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
789 <div class="doc_code">
791 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>] [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] <ResultType> @<FunctionName> ([argument list]) [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N] [<a href="#gc">gc</a>] { ... }
798 <!-- ======================================================================= -->
799 <div class="doc_subsection">
800 <a name="aliasstructure">Aliases</a>
802 <div class="doc_text">
803 <p>Aliases act as "second name" for the aliasee value (which can be either
804 function, global variable, another alias or bitcast of global value). Aliases
805 may have an optional <a href="#linkage">linkage type</a>, and an
806 optional <a href="#visibility">visibility style</a>.</p>
810 <div class="doc_code">
812 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
820 <!-- ======================================================================= -->
821 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
822 <div class="doc_text">
823 <p>The return type and each parameter of a function type may have a set of
824 <i>parameter attributes</i> associated with them. Parameter attributes are
825 used to communicate additional information about the result or parameters of
826 a function. Parameter attributes are considered to be part of the function,
827 not of the function type, so functions with different parameter attributes
828 can have the same function type.</p>
830 <p>Parameter attributes are simple keywords that follow the type specified. If
831 multiple parameter attributes are needed, they are space separated. For
834 <div class="doc_code">
836 declare i32 @printf(i8* noalias , ...)
837 declare i32 @atoi(i8 zeroext)
838 declare signext i8 @returns_signed_char()
842 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
843 <tt>readonly</tt>) come immediately after the argument list.</p>
845 <p>Currently, only the following parameter attributes are defined:</p>
847 <dt><tt>zeroext</tt></dt>
848 <dd>This indicates to the code generator that the parameter or return value
849 should be zero-extended to a 32-bit value by the caller (for a parameter)
850 or the callee (for a return value).</dd>
852 <dt><tt>signext</tt></dt>
853 <dd>This indicates to the code generator that the parameter or return value
854 should be sign-extended to a 32-bit value by the caller (for a parameter)
855 or the callee (for a return value).</dd>
857 <dt><tt>inreg</tt></dt>
858 <dd>This indicates that this parameter or return value should be treated
859 in a special target-dependent fashion during while emitting code for a
860 function call or return (usually, by putting it in a register as opposed
861 to memory, though some targets use it to distinguish between two different
862 kinds of registers). Use of this attribute is target-specific.</dd>
864 <dt><tt><a name="byval">byval</a></tt></dt>
865 <dd>This indicates that the pointer parameter should really be passed by
866 value to the function. The attribute implies that a hidden copy of the
867 pointee is made between the caller and the callee, so the callee is unable
868 to modify the value in the callee. This attribute is only valid on LLVM
869 pointer arguments. It is generally used to pass structs and arrays by
870 value, but is also valid on pointers to scalars. The copy is considered to
871 belong to the caller not the callee (for example,
872 <tt><a href="#readonly">readonly</a></tt> functions should not write to
873 <tt>byval</tt> parameters). This is not a valid attribute for return
876 <dt><tt>sret</tt></dt>
877 <dd>This indicates that the pointer parameter specifies the address of a
878 structure that is the return value of the function in the source program.
879 This pointer must be guaranteed by the caller to be valid: loads and stores
880 to the structure may be assumed by the callee to not to trap. This may only
881 be applied to the first parameter. This is not a valid attribute for
884 <dt><tt>noalias</tt></dt>
885 <dd>This indicates that the parameter does not alias any global or any other
886 parameter. The caller is responsible for ensuring that this is the case,
887 usually by placing the value in a stack allocation. This is not a valid
888 attribute for return values.</dd>
890 <dt><tt>nest</tt></dt>
891 <dd>This indicates that the pointer parameter can be excised using the
892 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
893 attribute for return values.</dd>
898 <!-- ======================================================================= -->
899 <div class="doc_subsection">
900 <a name="gc">Garbage Collector Names</a>
903 <div class="doc_text">
904 <p>Each function may specify a garbage collector name, which is simply a
907 <div class="doc_code"><pre
908 >define void @f() gc "name" { ...</pre></div>
910 <p>The compiler declares the supported values of <i>name</i>. Specifying a
911 collector which will cause the compiler to alter its output in order to support
912 the named garbage collection algorithm.</p>
915 <!-- ======================================================================= -->
916 <div class="doc_subsection">
917 <a name="fnattrs">Function Attributes</a>
920 <div class="doc_text">
922 <p>Function attributes are set to communicate additional information about
923 a function. Function attributes are considered to be part of the function,
924 not of the function type, so functions with different parameter attributes
925 can have the same function type.</p>
927 <p>Function attributes are simple keywords that follow the type specified. If
928 multiple attributes are needed, they are space separated. For
931 <div class="doc_code">
933 define void @f() noinline { ... }
934 define void @f() alwaysinline { ... }
935 define void @f() alwaysinline optsize { ... }
936 define void @f() optsize
941 <dt><tt>alwaysinline</tt></dt>
942 <dd>This attribute indicates that the inliner should attempt to inline this
943 function into callers whenever possible, ignoring any active inlining size
944 threshold for this caller.</dd>
946 <dt><tt>noinline</tt></dt>
947 <dd>This attribute indicates that the inliner should never inline this function
948 in any situation. This attribute may not be used together with the
949 <tt>alwaysinline</tt> attribute.</dd>
951 <dt><tt>optsize</tt></dt>
952 <dd>This attribute suggests that optimization passes and code generator passes
953 make choices that keep the code size of this function low, and otherwise do
954 optimizations specifically to reduce code size.</dd>
956 <dt><tt>noreturn</tt></dt>
957 <dd>This function attribute indicates that the function never returns normally.
958 This produces undefined behavior at runtime if the function ever does
959 dynamically return.</dd>
961 <dt><tt>nounwind</tt></dt>
962 <dd>This function attribute indicates that the function never returns with an
963 unwind or exceptional control flow. If the function does unwind, its runtime
964 behavior is undefined.</dd>
966 <dt><tt>readnone</tt></dt>
967 <dd>This attribute indicates that the function computes its result (or the
968 exception it throws) based strictly on its arguments, without dereferencing any
969 pointer arguments or otherwise accessing any mutable state (e.g. memory, control
970 registers, etc) visible to caller functions. It does not write through any
971 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments) and
972 never changes any state visible to callers.</dd>
974 <dt><tt><a name="readonly">readonly</a></tt></dt>
975 <dd>This attribute indicates that the function does not write through any
976 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments)
977 or otherwise modify any state (e.g. memory, control registers, etc) visible to
978 caller functions. It may dereference pointer arguments and read state that may
979 be set in the caller. A readonly function always returns the same value (or
980 throws the same exception) when called with the same set of arguments and global
986 <!-- ======================================================================= -->
987 <div class="doc_subsection">
988 <a name="moduleasm">Module-Level Inline Assembly</a>
991 <div class="doc_text">
993 Modules may contain "module-level inline asm" blocks, which corresponds to the
994 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
995 LLVM and treated as a single unit, but may be separated in the .ll file if
996 desired. The syntax is very simple:
999 <div class="doc_code">
1001 module asm "inline asm code goes here"
1002 module asm "more can go here"
1006 <p>The strings can contain any character by escaping non-printable characters.
1007 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1012 The inline asm code is simply printed to the machine code .s file when
1013 assembly code is generated.
1017 <!-- ======================================================================= -->
1018 <div class="doc_subsection">
1019 <a name="datalayout">Data Layout</a>
1022 <div class="doc_text">
1023 <p>A module may specify a target specific data layout string that specifies how
1024 data is to be laid out in memory. The syntax for the data layout is simply:</p>
1025 <pre> target datalayout = "<i>layout specification</i>"</pre>
1026 <p>The <i>layout specification</i> consists of a list of specifications
1027 separated by the minus sign character ('-'). Each specification starts with a
1028 letter and may include other information after the letter to define some
1029 aspect of the data layout. The specifications accepted are as follows: </p>
1032 <dd>Specifies that the target lays out data in big-endian form. That is, the
1033 bits with the most significance have the lowest address location.</dd>
1035 <dd>Specifies that the target lays out data in little-endian form. That is,
1036 the bits with the least significance have the lowest address location.</dd>
1037 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1038 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1039 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
1040 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
1042 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1043 <dd>This specifies the alignment for an integer type of a given bit
1044 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1045 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1046 <dd>This specifies the alignment for a vector type of a given bit
1048 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1049 <dd>This specifies the alignment for a floating point type of a given bit
1050 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1052 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1053 <dd>This specifies the alignment for an aggregate type of a given bit
1056 <p>When constructing the data layout for a given target, LLVM starts with a
1057 default set of specifications which are then (possibly) overriden by the
1058 specifications in the <tt>datalayout</tt> keyword. The default specifications
1059 are given in this list:</p>
1061 <li><tt>E</tt> - big endian</li>
1062 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1063 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1064 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1065 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1066 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1067 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1068 alignment of 64-bits</li>
1069 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1070 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1071 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1072 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1073 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1075 <p>When LLVM is determining the alignment for a given type, it uses the
1078 <li>If the type sought is an exact match for one of the specifications, that
1079 specification is used.</li>
1080 <li>If no match is found, and the type sought is an integer type, then the
1081 smallest integer type that is larger than the bitwidth of the sought type is
1082 used. If none of the specifications are larger than the bitwidth then the the
1083 largest integer type is used. For example, given the default specifications
1084 above, the i7 type will use the alignment of i8 (next largest) while both
1085 i65 and i256 will use the alignment of i64 (largest specified).</li>
1086 <li>If no match is found, and the type sought is a vector type, then the
1087 largest vector type that is smaller than the sought vector type will be used
1088 as a fall back. This happens because <128 x double> can be implemented in
1089 terms of 64 <2 x double>, for example.</li>
1093 <!-- *********************************************************************** -->
1094 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1095 <!-- *********************************************************************** -->
1097 <div class="doc_text">
1099 <p>The LLVM type system is one of the most important features of the
1100 intermediate representation. Being typed enables a number of
1101 optimizations to be performed on the intermediate representation directly,
1102 without having to do
1103 extra analyses on the side before the transformation. A strong type
1104 system makes it easier to read the generated code and enables novel
1105 analyses and transformations that are not feasible to perform on normal
1106 three address code representations.</p>
1110 <!-- ======================================================================= -->
1111 <div class="doc_subsection"> <a name="t_classifications">Type
1112 Classifications</a> </div>
1113 <div class="doc_text">
1114 <p>The types fall into a few useful
1115 classifications:</p>
1117 <table border="1" cellspacing="0" cellpadding="4">
1119 <tr><th>Classification</th><th>Types</th></tr>
1121 <td><a href="#t_integer">integer</a></td>
1122 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1125 <td><a href="#t_floating">floating point</a></td>
1126 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1129 <td><a name="t_firstclass">first class</a></td>
1130 <td><a href="#t_integer">integer</a>,
1131 <a href="#t_floating">floating point</a>,
1132 <a href="#t_pointer">pointer</a>,
1133 <a href="#t_vector">vector</a>,
1134 <a href="#t_struct">structure</a>,
1135 <a href="#t_array">array</a>,
1136 <a href="#t_label">label</a>.
1140 <td><a href="#t_primitive">primitive</a></td>
1141 <td><a href="#t_label">label</a>,
1142 <a href="#t_void">void</a>,
1143 <a href="#t_floating">floating point</a>.</td>
1146 <td><a href="#t_derived">derived</a></td>
1147 <td><a href="#t_integer">integer</a>,
1148 <a href="#t_array">array</a>,
1149 <a href="#t_function">function</a>,
1150 <a href="#t_pointer">pointer</a>,
1151 <a href="#t_struct">structure</a>,
1152 <a href="#t_pstruct">packed structure</a>,
1153 <a href="#t_vector">vector</a>,
1154 <a href="#t_opaque">opaque</a>.
1159 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1160 most important. Values of these types are the only ones which can be
1161 produced by instructions, passed as arguments, or used as operands to
1165 <!-- ======================================================================= -->
1166 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1168 <div class="doc_text">
1169 <p>The primitive types are the fundamental building blocks of the LLVM
1174 <!-- _______________________________________________________________________ -->
1175 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1177 <div class="doc_text">
1180 <tr><th>Type</th><th>Description</th></tr>
1181 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1182 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1183 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1184 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1185 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1190 <!-- _______________________________________________________________________ -->
1191 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1193 <div class="doc_text">
1195 <p>The void type does not represent any value and has no size.</p>
1204 <!-- _______________________________________________________________________ -->
1205 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1207 <div class="doc_text">
1209 <p>The label type represents code labels.</p>
1219 <!-- ======================================================================= -->
1220 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1222 <div class="doc_text">
1224 <p>The real power in LLVM comes from the derived types in the system.
1225 This is what allows a programmer to represent arrays, functions,
1226 pointers, and other useful types. Note that these derived types may be
1227 recursive: For example, it is possible to have a two dimensional array.</p>
1231 <!-- _______________________________________________________________________ -->
1232 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1234 <div class="doc_text">
1237 <p>The integer type is a very simple derived type that simply specifies an
1238 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1239 2^23-1 (about 8 million) can be specified.</p>
1247 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1251 <table class="layout">
1254 <td><tt>i1</tt></td>
1255 <td>a single-bit integer.</td>
1257 <td><tt>i32</tt></td>
1258 <td>a 32-bit integer.</td>
1260 <td><tt>i1942652</tt></td>
1261 <td>a really big integer of over 1 million bits.</td>
1267 <!-- _______________________________________________________________________ -->
1268 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1270 <div class="doc_text">
1274 <p>The array type is a very simple derived type that arranges elements
1275 sequentially in memory. The array type requires a size (number of
1276 elements) and an underlying data type.</p>
1281 [<# elements> x <elementtype>]
1284 <p>The number of elements is a constant integer value; elementtype may
1285 be any type with a size.</p>
1288 <table class="layout">
1290 <td class="left"><tt>[40 x i32]</tt></td>
1291 <td class="left">Array of 40 32-bit integer values.</td>
1294 <td class="left"><tt>[41 x i32]</tt></td>
1295 <td class="left">Array of 41 32-bit integer values.</td>
1298 <td class="left"><tt>[4 x i8]</tt></td>
1299 <td class="left">Array of 4 8-bit integer values.</td>
1302 <p>Here are some examples of multidimensional arrays:</p>
1303 <table class="layout">
1305 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1306 <td class="left">3x4 array of 32-bit integer values.</td>
1309 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1310 <td class="left">12x10 array of single precision floating point values.</td>
1313 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1314 <td class="left">2x3x4 array of 16-bit integer values.</td>
1318 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1319 length array. Normally, accesses past the end of an array are undefined in
1320 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1321 As a special case, however, zero length arrays are recognized to be variable
1322 length. This allows implementation of 'pascal style arrays' with the LLVM
1323 type "{ i32, [0 x float]}", for example.</p>
1327 <!-- _______________________________________________________________________ -->
1328 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1329 <div class="doc_text">
1333 <p>The function type can be thought of as a function signature. It
1334 consists of a return type and a list of formal parameter types. The
1335 return type of a function type is a scalar type, a void type, or a struct type.
1336 If the return type is a struct type then all struct elements must be of first
1337 class types, and the struct must have at least one element.</p>
1342 <returntype list> (<parameter list>)
1345 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1346 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1347 which indicates that the function takes a variable number of arguments.
1348 Variable argument functions can access their arguments with the <a
1349 href="#int_varargs">variable argument handling intrinsic</a> functions.
1350 '<tt><returntype list></tt>' is a comma-separated list of
1351 <a href="#t_firstclass">first class</a> type specifiers.</p>
1354 <table class="layout">
1356 <td class="left"><tt>i32 (i32)</tt></td>
1357 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1359 </tr><tr class="layout">
1360 <td class="left"><tt>float (i16 signext, i32 *) *
1362 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1363 an <tt>i16</tt> that should be sign extended and a
1364 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1367 </tr><tr class="layout">
1368 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1369 <td class="left">A vararg function that takes at least one
1370 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1371 which returns an integer. This is the signature for <tt>printf</tt> in
1374 </tr><tr class="layout">
1375 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1376 <td class="left">A function taking an <tt>i32></tt>, returning two
1377 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1383 <!-- _______________________________________________________________________ -->
1384 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1385 <div class="doc_text">
1387 <p>The structure type is used to represent a collection of data members
1388 together in memory. The packing of the field types is defined to match
1389 the ABI of the underlying processor. The elements of a structure may
1390 be any type that has a size.</p>
1391 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1392 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1393 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1396 <pre> { <type list> }<br></pre>
1398 <table class="layout">
1400 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1401 <td class="left">A triple of three <tt>i32</tt> values</td>
1402 </tr><tr class="layout">
1403 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1404 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1405 second element is a <a href="#t_pointer">pointer</a> to a
1406 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1407 an <tt>i32</tt>.</td>
1412 <!-- _______________________________________________________________________ -->
1413 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1415 <div class="doc_text">
1417 <p>The packed structure type is used to represent a collection of data members
1418 together in memory. There is no padding between fields. Further, the alignment
1419 of a packed structure is 1 byte. The elements of a packed structure may
1420 be any type that has a size.</p>
1421 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1422 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1423 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1426 <pre> < { <type list> } > <br></pre>
1428 <table class="layout">
1430 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1431 <td class="left">A triple of three <tt>i32</tt> values</td>
1432 </tr><tr class="layout">
1434 <tt>< { float, i32 (i32)* } ></tt></td>
1435 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1436 second element is a <a href="#t_pointer">pointer</a> to a
1437 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1438 an <tt>i32</tt>.</td>
1443 <!-- _______________________________________________________________________ -->
1444 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1445 <div class="doc_text">
1447 <p>As in many languages, the pointer type represents a pointer or
1448 reference to another object, which must live in memory. Pointer types may have
1449 an optional address space attribute defining the target-specific numbered
1450 address space where the pointed-to object resides. The default address space is
1453 <pre> <type> *<br></pre>
1455 <table class="layout">
1457 <td class="left"><tt>[4x i32]*</tt></td>
1458 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1459 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1462 <td class="left"><tt>i32 (i32 *) *</tt></td>
1463 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1464 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1468 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1469 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1470 that resides in address space #5.</td>
1475 <!-- _______________________________________________________________________ -->
1476 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1477 <div class="doc_text">
1481 <p>A vector type is a simple derived type that represents a vector
1482 of elements. Vector types are used when multiple primitive data
1483 are operated in parallel using a single instruction (SIMD).
1484 A vector type requires a size (number of
1485 elements) and an underlying primitive data type. Vectors must have a power
1486 of two length (1, 2, 4, 8, 16 ...). Vector types are
1487 considered <a href="#t_firstclass">first class</a>.</p>
1492 < <# elements> x <elementtype> >
1495 <p>The number of elements is a constant integer value; elementtype may
1496 be any integer or floating point type.</p>
1500 <table class="layout">
1502 <td class="left"><tt><4 x i32></tt></td>
1503 <td class="left">Vector of 4 32-bit integer values.</td>
1506 <td class="left"><tt><8 x float></tt></td>
1507 <td class="left">Vector of 8 32-bit floating-point values.</td>
1510 <td class="left"><tt><2 x i64></tt></td>
1511 <td class="left">Vector of 2 64-bit integer values.</td>
1516 <!-- _______________________________________________________________________ -->
1517 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1518 <div class="doc_text">
1522 <p>Opaque types are used to represent unknown types in the system. This
1523 corresponds (for example) to the C notion of a forward declared structure type.
1524 In LLVM, opaque types can eventually be resolved to any type (not just a
1525 structure type).</p>
1535 <table class="layout">
1537 <td class="left"><tt>opaque</tt></td>
1538 <td class="left">An opaque type.</td>
1544 <!-- *********************************************************************** -->
1545 <div class="doc_section"> <a name="constants">Constants</a> </div>
1546 <!-- *********************************************************************** -->
1548 <div class="doc_text">
1550 <p>LLVM has several different basic types of constants. This section describes
1551 them all and their syntax.</p>
1555 <!-- ======================================================================= -->
1556 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1558 <div class="doc_text">
1561 <dt><b>Boolean constants</b></dt>
1563 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1564 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1567 <dt><b>Integer constants</b></dt>
1569 <dd>Standard integers (such as '4') are constants of the <a
1570 href="#t_integer">integer</a> type. Negative numbers may be used with
1574 <dt><b>Floating point constants</b></dt>
1576 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1577 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1578 notation (see below). The assembler requires the exact decimal value of
1579 a floating-point constant. For example, the assembler accepts 1.25 but
1580 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1581 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1583 <dt><b>Null pointer constants</b></dt>
1585 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1586 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1590 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1591 of floating point constants. For example, the form '<tt>double
1592 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1593 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1594 (and the only time that they are generated by the disassembler) is when a
1595 floating point constant must be emitted but it cannot be represented as a
1596 decimal floating point number. For example, NaN's, infinities, and other
1597 special values are represented in their IEEE hexadecimal format so that
1598 assembly and disassembly do not cause any bits to change in the constants.</p>
1602 <!-- ======================================================================= -->
1603 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1606 <div class="doc_text">
1607 <p>Aggregate constants arise from aggregation of simple constants
1608 and smaller aggregate constants.</p>
1611 <dt><b>Structure constants</b></dt>
1613 <dd>Structure constants are represented with notation similar to structure
1614 type definitions (a comma separated list of elements, surrounded by braces
1615 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1616 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1617 must have <a href="#t_struct">structure type</a>, and the number and
1618 types of elements must match those specified by the type.
1621 <dt><b>Array constants</b></dt>
1623 <dd>Array constants are represented with notation similar to array type
1624 definitions (a comma separated list of elements, surrounded by square brackets
1625 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1626 constants must have <a href="#t_array">array type</a>, and the number and
1627 types of elements must match those specified by the type.
1630 <dt><b>Vector constants</b></dt>
1632 <dd>Vector constants are represented with notation similar to vector type
1633 definitions (a comma separated list of elements, surrounded by
1634 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1635 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1636 href="#t_vector">vector type</a>, and the number and types of elements must
1637 match those specified by the type.
1640 <dt><b>Zero initialization</b></dt>
1642 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1643 value to zero of <em>any</em> type, including scalar and aggregate types.
1644 This is often used to avoid having to print large zero initializers (e.g. for
1645 large arrays) and is always exactly equivalent to using explicit zero
1652 <!-- ======================================================================= -->
1653 <div class="doc_subsection">
1654 <a name="globalconstants">Global Variable and Function Addresses</a>
1657 <div class="doc_text">
1659 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1660 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1661 constants. These constants are explicitly referenced when the <a
1662 href="#identifiers">identifier for the global</a> is used and always have <a
1663 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1666 <div class="doc_code">
1670 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1676 <!-- ======================================================================= -->
1677 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1678 <div class="doc_text">
1679 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1680 no specific value. Undefined values may be of any type and be used anywhere
1681 a constant is permitted.</p>
1683 <p>Undefined values indicate to the compiler that the program is well defined
1684 no matter what value is used, giving the compiler more freedom to optimize.
1688 <!-- ======================================================================= -->
1689 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1692 <div class="doc_text">
1694 <p>Constant expressions are used to allow expressions involving other constants
1695 to be used as constants. Constant expressions may be of any <a
1696 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1697 that does not have side effects (e.g. load and call are not supported). The
1698 following is the syntax for constant expressions:</p>
1701 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1702 <dd>Truncate a constant to another type. The bit size of CST must be larger
1703 than the bit size of TYPE. Both types must be integers.</dd>
1705 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1706 <dd>Zero extend a constant to another type. The bit size of CST must be
1707 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1709 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1710 <dd>Sign extend a constant to another type. The bit size of CST must be
1711 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1713 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1714 <dd>Truncate a floating point constant to another floating point type. The
1715 size of CST must be larger than the size of TYPE. Both types must be
1716 floating point.</dd>
1718 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1719 <dd>Floating point extend a constant to another type. The size of CST must be
1720 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1722 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1723 <dd>Convert a floating point constant to the corresponding unsigned integer
1724 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1725 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1726 of the same number of elements. If the value won't fit in the integer type,
1727 the results are undefined.</dd>
1729 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1730 <dd>Convert a floating point constant to the corresponding signed integer
1731 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1732 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1733 of the same number of elements. If the value won't fit in the integer type,
1734 the results are undefined.</dd>
1736 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1737 <dd>Convert an unsigned integer constant to the corresponding floating point
1738 constant. TYPE must be a scalar or vector floating point type. CST must be of
1739 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1740 of the same number of elements. If the value won't fit in the floating point
1741 type, the results are undefined.</dd>
1743 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1744 <dd>Convert a signed integer constant to the corresponding floating point
1745 constant. TYPE must be a scalar or vector floating point type. CST must be of
1746 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1747 of the same number of elements. If the value won't fit in the floating point
1748 type, the results are undefined.</dd>
1750 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1751 <dd>Convert a pointer typed constant to the corresponding integer constant
1752 TYPE must be an integer type. CST must be of pointer type. The CST value is
1753 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1755 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1756 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1757 pointer type. CST must be of integer type. The CST value is zero extended,
1758 truncated, or unchanged to make it fit in a pointer size. This one is
1759 <i>really</i> dangerous!</dd>
1761 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1762 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1763 identical (same number of bits). The conversion is done as if the CST value
1764 was stored to memory and read back as TYPE. In other words, no bits change
1765 with this operator, just the type. This can be used for conversion of
1766 vector types to any other type, as long as they have the same bit width. For
1767 pointers it is only valid to cast to another pointer type. It is not valid
1768 to bitcast to or from an aggregate type.
1771 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1773 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1774 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1775 instruction, the index list may have zero or more indexes, which are required
1776 to make sense for the type of "CSTPTR".</dd>
1778 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1780 <dd>Perform the <a href="#i_select">select operation</a> on
1783 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1784 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1786 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1787 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1789 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1790 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1792 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1793 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1795 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1797 <dd>Perform the <a href="#i_extractelement">extractelement
1798 operation</a> on constants.
1800 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1802 <dd>Perform the <a href="#i_insertelement">insertelement
1803 operation</a> on constants.</dd>
1806 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1808 <dd>Perform the <a href="#i_shufflevector">shufflevector
1809 operation</a> on constants.</dd>
1811 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1813 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1814 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1815 binary</a> operations. The constraints on operands are the same as those for
1816 the corresponding instruction (e.g. no bitwise operations on floating point
1817 values are allowed).</dd>
1821 <!-- *********************************************************************** -->
1822 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1823 <!-- *********************************************************************** -->
1825 <!-- ======================================================================= -->
1826 <div class="doc_subsection">
1827 <a name="inlineasm">Inline Assembler Expressions</a>
1830 <div class="doc_text">
1833 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1834 Module-Level Inline Assembly</a>) through the use of a special value. This
1835 value represents the inline assembler as a string (containing the instructions
1836 to emit), a list of operand constraints (stored as a string), and a flag that
1837 indicates whether or not the inline asm expression has side effects. An example
1838 inline assembler expression is:
1841 <div class="doc_code">
1843 i32 (i32) asm "bswap $0", "=r,r"
1848 Inline assembler expressions may <b>only</b> be used as the callee operand of
1849 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1852 <div class="doc_code">
1854 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1859 Inline asms with side effects not visible in the constraint list must be marked
1860 as having side effects. This is done through the use of the
1861 '<tt>sideeffect</tt>' keyword, like so:
1864 <div class="doc_code">
1866 call void asm sideeffect "eieio", ""()
1870 <p>TODO: The format of the asm and constraints string still need to be
1871 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1872 need to be documented). This is probably best done by reference to another
1873 document that covers inline asm from a holistic perspective.
1878 <!-- *********************************************************************** -->
1879 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1880 <!-- *********************************************************************** -->
1882 <div class="doc_text">
1884 <p>The LLVM instruction set consists of several different
1885 classifications of instructions: <a href="#terminators">terminator
1886 instructions</a>, <a href="#binaryops">binary instructions</a>,
1887 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1888 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1889 instructions</a>.</p>
1893 <!-- ======================================================================= -->
1894 <div class="doc_subsection"> <a name="terminators">Terminator
1895 Instructions</a> </div>
1897 <div class="doc_text">
1899 <p>As mentioned <a href="#functionstructure">previously</a>, every
1900 basic block in a program ends with a "Terminator" instruction, which
1901 indicates which block should be executed after the current block is
1902 finished. These terminator instructions typically yield a '<tt>void</tt>'
1903 value: they produce control flow, not values (the one exception being
1904 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1905 <p>There are six different terminator instructions: the '<a
1906 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1907 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1908 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1909 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1910 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1914 <!-- _______________________________________________________________________ -->
1915 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1916 Instruction</a> </div>
1917 <div class="doc_text">
1920 ret <type> <value> <i>; Return a value from a non-void function</i>
1921 ret void <i>; Return from void function</i>
1926 <p>The '<tt>ret</tt>' instruction is used to return control flow (and
1927 optionally a value) from a function back to the caller.</p>
1928 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1929 returns a value and then causes control flow, and one that just causes
1930 control flow to occur.</p>
1934 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument,
1935 the return value. The type of the return value must be a
1936 '<a href="#t_firstclass">first class</a>' type.</p>
1938 <p>A function is not <a href="#wellformed">well formed</a> if
1939 it it has a non-void return type and contains a '<tt>ret</tt>'
1940 instruction with no return value or a return value with a type that
1941 does not match its type, or if it has a void return type and contains
1942 a '<tt>ret</tt>' instruction with a return value.</p>
1946 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1947 returns back to the calling function's context. If the caller is a "<a
1948 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1949 the instruction after the call. If the caller was an "<a
1950 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1951 at the beginning of the "normal" destination block. If the instruction
1952 returns a value, that value shall set the call or invoke instruction's
1958 ret i32 5 <i>; Return an integer value of 5</i>
1959 ret void <i>; Return from a void function</i>
1960 ret { i32, i8 } { i32 4, i8 2 } <i>; Return an aggregate of values 4 and 2</i>
1963 <!-- _______________________________________________________________________ -->
1964 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1965 <div class="doc_text">
1967 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1970 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1971 transfer to a different basic block in the current function. There are
1972 two forms of this instruction, corresponding to a conditional branch
1973 and an unconditional branch.</p>
1975 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1976 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1977 unconditional form of the '<tt>br</tt>' instruction takes a single
1978 '<tt>label</tt>' value as a target.</p>
1980 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1981 argument is evaluated. If the value is <tt>true</tt>, control flows
1982 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1983 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1985 <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
1986 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1988 <!-- _______________________________________________________________________ -->
1989 <div class="doc_subsubsection">
1990 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1993 <div class="doc_text">
1997 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2002 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2003 several different places. It is a generalization of the '<tt>br</tt>'
2004 instruction, allowing a branch to occur to one of many possible
2010 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2011 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
2012 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
2013 table is not allowed to contain duplicate constant entries.</p>
2017 <p>The <tt>switch</tt> instruction specifies a table of values and
2018 destinations. When the '<tt>switch</tt>' instruction is executed, this
2019 table is searched for the given value. If the value is found, control flow is
2020 transfered to the corresponding destination; otherwise, control flow is
2021 transfered to the default destination.</p>
2023 <h5>Implementation:</h5>
2025 <p>Depending on properties of the target machine and the particular
2026 <tt>switch</tt> instruction, this instruction may be code generated in different
2027 ways. For example, it could be generated as a series of chained conditional
2028 branches or with a lookup table.</p>
2033 <i>; Emulate a conditional br instruction</i>
2034 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2035 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
2037 <i>; Emulate an unconditional br instruction</i>
2038 switch i32 0, label %dest [ ]
2040 <i>; Implement a jump table:</i>
2041 switch i32 %val, label %otherwise [ i32 0, label %onzero
2043 i32 2, label %ontwo ]
2047 <!-- _______________________________________________________________________ -->
2048 <div class="doc_subsubsection">
2049 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2052 <div class="doc_text">
2057 <result> = invoke [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] <ptr to function ty> <function ptr val>(<function args>) [<a href="#fnattrs">fn attrs</a>]
2058 to label <normal label> unwind label <exception label>
2063 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2064 function, with the possibility of control flow transfer to either the
2065 '<tt>normal</tt>' label or the
2066 '<tt>exception</tt>' label. If the callee function returns with the
2067 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
2068 "normal" label. If the callee (or any indirect callees) returns with the "<a
2069 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
2070 continued at the dynamically nearest "exception" label.
2074 <p>This instruction requires several arguments:</p>
2078 The optional "cconv" marker indicates which <a href="#callingconv">calling
2079 convention</a> the call should use. If none is specified, the call defaults
2080 to using C calling conventions.
2083 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2084 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
2085 and '<tt>inreg</tt>' attributes are valid here.</li>
2087 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2088 function value being invoked. In most cases, this is a direct function
2089 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2090 an arbitrary pointer to function value.
2093 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2094 function to be invoked. </li>
2096 <li>'<tt>function args</tt>': argument list whose types match the function
2097 signature argument types. If the function signature indicates the function
2098 accepts a variable number of arguments, the extra arguments can be
2101 <li>'<tt>normal label</tt>': the label reached when the called function
2102 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2104 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2105 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2107 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2108 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2109 '<tt>readnone</tt>' attributes are valid here.</li>
2114 <p>This instruction is designed to operate as a standard '<tt><a
2115 href="#i_call">call</a></tt>' instruction in most regards. The primary
2116 difference is that it establishes an association with a label, which is used by
2117 the runtime library to unwind the stack.</p>
2119 <p>This instruction is used in languages with destructors to ensure that proper
2120 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2121 exception. Additionally, this is important for implementation of
2122 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2126 %retval = invoke i32 @Test(i32 15) to label %Continue
2127 unwind label %TestCleanup <i>; {i32}:retval set</i>
2128 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2129 unwind label %TestCleanup <i>; {i32}:retval set</i>
2134 <!-- _______________________________________________________________________ -->
2136 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2137 Instruction</a> </div>
2139 <div class="doc_text">
2148 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2149 at the first callee in the dynamic call stack which used an <a
2150 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2151 primarily used to implement exception handling.</p>
2155 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2156 immediately halt. The dynamic call stack is then searched for the first <a
2157 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2158 execution continues at the "exceptional" destination block specified by the
2159 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2160 dynamic call chain, undefined behavior results.</p>
2163 <!-- _______________________________________________________________________ -->
2165 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2166 Instruction</a> </div>
2168 <div class="doc_text">
2177 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2178 instruction is used to inform the optimizer that a particular portion of the
2179 code is not reachable. This can be used to indicate that the code after a
2180 no-return function cannot be reached, and other facts.</p>
2184 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2189 <!-- ======================================================================= -->
2190 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2191 <div class="doc_text">
2192 <p>Binary operators are used to do most of the computation in a
2193 program. They require two operands of the same type, execute an operation on them, and
2194 produce a single value. The operands might represent
2195 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2196 The result value has the same type as its operands.</p>
2197 <p>There are several different binary operators:</p>
2199 <!-- _______________________________________________________________________ -->
2200 <div class="doc_subsubsection">
2201 <a name="i_add">'<tt>add</tt>' Instruction</a>
2204 <div class="doc_text">
2209 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2214 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2218 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2219 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2220 <a href="#t_vector">vector</a> values. Both arguments must have identical
2225 <p>The value produced is the integer or floating point sum of the two
2228 <p>If an integer sum has unsigned overflow, the result returned is the
2229 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2232 <p>Because LLVM integers use a two's complement representation, this
2233 instruction is appropriate for both signed and unsigned integers.</p>
2238 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2241 <!-- _______________________________________________________________________ -->
2242 <div class="doc_subsubsection">
2243 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2246 <div class="doc_text">
2251 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2256 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2259 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2260 '<tt>neg</tt>' instruction present in most other intermediate
2261 representations.</p>
2265 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2266 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2267 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2272 <p>The value produced is the integer or floating point difference of
2273 the two operands.</p>
2275 <p>If an integer difference has unsigned overflow, the result returned is the
2276 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2279 <p>Because LLVM integers use a two's complement representation, this
2280 instruction is appropriate for both signed and unsigned integers.</p>
2284 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2285 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2289 <!-- _______________________________________________________________________ -->
2290 <div class="doc_subsubsection">
2291 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2294 <div class="doc_text">
2297 <pre> <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2300 <p>The '<tt>mul</tt>' instruction returns the product of its two
2305 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2306 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2307 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2312 <p>The value produced is the integer or floating point product of the
2315 <p>If the result of an integer multiplication has unsigned overflow,
2316 the result returned is the mathematical result modulo
2317 2<sup>n</sup>, where n is the bit width of the result.</p>
2318 <p>Because LLVM integers use a two's complement representation, and the
2319 result is the same width as the operands, this instruction returns the
2320 correct result for both signed and unsigned integers. If a full product
2321 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2322 should be sign-extended or zero-extended as appropriate to the
2323 width of the full product.</p>
2325 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2329 <!-- _______________________________________________________________________ -->
2330 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2332 <div class="doc_text">
2334 <pre> <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2337 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2342 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2343 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2344 values. Both arguments must have identical types.</p>
2348 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2349 <p>Note that unsigned integer division and signed integer division are distinct
2350 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2351 <p>Division by zero leads to undefined behavior.</p>
2353 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2356 <!-- _______________________________________________________________________ -->
2357 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2359 <div class="doc_text">
2362 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2367 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2372 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2373 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2374 values. Both arguments must have identical types.</p>
2377 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2378 <p>Note that signed integer division and unsigned integer division are distinct
2379 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2380 <p>Division by zero leads to undefined behavior. Overflow also leads to
2381 undefined behavior; this is a rare case, but can occur, for example,
2382 by doing a 32-bit division of -2147483648 by -1.</p>
2384 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2387 <!-- _______________________________________________________________________ -->
2388 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2389 Instruction</a> </div>
2390 <div class="doc_text">
2393 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2397 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2402 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2403 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2404 of floating point values. Both arguments must have identical types.</p>
2408 <p>The value produced is the floating point quotient of the two operands.</p>
2413 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2417 <!-- _______________________________________________________________________ -->
2418 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2420 <div class="doc_text">
2422 <pre> <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2425 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2426 unsigned division of its two arguments.</p>
2428 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2429 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2430 values. Both arguments must have identical types.</p>
2432 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2433 This instruction always performs an unsigned division to get the remainder.</p>
2434 <p>Note that unsigned integer remainder and signed integer remainder are
2435 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2436 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2438 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2442 <!-- _______________________________________________________________________ -->
2443 <div class="doc_subsubsection">
2444 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2447 <div class="doc_text">
2452 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2457 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2458 signed division of its two operands. This instruction can also take
2459 <a href="#t_vector">vector</a> versions of the values in which case
2460 the elements must be integers.</p>
2464 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2465 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2466 values. Both arguments must have identical types.</p>
2470 <p>This instruction returns the <i>remainder</i> of a division (where the result
2471 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2472 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
2473 a value. For more information about the difference, see <a
2474 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2475 Math Forum</a>. For a table of how this is implemented in various languages,
2476 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2477 Wikipedia: modulo operation</a>.</p>
2478 <p>Note that signed integer remainder and unsigned integer remainder are
2479 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2480 <p>Taking the remainder of a division by zero leads to undefined behavior.
2481 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2482 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2483 (The remainder doesn't actually overflow, but this rule lets srem be
2484 implemented using instructions that return both the result of the division
2485 and the remainder.)</p>
2487 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2491 <!-- _______________________________________________________________________ -->
2492 <div class="doc_subsubsection">
2493 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2495 <div class="doc_text">
2498 <pre> <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2501 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2502 division of its two operands.</p>
2504 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2505 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2506 of floating point values. Both arguments must have identical types.</p>
2510 <p>This instruction returns the <i>remainder</i> of a division.
2511 The remainder has the same sign as the dividend.</p>
2516 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2520 <!-- ======================================================================= -->
2521 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2522 Operations</a> </div>
2523 <div class="doc_text">
2524 <p>Bitwise binary operators are used to do various forms of
2525 bit-twiddling in a program. They are generally very efficient
2526 instructions and can commonly be strength reduced from other
2527 instructions. They require two operands of the same type, execute an operation on them,
2528 and produce a single value. The resulting value is the same type as its operands.</p>
2531 <!-- _______________________________________________________________________ -->
2532 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2533 Instruction</a> </div>
2534 <div class="doc_text">
2536 <pre> <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2541 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2542 the left a specified number of bits.</p>
2546 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2547 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2548 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2552 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 2<sup>n</sup>,
2553 where n is the width of the result. If <tt>op2</tt> is (statically or dynamically) negative or
2554 equal to or larger than the number of bits in <tt>op1</tt>, the result is undefined.</p>
2556 <h5>Example:</h5><pre>
2557 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2558 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2559 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2560 <result> = shl i32 1, 32 <i>; undefined</i>
2563 <!-- _______________________________________________________________________ -->
2564 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2565 Instruction</a> </div>
2566 <div class="doc_text">
2568 <pre> <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2572 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2573 operand shifted to the right a specified number of bits with zero fill.</p>
2576 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2577 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2578 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2582 <p>This instruction always performs a logical shift right operation. The most
2583 significant bits of the result will be filled with zero bits after the
2584 shift. If <tt>op2</tt> is (statically or dynamically) equal to or larger than
2585 the number of bits in <tt>op1</tt>, the result is undefined.</p>
2589 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2590 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2591 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2592 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2593 <result> = lshr i32 1, 32 <i>; undefined</i>
2597 <!-- _______________________________________________________________________ -->
2598 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2599 Instruction</a> </div>
2600 <div class="doc_text">
2603 <pre> <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2607 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2608 operand shifted to the right a specified number of bits with sign extension.</p>
2611 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2612 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2613 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2616 <p>This instruction always performs an arithmetic shift right operation,
2617 The most significant bits of the result will be filled with the sign bit
2618 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
2619 larger than the number of bits in <tt>op1</tt>, the result is undefined.
2624 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2625 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2626 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2627 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2628 <result> = ashr i32 1, 32 <i>; undefined</i>
2632 <!-- _______________________________________________________________________ -->
2633 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2634 Instruction</a> </div>
2636 <div class="doc_text">
2641 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2646 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2647 its two operands.</p>
2651 <p>The two arguments to the '<tt>and</tt>' instruction must be
2652 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2653 values. Both arguments must have identical types.</p>
2656 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2659 <table border="1" cellspacing="0" cellpadding="4">
2691 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2692 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2693 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2696 <!-- _______________________________________________________________________ -->
2697 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2698 <div class="doc_text">
2700 <pre> <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2703 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2704 or of its two operands.</p>
2707 <p>The two arguments to the '<tt>or</tt>' instruction must be
2708 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2709 values. Both arguments must have identical types.</p>
2711 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2714 <table border="1" cellspacing="0" cellpadding="4">
2745 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2746 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2747 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2750 <!-- _______________________________________________________________________ -->
2751 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2752 Instruction</a> </div>
2753 <div class="doc_text">
2755 <pre> <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2758 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2759 or of its two operands. The <tt>xor</tt> is used to implement the
2760 "one's complement" operation, which is the "~" operator in C.</p>
2762 <p>The two arguments to the '<tt>xor</tt>' instruction must be
2763 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2764 values. Both arguments must have identical types.</p>
2768 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2771 <table border="1" cellspacing="0" cellpadding="4">
2803 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2804 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2805 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2806 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2810 <!-- ======================================================================= -->
2811 <div class="doc_subsection">
2812 <a name="vectorops">Vector Operations</a>
2815 <div class="doc_text">
2817 <p>LLVM supports several instructions to represent vector operations in a
2818 target-independent manner. These instructions cover the element-access and
2819 vector-specific operations needed to process vectors effectively. While LLVM
2820 does directly support these vector operations, many sophisticated algorithms
2821 will want to use target-specific intrinsics to take full advantage of a specific
2826 <!-- _______________________________________________________________________ -->
2827 <div class="doc_subsubsection">
2828 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2831 <div class="doc_text">
2836 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2842 The '<tt>extractelement</tt>' instruction extracts a single scalar
2843 element from a vector at a specified index.
2850 The first operand of an '<tt>extractelement</tt>' instruction is a
2851 value of <a href="#t_vector">vector</a> type. The second operand is
2852 an index indicating the position from which to extract the element.
2853 The index may be a variable.</p>
2858 The result is a scalar of the same type as the element type of
2859 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2860 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2861 results are undefined.
2867 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2872 <!-- _______________________________________________________________________ -->
2873 <div class="doc_subsubsection">
2874 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2877 <div class="doc_text">
2882 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2888 The '<tt>insertelement</tt>' instruction inserts a scalar
2889 element into a vector at a specified index.
2896 The first operand of an '<tt>insertelement</tt>' instruction is a
2897 value of <a href="#t_vector">vector</a> type. The second operand is a
2898 scalar value whose type must equal the element type of the first
2899 operand. The third operand is an index indicating the position at
2900 which to insert the value. The index may be a variable.</p>
2905 The result is a vector of the same type as <tt>val</tt>. Its
2906 element values are those of <tt>val</tt> except at position
2907 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2908 exceeds the length of <tt>val</tt>, the results are undefined.
2914 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2918 <!-- _______________________________________________________________________ -->
2919 <div class="doc_subsubsection">
2920 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2923 <div class="doc_text">
2928 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2934 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2935 from two input vectors, returning a vector of the same type.
2941 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2942 with types that match each other and types that match the result of the
2943 instruction. The third argument is a shuffle mask, which has the same number
2944 of elements as the other vector type, but whose element type is always 'i32'.
2948 The shuffle mask operand is required to be a constant vector with either
2949 constant integer or undef values.
2955 The elements of the two input vectors are numbered from left to right across
2956 both of the vectors. The shuffle mask operand specifies, for each element of
2957 the result vector, which element of the two input registers the result element
2958 gets. The element selector may be undef (meaning "don't care") and the second
2959 operand may be undef if performing a shuffle from only one vector.
2965 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2966 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2967 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2968 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2973 <!-- ======================================================================= -->
2974 <div class="doc_subsection">
2975 <a name="aggregateops">Aggregate Operations</a>
2978 <div class="doc_text">
2980 <p>LLVM supports several instructions for working with aggregate values.
2985 <!-- _______________________________________________________________________ -->
2986 <div class="doc_subsubsection">
2987 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
2990 <div class="doc_text">
2995 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3001 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3002 or array element from an aggregate value.
3009 The first operand of an '<tt>extractvalue</tt>' instruction is a
3010 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
3011 type. The operands are constant indices to specify which value to extract
3012 in a similar manner as indices in a
3013 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3019 The result is the value at the position in the aggregate specified by
3026 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3031 <!-- _______________________________________________________________________ -->
3032 <div class="doc_subsubsection">
3033 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3036 <div class="doc_text">
3041 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3047 The '<tt>insertvalue</tt>' instruction inserts a value
3048 into a struct field or array element in an aggregate.
3055 The first operand of an '<tt>insertvalue</tt>' instruction is a
3056 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
3057 The second operand is a first-class value to insert.
3058 The following operands are constant indices
3059 indicating the position at which to insert the value in a similar manner as
3061 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3062 The value to insert must have the same type as the value identified
3068 The result is an aggregate of the same type as <tt>val</tt>. Its
3069 value is that of <tt>val</tt> except that the value at the position
3070 specified by the indices is that of <tt>elt</tt>.
3076 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3081 <!-- ======================================================================= -->
3082 <div class="doc_subsection">
3083 <a name="memoryops">Memory Access and Addressing Operations</a>
3086 <div class="doc_text">
3088 <p>A key design point of an SSA-based representation is how it
3089 represents memory. In LLVM, no memory locations are in SSA form, which
3090 makes things very simple. This section describes how to read, write,
3091 allocate, and free memory in LLVM.</p>
3095 <!-- _______________________________________________________________________ -->
3096 <div class="doc_subsubsection">
3097 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3100 <div class="doc_text">
3105 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3110 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3111 heap and returns a pointer to it. The object is always allocated in the generic
3112 address space (address space zero).</p>
3116 <p>The '<tt>malloc</tt>' instruction allocates
3117 <tt>sizeof(<type>)*NumElements</tt>
3118 bytes of memory from the operating system and returns a pointer of the
3119 appropriate type to the program. If "NumElements" is specified, it is the
3120 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3121 If a constant alignment is specified, the value result of the allocation is guaranteed to
3122 be aligned to at least that boundary. If not specified, or if zero, the target can
3123 choose to align the allocation on any convenient boundary.</p>
3125 <p>'<tt>type</tt>' must be a sized type.</p>
3129 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3130 a pointer is returned. The result of a zero byte allocattion is undefined. The
3131 result is null if there is insufficient memory available.</p>
3136 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
3138 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3139 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3140 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3141 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3142 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3146 <!-- _______________________________________________________________________ -->
3147 <div class="doc_subsubsection">
3148 <a name="i_free">'<tt>free</tt>' Instruction</a>
3151 <div class="doc_text">
3156 free <type> <value> <i>; yields {void}</i>
3161 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3162 memory heap to be reallocated in the future.</p>
3166 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3167 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3172 <p>Access to the memory pointed to by the pointer is no longer defined
3173 after this instruction executes. If the pointer is null, the operation
3179 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3180 free [4 x i8]* %array
3184 <!-- _______________________________________________________________________ -->
3185 <div class="doc_subsubsection">
3186 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3189 <div class="doc_text">
3194 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3199 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3200 currently executing function, to be automatically released when this function
3201 returns to its caller. The object is always allocated in the generic address
3202 space (address space zero).</p>
3206 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3207 bytes of memory on the runtime stack, returning a pointer of the
3208 appropriate type to the program. If "NumElements" is specified, it is the
3209 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3210 If a constant alignment is specified, the value result of the allocation is guaranteed
3211 to be aligned to at least that boundary. If not specified, or if zero, the target
3212 can choose to align the allocation on any convenient boundary.</p>
3214 <p>'<tt>type</tt>' may be any sized type.</p>
3218 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3219 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3220 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3221 instruction is commonly used to represent automatic variables that must
3222 have an address available. When the function returns (either with the <tt><a
3223 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3224 instructions), the memory is reclaimed. Allocating zero bytes
3225 is legal, but the result is undefined.</p>
3230 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3231 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3232 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3233 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3237 <!-- _______________________________________________________________________ -->
3238 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3239 Instruction</a> </div>
3240 <div class="doc_text">
3242 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3244 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3246 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3247 address from which to load. The pointer must point to a <a
3248 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3249 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3250 the number or order of execution of this <tt>load</tt> with other
3251 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3254 The optional constant "align" argument specifies the alignment of the operation
3255 (that is, the alignment of the memory address). A value of 0 or an
3256 omitted "align" argument means that the operation has the preferential
3257 alignment for the target. It is the responsibility of the code emitter
3258 to ensure that the alignment information is correct. Overestimating
3259 the alignment results in an undefined behavior. Underestimating the
3260 alignment may produce less efficient code. An alignment of 1 is always
3264 <p>The location of memory pointed to is loaded.</p>
3266 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3268 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3269 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3272 <!-- _______________________________________________________________________ -->
3273 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3274 Instruction</a> </div>
3275 <div class="doc_text">
3277 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3278 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3281 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3283 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3284 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3285 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3286 of the '<tt><value></tt>'
3287 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3288 optimizer is not allowed to modify the number or order of execution of
3289 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3290 href="#i_store">store</a></tt> instructions.</p>
3292 The optional constant "align" argument specifies the alignment of the operation
3293 (that is, the alignment of the memory address). A value of 0 or an
3294 omitted "align" argument means that the operation has the preferential
3295 alignment for the target. It is the responsibility of the code emitter
3296 to ensure that the alignment information is correct. Overestimating
3297 the alignment results in an undefined behavior. Underestimating the
3298 alignment may produce less efficient code. An alignment of 1 is always
3302 <p>The contents of memory are updated to contain '<tt><value></tt>'
3303 at the location specified by the '<tt><pointer></tt>' operand.</p>
3305 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3306 store i32 3, i32* %ptr <i>; yields {void}</i>
3307 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3311 <!-- _______________________________________________________________________ -->
3312 <div class="doc_subsubsection">
3313 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3316 <div class="doc_text">
3319 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
3325 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3326 subelement of an aggregate data structure. It performs address calculation only
3327 and does not access memory.</p>
3331 <p>The first argument is always a pointer, and forms the basis of the
3332 calculation. The remaining arguments are indices, that indicate which of the
3333 elements of the aggregate object are indexed. The interpretation of each index
3334 is dependent on the type being indexed into. The first index always indexes the
3335 pointer value given as the first argument, the second index indexes a value of
3336 the type pointed to (not necessarily the value directly pointed to, since the
3337 first index can be non-zero), etc. The first type indexed into must be a pointer
3338 value, subsequent types can be arrays, vectors and structs. Note that subsequent
3339 types being indexed into can never be pointers, since that would require loading
3340 the pointer before continuing calculation.</p>
3342 <p>The type of each index argument depends on the type it is indexing into.
3343 When indexing into a (packed) structure, only <tt>i32</tt> integer
3344 <b>constants</b> are allowed. When indexing into an array, pointer or vector,
3345 only integers of 32 or 64 bits are allowed (also non-constants). 32-bit values
3346 will be sign extended to 64-bits if required.</p>
3348 <p>For example, let's consider a C code fragment and how it gets
3349 compiled to LLVM:</p>
3351 <div class="doc_code">
3364 int *foo(struct ST *s) {
3365 return &s[1].Z.B[5][13];
3370 <p>The LLVM code generated by the GCC frontend is:</p>
3372 <div class="doc_code">
3374 %RT = type { i8 , [10 x [20 x i32]], i8 }
3375 %ST = type { i32, double, %RT }
3377 define i32* %foo(%ST* %s) {
3379 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3387 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3388 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3389 }</tt>' type, a structure. The second index indexes into the third element of
3390 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3391 i8 }</tt>' type, another structure. The third index indexes into the second
3392 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3393 array. The two dimensions of the array are subscripted into, yielding an
3394 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3395 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3397 <p>Note that it is perfectly legal to index partially through a
3398 structure, returning a pointer to an inner element. Because of this,
3399 the LLVM code for the given testcase is equivalent to:</p>
3402 define i32* %foo(%ST* %s) {
3403 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3404 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3405 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3406 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3407 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3412 <p>Note that it is undefined to access an array out of bounds: array and
3413 pointer indexes must always be within the defined bounds of the array type.
3414 The one exception for this rule is zero length arrays. These arrays are
3415 defined to be accessible as variable length arrays, which requires access
3416 beyond the zero'th element.</p>
3418 <p>The getelementptr instruction is often confusing. For some more insight
3419 into how it works, see <a href="GetElementPtr.html">the getelementptr
3425 <i>; yields [12 x i8]*:aptr</i>
3426 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
3427 <i>; yields i8*:vptr</i>
3428 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
3429 <i>; yields i8*:eptr</i>
3430 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
3434 <!-- ======================================================================= -->
3435 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3437 <div class="doc_text">
3438 <p>The instructions in this category are the conversion instructions (casting)
3439 which all take a single operand and a type. They perform various bit conversions
3443 <!-- _______________________________________________________________________ -->
3444 <div class="doc_subsubsection">
3445 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3447 <div class="doc_text">
3451 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3456 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3461 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3462 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3463 and type of the result, which must be an <a href="#t_integer">integer</a>
3464 type. The bit size of <tt>value</tt> must be larger than the bit size of
3465 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3469 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3470 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3471 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3472 It will always truncate bits.</p>
3476 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3477 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3478 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3482 <!-- _______________________________________________________________________ -->
3483 <div class="doc_subsubsection">
3484 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3486 <div class="doc_text">
3490 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3494 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3499 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3500 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3501 also be of <a href="#t_integer">integer</a> type. The bit size of the
3502 <tt>value</tt> must be smaller than the bit size of the destination type,
3506 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3507 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3509 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3513 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3514 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3518 <!-- _______________________________________________________________________ -->
3519 <div class="doc_subsubsection">
3520 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3522 <div class="doc_text">
3526 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3530 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3534 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3535 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3536 also be of <a href="#t_integer">integer</a> type. The bit size of the
3537 <tt>value</tt> must be smaller than the bit size of the destination type,
3542 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3543 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3544 the type <tt>ty2</tt>.</p>
3546 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3550 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3551 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3555 <!-- _______________________________________________________________________ -->
3556 <div class="doc_subsubsection">
3557 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3560 <div class="doc_text">
3565 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3569 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3574 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3575 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3576 cast it to. The size of <tt>value</tt> must be larger than the size of
3577 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3578 <i>no-op cast</i>.</p>
3581 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3582 <a href="#t_floating">floating point</a> type to a smaller
3583 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3584 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3588 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3589 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3593 <!-- _______________________________________________________________________ -->
3594 <div class="doc_subsubsection">
3595 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3597 <div class="doc_text">
3601 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3605 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3606 floating point value.</p>
3609 <p>The '<tt>fpext</tt>' instruction takes a
3610 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3611 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3612 type must be smaller than the destination type.</p>
3615 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3616 <a href="#t_floating">floating point</a> type to a larger
3617 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3618 used to make a <i>no-op cast</i> because it always changes bits. Use
3619 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3623 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3624 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3628 <!-- _______________________________________________________________________ -->
3629 <div class="doc_subsubsection">
3630 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3632 <div class="doc_text">
3636 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3640 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3641 unsigned integer equivalent of type <tt>ty2</tt>.
3645 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3646 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3647 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3648 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3649 vector integer type with the same number of elements as <tt>ty</tt></p>
3652 <p> The '<tt>fptoui</tt>' instruction converts its
3653 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3654 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3655 the results are undefined.</p>
3659 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3660 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3661 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3665 <!-- _______________________________________________________________________ -->
3666 <div class="doc_subsubsection">
3667 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3669 <div class="doc_text">
3673 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3677 <p>The '<tt>fptosi</tt>' instruction converts
3678 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3682 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3683 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3684 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3685 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3686 vector integer type with the same number of elements as <tt>ty</tt></p>
3689 <p>The '<tt>fptosi</tt>' instruction converts its
3690 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3691 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3692 the results are undefined.</p>
3696 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3697 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3698 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3702 <!-- _______________________________________________________________________ -->
3703 <div class="doc_subsubsection">
3704 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3706 <div class="doc_text">
3710 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3714 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3715 integer and converts that value to the <tt>ty2</tt> type.</p>
3718 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3719 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3720 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3721 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3722 floating point type with the same number of elements as <tt>ty</tt></p>
3725 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3726 integer quantity and converts it to the corresponding floating point value. If
3727 the value cannot fit in the floating point value, the results are undefined.</p>
3731 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3732 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3736 <!-- _______________________________________________________________________ -->
3737 <div class="doc_subsubsection">
3738 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3740 <div class="doc_text">
3744 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3748 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3749 integer and converts that value to the <tt>ty2</tt> type.</p>
3752 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3753 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3754 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3755 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3756 floating point type with the same number of elements as <tt>ty</tt></p>
3759 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3760 integer quantity and converts it to the corresponding floating point value. If
3761 the value cannot fit in the floating point value, the results are undefined.</p>
3765 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3766 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3770 <!-- _______________________________________________________________________ -->
3771 <div class="doc_subsubsection">
3772 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3774 <div class="doc_text">
3778 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3782 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3783 the integer type <tt>ty2</tt>.</p>
3786 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3787 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3788 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3791 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3792 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3793 truncating or zero extending that value to the size of the integer type. If
3794 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3795 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3796 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3801 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3802 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3806 <!-- _______________________________________________________________________ -->
3807 <div class="doc_subsubsection">
3808 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3810 <div class="doc_text">
3814 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3818 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3819 a pointer type, <tt>ty2</tt>.</p>
3822 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3823 value to cast, and a type to cast it to, which must be a
3824 <a href="#t_pointer">pointer</a> type.
3827 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3828 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3829 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3830 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3831 the size of a pointer then a zero extension is done. If they are the same size,
3832 nothing is done (<i>no-op cast</i>).</p>
3836 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3837 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3838 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3842 <!-- _______________________________________________________________________ -->
3843 <div class="doc_subsubsection">
3844 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3846 <div class="doc_text">
3850 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3855 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3856 <tt>ty2</tt> without changing any bits.</p>
3860 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3861 a non-aggregate first class value, and a type to cast it to, which must also be
3862 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes of
3864 and the destination type, <tt>ty2</tt>, must be identical. If the source
3865 type is a pointer, the destination type must also be a pointer. This
3866 instruction supports bitwise conversion of vectors to integers and to vectors
3867 of other types (as long as they have the same size).</p>
3870 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3871 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3872 this conversion. The conversion is done as if the <tt>value</tt> had been
3873 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3874 converted to other pointer types with this instruction. To convert pointers to
3875 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3876 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3880 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3881 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3882 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3886 <!-- ======================================================================= -->
3887 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3888 <div class="doc_text">
3889 <p>The instructions in this category are the "miscellaneous"
3890 instructions, which defy better classification.</p>
3893 <!-- _______________________________________________________________________ -->
3894 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3896 <div class="doc_text">
3898 <pre> <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
3901 <p>The '<tt>icmp</tt>' instruction returns a boolean value or
3902 a vector of boolean values based on comparison
3903 of its two integer, integer vector, or pointer operands.</p>
3905 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3906 the condition code indicating the kind of comparison to perform. It is not
3907 a value, just a keyword. The possible condition code are:
3909 <li><tt>eq</tt>: equal</li>
3910 <li><tt>ne</tt>: not equal </li>
3911 <li><tt>ugt</tt>: unsigned greater than</li>
3912 <li><tt>uge</tt>: unsigned greater or equal</li>
3913 <li><tt>ult</tt>: unsigned less than</li>
3914 <li><tt>ule</tt>: unsigned less or equal</li>
3915 <li><tt>sgt</tt>: signed greater than</li>
3916 <li><tt>sge</tt>: signed greater or equal</li>
3917 <li><tt>slt</tt>: signed less than</li>
3918 <li><tt>sle</tt>: signed less or equal</li>
3920 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3921 <a href="#t_pointer">pointer</a>
3922 or integer <a href="#t_vector">vector</a> typed.
3923 They must also be identical types.</p>
3925 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to
3926 the condition code given as <tt>cond</tt>. The comparison performed always
3927 yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
3929 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3930 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3932 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3933 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3934 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3935 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3936 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3937 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3938 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3939 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
3940 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3941 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3942 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3943 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3944 <li><tt>sge</tt>: interprets the operands as signed values and yields
3945 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3946 <li><tt>slt</tt>: interprets the operands as signed values and yields
3947 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
3948 <li><tt>sle</tt>: interprets the operands as signed values and yields
3949 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3951 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3952 values are compared as if they were integers.</p>
3953 <p>If the operands are integer vectors, then they are compared
3954 element by element. The result is an <tt>i1</tt> vector with
3955 the same number of elements as the values being compared.
3956 Otherwise, the result is an <tt>i1</tt>.
3960 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3961 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3962 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3963 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3964 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3965 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3969 <!-- _______________________________________________________________________ -->
3970 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3972 <div class="doc_text">
3974 <pre> <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
3977 <p>The '<tt>fcmp</tt>' instruction returns a boolean value
3978 or vector of boolean values based on comparison
3981 If the operands are floating point scalars, then the result
3982 type is a boolean (<a href="#t_primitive"><tt>i1</tt></a>).
3984 <p>If the operands are floating point vectors, then the result type
3985 is a vector of boolean with the same number of elements as the
3986 operands being compared.</p>
3988 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3989 the condition code indicating the kind of comparison to perform. It is not
3990 a value, just a keyword. The possible condition code are:
3992 <li><tt>false</tt>: no comparison, always returns false</li>
3993 <li><tt>oeq</tt>: ordered and equal</li>
3994 <li><tt>ogt</tt>: ordered and greater than </li>
3995 <li><tt>oge</tt>: ordered and greater than or equal</li>
3996 <li><tt>olt</tt>: ordered and less than </li>
3997 <li><tt>ole</tt>: ordered and less than or equal</li>
3998 <li><tt>one</tt>: ordered and not equal</li>
3999 <li><tt>ord</tt>: ordered (no nans)</li>
4000 <li><tt>ueq</tt>: unordered or equal</li>
4001 <li><tt>ugt</tt>: unordered or greater than </li>
4002 <li><tt>uge</tt>: unordered or greater than or equal</li>
4003 <li><tt>ult</tt>: unordered or less than </li>
4004 <li><tt>ule</tt>: unordered or less than or equal</li>
4005 <li><tt>une</tt>: unordered or not equal</li>
4006 <li><tt>uno</tt>: unordered (either nans)</li>
4007 <li><tt>true</tt>: no comparison, always returns true</li>
4009 <p><i>Ordered</i> means that neither operand is a QNAN while
4010 <i>unordered</i> means that either operand may be a QNAN.</p>
4011 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be
4012 either a <a href="#t_floating">floating point</a> type
4013 or a <a href="#t_vector">vector</a> of floating point type.
4014 They must have identical types.</p>
4016 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4017 according to the condition code given as <tt>cond</tt>.
4018 If the operands are vectors, then the vectors are compared
4020 Each comparison performed
4021 always yields an <a href="#t_primitive">i1</a> result, as follows:
4023 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4024 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4025 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4026 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4027 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4028 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4029 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4030 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4031 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4032 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4033 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4034 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4035 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4036 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4037 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4038 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4039 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4040 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4041 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4042 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4043 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4044 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4045 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4046 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4047 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4048 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4049 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4050 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4054 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4055 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4056 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4057 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4061 <!-- _______________________________________________________________________ -->
4062 <div class="doc_subsubsection">
4063 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
4065 <div class="doc_text">
4067 <pre> <result> = vicmp <cond> <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4070 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
4071 element-wise comparison of its two integer vector operands.</p>
4073 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
4074 the condition code indicating the kind of comparison to perform. It is not
4075 a value, just a keyword. The possible condition code are:
4077 <li><tt>eq</tt>: equal</li>
4078 <li><tt>ne</tt>: not equal </li>
4079 <li><tt>ugt</tt>: unsigned greater than</li>
4080 <li><tt>uge</tt>: unsigned greater or equal</li>
4081 <li><tt>ult</tt>: unsigned less than</li>
4082 <li><tt>ule</tt>: unsigned less or equal</li>
4083 <li><tt>sgt</tt>: signed greater than</li>
4084 <li><tt>sge</tt>: signed greater or equal</li>
4085 <li><tt>slt</tt>: signed less than</li>
4086 <li><tt>sle</tt>: signed less or equal</li>
4088 <p>The remaining two arguments must be <a href="#t_vector">vector</a> or
4089 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
4091 <p>The '<tt>vicmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4092 according to the condition code given as <tt>cond</tt>. The comparison yields a
4093 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
4094 identical type as the values being compared. The most significant bit in each
4095 element is 1 if the element-wise comparison evaluates to true, and is 0
4096 otherwise. All other bits of the result are undefined. The condition codes
4097 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
4102 <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>
4103 <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>
4107 <!-- _______________________________________________________________________ -->
4108 <div class="doc_subsubsection">
4109 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
4111 <div class="doc_text">
4113 <pre> <result> = vfcmp <cond> <ty> <op1>, <op2></pre>
4115 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4116 element-wise comparison of its two floating point vector operands. The output
4117 elements have the same width as the input elements.</p>
4119 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4120 the condition code indicating the kind of comparison to perform. It is not
4121 a value, just a keyword. The possible condition code are:
4123 <li><tt>false</tt>: no comparison, always returns false</li>
4124 <li><tt>oeq</tt>: ordered and equal</li>
4125 <li><tt>ogt</tt>: ordered and greater than </li>
4126 <li><tt>oge</tt>: ordered and greater than or equal</li>
4127 <li><tt>olt</tt>: ordered and less than </li>
4128 <li><tt>ole</tt>: ordered and less than or equal</li>
4129 <li><tt>one</tt>: ordered and not equal</li>
4130 <li><tt>ord</tt>: ordered (no nans)</li>
4131 <li><tt>ueq</tt>: unordered or equal</li>
4132 <li><tt>ugt</tt>: unordered or greater than </li>
4133 <li><tt>uge</tt>: unordered or greater than or equal</li>
4134 <li><tt>ult</tt>: unordered or less than </li>
4135 <li><tt>ule</tt>: unordered or less than or equal</li>
4136 <li><tt>une</tt>: unordered or not equal</li>
4137 <li><tt>uno</tt>: unordered (either nans)</li>
4138 <li><tt>true</tt>: no comparison, always returns true</li>
4140 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4141 <a href="#t_floating">floating point</a> typed. They must also be identical
4144 <p>The '<tt>vfcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4145 according to the condition code given as <tt>cond</tt>. The comparison yields a
4146 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4147 an identical number of elements as the values being compared, and each element
4148 having identical with to the width of the floating point elements. The most
4149 significant bit in each element is 1 if the element-wise comparison evaluates to
4150 true, and is 0 otherwise. All other bits of the result are undefined. The
4151 condition codes are evaluated identically to the
4152 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.
4156 <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>
4157 <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>
4161 <!-- _______________________________________________________________________ -->
4162 <div class="doc_subsubsection">
4163 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4166 <div class="doc_text">
4170 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4172 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4173 the SSA graph representing the function.</p>
4176 <p>The type of the incoming values is specified with the first type
4177 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4178 as arguments, with one pair for each predecessor basic block of the
4179 current block. Only values of <a href="#t_firstclass">first class</a>
4180 type may be used as the value arguments to the PHI node. Only labels
4181 may be used as the label arguments.</p>
4183 <p>There must be no non-phi instructions between the start of a basic
4184 block and the PHI instructions: i.e. PHI instructions must be first in
4189 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4190 specified by the pair corresponding to the predecessor basic block that executed
4191 just prior to the current block.</p>
4195 Loop: ; Infinite loop that counts from 0 on up...
4196 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4197 %nextindvar = add i32 %indvar, 1
4202 <!-- _______________________________________________________________________ -->
4203 <div class="doc_subsubsection">
4204 <a name="i_select">'<tt>select</tt>' Instruction</a>
4207 <div class="doc_text">
4212 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4214 <i>selty</i> is either i1 or {<N x i1>}
4220 The '<tt>select</tt>' instruction is used to choose one value based on a
4221 condition, without branching.
4228 The '<tt>select</tt>' instruction requires an 'i1' value or
4229 a vector of 'i1' values indicating the
4230 condition, and two values of the same <a href="#t_firstclass">first class</a>
4231 type. If the val1/val2 are vectors and
4232 the condition is a scalar, then entire vectors are selected, not
4233 individual elements.
4239 If the condition is an i1 and it evaluates to 1, the instruction returns the first
4240 value argument; otherwise, it returns the second value argument.
4243 If the condition is a vector of i1, then the value arguments must
4244 be vectors of the same size, and the selection is done element
4251 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4256 <!-- _______________________________________________________________________ -->
4257 <div class="doc_subsubsection">
4258 <a name="i_call">'<tt>call</tt>' Instruction</a>
4261 <div class="doc_text">
4265 <result> = [tail] call [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] <ty> [<fnty>*] <fnptrval>(<function args>) [<a href="#fnattrs">fn attrs</a>]
4270 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4274 <p>This instruction requires several arguments:</p>
4278 <p>The optional "tail" marker indicates whether the callee function accesses
4279 any allocas or varargs in the caller. If the "tail" marker is present, the
4280 function call is eligible for tail call optimization. Note that calls may
4281 be marked "tail" even if they do not occur before a <a
4282 href="#i_ret"><tt>ret</tt></a> instruction.
4285 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4286 convention</a> the call should use. If none is specified, the call defaults
4287 to using C calling conventions.
4291 <p>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4292 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
4293 and '<tt>inreg</tt>' attributes are valid here.</p>
4297 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4298 the type of the return value. Functions that return no value are marked
4299 <tt><a href="#t_void">void</a></tt>.</p>
4302 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4303 value being invoked. The argument types must match the types implied by
4304 this signature. This type can be omitted if the function is not varargs
4305 and if the function type does not return a pointer to a function.</p>
4308 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4309 be invoked. In most cases, this is a direct function invocation, but
4310 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4311 to function value.</p>
4314 <p>'<tt>function args</tt>': argument list whose types match the
4315 function signature argument types. All arguments must be of
4316 <a href="#t_firstclass">first class</a> type. If the function signature
4317 indicates the function accepts a variable number of arguments, the extra
4318 arguments can be specified.</p>
4321 <p>The optional <a href="#fnattrs">function attributes</a> list. Only
4322 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4323 '<tt>readnone</tt>' attributes are valid here.</p>
4329 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4330 transfer to a specified function, with its incoming arguments bound to
4331 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4332 instruction in the called function, control flow continues with the
4333 instruction after the function call, and the return value of the
4334 function is bound to the result argument.
4339 %retval = call i32 @test(i32 %argc)
4340 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4341 %X = tail call i32 @foo() <i>; yields i32</i>
4342 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4343 call void %foo(i8 97 signext)
4345 %struct.A = type { i32, i8 }
4346 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4347 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4348 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
4349 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
4350 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
4355 <!-- _______________________________________________________________________ -->
4356 <div class="doc_subsubsection">
4357 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4360 <div class="doc_text">
4365 <resultval> = va_arg <va_list*> <arglist>, <argty>
4370 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4371 the "variable argument" area of a function call. It is used to implement the
4372 <tt>va_arg</tt> macro in C.</p>
4376 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4377 the argument. It returns a value of the specified argument type and
4378 increments the <tt>va_list</tt> to point to the next argument. The
4379 actual type of <tt>va_list</tt> is target specific.</p>
4383 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4384 type from the specified <tt>va_list</tt> and causes the
4385 <tt>va_list</tt> to point to the next argument. For more information,
4386 see the variable argument handling <a href="#int_varargs">Intrinsic
4389 <p>It is legal for this instruction to be called in a function which does not
4390 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4393 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4394 href="#intrinsics">intrinsic function</a> because it takes a type as an
4399 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4403 <!-- *********************************************************************** -->
4404 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4405 <!-- *********************************************************************** -->
4407 <div class="doc_text">
4409 <p>LLVM supports the notion of an "intrinsic function". These functions have
4410 well known names and semantics and are required to follow certain restrictions.
4411 Overall, these intrinsics represent an extension mechanism for the LLVM
4412 language that does not require changing all of the transformations in LLVM when
4413 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4415 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4416 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4417 begin with this prefix. Intrinsic functions must always be external functions:
4418 you cannot define the body of intrinsic functions. Intrinsic functions may
4419 only be used in call or invoke instructions: it is illegal to take the address
4420 of an intrinsic function. Additionally, because intrinsic functions are part
4421 of the LLVM language, it is required if any are added that they be documented
4424 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4425 a family of functions that perform the same operation but on different data
4426 types. Because LLVM can represent over 8 million different integer types,
4427 overloading is used commonly to allow an intrinsic function to operate on any
4428 integer type. One or more of the argument types or the result type can be
4429 overloaded to accept any integer type. Argument types may also be defined as
4430 exactly matching a previous argument's type or the result type. This allows an
4431 intrinsic function which accepts multiple arguments, but needs all of them to
4432 be of the same type, to only be overloaded with respect to a single argument or
4435 <p>Overloaded intrinsics will have the names of its overloaded argument types
4436 encoded into its function name, each preceded by a period. Only those types
4437 which are overloaded result in a name suffix. Arguments whose type is matched
4438 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4439 take an integer of any width and returns an integer of exactly the same integer
4440 width. This leads to a family of functions such as
4441 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4442 Only one type, the return type, is overloaded, and only one type suffix is
4443 required. Because the argument's type is matched against the return type, it
4444 does not require its own name suffix.</p>
4446 <p>To learn how to add an intrinsic function, please see the
4447 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4452 <!-- ======================================================================= -->
4453 <div class="doc_subsection">
4454 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4457 <div class="doc_text">
4459 <p>Variable argument support is defined in LLVM with the <a
4460 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4461 intrinsic functions. These functions are related to the similarly
4462 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4464 <p>All of these functions operate on arguments that use a
4465 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4466 language reference manual does not define what this type is, so all
4467 transformations should be prepared to handle these functions regardless of
4470 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4471 instruction and the variable argument handling intrinsic functions are
4474 <div class="doc_code">
4476 define i32 @test(i32 %X, ...) {
4477 ; Initialize variable argument processing
4479 %ap2 = bitcast i8** %ap to i8*
4480 call void @llvm.va_start(i8* %ap2)
4482 ; Read a single integer argument
4483 %tmp = va_arg i8** %ap, i32
4485 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4487 %aq2 = bitcast i8** %aq to i8*
4488 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4489 call void @llvm.va_end(i8* %aq2)
4491 ; Stop processing of arguments.
4492 call void @llvm.va_end(i8* %ap2)
4496 declare void @llvm.va_start(i8*)
4497 declare void @llvm.va_copy(i8*, i8*)
4498 declare void @llvm.va_end(i8*)
4504 <!-- _______________________________________________________________________ -->
4505 <div class="doc_subsubsection">
4506 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4510 <div class="doc_text">
4512 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4514 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
4515 <tt>*<arglist></tt> for subsequent use by <tt><a
4516 href="#i_va_arg">va_arg</a></tt>.</p>
4520 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4524 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4525 macro available in C. In a target-dependent way, it initializes the
4526 <tt>va_list</tt> element to which the argument points, so that the next call to
4527 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4528 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4529 last argument of the function as the compiler can figure that out.</p>
4533 <!-- _______________________________________________________________________ -->
4534 <div class="doc_subsubsection">
4535 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4538 <div class="doc_text">
4540 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4543 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4544 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4545 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4549 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4553 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4554 macro available in C. In a target-dependent way, it destroys the
4555 <tt>va_list</tt> element to which the argument points. Calls to <a
4556 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4557 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4558 <tt>llvm.va_end</tt>.</p>
4562 <!-- _______________________________________________________________________ -->
4563 <div class="doc_subsubsection">
4564 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4567 <div class="doc_text">
4572 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4577 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4578 from the source argument list to the destination argument list.</p>
4582 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4583 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4588 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4589 macro available in C. In a target-dependent way, it copies the source
4590 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4591 intrinsic is necessary because the <tt><a href="#int_va_start">
4592 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4593 example, memory allocation.</p>
4597 <!-- ======================================================================= -->
4598 <div class="doc_subsection">
4599 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4602 <div class="doc_text">
4605 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4606 Collection</a> (GC) requires the implementation and generation of these
4608 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4609 stack</a>, as well as garbage collector implementations that require <a
4610 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4611 Front-ends for type-safe garbage collected languages should generate these
4612 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4613 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4616 <p>The garbage collection intrinsics only operate on objects in the generic
4617 address space (address space zero).</p>
4621 <!-- _______________________________________________________________________ -->
4622 <div class="doc_subsubsection">
4623 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4626 <div class="doc_text">
4631 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4636 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4637 the code generator, and allows some metadata to be associated with it.</p>
4641 <p>The first argument specifies the address of a stack object that contains the
4642 root pointer. The second pointer (which must be either a constant or a global
4643 value address) contains the meta-data to be associated with the root.</p>
4647 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4648 location. At compile-time, the code generator generates information to allow
4649 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4650 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4656 <!-- _______________________________________________________________________ -->
4657 <div class="doc_subsubsection">
4658 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4661 <div class="doc_text">
4666 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4671 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4672 locations, allowing garbage collector implementations that require read
4677 <p>The second argument is the address to read from, which should be an address
4678 allocated from the garbage collector. The first object is a pointer to the
4679 start of the referenced object, if needed by the language runtime (otherwise
4684 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4685 instruction, but may be replaced with substantially more complex code by the
4686 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4687 may only be used in a function which <a href="#gc">specifies a GC
4693 <!-- _______________________________________________________________________ -->
4694 <div class="doc_subsubsection">
4695 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4698 <div class="doc_text">
4703 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4708 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4709 locations, allowing garbage collector implementations that require write
4710 barriers (such as generational or reference counting collectors).</p>
4714 <p>The first argument is the reference to store, the second is the start of the
4715 object to store it to, and the third is the address of the field of Obj to
4716 store to. If the runtime does not require a pointer to the object, Obj may be
4721 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4722 instruction, but may be replaced with substantially more complex code by the
4723 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4724 may only be used in a function which <a href="#gc">specifies a GC
4731 <!-- ======================================================================= -->
4732 <div class="doc_subsection">
4733 <a name="int_codegen">Code Generator Intrinsics</a>
4736 <div class="doc_text">
4738 These intrinsics are provided by LLVM to expose special features that may only
4739 be implemented with code generator support.
4744 <!-- _______________________________________________________________________ -->
4745 <div class="doc_subsubsection">
4746 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4749 <div class="doc_text">
4753 declare i8 *@llvm.returnaddress(i32 <level>)
4759 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4760 target-specific value indicating the return address of the current function
4761 or one of its callers.
4767 The argument to this intrinsic indicates which function to return the address
4768 for. Zero indicates the calling function, one indicates its caller, etc. The
4769 argument is <b>required</b> to be a constant integer value.
4775 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4776 the return address of the specified call frame, or zero if it cannot be
4777 identified. The value returned by this intrinsic is likely to be incorrect or 0
4778 for arguments other than zero, so it should only be used for debugging purposes.
4782 Note that calling this intrinsic does not prevent function inlining or other
4783 aggressive transformations, so the value returned may not be that of the obvious
4784 source-language caller.
4789 <!-- _______________________________________________________________________ -->
4790 <div class="doc_subsubsection">
4791 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4794 <div class="doc_text">
4798 declare i8 *@llvm.frameaddress(i32 <level>)
4804 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4805 target-specific frame pointer value for the specified stack frame.
4811 The argument to this intrinsic indicates which function to return the frame
4812 pointer for. Zero indicates the calling function, one indicates its caller,
4813 etc. The argument is <b>required</b> to be a constant integer value.
4819 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4820 the frame address of the specified call frame, or zero if it cannot be
4821 identified. The value returned by this intrinsic is likely to be incorrect or 0
4822 for arguments other than zero, so it should only be used for debugging purposes.
4826 Note that calling this intrinsic does not prevent function inlining or other
4827 aggressive transformations, so the value returned may not be that of the obvious
4828 source-language caller.
4832 <!-- _______________________________________________________________________ -->
4833 <div class="doc_subsubsection">
4834 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4837 <div class="doc_text">
4841 declare i8 *@llvm.stacksave()
4847 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4848 the function stack, for use with <a href="#int_stackrestore">
4849 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4850 features like scoped automatic variable sized arrays in C99.
4856 This intrinsic returns a opaque pointer value that can be passed to <a
4857 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4858 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4859 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4860 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4861 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4862 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4867 <!-- _______________________________________________________________________ -->
4868 <div class="doc_subsubsection">
4869 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4872 <div class="doc_text">
4876 declare void @llvm.stackrestore(i8 * %ptr)
4882 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4883 the function stack to the state it was in when the corresponding <a
4884 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4885 useful for implementing language features like scoped automatic variable sized
4892 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4898 <!-- _______________________________________________________________________ -->
4899 <div class="doc_subsubsection">
4900 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4903 <div class="doc_text">
4907 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4914 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4915 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4917 effect on the behavior of the program but can change its performance
4924 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4925 determining if the fetch should be for a read (0) or write (1), and
4926 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4927 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4928 <tt>locality</tt> arguments must be constant integers.
4934 This intrinsic does not modify the behavior of the program. In particular,
4935 prefetches cannot trap and do not produce a value. On targets that support this
4936 intrinsic, the prefetch can provide hints to the processor cache for better
4942 <!-- _______________________________________________________________________ -->
4943 <div class="doc_subsubsection">
4944 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4947 <div class="doc_text">
4951 declare void @llvm.pcmarker(i32 <id>)
4958 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4960 code to simulators and other tools. The method is target specific, but it is
4961 expected that the marker will use exported symbols to transmit the PC of the
4963 The marker makes no guarantees that it will remain with any specific instruction
4964 after optimizations. It is possible that the presence of a marker will inhibit
4965 optimizations. The intended use is to be inserted after optimizations to allow
4966 correlations of simulation runs.
4972 <tt>id</tt> is a numerical id identifying the marker.
4978 This intrinsic does not modify the behavior of the program. Backends that do not
4979 support this intrinisic may ignore it.
4984 <!-- _______________________________________________________________________ -->
4985 <div class="doc_subsubsection">
4986 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4989 <div class="doc_text">
4993 declare i64 @llvm.readcyclecounter( )
5000 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5001 counter register (or similar low latency, high accuracy clocks) on those targets
5002 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
5003 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
5004 should only be used for small timings.
5010 When directly supported, reading the cycle counter should not modify any memory.
5011 Implementations are allowed to either return a application specific value or a
5012 system wide value. On backends without support, this is lowered to a constant 0.
5017 <!-- ======================================================================= -->
5018 <div class="doc_subsection">
5019 <a name="int_libc">Standard C Library Intrinsics</a>
5022 <div class="doc_text">
5024 LLVM provides intrinsics for a few important standard C library functions.
5025 These intrinsics allow source-language front-ends to pass information about the
5026 alignment of the pointer arguments to the code generator, providing opportunity
5027 for more efficient code generation.
5032 <!-- _______________________________________________________________________ -->
5033 <div class="doc_subsubsection">
5034 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5037 <div class="doc_text">
5041 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5042 i32 <len>, i32 <align>)
5043 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5044 i64 <len>, i32 <align>)
5050 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5051 location to the destination location.
5055 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5056 intrinsics do not return a value, and takes an extra alignment argument.
5062 The first argument is a pointer to the destination, the second is a pointer to
5063 the source. The third argument is an integer argument
5064 specifying the number of bytes to copy, and the fourth argument is the alignment
5065 of the source and destination locations.
5069 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5070 the caller guarantees that both the source and destination pointers are aligned
5077 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5078 location to the destination location, which are not allowed to overlap. It
5079 copies "len" bytes of memory over. If the argument is known to be aligned to
5080 some boundary, this can be specified as the fourth argument, otherwise it should
5086 <!-- _______________________________________________________________________ -->
5087 <div class="doc_subsubsection">
5088 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5091 <div class="doc_text">
5095 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5096 i32 <len>, i32 <align>)
5097 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5098 i64 <len>, i32 <align>)
5104 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5105 location to the destination location. It is similar to the
5106 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5110 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5111 intrinsics do not return a value, and takes an extra alignment argument.
5117 The first argument is a pointer to the destination, the second is a pointer to
5118 the source. The third argument is an integer argument
5119 specifying the number of bytes to copy, and the fourth argument is the alignment
5120 of the source and destination locations.
5124 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5125 the caller guarantees that the source and destination pointers are aligned to
5132 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5133 location to the destination location, which may overlap. It
5134 copies "len" bytes of memory over. If the argument is known to be aligned to
5135 some boundary, this can be specified as the fourth argument, otherwise it should
5141 <!-- _______________________________________________________________________ -->
5142 <div class="doc_subsubsection">
5143 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5146 <div class="doc_text">
5150 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5151 i32 <len>, i32 <align>)
5152 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5153 i64 <len>, i32 <align>)
5159 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5164 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5165 does not return a value, and takes an extra alignment argument.
5171 The first argument is a pointer to the destination to fill, the second is the
5172 byte value to fill it with, the third argument is an integer
5173 argument specifying the number of bytes to fill, and the fourth argument is the
5174 known alignment of destination location.
5178 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5179 the caller guarantees that the destination pointer is aligned to that boundary.
5185 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5187 destination location. If the argument is known to be aligned to some boundary,
5188 this can be specified as the fourth argument, otherwise it should be set to 0 or
5194 <!-- _______________________________________________________________________ -->
5195 <div class="doc_subsubsection">
5196 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5199 <div class="doc_text">
5202 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5203 floating point or vector of floating point type. Not all targets support all
5206 declare float @llvm.sqrt.f32(float %Val)
5207 declare double @llvm.sqrt.f64(double %Val)
5208 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5209 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5210 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5216 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5217 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5218 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5219 negative numbers other than -0.0 (which allows for better optimization, because
5220 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5221 defined to return -0.0 like IEEE sqrt.
5227 The argument and return value are floating point numbers of the same type.
5233 This function returns the sqrt of the specified operand if it is a nonnegative
5234 floating point number.
5238 <!-- _______________________________________________________________________ -->
5239 <div class="doc_subsubsection">
5240 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5243 <div class="doc_text">
5246 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5247 floating point or vector of floating point type. Not all targets support all
5250 declare float @llvm.powi.f32(float %Val, i32 %power)
5251 declare double @llvm.powi.f64(double %Val, i32 %power)
5252 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5253 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5254 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5260 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5261 specified (positive or negative) power. The order of evaluation of
5262 multiplications is not defined. When a vector of floating point type is
5263 used, the second argument remains a scalar integer value.
5269 The second argument is an integer power, and the first is a value to raise to
5276 This function returns the first value raised to the second power with an
5277 unspecified sequence of rounding operations.</p>
5280 <!-- _______________________________________________________________________ -->
5281 <div class="doc_subsubsection">
5282 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5285 <div class="doc_text">
5288 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5289 floating point or vector of floating point type. Not all targets support all
5292 declare float @llvm.sin.f32(float %Val)
5293 declare double @llvm.sin.f64(double %Val)
5294 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5295 declare fp128 @llvm.sin.f128(fp128 %Val)
5296 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5302 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5308 The argument and return value are floating point numbers of the same type.
5314 This function returns the sine of the specified operand, returning the
5315 same values as the libm <tt>sin</tt> functions would, and handles error
5316 conditions in the same way.</p>
5319 <!-- _______________________________________________________________________ -->
5320 <div class="doc_subsubsection">
5321 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5324 <div class="doc_text">
5327 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5328 floating point or vector of floating point type. Not all targets support all
5331 declare float @llvm.cos.f32(float %Val)
5332 declare double @llvm.cos.f64(double %Val)
5333 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5334 declare fp128 @llvm.cos.f128(fp128 %Val)
5335 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5341 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5347 The argument and return value are floating point numbers of the same type.
5353 This function returns the cosine of the specified operand, returning the
5354 same values as the libm <tt>cos</tt> functions would, and handles error
5355 conditions in the same way.</p>
5358 <!-- _______________________________________________________________________ -->
5359 <div class="doc_subsubsection">
5360 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5363 <div class="doc_text">
5366 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5367 floating point or vector of floating point type. Not all targets support all
5370 declare float @llvm.pow.f32(float %Val, float %Power)
5371 declare double @llvm.pow.f64(double %Val, double %Power)
5372 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5373 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5374 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5380 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5381 specified (positive or negative) power.
5387 The second argument is a floating point power, and the first is a value to
5388 raise to that power.
5394 This function returns the first value raised to the second power,
5396 same values as the libm <tt>pow</tt> functions would, and handles error
5397 conditions in the same way.</p>
5401 <!-- ======================================================================= -->
5402 <div class="doc_subsection">
5403 <a name="int_manip">Bit Manipulation Intrinsics</a>
5406 <div class="doc_text">
5408 LLVM provides intrinsics for a few important bit manipulation operations.
5409 These allow efficient code generation for some algorithms.
5414 <!-- _______________________________________________________________________ -->
5415 <div class="doc_subsubsection">
5416 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5419 <div class="doc_text">
5422 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5423 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
5425 declare i16 @llvm.bswap.i16(i16 <id>)
5426 declare i32 @llvm.bswap.i32(i32 <id>)
5427 declare i64 @llvm.bswap.i64(i64 <id>)
5433 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5434 values with an even number of bytes (positive multiple of 16 bits). These are
5435 useful for performing operations on data that is not in the target's native
5442 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5443 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5444 intrinsic returns an i32 value that has the four bytes of the input i32
5445 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5446 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5447 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5448 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5453 <!-- _______________________________________________________________________ -->
5454 <div class="doc_subsubsection">
5455 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5458 <div class="doc_text">
5461 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5462 width. Not all targets support all bit widths however.
5464 declare i8 @llvm.ctpop.i8 (i8 <src>)
5465 declare i16 @llvm.ctpop.i16(i16 <src>)
5466 declare i32 @llvm.ctpop.i32(i32 <src>)
5467 declare i64 @llvm.ctpop.i64(i64 <src>)
5468 declare i256 @llvm.ctpop.i256(i256 <src>)
5474 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5481 The only argument is the value to be counted. The argument may be of any
5482 integer type. The return type must match the argument type.
5488 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5492 <!-- _______________________________________________________________________ -->
5493 <div class="doc_subsubsection">
5494 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5497 <div class="doc_text">
5500 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5501 integer bit width. Not all targets support all bit widths however.
5503 declare i8 @llvm.ctlz.i8 (i8 <src>)
5504 declare i16 @llvm.ctlz.i16(i16 <src>)
5505 declare i32 @llvm.ctlz.i32(i32 <src>)
5506 declare i64 @llvm.ctlz.i64(i64 <src>)
5507 declare i256 @llvm.ctlz.i256(i256 <src>)
5513 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5514 leading zeros in a variable.
5520 The only argument is the value to be counted. The argument may be of any
5521 integer type. The return type must match the argument type.
5527 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5528 in a variable. If the src == 0 then the result is the size in bits of the type
5529 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5535 <!-- _______________________________________________________________________ -->
5536 <div class="doc_subsubsection">
5537 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5540 <div class="doc_text">
5543 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5544 integer bit width. Not all targets support all bit widths however.
5546 declare i8 @llvm.cttz.i8 (i8 <src>)
5547 declare i16 @llvm.cttz.i16(i16 <src>)
5548 declare i32 @llvm.cttz.i32(i32 <src>)
5549 declare i64 @llvm.cttz.i64(i64 <src>)
5550 declare i256 @llvm.cttz.i256(i256 <src>)
5556 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5563 The only argument is the value to be counted. The argument may be of any
5564 integer type. The return type must match the argument type.
5570 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5571 in a variable. If the src == 0 then the result is the size in bits of the type
5572 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5576 <!-- _______________________________________________________________________ -->
5577 <div class="doc_subsubsection">
5578 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5581 <div class="doc_text">
5584 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5585 on any integer bit width.
5587 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5588 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5592 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5593 range of bits from an integer value and returns them in the same bit width as
5594 the original value.</p>
5597 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5598 any bit width but they must have the same bit width. The second and third
5599 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5602 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5603 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5604 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5605 operates in forward mode.</p>
5606 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5607 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5608 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5610 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5611 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5612 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5613 to determine the number of bits to retain.</li>
5614 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5615 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5617 <p>In reverse mode, a similar computation is made except that the bits are
5618 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5619 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5620 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5621 <tt>i16 0x0026 (000000100110)</tt>.</p>
5624 <div class="doc_subsubsection">
5625 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5628 <div class="doc_text">
5631 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5632 on any integer bit width.
5634 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5635 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5639 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5640 of bits in an integer value with another integer value. It returns the integer
5641 with the replaced bits.</p>
5644 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5645 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5646 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5647 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5648 type since they specify only a bit index.</p>
5651 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5652 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5653 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5654 operates in forward mode.</p>
5655 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5656 truncating it down to the size of the replacement area or zero extending it
5657 up to that size.</p>
5658 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5659 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5660 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5661 to the <tt>%hi</tt>th bit.
5662 <p>In reverse mode, a similar computation is made except that the bits are
5663 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5664 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5667 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5668 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5669 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5670 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5671 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5675 <!-- ======================================================================= -->
5676 <div class="doc_subsection">
5677 <a name="int_debugger">Debugger Intrinsics</a>
5680 <div class="doc_text">
5682 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5683 are described in the <a
5684 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5685 Debugging</a> document.
5690 <!-- ======================================================================= -->
5691 <div class="doc_subsection">
5692 <a name="int_eh">Exception Handling Intrinsics</a>
5695 <div class="doc_text">
5696 <p> The LLVM exception handling intrinsics (which all start with
5697 <tt>llvm.eh.</tt> prefix), are described in the <a
5698 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5699 Handling</a> document. </p>
5702 <!-- ======================================================================= -->
5703 <div class="doc_subsection">
5704 <a name="int_trampoline">Trampoline Intrinsic</a>
5707 <div class="doc_text">
5709 This intrinsic makes it possible to excise one parameter, marked with
5710 the <tt>nest</tt> attribute, from a function. The result is a callable
5711 function pointer lacking the nest parameter - the caller does not need
5712 to provide a value for it. Instead, the value to use is stored in
5713 advance in a "trampoline", a block of memory usually allocated
5714 on the stack, which also contains code to splice the nest value into the
5715 argument list. This is used to implement the GCC nested function address
5719 For example, if the function is
5720 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5721 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5723 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5724 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5725 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5726 %fp = bitcast i8* %p to i32 (i32, i32)*
5728 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5729 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5732 <!-- _______________________________________________________________________ -->
5733 <div class="doc_subsubsection">
5734 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5736 <div class="doc_text">
5739 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5743 This fills the memory pointed to by <tt>tramp</tt> with code
5744 and returns a function pointer suitable for executing it.
5748 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5749 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5750 and sufficiently aligned block of memory; this memory is written to by the
5751 intrinsic. Note that the size and the alignment are target-specific - LLVM
5752 currently provides no portable way of determining them, so a front-end that
5753 generates this intrinsic needs to have some target-specific knowledge.
5754 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5758 The block of memory pointed to by <tt>tramp</tt> is filled with target
5759 dependent code, turning it into a function. A pointer to this function is
5760 returned, but needs to be bitcast to an
5761 <a href="#int_trampoline">appropriate function pointer type</a>
5762 before being called. The new function's signature is the same as that of
5763 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5764 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5765 of pointer type. Calling the new function is equivalent to calling
5766 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5767 missing <tt>nest</tt> argument. If, after calling
5768 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5769 modified, then the effect of any later call to the returned function pointer is
5774 <!-- ======================================================================= -->
5775 <div class="doc_subsection">
5776 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5779 <div class="doc_text">
5781 These intrinsic functions expand the "universal IR" of LLVM to represent
5782 hardware constructs for atomic operations and memory synchronization. This
5783 provides an interface to the hardware, not an interface to the programmer. It
5784 is aimed at a low enough level to allow any programming models or APIs
5785 (Application Programming Interfaces) which
5786 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5787 hardware behavior. Just as hardware provides a "universal IR" for source
5788 languages, it also provides a starting point for developing a "universal"
5789 atomic operation and synchronization IR.
5792 These do <em>not</em> form an API such as high-level threading libraries,
5793 software transaction memory systems, atomic primitives, and intrinsic
5794 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5795 application libraries. The hardware interface provided by LLVM should allow
5796 a clean implementation of all of these APIs and parallel programming models.
5797 No one model or paradigm should be selected above others unless the hardware
5798 itself ubiquitously does so.
5803 <!-- _______________________________________________________________________ -->
5804 <div class="doc_subsubsection">
5805 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5807 <div class="doc_text">
5810 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5816 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5817 specific pairs of memory access types.
5821 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5822 The first four arguments enables a specific barrier as listed below. The fith
5823 argument specifies that the barrier applies to io or device or uncached memory.
5827 <li><tt>ll</tt>: load-load barrier</li>
5828 <li><tt>ls</tt>: load-store barrier</li>
5829 <li><tt>sl</tt>: store-load barrier</li>
5830 <li><tt>ss</tt>: store-store barrier</li>
5831 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5835 This intrinsic causes the system to enforce some ordering constraints upon
5836 the loads and stores of the program. This barrier does not indicate
5837 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5838 which they occur. For any of the specified pairs of load and store operations
5839 (f.ex. load-load, or store-load), all of the first operations preceding the
5840 barrier will complete before any of the second operations succeeding the
5841 barrier begin. Specifically the semantics for each pairing is as follows:
5844 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5845 after the barrier begins.</li>
5847 <li><tt>ls</tt>: All loads before the barrier must complete before any
5848 store after the barrier begins.</li>
5849 <li><tt>ss</tt>: All stores before the barrier must complete before any
5850 store after the barrier begins.</li>
5851 <li><tt>sl</tt>: All stores before the barrier must complete before any
5852 load after the barrier begins.</li>
5855 These semantics are applied with a logical "and" behavior when more than one
5856 is enabled in a single memory barrier intrinsic.
5859 Backends may implement stronger barriers than those requested when they do not
5860 support as fine grained a barrier as requested. Some architectures do not
5861 need all types of barriers and on such architectures, these become noops.
5868 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5869 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5870 <i>; guarantee the above finishes</i>
5871 store i32 8, %ptr <i>; before this begins</i>
5875 <!-- _______________________________________________________________________ -->
5876 <div class="doc_subsubsection">
5877 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
5879 <div class="doc_text">
5882 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
5883 any integer bit width and for different address spaces. Not all targets
5884 support all bit widths however.</p>
5887 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5888 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5889 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5890 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5895 This loads a value in memory and compares it to a given value. If they are
5896 equal, it stores a new value into the memory.
5900 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
5901 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5902 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5903 this integer type. While any bit width integer may be used, targets may only
5904 lower representations they support in hardware.
5909 This entire intrinsic must be executed atomically. It first loads the value
5910 in memory pointed to by <tt>ptr</tt> and compares it with the value
5911 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5912 loaded value is yielded in all cases. This provides the equivalent of an
5913 atomic compare-and-swap operation within the SSA framework.
5921 %val1 = add i32 4, 4
5922 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
5923 <i>; yields {i32}:result1 = 4</i>
5924 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5925 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5927 %val2 = add i32 1, 1
5928 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
5929 <i>; yields {i32}:result2 = 8</i>
5930 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5932 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5936 <!-- _______________________________________________________________________ -->
5937 <div class="doc_subsubsection">
5938 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5940 <div class="doc_text">
5944 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5945 integer bit width. Not all targets support all bit widths however.</p>
5947 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
5948 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
5949 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
5950 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
5955 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5956 the value from memory. It then stores the value in <tt>val</tt> in the memory
5962 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
5963 <tt>val</tt> argument and the result must be integers of the same bit width.
5964 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5965 integer type. The targets may only lower integer representations they
5970 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5971 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5972 equivalent of an atomic swap operation within the SSA framework.
5980 %val1 = add i32 4, 4
5981 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
5982 <i>; yields {i32}:result1 = 4</i>
5983 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5984 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5986 %val2 = add i32 1, 1
5987 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
5988 <i>; yields {i32}:result2 = 8</i>
5990 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5991 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5995 <!-- _______________________________________________________________________ -->
5996 <div class="doc_subsubsection">
5997 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6000 <div class="doc_text">
6003 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on any
6004 integer bit width. Not all targets support all bit widths however.</p>
6006 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6007 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6008 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6009 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6014 This intrinsic adds <tt>delta</tt> to the value stored in memory at
6015 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6020 The intrinsic takes two arguments, the first a pointer to an integer value
6021 and the second an integer value. The result is also an integer value. These
6022 integer types can have any bit width, but they must all have the same bit
6023 width. The targets may only lower integer representations they support.
6027 This intrinsic does a series of operations atomically. It first loads the
6028 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6029 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6036 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6037 <i>; yields {i32}:result1 = 4</i>
6038 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6039 <i>; yields {i32}:result2 = 8</i>
6040 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6041 <i>; yields {i32}:result3 = 10</i>
6042 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6046 <!-- _______________________________________________________________________ -->
6047 <div class="doc_subsubsection">
6048 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6051 <div class="doc_text">
6054 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6055 any integer bit width and for different address spaces. Not all targets
6056 support all bit widths however.</p>
6058 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6059 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6060 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6061 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6066 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6067 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6072 The intrinsic takes two arguments, the first a pointer to an integer value
6073 and the second an integer value. The result is also an integer value. These
6074 integer types can have any bit width, but they must all have the same bit
6075 width. The targets may only lower integer representations they support.
6079 This intrinsic does a series of operations atomically. It first loads the
6080 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6081 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6088 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6089 <i>; yields {i32}:result1 = 8</i>
6090 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6091 <i>; yields {i32}:result2 = 4</i>
6092 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6093 <i>; yields {i32}:result3 = 2</i>
6094 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6098 <!-- _______________________________________________________________________ -->
6099 <div class="doc_subsubsection">
6100 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6101 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6102 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6103 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6106 <div class="doc_text">
6109 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6110 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
6111 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6112 address spaces. Not all targets support all bit widths however.</p>
6114 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6115 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6116 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6117 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6122 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6123 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6124 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6125 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6130 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6131 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6132 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6133 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6138 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6139 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6140 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6141 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6146 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6147 the value stored in memory at <tt>ptr</tt>. It yields the original value
6153 These intrinsics take two arguments, the first a pointer to an integer value
6154 and the second an integer value. The result is also an integer value. These
6155 integer types can have any bit width, but they must all have the same bit
6156 width. The targets may only lower integer representations they support.
6160 These intrinsics does a series of operations atomically. They first load the
6161 value stored at <tt>ptr</tt>. They then do the bitwise operation
6162 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6163 value stored at <tt>ptr</tt>.
6169 store i32 0x0F0F, %ptr
6170 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6171 <i>; yields {i32}:result0 = 0x0F0F</i>
6172 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6173 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6174 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6175 <i>; yields {i32}:result2 = 0xF0</i>
6176 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6177 <i>; yields {i32}:result3 = FF</i>
6178 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6183 <!-- _______________________________________________________________________ -->
6184 <div class="doc_subsubsection">
6185 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6186 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6187 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6188 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6191 <div class="doc_text">
6194 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6195 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6196 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6197 address spaces. Not all targets
6198 support all bit widths however.</p>
6200 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6201 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6202 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6203 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6208 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6209 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6210 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6211 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6216 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6217 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6218 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6219 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6224 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6225 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6226 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6227 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6232 These intrinsics takes the signed or unsigned minimum or maximum of
6233 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6234 original value at <tt>ptr</tt>.
6239 These intrinsics take two arguments, the first a pointer to an integer value
6240 and the second an integer value. The result is also an integer value. These
6241 integer types can have any bit width, but they must all have the same bit
6242 width. The targets may only lower integer representations they support.
6246 These intrinsics does a series of operations atomically. They first load the
6247 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
6248 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
6249 the original value stored at <tt>ptr</tt>.
6256 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6257 <i>; yields {i32}:result0 = 7</i>
6258 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6259 <i>; yields {i32}:result1 = -2</i>
6260 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6261 <i>; yields {i32}:result2 = 8</i>
6262 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6263 <i>; yields {i32}:result3 = 8</i>
6264 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6268 <!-- ======================================================================= -->
6269 <div class="doc_subsection">
6270 <a name="int_general">General Intrinsics</a>
6273 <div class="doc_text">
6274 <p> This class of intrinsics is designed to be generic and has
6275 no specific purpose. </p>
6278 <!-- _______________________________________________________________________ -->
6279 <div class="doc_subsubsection">
6280 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6283 <div class="doc_text">
6287 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6293 The '<tt>llvm.var.annotation</tt>' intrinsic
6299 The first argument is a pointer to a value, the second is a pointer to a
6300 global string, the third is a pointer to a global string which is the source
6301 file name, and the last argument is the line number.
6307 This intrinsic allows annotation of local variables with arbitrary strings.
6308 This can be useful for special purpose optimizations that want to look for these
6309 annotations. These have no other defined use, they are ignored by code
6310 generation and optimization.
6314 <!-- _______________________________________________________________________ -->
6315 <div class="doc_subsubsection">
6316 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6319 <div class="doc_text">
6322 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6323 any integer bit width.
6326 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6327 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6328 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6329 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6330 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6336 The '<tt>llvm.annotation</tt>' intrinsic.
6342 The first argument is an integer value (result of some expression),
6343 the second is a pointer to a global string, the third is a pointer to a global
6344 string which is the source file name, and the last argument is the line number.
6345 It returns the value of the first argument.
6351 This intrinsic allows annotations to be put on arbitrary expressions
6352 with arbitrary strings. This can be useful for special purpose optimizations
6353 that want to look for these annotations. These have no other defined use, they
6354 are ignored by code generation and optimization.
6357 <!-- _______________________________________________________________________ -->
6358 <div class="doc_subsubsection">
6359 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6362 <div class="doc_text">
6366 declare void @llvm.trap()
6372 The '<tt>llvm.trap</tt>' intrinsic
6384 This intrinsics is lowered to the target dependent trap instruction. If the
6385 target does not have a trap instruction, this intrinsic will be lowered to the
6386 call of the abort() function.
6390 <!-- *********************************************************************** -->
6393 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6394 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
6395 <a href="http://validator.w3.org/check/referer"><img
6396 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
6398 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6399 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6400 Last modified: $Date$