 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3200 bytes of memory on the runtime stack, returning a pointer of the
3201 appropriate type to the program. If "NumElements" is specified, it is the
3202 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3203 If a constant alignment is specified, the value result of the allocation is guaranteed
3204 to be aligned to at least that boundary. If not specified, or if zero, the target
3205 can choose to align the allocation on any convenient boundary.</p>
3207 <p>'<tt>type</tt>' may be any sized type.</p>
3211 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3212 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3213 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3214 instruction is commonly used to represent automatic variables that must
3215 have an address available. When the function returns (either with the <tt><a
3216 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3217 instructions), the memory is reclaimed. Allocating zero bytes
3218 is legal, but the result is undefined.</p>
3223 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3224 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3225 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3226 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3230 <!-- _______________________________________________________________________ -->
3231 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3232 Instruction</a> </div>
3233 <div class="doc_text">
3235 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3237 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3239 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3240 address from which to load. The pointer must point to a <a
3241 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3242 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3243 the number or order of execution of this <tt>load</tt> with other
3244 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3247 The optional constant "align" argument specifies the alignment of the operation
3248 (that is, the alignment of the memory address). A value of 0 or an
3249 omitted "align" argument means that the operation has the preferential
3250 alignment for the target. It is the responsibility of the code emitter
3251 to ensure that the alignment information is correct. Overestimating
3252 the alignment results in an undefined behavior. Underestimating the
3253 alignment may produce less efficient code. An alignment of 1 is always
3257 <p>The location of memory pointed to is loaded.</p>
3259 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3261 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3262 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3265 <!-- _______________________________________________________________________ -->
3266 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3267 Instruction</a> </div>
3268 <div class="doc_text">
3270 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3271 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3274 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3276 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3277 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3278 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3279 of the '<tt><value></tt>'
3280 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3281 optimizer is not allowed to modify the number or order of execution of
3282 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3283 href="#i_store">store</a></tt> instructions.</p>
3285 The optional constant "align" argument specifies the alignment of the operation
3286 (that is, the alignment of the memory address). A value of 0 or an
3287 omitted "align" argument means that the operation has the preferential
3288 alignment for the target. It is the responsibility of the code emitter
3289 to ensure that the alignment information is correct. Overestimating
3290 the alignment results in an undefined behavior. Underestimating the
3291 alignment may produce less efficient code. An alignment of 1 is always
3295 <p>The contents of memory are updated to contain '<tt><value></tt>'
3296 at the location specified by the '<tt><pointer></tt>' operand.</p>
3298 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3299 store i32 3, i32* %ptr <i>; yields {void}</i>
3300 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3304 <!-- _______________________________________________________________________ -->
3305 <div class="doc_subsubsection">
3306 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3309 <div class="doc_text">
3312 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
3318 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3319 subelement of an aggregate data structure.</p>
3323 <p>This instruction takes a list of integer operands that indicate what
3324 elements of the aggregate object to index to. The actual types of the arguments
3325 provided depend on the type of the first pointer argument. The
3326 '<tt>getelementptr</tt>' instruction is used to index down through the type
3327 levels of a structure or to a specific index in an array. When indexing into a
3328 structure, only <tt>i32</tt> integer constants are allowed. When indexing
3329 into an array or pointer, only integers of 32 or 64 bits are allowed; 32-bit
3330 values will be sign extended to 64-bits if required.</p>
3332 <p>For example, let's consider a C code fragment and how it gets
3333 compiled to LLVM:</p>
3335 <div class="doc_code">
3348 int *foo(struct ST *s) {
3349 return &s[1].Z.B[5][13];
3354 <p>The LLVM code generated by the GCC frontend is:</p>
3356 <div class="doc_code">
3358 %RT = type { i8 , [10 x [20 x i32]], i8 }
3359 %ST = type { i32, double, %RT }
3361 define i32* %foo(%ST* %s) {
3363 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3371 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3372 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3373 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3374 <a href="#t_integer">integer</a> type but the value will always be sign extended
3375 to 64-bits. <a href="#t_struct">Structure</a> and <a href="#t_pstruct">packed
3376 structure</a> types require <tt>i32</tt> <b>constants</b>.</p>
3378 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3379 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3380 }</tt>' type, a structure. The second index indexes into the third element of
3381 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3382 i8 }</tt>' type, another structure. The third index indexes into the second
3383 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3384 array. The two dimensions of the array are subscripted into, yielding an
3385 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3386 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3388 <p>Note that it is perfectly legal to index partially through a
3389 structure, returning a pointer to an inner element. Because of this,
3390 the LLVM code for the given testcase is equivalent to:</p>
3393 define i32* %foo(%ST* %s) {
3394 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3395 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3396 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3397 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3398 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3403 <p>Note that it is undefined to access an array out of bounds: array and
3404 pointer indexes must always be within the defined bounds of the array type.
3405 The one exception for this rule is zero length arrays. These arrays are
3406 defined to be accessible as variable length arrays, which requires access
3407 beyond the zero'th element.</p>
3409 <p>The getelementptr instruction is often confusing. For some more insight
3410 into how it works, see <a href="GetElementPtr.html">the getelementptr
3416 <i>; yields [12 x i8]*:aptr</i>
3417 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3421 <!-- ======================================================================= -->
3422 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3424 <div class="doc_text">
3425 <p>The instructions in this category are the conversion instructions (casting)
3426 which all take a single operand and a type. They perform various bit conversions
3430 <!-- _______________________________________________________________________ -->
3431 <div class="doc_subsubsection">
3432 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3434 <div class="doc_text">
3438 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3443 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3448 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3449 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3450 and type of the result, which must be an <a href="#t_integer">integer</a>
3451 type. The bit size of <tt>value</tt> must be larger than the bit size of
3452 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3456 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3457 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3458 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3459 It will always truncate bits.</p>
3463 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3464 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3465 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3469 <!-- _______________________________________________________________________ -->
3470 <div class="doc_subsubsection">
3471 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3473 <div class="doc_text">
3477 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3481 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3486 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3487 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3488 also be of <a href="#t_integer">integer</a> type. The bit size of the
3489 <tt>value</tt> must be smaller than the bit size of the destination type,
3493 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3494 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3496 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3500 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3501 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3505 <!-- _______________________________________________________________________ -->
3506 <div class="doc_subsubsection">
3507 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3509 <div class="doc_text">
3513 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3517 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3521 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3522 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3523 also be of <a href="#t_integer">integer</a> type. The bit size of the
3524 <tt>value</tt> must be smaller than the bit size of the destination type,
3529 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3530 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3531 the type <tt>ty2</tt>.</p>
3533 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3537 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3538 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3542 <!-- _______________________________________________________________________ -->
3543 <div class="doc_subsubsection">
3544 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3547 <div class="doc_text">
3552 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3556 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3561 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3562 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3563 cast it to. The size of <tt>value</tt> must be larger than the size of
3564 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3565 <i>no-op cast</i>.</p>
3568 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3569 <a href="#t_floating">floating point</a> type to a smaller
3570 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3571 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3575 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3576 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3580 <!-- _______________________________________________________________________ -->
3581 <div class="doc_subsubsection">
3582 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3584 <div class="doc_text">
3588 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3592 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3593 floating point value.</p>
3596 <p>The '<tt>fpext</tt>' instruction takes a
3597 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3598 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3599 type must be smaller than the destination type.</p>
3602 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3603 <a href="#t_floating">floating point</a> type to a larger
3604 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3605 used to make a <i>no-op cast</i> because it always changes bits. Use
3606 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3610 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3611 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3615 <!-- _______________________________________________________________________ -->
3616 <div class="doc_subsubsection">
3617 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3619 <div class="doc_text">
3623 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3627 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3628 unsigned integer equivalent of type <tt>ty2</tt>.
3632 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3633 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3634 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3635 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3636 vector integer type with the same number of elements as <tt>ty</tt></p>
3639 <p> The '<tt>fptoui</tt>' instruction converts its
3640 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3641 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3642 the results are undefined.</p>
3646 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3647 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3648 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3652 <!-- _______________________________________________________________________ -->
3653 <div class="doc_subsubsection">
3654 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3656 <div class="doc_text">
3660 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3664 <p>The '<tt>fptosi</tt>' instruction converts
3665 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3669 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3670 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3671 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3672 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3673 vector integer type with the same number of elements as <tt>ty</tt></p>
3676 <p>The '<tt>fptosi</tt>' instruction converts its
3677 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3678 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3679 the results are undefined.</p>
3683 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3684 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3685 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3689 <!-- _______________________________________________________________________ -->
3690 <div class="doc_subsubsection">
3691 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3693 <div class="doc_text">
3697 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3701 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3702 integer and converts that value to the <tt>ty2</tt> type.</p>
3705 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3706 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3707 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3708 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3709 floating point type with the same number of elements as <tt>ty</tt></p>
3712 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3713 integer quantity and converts it to the corresponding floating point value. If
3714 the value cannot fit in the floating point value, the results are undefined.</p>
3718 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3719 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3723 <!-- _______________________________________________________________________ -->
3724 <div class="doc_subsubsection">
3725 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3727 <div class="doc_text">
3731 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3735 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3736 integer and converts that value to the <tt>ty2</tt> type.</p>
3739 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3740 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3741 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3742 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3743 floating point type with the same number of elements as <tt>ty</tt></p>
3746 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3747 integer quantity and converts it to the corresponding floating point value. If
3748 the value cannot fit in the floating point value, the results are undefined.</p>
3752 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3753 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3757 <!-- _______________________________________________________________________ -->
3758 <div class="doc_subsubsection">
3759 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3761 <div class="doc_text">
3765 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3769 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3770 the integer type <tt>ty2</tt>.</p>
3773 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3774 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3775 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3778 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3779 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3780 truncating or zero extending that value to the size of the integer type. If
3781 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3782 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3783 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3788 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3789 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3793 <!-- _______________________________________________________________________ -->
3794 <div class="doc_subsubsection">
3795 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3797 <div class="doc_text">
3801 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3805 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3806 a pointer type, <tt>ty2</tt>.</p>
3809 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3810 value to cast, and a type to cast it to, which must be a
3811 <a href="#t_pointer">pointer</a> type.
3814 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3815 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3816 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3817 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3818 the size of a pointer then a zero extension is done. If they are the same size,
3819 nothing is done (<i>no-op cast</i>).</p>
3823 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3824 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3825 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3829 <!-- _______________________________________________________________________ -->
3830 <div class="doc_subsubsection">
3831 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3833 <div class="doc_text">
3837 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3842 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3843 <tt>ty2</tt> without changing any bits.</p>
3847 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3848 a non-aggregate first class value, and a type to cast it to, which must also be
3849 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes of
3851 and the destination type, <tt>ty2</tt>, must be identical. If the source
3852 type is a pointer, the destination type must also be a pointer. This
3853 instruction supports bitwise conversion of vectors to integers and to vectors
3854 of other types (as long as they have the same size).</p>
3857 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3858 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3859 this conversion. The conversion is done as if the <tt>value</tt> had been
3860 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3861 converted to other pointer types with this instruction. To convert pointers to
3862 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3863 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3867 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3868 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3869 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3873 <!-- ======================================================================= -->
3874 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3875 <div class="doc_text">
3876 <p>The instructions in this category are the "miscellaneous"
3877 instructions, which defy better classification.</p>
3880 <!-- _______________________________________________________________________ -->
3881 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3883 <div class="doc_text">
3885 <pre> <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
3888 <p>The '<tt>icmp</tt>' instruction returns a boolean value or
3889 a vector of boolean values based on comparison
3890 of its two integer, integer vector, or pointer operands.</p>
3892 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3893 the condition code indicating the kind of comparison to perform. It is not
3894 a value, just a keyword. The possible condition code are:
3896 <li><tt>eq</tt>: equal</li>
3897 <li><tt>ne</tt>: not equal </li>
3898 <li><tt>ugt</tt>: unsigned greater than</li>
3899 <li><tt>uge</tt>: unsigned greater or equal</li>
3900 <li><tt>ult</tt>: unsigned less than</li>
3901 <li><tt>ule</tt>: unsigned less or equal</li>
3902 <li><tt>sgt</tt>: signed greater than</li>
3903 <li><tt>sge</tt>: signed greater or equal</li>
3904 <li><tt>slt</tt>: signed less than</li>
3905 <li><tt>sle</tt>: signed less or equal</li>
3907 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3908 <a href="#t_pointer">pointer</a>
3909 or integer <a href="#t_vector">vector</a> typed.
3910 They must also be identical types.</p>
3912 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to
3913 the condition code given as <tt>cond</tt>. The comparison performed always
3914 yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
3916 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3917 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3919 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3920 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3921 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3922 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3923 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3924 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3925 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3926 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
3927 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3928 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3929 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3930 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3931 <li><tt>sge</tt>: interprets the operands as signed values and yields
3932 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3933 <li><tt>slt</tt>: interprets the operands as signed values and yields
3934 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
3935 <li><tt>sle</tt>: interprets the operands as signed values and yields
3936 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3938 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3939 values are compared as if they were integers.</p>
3940 <p>If the operands are integer vectors, then they are compared
3941 element by element. The result is an <tt>i1</tt> vector with
3942 the same number of elements as the values being compared.
3943 Otherwise, the result is an <tt>i1</tt>.
3947 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3948 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3949 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3950 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3951 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3952 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3956 <!-- _______________________________________________________________________ -->
3957 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3959 <div class="doc_text">
3961 <pre> <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
3964 <p>The '<tt>fcmp</tt>' instruction returns a boolean value
3965 or vector of boolean values based on comparison
3968 If the operands are floating point scalars, then the result
3969 type is a boolean (<a href="#t_primitive"><tt>i1</tt></a>).
3971 <p>If the operands are floating point vectors, then the result type
3972 is a vector of boolean with the same number of elements as the
3973 operands being compared.</p>
3975 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3976 the condition code indicating the kind of comparison to perform. It is not
3977 a value, just a keyword. The possible condition code are:
3979 <li><tt>false</tt>: no comparison, always returns false</li>
3980 <li><tt>oeq</tt>: ordered and equal</li>
3981 <li><tt>ogt</tt>: ordered and greater than </li>
3982 <li><tt>oge</tt>: ordered and greater than or equal</li>
3983 <li><tt>olt</tt>: ordered and less than </li>
3984 <li><tt>ole</tt>: ordered and less than or equal</li>
3985 <li><tt>one</tt>: ordered and not equal</li>
3986 <li><tt>ord</tt>: ordered (no nans)</li>
3987 <li><tt>ueq</tt>: unordered or equal</li>
3988 <li><tt>ugt</tt>: unordered or greater than </li>
3989 <li><tt>uge</tt>: unordered or greater than or equal</li>
3990 <li><tt>ult</tt>: unordered or less than </li>
3991 <li><tt>ule</tt>: unordered or less than or equal</li>
3992 <li><tt>une</tt>: unordered or not equal</li>
3993 <li><tt>uno</tt>: unordered (either nans)</li>
3994 <li><tt>true</tt>: no comparison, always returns true</li>
3996 <p><i>Ordered</i> means that neither operand is a QNAN while
3997 <i>unordered</i> means that either operand may be a QNAN.</p>
3998 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be
3999 either a <a href="#t_floating">floating point</a> type
4000 or a <a href="#t_vector">vector</a> of floating point type.
4001 They must have identical types.</p>
4003 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4004 according to the condition code given as <tt>cond</tt>.
4005 If the operands are vectors, then the vectors are compared
4007 Each comparison performed
4008 always yields an <a href="#t_primitive">i1</a> result, as follows:
4010 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4011 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4012 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4013 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4014 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4015 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4016 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4017 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4018 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4019 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4020 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4021 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4022 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4023 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4024 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4025 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4026 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4027 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4028 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4029 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4030 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4031 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4032 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4033 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4034 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4035 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4036 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4037 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4041 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4042 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4043 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4044 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4048 <!-- _______________________________________________________________________ -->
4049 <div class="doc_subsubsection">
4050 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
4052 <div class="doc_text">
4054 <pre> <result> = vicmp <cond> <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4057 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
4058 element-wise comparison of its two integer vector operands.</p>
4060 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
4061 the condition code indicating the kind of comparison to perform. It is not
4062 a value, just a keyword. The possible condition code are:
4064 <li><tt>eq</tt>: equal</li>
4065 <li><tt>ne</tt>: not equal </li>
4066 <li><tt>ugt</tt>: unsigned greater than</li>
4067 <li><tt>uge</tt>: unsigned greater or equal</li>
4068 <li><tt>ult</tt>: unsigned less than</li>
4069 <li><tt>ule</tt>: unsigned less or equal</li>
4070 <li><tt>sgt</tt>: signed greater than</li>
4071 <li><tt>sge</tt>: signed greater or equal</li>
4072 <li><tt>slt</tt>: signed less than</li>
4073 <li><tt>sle</tt>: signed less or equal</li>
4075 <p>The remaining two arguments must be <a href="#t_vector">vector</a> or
4076 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
4078 <p>The '<tt>vicmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4079 according to the condition code given as <tt>cond</tt>. The comparison yields a
4080 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
4081 identical type as the values being compared. The most significant bit in each
4082 element is 1 if the element-wise comparison evaluates to true, and is 0
4083 otherwise. All other bits of the result are undefined. The condition codes
4084 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
4089 <result> = vicmp eq <2 x i32> < i32 4, i32 0>, < i32 5, i32 0> <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4090 <result> = vicmp ult <2 x i8 > < i8 1, i8 2>, < i8 2, i8 2 > <i>; yields: result=<2 x i8> < i8 -1, i8 0 ></i>
4094 <!-- _______________________________________________________________________ -->
4095 <div class="doc_subsubsection">
4096 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
4098 <div class="doc_text">
4100 <pre> <result> = vfcmp <cond> <ty> <op1>, <op2></pre>
4102 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4103 element-wise comparison of its two floating point vector operands. The output
4104 elements have the same width as the input elements.</p>
4106 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4107 the condition code indicating the kind of comparison to perform. It is not
4108 a value, just a keyword. The possible condition code are:
4110 <li><tt>false</tt>: no comparison, always returns false</li>
4111 <li><tt>oeq</tt>: ordered and equal</li>
4112 <li><tt>ogt</tt>: ordered and greater than </li>
4113 <li><tt>oge</tt>: ordered and greater than or equal</li>
4114 <li><tt>olt</tt>: ordered and less than </li>
4115 <li><tt>ole</tt>: ordered and less than or equal</li>
4116 <li><tt>one</tt>: ordered and not equal</li>
4117 <li><tt>ord</tt>: ordered (no nans)</li>
4118 <li><tt>ueq</tt>: unordered or equal</li>
4119 <li><tt>ugt</tt>: unordered or greater than </li>
4120 <li><tt>uge</tt>: unordered or greater than or equal</li>
4121 <li><tt>ult</tt>: unordered or less than </li>
4122 <li><tt>ule</tt>: unordered or less than or equal</li>
4123 <li><tt>une</tt>: unordered or not equal</li>
4124 <li><tt>uno</tt>: unordered (either nans)</li>
4125 <li><tt>true</tt>: no comparison, always returns true</li>
4127 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4128 <a href="#t_floating">floating point</a> typed. They must also be identical
4131 <p>The '<tt>vfcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4132 according to the condition code given as <tt>cond</tt>. The comparison yields a
4133 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4134 an identical number of elements as the values being compared, and each element
4135 having identical with to the width of the floating point elements. The most
4136 significant bit in each element is 1 if the element-wise comparison evaluates to
4137 true, and is 0 otherwise. All other bits of the result are undefined. The
4138 condition codes are evaluated identically to the
4139 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.
4143 <result> = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 > <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4144 <result> = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2> <i>; yields: result=<2 x i64> < i64 -1, i64 0 ></i>
4148 <!-- _______________________________________________________________________ -->
4149 <div class="doc_subsubsection">
4150 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4153 <div class="doc_text">
4157 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4159 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4160 the SSA graph representing the function.</p>
4163 <p>The type of the incoming values is specified with the first type
4164 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4165 as arguments, with one pair for each predecessor basic block of the
4166 current block. Only values of <a href="#t_firstclass">first class</a>
4167 type may be used as the value arguments to the PHI node. Only labels
4168 may be used as the label arguments.</p>
4170 <p>There must be no non-phi instructions between the start of a basic
4171 block and the PHI instructions: i.e. PHI instructions must be first in
4176 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4177 specified by the pair corresponding to the predecessor basic block that executed
4178 just prior to the current block.</p>
4182 Loop: ; Infinite loop that counts from 0 on up...
4183 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4184 %nextindvar = add i32 %indvar, 1
4189 <!-- _______________________________________________________________________ -->
4190 <div class="doc_subsubsection">
4191 <a name="i_select">'<tt>select</tt>' Instruction</a>
4194 <div class="doc_text">
4199 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4201 <i>selty</i> is either i1 or {<N x i1>}
4207 The '<tt>select</tt>' instruction is used to choose one value based on a
4208 condition, without branching.
4215 The '<tt>select</tt>' instruction requires an 'i1' value or
4216 a vector of 'i1' values indicating the
4217 condition, and two values of the same <a href="#t_firstclass">first class</a>
4218 type. If the val1/val2 are vectors and
4219 the condition is a scalar, then entire vectors are selected, not
4220 individual elements.
4226 If the condition is an i1 and it evaluates to 1, the instruction returns the first
4227 value argument; otherwise, it returns the second value argument.
4230 If the condition is a vector of i1, then the value arguments must
4231 be vectors of the same size, and the selection is done element
4238 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4243 <!-- _______________________________________________________________________ -->
4244 <div class="doc_subsubsection">
4245 <a name="i_call">'<tt>call</tt>' Instruction</a>
4248 <div class="doc_text">
4252 <result> = [tail] call [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">RetAttrs</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
4257 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4261 <p>This instruction requires several arguments:</p>
4265 <p>The optional "tail" marker indicates whether the callee function accesses
4266 any allocas or varargs in the caller. If the "tail" marker is present, the
4267 function call is eligible for tail call optimization. Note that calls may
4268 be marked "tail" even if they do not occur before a <a
4269 href="#i_ret"><tt>ret</tt></a> instruction.
4272 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4273 convention</a> the call should use. If none is specified, the call defaults
4274 to using C calling conventions.
4278 <p>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4279 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
4280 and '<tt>inreg</tt>' attributes are valid here.</p>
4284 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4285 the type of the return value. Functions that return no value are marked
4286 <tt><a href="#t_void">void</a></tt>.</p>
4289 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4290 value being invoked. The argument types must match the types implied by
4291 this signature. This type can be omitted if the function is not varargs
4292 and if the function type does not return a pointer to a function.</p>
4295 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4296 be invoked. In most cases, this is a direct function invocation, but
4297 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4298 to function value.</p>
4301 <p>'<tt>function args</tt>': argument list whose types match the
4302 function signature argument types. All arguments must be of
4303 <a href="#t_firstclass">first class</a> type. If the function signature
4304 indicates the function accepts a variable number of arguments, the extra
4305 arguments can be specified.</p>
4308 <p>The optional <a href="fnattrs">function attributes</a> list. Only
4309 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4310 '<tt>readnone</tt>' attributes are valid here.</p>
4316 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4317 transfer to a specified function, with its incoming arguments bound to
4318 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4319 instruction in the called function, control flow continues with the
4320 instruction after the function call, and the return value of the
4321 function is bound to the result argument.
4326 %retval = call i32 @test(i32 %argc)
4327 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4328 %X = tail call i32 @foo() <i>; yields i32</i>
4329 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4330 call void %foo(i8 97 signext)
4332 %struct.A = type { i32, i8 }
4333 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4334 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4335 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
4336 %Z = call void @foo() noreturn <i>; indicates that foo never returns nomrally
4337 %ZZ = call zeroext i32 @bar() <i>; Return value is zero extended
4342 <!-- _______________________________________________________________________ -->
4343 <div class="doc_subsubsection">
4344 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4347 <div class="doc_text">
4352 <resultval> = va_arg <va_list*> <arglist>, <argty>
4357 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4358 the "variable argument" area of a function call. It is used to implement the
4359 <tt>va_arg</tt> macro in C.</p>
4363 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4364 the argument. It returns a value of the specified argument type and
4365 increments the <tt>va_list</tt> to point to the next argument. The
4366 actual type of <tt>va_list</tt> is target specific.</p>
4370 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4371 type from the specified <tt>va_list</tt> and causes the
4372 <tt>va_list</tt> to point to the next argument. For more information,
4373 see the variable argument handling <a href="#int_varargs">Intrinsic
4376 <p>It is legal for this instruction to be called in a function which does not
4377 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4380 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4381 href="#intrinsics">intrinsic function</a> because it takes a type as an
4386 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4390 <!-- *********************************************************************** -->
4391 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4392 <!-- *********************************************************************** -->
4394 <div class="doc_text">
4396 <p>LLVM supports the notion of an "intrinsic function". These functions have
4397 well known names and semantics and are required to follow certain restrictions.
4398 Overall, these intrinsics represent an extension mechanism for the LLVM
4399 language that does not require changing all of the transformations in LLVM when
4400 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4402 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4403 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4404 begin with this prefix. Intrinsic functions must always be external functions:
4405 you cannot define the body of intrinsic functions. Intrinsic functions may
4406 only be used in call or invoke instructions: it is illegal to take the address
4407 of an intrinsic function. Additionally, because intrinsic functions are part
4408 of the LLVM language, it is required if any are added that they be documented
4411 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4412 a family of functions that perform the same operation but on different data
4413 types. Because LLVM can represent over 8 million different integer types,
4414 overloading is used commonly to allow an intrinsic function to operate on any
4415 integer type. One or more of the argument types or the result type can be
4416 overloaded to accept any integer type. Argument types may also be defined as
4417 exactly matching a previous argument's type or the result type. This allows an
4418 intrinsic function which accepts multiple arguments, but needs all of them to
4419 be of the same type, to only be overloaded with respect to a single argument or
4422 <p>Overloaded intrinsics will have the names of its overloaded argument types
4423 encoded into its function name, each preceded by a period. Only those types
4424 which are overloaded result in a name suffix. Arguments whose type is matched
4425 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4426 take an integer of any width and returns an integer of exactly the same integer
4427 width. This leads to a family of functions such as
4428 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4429 Only one type, the return type, is overloaded, and only one type suffix is
4430 required. Because the argument's type is matched against the return type, it
4431 does not require its own name suffix.</p>
4433 <p>To learn how to add an intrinsic function, please see the
4434 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4439 <!-- ======================================================================= -->
4440 <div class="doc_subsection">
4441 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4444 <div class="doc_text">
4446 <p>Variable argument support is defined in LLVM with the <a
4447 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4448 intrinsic functions. These functions are related to the similarly
4449 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4451 <p>All of these functions operate on arguments that use a
4452 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4453 language reference manual does not define what this type is, so all
4454 transformations should be prepared to handle these functions regardless of
4457 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4458 instruction and the variable argument handling intrinsic functions are
4461 <div class="doc_code">
4463 define i32 @test(i32 %X, ...) {
4464 ; Initialize variable argument processing
4466 %ap2 = bitcast i8** %ap to i8*
4467 call void @llvm.va_start(i8* %ap2)
4469 ; Read a single integer argument
4470 %tmp = va_arg i8** %ap, i32
4472 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4474 %aq2 = bitcast i8** %aq to i8*
4475 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4476 call void @llvm.va_end(i8* %aq2)
4478 ; Stop processing of arguments.
4479 call void @llvm.va_end(i8* %ap2)
4483 declare void @llvm.va_start(i8*)
4484 declare void @llvm.va_copy(i8*, i8*)
4485 declare void @llvm.va_end(i8*)
4491 <!-- _______________________________________________________________________ -->
4492 <div class="doc_subsubsection">
4493 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4497 <div class="doc_text">
4499 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4501 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
4502 <tt>*<arglist></tt> for subsequent use by <tt><a
4503 href="#i_va_arg">va_arg</a></tt>.</p>
4507 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4511 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4512 macro available in C. In a target-dependent way, it initializes the
4513 <tt>va_list</tt> element to which the argument points, so that the next call to
4514 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4515 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4516 last argument of the function as the compiler can figure that out.</p>
4520 <!-- _______________________________________________________________________ -->
4521 <div class="doc_subsubsection">
4522 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4525 <div class="doc_text">
4527 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4530 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4531 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4532 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4536 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4540 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4541 macro available in C. In a target-dependent way, it destroys the
4542 <tt>va_list</tt> element to which the argument points. Calls to <a
4543 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4544 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4545 <tt>llvm.va_end</tt>.</p>
4549 <!-- _______________________________________________________________________ -->
4550 <div class="doc_subsubsection">
4551 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4554 <div class="doc_text">
4559 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4564 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4565 from the source argument list to the destination argument list.</p>
4569 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4570 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4575 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4576 macro available in C. In a target-dependent way, it copies the source
4577 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4578 intrinsic is necessary because the <tt><a href="#int_va_start">
4579 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4580 example, memory allocation.</p>
4584 <!-- ======================================================================= -->
4585 <div class="doc_subsection">
4586 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4589 <div class="doc_text">
4592 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4593 Collection</a> (GC) requires the implementation and generation of these
4595 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4596 stack</a>, as well as garbage collector implementations that require <a
4597 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4598 Front-ends for type-safe garbage collected languages should generate these
4599 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4600 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4603 <p>The garbage collection intrinsics only operate on objects in the generic
4604 address space (address space zero).</p>
4608 <!-- _______________________________________________________________________ -->
4609 <div class="doc_subsubsection">
4610 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4613 <div class="doc_text">
4618 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4623 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4624 the code generator, and allows some metadata to be associated with it.</p>
4628 <p>The first argument specifies the address of a stack object that contains the
4629 root pointer. The second pointer (which must be either a constant or a global
4630 value address) contains the meta-data to be associated with the root.</p>
4634 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4635 location. At compile-time, the code generator generates information to allow
4636 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4637 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4643 <!-- _______________________________________________________________________ -->
4644 <div class="doc_subsubsection">
4645 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4648 <div class="doc_text">
4653 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4658 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4659 locations, allowing garbage collector implementations that require read
4664 <p>The second argument is the address to read from, which should be an address
4665 allocated from the garbage collector. The first object is a pointer to the
4666 start of the referenced object, if needed by the language runtime (otherwise
4671 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4672 instruction, but may be replaced with substantially more complex code by the
4673 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4674 may only be used in a function which <a href="#gc">specifies a GC
4680 <!-- _______________________________________________________________________ -->
4681 <div class="doc_subsubsection">
4682 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4685 <div class="doc_text">
4690 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4695 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4696 locations, allowing garbage collector implementations that require write
4697 barriers (such as generational or reference counting collectors).</p>
4701 <p>The first argument is the reference to store, the second is the start of the
4702 object to store it to, and the third is the address of the field of Obj to
4703 store to. If the runtime does not require a pointer to the object, Obj may be
4708 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4709 instruction, but may be replaced with substantially more complex code by the
4710 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4711 may only be used in a function which <a href="#gc">specifies a GC
4718 <!-- ======================================================================= -->
4719 <div class="doc_subsection">
4720 <a name="int_codegen">Code Generator Intrinsics</a>
4723 <div class="doc_text">
4725 These intrinsics are provided by LLVM to expose special features that may only
4726 be implemented with code generator support.
4731 <!-- _______________________________________________________________________ -->
4732 <div class="doc_subsubsection">
4733 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4736 <div class="doc_text">
4740 declare i8 *@llvm.returnaddress(i32 <level>)
4746 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4747 target-specific value indicating the return address of the current function
4748 or one of its callers.
4754 The argument to this intrinsic indicates which function to return the address
4755 for. Zero indicates the calling function, one indicates its caller, etc. The
4756 argument is <b>required</b> to be a constant integer value.
4762 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4763 the return address of the specified call frame, or zero if it cannot be
4764 identified. The value returned by this intrinsic is likely to be incorrect or 0
4765 for arguments other than zero, so it should only be used for debugging purposes.
4769 Note that calling this intrinsic does not prevent function inlining or other
4770 aggressive transformations, so the value returned may not be that of the obvious
4771 source-language caller.
4776 <!-- _______________________________________________________________________ -->
4777 <div class="doc_subsubsection">
4778 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4781 <div class="doc_text">
4785 declare i8 *@llvm.frameaddress(i32 <level>)
4791 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4792 target-specific frame pointer value for the specified stack frame.
4798 The argument to this intrinsic indicates which function to return the frame
4799 pointer for. Zero indicates the calling function, one indicates its caller,
4800 etc. The argument is <b>required</b> to be a constant integer value.
4806 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4807 the frame address of the specified call frame, or zero if it cannot be
4808 identified. The value returned by this intrinsic is likely to be incorrect or 0
4809 for arguments other than zero, so it should only be used for debugging purposes.
4813 Note that calling this intrinsic does not prevent function inlining or other
4814 aggressive transformations, so the value returned may not be that of the obvious
4815 source-language caller.
4819 <!-- _______________________________________________________________________ -->
4820 <div class="doc_subsubsection">
4821 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4824 <div class="doc_text">
4828 declare i8 *@llvm.stacksave()
4834 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4835 the function stack, for use with <a href="#int_stackrestore">
4836 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4837 features like scoped automatic variable sized arrays in C99.
4843 This intrinsic returns a opaque pointer value that can be passed to <a
4844 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4845 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4846 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4847 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4848 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4849 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4854 <!-- _______________________________________________________________________ -->
4855 <div class="doc_subsubsection">
4856 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4859 <div class="doc_text">
4863 declare void @llvm.stackrestore(i8 * %ptr)
4869 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4870 the function stack to the state it was in when the corresponding <a
4871 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4872 useful for implementing language features like scoped automatic variable sized
4879 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4885 <!-- _______________________________________________________________________ -->
4886 <div class="doc_subsubsection">
4887 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4890 <div class="doc_text">
4894 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4901 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4902 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4904 effect on the behavior of the program but can change its performance
4911 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4912 determining if the fetch should be for a read (0) or write (1), and
4913 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4914 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4915 <tt>locality</tt> arguments must be constant integers.
4921 This intrinsic does not modify the behavior of the program. In particular,
4922 prefetches cannot trap and do not produce a value. On targets that support this
4923 intrinsic, the prefetch can provide hints to the processor cache for better
4929 <!-- _______________________________________________________________________ -->
4930 <div class="doc_subsubsection">
4931 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4934 <div class="doc_text">
4938 declare void @llvm.pcmarker(i32 <id>)
4945 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4947 code to simulators and other tools. The method is target specific, but it is
4948 expected that the marker will use exported symbols to transmit the PC of the
4950 The marker makes no guarantees that it will remain with any specific instruction
4951 after optimizations. It is possible that the presence of a marker will inhibit
4952 optimizations. The intended use is to be inserted after optimizations to allow
4953 correlations of simulation runs.
4959 <tt>id</tt> is a numerical id identifying the marker.
4965 This intrinsic does not modify the behavior of the program. Backends that do not
4966 support this intrinisic may ignore it.
4971 <!-- _______________________________________________________________________ -->
4972 <div class="doc_subsubsection">
4973 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4976 <div class="doc_text">
4980 declare i64 @llvm.readcyclecounter( )
4987 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4988 counter register (or similar low latency, high accuracy clocks) on those targets
4989 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4990 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4991 should only be used for small timings.
4997 When directly supported, reading the cycle counter should not modify any memory.
4998 Implementations are allowed to either return a application specific value or a
4999 system wide value. On backends without support, this is lowered to a constant 0.
5004 <!-- ======================================================================= -->
5005 <div class="doc_subsection">
5006 <a name="int_libc">Standard C Library Intrinsics</a>
5009 <div class="doc_text">
5011 LLVM provides intrinsics for a few important standard C library functions.
5012 These intrinsics allow source-language front-ends to pass information about the
5013 alignment of the pointer arguments to the code generator, providing opportunity
5014 for more efficient code generation.
5019 <!-- _______________________________________________________________________ -->
5020 <div class="doc_subsubsection">
5021 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5024 <div class="doc_text">
5028 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5029 i32 <len>, i32 <align>)
5030 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5031 i64 <len>, i32 <align>)
5037 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5038 location to the destination location.
5042 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5043 intrinsics do not return a value, and takes an extra alignment argument.
5049 The first argument is a pointer to the destination, the second is a pointer to
5050 the source. The third argument is an integer argument
5051 specifying the number of bytes to copy, and the fourth argument is the alignment
5052 of the source and destination locations.
5056 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5057 the caller guarantees that both the source and destination pointers are aligned
5064 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5065 location to the destination location, which are not allowed to overlap. It
5066 copies "len" bytes of memory over. If the argument is known to be aligned to
5067 some boundary, this can be specified as the fourth argument, otherwise it should
5073 <!-- _______________________________________________________________________ -->
5074 <div class="doc_subsubsection">
5075 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5078 <div class="doc_text">
5082 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5083 i32 <len>, i32 <align>)
5084 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5085 i64 <len>, i32 <align>)
5091 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5092 location to the destination location. It is similar to the
5093 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5097 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5098 intrinsics do not return a value, and takes an extra alignment argument.
5104 The first argument is a pointer to the destination, the second is a pointer to
5105 the source. The third argument is an integer argument
5106 specifying the number of bytes to copy, and the fourth argument is the alignment
5107 of the source and destination locations.
5111 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5112 the caller guarantees that the source and destination pointers are aligned to
5119 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5120 location to the destination location, which may overlap. It
5121 copies "len" bytes of memory over. If the argument is known to be aligned to
5122 some boundary, this can be specified as the fourth argument, otherwise it should
5128 <!-- _______________________________________________________________________ -->
5129 <div class="doc_subsubsection">
5130 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5133 <div class="doc_text">
5137 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5138 i32 <len>, i32 <align>)
5139 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5140 i64 <len>, i32 <align>)
5146 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5151 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5152 does not return a value, and takes an extra alignment argument.
5158 The first argument is a pointer to the destination to fill, the second is the
5159 byte value to fill it with, the third argument is an integer
5160 argument specifying the number of bytes to fill, and the fourth argument is the
5161 known alignment of destination location.
5165 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5166 the caller guarantees that the destination pointer is aligned to that boundary.
5172 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5174 destination location. If the argument is known to be aligned to some boundary,
5175 this can be specified as the fourth argument, otherwise it should be set to 0 or
5181 <!-- _______________________________________________________________________ -->
5182 <div class="doc_subsubsection">
5183 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5186 <div class="doc_text">
5189 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5190 floating point or vector of floating point type. Not all targets support all
5193 declare float @llvm.sqrt.f32(float %Val)
5194 declare double @llvm.sqrt.f64(double %Val)
5195 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5196 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5197 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5203 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5204 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5205 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5206 negative numbers other than -0.0 (which allows for better optimization, because
5207 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5208 defined to return -0.0 like IEEE sqrt.
5214 The argument and return value are floating point numbers of the same type.
5220 This function returns the sqrt of the specified operand if it is a nonnegative
5221 floating point number.
5225 <!-- _______________________________________________________________________ -->
5226 <div class="doc_subsubsection">
5227 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5230 <div class="doc_text">
5233 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5234 floating point or vector of floating point type. Not all targets support all
5237 declare float @llvm.powi.f32(float %Val, i32 %power)
5238 declare double @llvm.powi.f64(double %Val, i32 %power)
5239 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5240 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5241 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5247 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5248 specified (positive or negative) power. The order of evaluation of
5249 multiplications is not defined. When a vector of floating point type is
5250 used, the second argument remains a scalar integer value.
5256 The second argument is an integer power, and the first is a value to raise to
5263 This function returns the first value raised to the second power with an
5264 unspecified sequence of rounding operations.</p>
5267 <!-- _______________________________________________________________________ -->
5268 <div class="doc_subsubsection">
5269 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5272 <div class="doc_text">
5275 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5276 floating point or vector of floating point type. Not all targets support all
5279 declare float @llvm.sin.f32(float %Val)
5280 declare double @llvm.sin.f64(double %Val)
5281 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5282 declare fp128 @llvm.sin.f128(fp128 %Val)
5283 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5289 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5295 The argument and return value are floating point numbers of the same type.
5301 This function returns the sine of the specified operand, returning the
5302 same values as the libm <tt>sin</tt> functions would, and handles error
5303 conditions in the same way.</p>
5306 <!-- _______________________________________________________________________ -->
5307 <div class="doc_subsubsection">
5308 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5311 <div class="doc_text">
5314 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5315 floating point or vector of floating point type. Not all targets support all
5318 declare float @llvm.cos.f32(float %Val)
5319 declare double @llvm.cos.f64(double %Val)
5320 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5321 declare fp128 @llvm.cos.f128(fp128 %Val)
5322 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5328 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5334 The argument and return value are floating point numbers of the same type.
5340 This function returns the cosine of the specified operand, returning the
5341 same values as the libm <tt>cos</tt> functions would, and handles error
5342 conditions in the same way.</p>
5345 <!-- _______________________________________________________________________ -->
5346 <div class="doc_subsubsection">
5347 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5350 <div class="doc_text">
5353 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5354 floating point or vector of floating point type. Not all targets support all
5357 declare float @llvm.pow.f32(float %Val, float %Power)
5358 declare double @llvm.pow.f64(double %Val, double %Power)
5359 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5360 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5361 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5367 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5368 specified (positive or negative) power.
5374 The second argument is a floating point power, and the first is a value to
5375 raise to that power.
5381 This function returns the first value raised to the second power,
5383 same values as the libm <tt>pow</tt> functions would, and handles error
5384 conditions in the same way.</p>
5388 <!-- ======================================================================= -->
5389 <div class="doc_subsection">
5390 <a name="int_manip">Bit Manipulation Intrinsics</a>
5393 <div class="doc_text">
5395 LLVM provides intrinsics for a few important bit manipulation operations.
5396 These allow efficient code generation for some algorithms.
5401 <!-- _______________________________________________________________________ -->
5402 <div class="doc_subsubsection">
5403 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5406 <div class="doc_text">
5409 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5410 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
5412 declare i16 @llvm.bswap.i16(i16 <id>)
5413 declare i32 @llvm.bswap.i32(i32 <id>)
5414 declare i64 @llvm.bswap.i64(i64 <id>)
5420 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5421 values with an even number of bytes (positive multiple of 16 bits). These are
5422 useful for performing operations on data that is not in the target's native
5429 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5430 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5431 intrinsic returns an i32 value that has the four bytes of the input i32
5432 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5433 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5434 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5435 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5440 <!-- _______________________________________________________________________ -->
5441 <div class="doc_subsubsection">
5442 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5445 <div class="doc_text">
5448 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5449 width. Not all targets support all bit widths however.
5451 declare i8 @llvm.ctpop.i8 (i8 <src>)
5452 declare i16 @llvm.ctpop.i16(i16 <src>)
5453 declare i32 @llvm.ctpop.i32(i32 <src>)
5454 declare i64 @llvm.ctpop.i64(i64 <src>)
5455 declare i256 @llvm.ctpop.i256(i256 <src>)
5461 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5468 The only argument is the value to be counted. The argument may be of any
5469 integer type. The return type must match the argument type.
5475 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5479 <!-- _______________________________________________________________________ -->
5480 <div class="doc_subsubsection">
5481 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5484 <div class="doc_text">
5487 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5488 integer bit width. Not all targets support all bit widths however.
5490 declare i8 @llvm.ctlz.i8 (i8 <src>)
5491 declare i16 @llvm.ctlz.i16(i16 <src>)
5492 declare i32 @llvm.ctlz.i32(i32 <src>)
5493 declare i64 @llvm.ctlz.i64(i64 <src>)
5494 declare i256 @llvm.ctlz.i256(i256 <src>)
5500 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5501 leading zeros in a variable.
5507 The only argument is the value to be counted. The argument may be of any
5508 integer type. The return type must match the argument type.
5514 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5515 in a variable. If the src == 0 then the result is the size in bits of the type
5516 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5522 <!-- _______________________________________________________________________ -->
5523 <div class="doc_subsubsection">
5524 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5527 <div class="doc_text">
5530 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5531 integer bit width. Not all targets support all bit widths however.
5533 declare i8 @llvm.cttz.i8 (i8 <src>)
5534 declare i16 @llvm.cttz.i16(i16 <src>)
5535 declare i32 @llvm.cttz.i32(i32 <src>)
5536 declare i64 @llvm.cttz.i64(i64 <src>)
5537 declare i256 @llvm.cttz.i256(i256 <src>)
5543 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5550 The only argument is the value to be counted. The argument may be of any
5551 integer type. The return type must match the argument type.
5557 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5558 in a variable. If the src == 0 then the result is the size in bits of the type
5559 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5563 <!-- _______________________________________________________________________ -->
5564 <div class="doc_subsubsection">
5565 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5568 <div class="doc_text">
5571 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5572 on any integer bit width.
5574 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5575 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5579 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5580 range of bits from an integer value and returns them in the same bit width as
5581 the original value.</p>
5584 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5585 any bit width but they must have the same bit width. The second and third
5586 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5589 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5590 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5591 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5592 operates in forward mode.</p>
5593 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5594 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5595 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5597 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5598 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5599 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5600 to determine the number of bits to retain.</li>
5601 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5602 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5604 <p>In reverse mode, a similar computation is made except that the bits are
5605 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5606 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5607 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5608 <tt>i16 0x0026 (000000100110)</tt>.</p>
5611 <div class="doc_subsubsection">
5612 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5615 <div class="doc_text">
5618 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5619 on any integer bit width.
5621 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5622 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5626 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5627 of bits in an integer value with another integer value. It returns the integer
5628 with the replaced bits.</p>
5631 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5632 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5633 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5634 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5635 type since they specify only a bit index.</p>
5638 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5639 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5640 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5641 operates in forward mode.</p>
5642 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5643 truncating it down to the size of the replacement area or zero extending it
5644 up to that size.</p>
5645 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5646 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5647 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5648 to the <tt>%hi</tt>th bit.
5649 <p>In reverse mode, a similar computation is made except that the bits are
5650 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5651 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5654 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5655 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5656 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5657 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5658 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5662 <!-- ======================================================================= -->
5663 <div class="doc_subsection">
5664 <a name="int_debugger">Debugger Intrinsics</a>
5667 <div class="doc_text">
5669 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5670 are described in the <a
5671 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5672 Debugging</a> document.
5677 <!-- ======================================================================= -->
5678 <div class="doc_subsection">
5679 <a name="int_eh">Exception Handling Intrinsics</a>
5682 <div class="doc_text">
5683 <p> The LLVM exception handling intrinsics (which all start with
5684 <tt>llvm.eh.</tt> prefix), are described in the <a
5685 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5686 Handling</a> document. </p>
5689 <!-- ======================================================================= -->
5690 <div class="doc_subsection">
5691 <a name="int_trampoline">Trampoline Intrinsic</a>
5694 <div class="doc_text">
5696 This intrinsic makes it possible to excise one parameter, marked with
5697 the <tt>nest</tt> attribute, from a function. The result is a callable
5698 function pointer lacking the nest parameter - the caller does not need
5699 to provide a value for it. Instead, the value to use is stored in
5700 advance in a "trampoline", a block of memory usually allocated
5701 on the stack, which also contains code to splice the nest value into the
5702 argument list. This is used to implement the GCC nested function address
5706 For example, if the function is
5707 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5708 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5710 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5711 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5712 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5713 %fp = bitcast i8* %p to i32 (i32, i32)*
5715 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5716 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5719 <!-- _______________________________________________________________________ -->
5720 <div class="doc_subsubsection">
5721 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5723 <div class="doc_text">
5726 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5730 This fills the memory pointed to by <tt>tramp</tt> with code
5731 and returns a function pointer suitable for executing it.
5735 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5736 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5737 and sufficiently aligned block of memory; this memory is written to by the
5738 intrinsic. Note that the size and the alignment are target-specific - LLVM
5739 currently provides no portable way of determining them, so a front-end that
5740 generates this intrinsic needs to have some target-specific knowledge.
5741 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5745 The block of memory pointed to by <tt>tramp</tt> is filled with target
5746 dependent code, turning it into a function. A pointer to this function is
5747 returned, but needs to be bitcast to an
5748 <a href="#int_trampoline">appropriate function pointer type</a>
5749 before being called. The new function's signature is the same as that of
5750 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5751 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5752 of pointer type. Calling the new function is equivalent to calling
5753 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5754 missing <tt>nest</tt> argument. If, after calling
5755 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5756 modified, then the effect of any later call to the returned function pointer is
5761 <!-- ======================================================================= -->
5762 <div class="doc_subsection">
5763 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5766 <div class="doc_text">
5768 These intrinsic functions expand the "universal IR" of LLVM to represent
5769 hardware constructs for atomic operations and memory synchronization. This
5770 provides an interface to the hardware, not an interface to the programmer. It
5771 is aimed at a low enough level to allow any programming models or APIs
5772 (Application Programming Interfaces) which
5773 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5774 hardware behavior. Just as hardware provides a "universal IR" for source
5775 languages, it also provides a starting point for developing a "universal"
5776 atomic operation and synchronization IR.
5779 These do <em>not</em> form an API such as high-level threading libraries,
5780 software transaction memory systems, atomic primitives, and intrinsic
5781 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5782 application libraries. The hardware interface provided by LLVM should allow
5783 a clean implementation of all of these APIs and parallel programming models.
5784 No one model or paradigm should be selected above others unless the hardware
5785 itself ubiquitously does so.
5790 <!-- _______________________________________________________________________ -->
5791 <div class="doc_subsubsection">
5792 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5794 <div class="doc_text">
5797 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5803 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5804 specific pairs of memory access types.
5808 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5809 The first four arguments enables a specific barrier as listed below. The fith
5810 argument specifies that the barrier applies to io or device or uncached memory.
5814 <li><tt>ll</tt>: load-load barrier</li>
5815 <li><tt>ls</tt>: load-store barrier</li>
5816 <li><tt>sl</tt>: store-load barrier</li>
5817 <li><tt>ss</tt>: store-store barrier</li>
5818 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5822 This intrinsic causes the system to enforce some ordering constraints upon
5823 the loads and stores of the program. This barrier does not indicate
5824 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5825 which they occur. For any of the specified pairs of load and store operations
5826 (f.ex. load-load, or store-load), all of the first operations preceding the
5827 barrier will complete before any of the second operations succeeding the
5828 barrier begin. Specifically the semantics for each pairing is as follows:
5831 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5832 after the barrier begins.</li>
5834 <li><tt>ls</tt>: All loads before the barrier must complete before any
5835 store after the barrier begins.</li>
5836 <li><tt>ss</tt>: All stores before the barrier must complete before any
5837 store after the barrier begins.</li>
5838 <li><tt>sl</tt>: All stores before the barrier must complete before any
5839 load after the barrier begins.</li>
5842 These semantics are applied with a logical "and" behavior when more than one
5843 is enabled in a single memory barrier intrinsic.
5846 Backends may implement stronger barriers than those requested when they do not
5847 support as fine grained a barrier as requested. Some architectures do not
5848 need all types of barriers and on such architectures, these become noops.
5855 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5856 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5857 <i>; guarantee the above finishes</i>
5858 store i32 8, %ptr <i>; before this begins</i>
5862 <!-- _______________________________________________________________________ -->
5863 <div class="doc_subsubsection">
5864 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
5866 <div class="doc_text">
5869 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
5870 any integer bit width and for different address spaces. Not all targets
5871 support all bit widths however.</p>
5874 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5875 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5876 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5877 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5882 This loads a value in memory and compares it to a given value. If they are
5883 equal, it stores a new value into the memory.
5887 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
5888 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5889 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5890 this integer type. While any bit width integer may be used, targets may only
5891 lower representations they support in hardware.
5896 This entire intrinsic must be executed atomically. It first loads the value
5897 in memory pointed to by <tt>ptr</tt> and compares it with the value
5898 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5899 loaded value is yielded in all cases. This provides the equivalent of an
5900 atomic compare-and-swap operation within the SSA framework.
5908 %val1 = add i32 4, 4
5909 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
5910 <i>; yields {i32}:result1 = 4</i>
5911 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5912 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5914 %val2 = add i32 1, 1
5915 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
5916 <i>; yields {i32}:result2 = 8</i>
5917 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5919 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5923 <!-- _______________________________________________________________________ -->
5924 <div class="doc_subsubsection">
5925 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5927 <div class="doc_text">
5931 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5932 integer bit width. Not all targets support all bit widths however.</p>
5934 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
5935 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
5936 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
5937 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
5942 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5943 the value from memory. It then stores the value in <tt>val</tt> in the memory
5949 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
5950 <tt>val</tt> argument and the result must be integers of the same bit width.
5951 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5952 integer type. The targets may only lower integer representations they
5957 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5958 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5959 equivalent of an atomic swap operation within the SSA framework.
5967 %val1 = add i32 4, 4
5968 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
5969 <i>; yields {i32}:result1 = 4</i>
5970 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5971 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5973 %val2 = add i32 1, 1
5974 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
5975 <i>; yields {i32}:result2 = 8</i>
5977 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5978 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5982 <!-- _______________________________________________________________________ -->
5983 <div class="doc_subsubsection">
5984 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
5987 <div class="doc_text">
5990 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on any
5991 integer bit width. Not all targets support all bit widths however.</p>
5993 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
5994 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
5995 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
5996 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6001 This intrinsic adds <tt>delta</tt> to the value stored in memory at
6002 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6007 The intrinsic takes two arguments, the first a pointer to an integer value
6008 and the second an integer value. The result is also an integer value. These
6009 integer types can have any bit width, but they must all have the same bit
6010 width. The targets may only lower integer representations they support.
6014 This intrinsic does a series of operations atomically. It first loads the
6015 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6016 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6023 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6024 <i>; yields {i32}:result1 = 4</i>
6025 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6026 <i>; yields {i32}:result2 = 8</i>
6027 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6028 <i>; yields {i32}:result3 = 10</i>
6029 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6033 <!-- _______________________________________________________________________ -->
6034 <div class="doc_subsubsection">
6035 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6038 <div class="doc_text">
6041 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6042 any integer bit width and for different address spaces. Not all targets
6043 support all bit widths however.</p>
6045 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6046 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6047 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6048 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6053 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6054 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6059 The intrinsic takes two arguments, the first a pointer to an integer value
6060 and the second an integer value. The result is also an integer value. These
6061 integer types can have any bit width, but they must all have the same bit
6062 width. The targets may only lower integer representations they support.
6066 This intrinsic does a series of operations atomically. It first loads the
6067 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6068 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6075 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6076 <i>; yields {i32}:result1 = 8</i>
6077 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6078 <i>; yields {i32}:result2 = 4</i>
6079 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6080 <i>; yields {i32}:result3 = 2</i>
6081 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6085 <!-- _______________________________________________________________________ -->
6086 <div class="doc_subsubsection">
6087 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6088 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6089 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6090 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6093 <div class="doc_text">
6096 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6097 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
6098 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6099 address spaces. Not all targets support all bit widths however.</p>
6101 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6102 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6103 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6104 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6109 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6110 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6111 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6112 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6117 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6118 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6119 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6120 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6125 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6126 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6127 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6128 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6133 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6134 the value stored in memory at <tt>ptr</tt>. It yields the original value
6140 These intrinsics take two arguments, the first a pointer to an integer value
6141 and the second an integer value. The result is also an integer value. These
6142 integer types can have any bit width, but they must all have the same bit
6143 width. The targets may only lower integer representations they support.
6147 These intrinsics does a series of operations atomically. They first load the
6148 value stored at <tt>ptr</tt>. They then do the bitwise operation
6149 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6150 value stored at <tt>ptr</tt>.
6156 store i32 0x0F0F, %ptr
6157 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6158 <i>; yields {i32}:result0 = 0x0F0F</i>
6159 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6160 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6161 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6162 <i>; yields {i32}:result2 = 0xF0</i>
6163 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6164 <i>; yields {i32}:result3 = FF</i>
6165 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6170 <!-- _______________________________________________________________________ -->
6171 <div class="doc_subsubsection">
6172 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6173 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6174 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6175 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6178 <div class="doc_text">
6181 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6182 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6183 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6184 address spaces. Not all targets
6185 support all bit widths however.</p>
6187 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6188 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6189 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6190 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6195 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6196 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6197 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6198 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6203 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6204 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6205 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6206 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6211 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6212 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6213 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6214 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6219 These intrinsics takes the signed or unsigned minimum or maximum of
6220 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6221 original value at <tt>ptr</tt>.
6226 These intrinsics take two arguments, the first a pointer to an integer value
6227 and the second an integer value. The result is also an integer value. These
6228 integer types can have any bit width, but they must all have the same bit
6229 width. The targets may only lower integer representations they support.
6233 These intrinsics does a series of operations atomically. They first load the
6234 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
6235 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
6236 the original value stored at <tt>ptr</tt>.
6243 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6244 <i>; yields {i32}:result0 = 7</i>
6245 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6246 <i>; yields {i32}:result1 = -2</i>
6247 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6248 <i>; yields {i32}:result2 = 8</i>
6249 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6250 <i>; yields {i32}:result3 = 8</i>
6251 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6255 <!-- ======================================================================= -->
6256 <div class="doc_subsection">
6257 <a name="int_general">General Intrinsics</a>
6260 <div class="doc_text">
6261 <p> This class of intrinsics is designed to be generic and has
6262 no specific purpose. </p>
6265 <!-- _______________________________________________________________________ -->
6266 <div class="doc_subsubsection">
6267 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6270 <div class="doc_text">
6274 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6280 The '<tt>llvm.var.annotation</tt>' intrinsic
6286 The first argument is a pointer to a value, the second is a pointer to a
6287 global string, the third is a pointer to a global string which is the source
6288 file name, and the last argument is the line number.
6294 This intrinsic allows annotation of local variables with arbitrary strings.
6295 This can be useful for special purpose optimizations that want to look for these
6296 annotations. These have no other defined use, they are ignored by code
6297 generation and optimization.
6301 <!-- _______________________________________________________________________ -->
6302 <div class="doc_subsubsection">
6303 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6306 <div class="doc_text">
6309 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6310 any integer bit width.
6313 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6314 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6315 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6316 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6317 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6323 The '<tt>llvm.annotation</tt>' intrinsic.
6329 The first argument is an integer value (result of some expression),
6330 the second is a pointer to a global string, the third is a pointer to a global
6331 string which is the source file name, and the last argument is the line number.
6332 It returns the value of the first argument.
6338 This intrinsic allows annotations to be put on arbitrary expressions
6339 with arbitrary strings. This can be useful for special purpose optimizations
6340 that want to look for these annotations. These have no other defined use, they
6341 are ignored by code generation and optimization.
6344 <!-- _______________________________________________________________________ -->
6345 <div class="doc_subsubsection">
6346 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6349 <div class="doc_text">
6353 declare void @llvm.trap()
6359 The '<tt>llvm.trap</tt>' intrinsic
6371 This intrinsics is lowered to the target dependent trap instruction. If the
6372 target does not have a trap instruction, this intrinsic will be lowered to the
6373 call of the abort() function.
6377 <!-- *********************************************************************** -->
6380 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6381 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
6382 <a href="http://validator.w3.org/check/referer"><img
6383 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
6385 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6386 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6387 Last modified: $Date$