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="#gc">Garbage Collector Names</a></li>
30 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
31 <li><a href="#datalayout">Data Layout</a></li>
34 <li><a href="#typesystem">Type System</a>
36 <li><a href="#t_classifications">Type Classifications</a></li>
37 <li><a href="#t_primitive">Primitive Types</a>
39 <li><a href="#t_floating">Floating Point Types</a></li>
40 <li><a href="#t_void">Void Type</a></li>
41 <li><a href="#t_label">Label Type</a></li>
44 <li><a href="#t_derived">Derived Types</a>
46 <li><a href="#t_integer">Integer Type</a></li>
47 <li><a href="#t_array">Array Type</a></li>
48 <li><a href="#t_function">Function Type</a></li>
49 <li><a href="#t_pointer">Pointer Type</a></li>
50 <li><a href="#t_struct">Structure Type</a></li>
51 <li><a href="#t_pstruct">Packed Structure Type</a></li>
52 <li><a href="#t_vector">Vector Type</a></li>
53 <li><a href="#t_opaque">Opaque Type</a></li>
58 <li><a href="#constants">Constants</a>
60 <li><a href="#simpleconstants">Simple Constants</a>
61 <li><a href="#aggregateconstants">Aggregate Constants</a>
62 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
63 <li><a href="#undefvalues">Undefined Values</a>
64 <li><a href="#constantexprs">Constant Expressions</a>
67 <li><a href="#othervalues">Other Values</a>
69 <li><a href="#inlineasm">Inline Assembler Expressions</a>
72 <li><a href="#instref">Instruction Reference</a>
74 <li><a href="#terminators">Terminator Instructions</a>
76 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
77 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
78 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
79 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
80 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
81 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
84 <li><a href="#binaryops">Binary Operations</a>
86 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
87 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
88 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
89 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
90 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
91 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
92 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
93 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
94 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
97 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
99 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
100 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
101 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
102 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
103 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
104 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
107 <li><a href="#vectorops">Vector Operations</a>
109 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
110 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
111 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
114 <li><a href="#aggregateops">Aggregate Operations</a>
116 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
117 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
120 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
122 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
123 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
124 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
125 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
126 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
127 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
130 <li><a href="#convertops">Conversion Operations</a>
132 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
133 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
134 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
135 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
136 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
137 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
138 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
139 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
140 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
141 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
142 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
143 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
145 <li><a href="#otherops">Other Operations</a>
147 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
148 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
149 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
150 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
151 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
152 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
153 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
154 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
155 <li><a href="#i_getresult">'<tt>getresult</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 an SSA based representation that provides type safety,
259 low-level operations, flexibility, and the capability of representing
260 '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 model: the symbol is weak
526 until linked, if not linked, the symbol becomes null instead of being an
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)
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. Implementations of
606 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 ELF, default visibility means that the declaration is visible to other
652 modules and, in shared libraries, means that the declared entity may be
653 overridden. On Darwin, default visibility means that the declaration is
654 visible to other modules. Default visibility corresponds to "external
655 linkage" in the language.
658 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
660 <dd>Two declarations of an object with hidden visibility refer to the same
661 object if they are in the same shared object. Usually, hidden visibility
662 indicates that the symbol will not be placed into the dynamic symbol table,
663 so no other module (executable or shared library) can reference it
667 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
669 <dd>On ELF, protected visibility indicates that the symbol will be placed in
670 the dynamic symbol table, but that references within the defining module will
671 bind to the local symbol. That is, the symbol cannot be overridden by another
678 <!-- ======================================================================= -->
679 <div class="doc_subsection">
680 <a name="globalvars">Global Variables</a>
683 <div class="doc_text">
685 <p>Global variables define regions of memory allocated at compilation time
686 instead of run-time. Global variables may optionally be initialized, may have
687 an explicit section to be placed in, and may have an optional explicit alignment
688 specified. A variable may be defined as "thread_local", which means that it
689 will not be shared by threads (each thread will have a separated copy of the
690 variable). A variable may be defined as a global "constant," which indicates
691 that the contents of the variable will <b>never</b> be modified (enabling better
692 optimization, allowing the global data to be placed in the read-only section of
693 an executable, etc). Note that variables that need runtime initialization
694 cannot be marked "constant" as there is a store to the variable.</p>
697 LLVM explicitly allows <em>declarations</em> of global variables to be marked
698 constant, even if the final definition of the global is not. This capability
699 can be used to enable slightly better optimization of the program, but requires
700 the language definition to guarantee that optimizations based on the
701 'constantness' are valid for the translation units that do not include the
705 <p>As SSA values, global variables define pointer values that are in
706 scope (i.e. they dominate) all basic blocks in the program. Global
707 variables always define a pointer to their "content" type because they
708 describe a region of memory, and all memory objects in LLVM are
709 accessed through pointers.</p>
711 <p>A global variable may be declared to reside in a target-specifc numbered
712 address space. For targets that support them, address spaces may affect how
713 optimizations are performed and/or what target instructions are used to access
714 the variable. The default address space is zero. The address space qualifier
715 must precede any other attributes.</p>
717 <p>LLVM allows an explicit section to be specified for globals. If the target
718 supports it, it will emit globals to the section specified.</p>
720 <p>An explicit alignment may be specified for a global. If not present, or if
721 the alignment is set to zero, the alignment of the global is set by the target
722 to whatever it feels convenient. If an explicit alignment is specified, the
723 global is forced to have at least that much alignment. All alignments must be
726 <p>For example, the following defines a global in a numbered address space with
727 an initializer, section, and alignment:</p>
729 <div class="doc_code">
731 @G = constant float 1.0 addrspace(5), section "foo", align 4
738 <!-- ======================================================================= -->
739 <div class="doc_subsection">
740 <a name="functionstructure">Functions</a>
743 <div class="doc_text">
745 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
746 an optional <a href="#linkage">linkage type</a>, an optional
747 <a href="#visibility">visibility style</a>, an optional
748 <a href="#callingconv">calling convention</a>, a return type, an optional
749 <a href="#paramattrs">parameter attribute</a> for the return type, a function
750 name, a (possibly empty) argument list (each with optional
751 <a href="#paramattrs">parameter attributes</a>), an optional section, an
752 optional alignment, an optional <a href="#gc">garbage collector name</a>, an
753 opening curly brace, a list of basic blocks, and a closing curly brace.
755 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
756 optional <a href="#linkage">linkage type</a>, an optional
757 <a href="#visibility">visibility style</a>, an optional
758 <a href="#callingconv">calling convention</a>, a return type, an optional
759 <a href="#paramattrs">parameter attribute</a> for the return type, a function
760 name, a possibly empty list of arguments, an optional alignment, and an optional
761 <a href="#gc">garbage collector name</a>.</p>
763 <p>A function definition contains a list of basic blocks, forming the CFG for
764 the function. Each basic block may optionally start with a label (giving the
765 basic block a symbol table entry), contains a list of instructions, and ends
766 with a <a href="#terminators">terminator</a> instruction (such as a branch or
767 function return).</p>
769 <p>The first basic block in a function is special in two ways: it is immediately
770 executed on entrance to the function, and it is not allowed to have predecessor
771 basic blocks (i.e. there can not be any branches to the entry block of a
772 function). Because the block can have no predecessors, it also cannot have any
773 <a href="#i_phi">PHI nodes</a>.</p>
775 <p>LLVM allows an explicit section to be specified for functions. If the target
776 supports it, it will emit functions to the section specified.</p>
778 <p>An explicit alignment may be specified for a function. If not present, or if
779 the alignment is set to zero, the alignment of the function is set by the target
780 to whatever it feels convenient. If an explicit alignment is specified, the
781 function is forced to have at least that much alignment. All alignments must be
787 <!-- ======================================================================= -->
788 <div class="doc_subsection">
789 <a name="aliasstructure">Aliases</a>
791 <div class="doc_text">
792 <p>Aliases act as "second name" for the aliasee value (which can be either
793 function, global variable, another alias or bitcast of global value). Aliases
794 may have an optional <a href="#linkage">linkage type</a>, and an
795 optional <a href="#visibility">visibility style</a>.</p>
799 <div class="doc_code">
801 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
809 <!-- ======================================================================= -->
810 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
811 <div class="doc_text">
812 <p>The return type and each parameter of a function type may have a set of
813 <i>parameter attributes</i> associated with them. Parameter attributes are
814 used to communicate additional information about the result or parameters of
815 a function. Parameter attributes are considered to be part of the function,
816 not of the function type, so functions with different parameter attributes
817 can have the same function type.</p>
819 <p>Parameter attributes are simple keywords that follow the type specified. If
820 multiple parameter attributes are needed, they are space separated. For
823 <div class="doc_code">
825 declare i32 @printf(i8* noalias , ...) nounwind
826 declare i32 @atoi(i8*) nounwind readonly
830 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
831 <tt>readonly</tt>) come immediately after the argument list.</p>
833 <p>Currently, only the following parameter attributes are defined:</p>
835 <dt><tt>zeroext</tt></dt>
836 <dd>This indicates that the parameter should be zero extended just before
837 a call to this function.</dd>
839 <dt><tt>signext</tt></dt>
840 <dd>This indicates that the parameter should be sign extended just before
841 a call to this function.</dd>
843 <dt><tt>inreg</tt></dt>
844 <dd>This indicates that the parameter should be placed in register (if
845 possible) during assembling function call. Support for this attribute is
848 <dt><tt>byval</tt></dt>
849 <dd>This indicates that the pointer parameter should really be passed by
850 value to the function. The attribute implies that a hidden copy of the
851 pointee is made between the caller and the callee, so the callee is unable
852 to modify the value in the callee. This attribute is only valid on llvm
853 pointer arguments. It is generally used to pass structs and arrays by
854 value, but is also valid on scalars (even though this is silly).</dd>
856 <dt><tt>sret</tt></dt>
857 <dd>This indicates that the pointer parameter specifies the address of a
858 structure that is the return value of the function in the source program.
859 Loads and stores to the structure are assumed not to trap.
860 May only be applied to the first parameter.</dd>
862 <dt><tt>noalias</tt></dt>
863 <dd>This indicates that the parameter does not alias any global or any other
864 parameter. The caller is responsible for ensuring that this is the case,
865 usually by placing the value in a stack allocation.</dd>
867 <dt><tt>noreturn</tt></dt>
868 <dd>This function attribute indicates that the function never returns. This
869 indicates to LLVM that every call to this function should be treated as if
870 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
872 <dt><tt>nounwind</tt></dt>
873 <dd>This function attribute indicates that no exceptions unwind out of the
874 function. Usually this is because the function makes no use of exceptions,
875 but it may also be that the function catches any exceptions thrown when
878 <dt><tt>nest</tt></dt>
879 <dd>This indicates that the pointer parameter can be excised using the
880 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
881 <dt><tt>readonly</tt></dt>
882 <dd>This function attribute indicates that the function has no side-effects
883 except for producing a return value or throwing an exception. The value
884 returned must only depend on the function arguments and/or global variables.
885 It may use values obtained by dereferencing pointers.</dd>
886 <dt><tt>readnone</tt></dt>
887 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
888 function, but in addition it is not allowed to dereference any pointer arguments
894 <!-- ======================================================================= -->
895 <div class="doc_subsection">
896 <a name="gc">Garbage Collector Names</a>
899 <div class="doc_text">
900 <p>Each function may specify a garbage collector name, which is simply a
903 <div class="doc_code"><pre
904 >define void @f() gc "name" { ...</pre></div>
906 <p>The compiler declares the supported values of <i>name</i>. Specifying a
907 collector which will cause the compiler to alter its output in order to support
908 the named garbage collection algorithm.</p>
911 <!-- ======================================================================= -->
912 <div class="doc_subsection">
913 <a name="moduleasm">Module-Level Inline Assembly</a>
916 <div class="doc_text">
918 Modules may contain "module-level inline asm" blocks, which corresponds to the
919 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
920 LLVM and treated as a single unit, but may be separated in the .ll file if
921 desired. The syntax is very simple:
924 <div class="doc_code">
926 module asm "inline asm code goes here"
927 module asm "more can go here"
931 <p>The strings can contain any character by escaping non-printable characters.
932 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
937 The inline asm code is simply printed to the machine code .s file when
938 assembly code is generated.
942 <!-- ======================================================================= -->
943 <div class="doc_subsection">
944 <a name="datalayout">Data Layout</a>
947 <div class="doc_text">
948 <p>A module may specify a target specific data layout string that specifies how
949 data is to be laid out in memory. The syntax for the data layout is simply:</p>
950 <pre> target datalayout = "<i>layout specification</i>"</pre>
951 <p>The <i>layout specification</i> consists of a list of specifications
952 separated by the minus sign character ('-'). Each specification starts with a
953 letter and may include other information after the letter to define some
954 aspect of the data layout. The specifications accepted are as follows: </p>
957 <dd>Specifies that the target lays out data in big-endian form. That is, the
958 bits with the most significance have the lowest address location.</dd>
960 <dd>Specifies that hte target lays out data in little-endian form. That is,
961 the bits with the least significance have the lowest address location.</dd>
962 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
963 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
964 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
965 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
967 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
968 <dd>This specifies the alignment for an integer type of a given bit
969 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
970 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
971 <dd>This specifies the alignment for a vector type of a given bit
973 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
974 <dd>This specifies the alignment for a floating point type of a given bit
975 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
977 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
978 <dd>This specifies the alignment for an aggregate type of a given bit
981 <p>When constructing the data layout for a given target, LLVM starts with a
982 default set of specifications which are then (possibly) overriden by the
983 specifications in the <tt>datalayout</tt> keyword. The default specifications
984 are given in this list:</p>
986 <li><tt>E</tt> - big endian</li>
987 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
988 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
989 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
990 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
991 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
992 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
993 alignment of 64-bits</li>
994 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
995 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
996 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
997 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
998 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1000 <p>When llvm is determining the alignment for a given type, it uses the
1003 <li>If the type sought is an exact match for one of the specifications, that
1004 specification is used.</li>
1005 <li>If no match is found, and the type sought is an integer type, then the
1006 smallest integer type that is larger than the bitwidth of the sought type is
1007 used. If none of the specifications are larger than the bitwidth then the the
1008 largest integer type is used. For example, given the default specifications
1009 above, the i7 type will use the alignment of i8 (next largest) while both
1010 i65 and i256 will use the alignment of i64 (largest specified).</li>
1011 <li>If no match is found, and the type sought is a vector type, then the
1012 largest vector type that is smaller than the sought vector type will be used
1013 as a fall back. This happens because <128 x double> can be implemented in
1014 terms of 64 <2 x double>, for example.</li>
1018 <!-- *********************************************************************** -->
1019 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1020 <!-- *********************************************************************** -->
1022 <div class="doc_text">
1024 <p>The LLVM type system is one of the most important features of the
1025 intermediate representation. Being typed enables a number of
1026 optimizations to be performed on the IR directly, without having to do
1027 extra analyses on the side before the transformation. A strong type
1028 system makes it easier to read the generated code and enables novel
1029 analyses and transformations that are not feasible to perform on normal
1030 three address code representations.</p>
1034 <!-- ======================================================================= -->
1035 <div class="doc_subsection"> <a name="t_classifications">Type
1036 Classifications</a> </div>
1037 <div class="doc_text">
1038 <p>The types fall into a few useful
1039 classifications:</p>
1041 <table border="1" cellspacing="0" cellpadding="4">
1043 <tr><th>Classification</th><th>Types</th></tr>
1045 <td><a href="#t_integer">integer</a></td>
1046 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1049 <td><a href="#t_floating">floating point</a></td>
1050 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1053 <td><a name="t_firstclass">first class</a></td>
1054 <td><a href="#t_integer">integer</a>,
1055 <a href="#t_floating">floating point</a>,
1056 <a href="#t_pointer">pointer</a>,
1057 <a href="#t_vector">vector</a>,
1058 <a href="#t_struct">structure</a>,
1059 <a href="#t_array">array</a>,
1060 <a href="#t_label">label</a>.
1064 <td><a href="#t_primitive">primitive</a></td>
1065 <td><a href="#t_label">label</a>,
1066 <a href="#t_void">void</a>,
1067 <a href="#t_floating">floating point</a>.</td>
1070 <td><a href="#t_derived">derived</a></td>
1071 <td><a href="#t_integer">integer</a>,
1072 <a href="#t_array">array</a>,
1073 <a href="#t_function">function</a>,
1074 <a href="#t_pointer">pointer</a>,
1075 <a href="#t_struct">structure</a>,
1076 <a href="#t_pstruct">packed structure</a>,
1077 <a href="#t_vector">vector</a>,
1078 <a href="#t_opaque">opaque</a>.
1083 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1084 most important. Values of these types are the only ones which can be
1085 produced by instructions, passed as arguments, or used as operands to
1089 <!-- ======================================================================= -->
1090 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1092 <div class="doc_text">
1093 <p>The primitive types are the fundamental building blocks of the LLVM
1098 <!-- _______________________________________________________________________ -->
1099 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1101 <div class="doc_text">
1104 <tr><th>Type</th><th>Description</th></tr>
1105 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1106 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1107 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1108 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1109 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1114 <!-- _______________________________________________________________________ -->
1115 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1117 <div class="doc_text">
1119 <p>The void type does not represent any value and has no size.</p>
1128 <!-- _______________________________________________________________________ -->
1129 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1131 <div class="doc_text">
1133 <p>The label type represents code labels.</p>
1143 <!-- ======================================================================= -->
1144 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1146 <div class="doc_text">
1148 <p>The real power in LLVM comes from the derived types in the system.
1149 This is what allows a programmer to represent arrays, functions,
1150 pointers, and other useful types. Note that these derived types may be
1151 recursive: For example, it is possible to have a two dimensional array.</p>
1155 <!-- _______________________________________________________________________ -->
1156 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1158 <div class="doc_text">
1161 <p>The integer type is a very simple derived type that simply specifies an
1162 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1163 2^23-1 (about 8 million) can be specified.</p>
1171 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1175 <table class="layout">
1178 <td><tt>i1</tt></td>
1179 <td>a single-bit integer.</td>
1181 <td><tt>i32</tt></td>
1182 <td>a 32-bit integer.</td>
1184 <td><tt>i1942652</tt></td>
1185 <td>a really big integer of over 1 million bits.</td>
1191 <!-- _______________________________________________________________________ -->
1192 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1194 <div class="doc_text">
1198 <p>The array type is a very simple derived type that arranges elements
1199 sequentially in memory. The array type requires a size (number of
1200 elements) and an underlying data type.</p>
1205 [<# elements> x <elementtype>]
1208 <p>The number of elements is a constant integer value; elementtype may
1209 be any type with a size.</p>
1212 <table class="layout">
1214 <td class="left"><tt>[40 x i32]</tt></td>
1215 <td class="left">Array of 40 32-bit integer values.</td>
1218 <td class="left"><tt>[41 x i32]</tt></td>
1219 <td class="left">Array of 41 32-bit integer values.</td>
1222 <td class="left"><tt>[4 x i8]</tt></td>
1223 <td class="left">Array of 4 8-bit integer values.</td>
1226 <p>Here are some examples of multidimensional arrays:</p>
1227 <table class="layout">
1229 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1230 <td class="left">3x4 array of 32-bit integer values.</td>
1233 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1234 <td class="left">12x10 array of single precision floating point values.</td>
1237 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1238 <td class="left">2x3x4 array of 16-bit integer values.</td>
1242 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1243 length array. Normally, accesses past the end of an array are undefined in
1244 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1245 As a special case, however, zero length arrays are recognized to be variable
1246 length. This allows implementation of 'pascal style arrays' with the LLVM
1247 type "{ i32, [0 x float]}", for example.</p>
1251 <!-- _______________________________________________________________________ -->
1252 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1253 <div class="doc_text">
1257 <p>The function type can be thought of as a function signature. It
1258 consists of a return type and a list of formal parameter types. The
1259 return type of a function type is a scalar type, a void type, or a struct type.
1260 If the return type is a struct type then all struct elements must be of first
1261 class types, and the struct must have at least one element.</p>
1266 <returntype list> (<parameter list>)
1269 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1270 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1271 which indicates that the function takes a variable number of arguments.
1272 Variable argument functions can access their arguments with the <a
1273 href="#int_varargs">variable argument handling intrinsic</a> functions.
1274 '<tt><returntype list></tt>' is a comma-separated list of
1275 <a href="#t_firstclass">first class</a> type specifiers.</p>
1278 <table class="layout">
1280 <td class="left"><tt>i32 (i32)</tt></td>
1281 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1283 </tr><tr class="layout">
1284 <td class="left"><tt>float (i16 signext, i32 *) *
1286 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1287 an <tt>i16</tt> that should be sign extended and a
1288 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1291 </tr><tr class="layout">
1292 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1293 <td class="left">A vararg function that takes at least one
1294 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1295 which returns an integer. This is the signature for <tt>printf</tt> in
1298 </tr><tr class="layout">
1299 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1300 <td class="left">A function taking an <tt>i32></tt>, returning two
1301 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1307 <!-- _______________________________________________________________________ -->
1308 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1309 <div class="doc_text">
1311 <p>The structure type is used to represent a collection of data members
1312 together in memory. The packing of the field types is defined to match
1313 the ABI of the underlying processor. The elements of a structure may
1314 be any type that has a size.</p>
1315 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1316 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1317 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1320 <pre> { <type list> }<br></pre>
1322 <table class="layout">
1324 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1325 <td class="left">A triple of three <tt>i32</tt> values</td>
1326 </tr><tr class="layout">
1327 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1328 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1329 second element is a <a href="#t_pointer">pointer</a> to a
1330 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1331 an <tt>i32</tt>.</td>
1336 <!-- _______________________________________________________________________ -->
1337 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1339 <div class="doc_text">
1341 <p>The packed structure type is used to represent a collection of data members
1342 together in memory. There is no padding between fields. Further, the alignment
1343 of a packed structure is 1 byte. The elements of a packed structure may
1344 be any type that has a size.</p>
1345 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1346 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1347 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1350 <pre> < { <type list> } > <br></pre>
1352 <table class="layout">
1354 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1355 <td class="left">A triple of three <tt>i32</tt> values</td>
1356 </tr><tr class="layout">
1357 <td class="left"><tt>< { float, i32 (i32)* } ></tt></td>
1358 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1359 second element is a <a href="#t_pointer">pointer</a> to a
1360 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1361 an <tt>i32</tt>.</td>
1366 <!-- _______________________________________________________________________ -->
1367 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1368 <div class="doc_text">
1370 <p>As in many languages, the pointer type represents a pointer or
1371 reference to another object, which must live in memory. Pointer types may have
1372 an optional address space attribute defining the target-specific numbered
1373 address space where the pointed-to object resides. The default address space is
1376 <pre> <type> *<br></pre>
1378 <table class="layout">
1380 <td class="left"><tt>[4x i32]*</tt></td>
1381 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1382 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1385 <td class="left"><tt>i32 (i32 *) *</tt></td>
1386 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1387 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1391 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1392 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1393 that resides in address space #5.</td>
1398 <!-- _______________________________________________________________________ -->
1399 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1400 <div class="doc_text">
1404 <p>A vector type is a simple derived type that represents a vector
1405 of elements. Vector types are used when multiple primitive data
1406 are operated in parallel using a single instruction (SIMD).
1407 A vector type requires a size (number of
1408 elements) and an underlying primitive data type. Vectors must have a power
1409 of two length (1, 2, 4, 8, 16 ...). Vector types are
1410 considered <a href="#t_firstclass">first class</a>.</p>
1415 < <# elements> x <elementtype> >
1418 <p>The number of elements is a constant integer value; elementtype may
1419 be any integer or floating point type.</p>
1423 <table class="layout">
1425 <td class="left"><tt><4 x i32></tt></td>
1426 <td class="left">Vector of 4 32-bit integer values.</td>
1429 <td class="left"><tt><8 x float></tt></td>
1430 <td class="left">Vector of 8 32-bit floating-point values.</td>
1433 <td class="left"><tt><2 x i64></tt></td>
1434 <td class="left">Vector of 2 64-bit integer values.</td>
1439 <!-- _______________________________________________________________________ -->
1440 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1441 <div class="doc_text">
1445 <p>Opaque types are used to represent unknown types in the system. This
1446 corresponds (for example) to the C notion of a forward declared structure type.
1447 In LLVM, opaque types can eventually be resolved to any type (not just a
1448 structure type).</p>
1458 <table class="layout">
1460 <td class="left"><tt>opaque</tt></td>
1461 <td class="left">An opaque type.</td>
1467 <!-- *********************************************************************** -->
1468 <div class="doc_section"> <a name="constants">Constants</a> </div>
1469 <!-- *********************************************************************** -->
1471 <div class="doc_text">
1473 <p>LLVM has several different basic types of constants. This section describes
1474 them all and their syntax.</p>
1478 <!-- ======================================================================= -->
1479 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1481 <div class="doc_text">
1484 <dt><b>Boolean constants</b></dt>
1486 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1487 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1490 <dt><b>Integer constants</b></dt>
1492 <dd>Standard integers (such as '4') are constants of the <a
1493 href="#t_integer">integer</a> type. Negative numbers may be used with
1497 <dt><b>Floating point constants</b></dt>
1499 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1500 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1501 notation (see below). The assembler requires the exact decimal value of
1502 a floating-point constant. For example, the assembler accepts 1.25 but
1503 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1504 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1506 <dt><b>Null pointer constants</b></dt>
1508 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1509 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1513 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1514 of floating point constants. For example, the form '<tt>double
1515 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1516 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1517 (and the only time that they are generated by the disassembler) is when a
1518 floating point constant must be emitted but it cannot be represented as a
1519 decimal floating point number. For example, NaN's, infinities, and other
1520 special values are represented in their IEEE hexadecimal format so that
1521 assembly and disassembly do not cause any bits to change in the constants.</p>
1525 <!-- ======================================================================= -->
1526 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1529 <div class="doc_text">
1530 <p>Aggregate constants arise from aggregation of simple constants
1531 and smaller aggregate constants.</p>
1534 <dt><b>Structure constants</b></dt>
1536 <dd>Structure constants are represented with notation similar to structure
1537 type definitions (a comma separated list of elements, surrounded by braces
1538 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1539 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1540 must have <a href="#t_struct">structure type</a>, and the number and
1541 types of elements must match those specified by the type.
1544 <dt><b>Array constants</b></dt>
1546 <dd>Array constants are represented with notation similar to array type
1547 definitions (a comma separated list of elements, surrounded by square brackets
1548 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1549 constants must have <a href="#t_array">array type</a>, and the number and
1550 types of elements must match those specified by the type.
1553 <dt><b>Vector constants</b></dt>
1555 <dd>Vector constants are represented with notation similar to vector type
1556 definitions (a comma separated list of elements, surrounded by
1557 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1558 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1559 href="#t_vector">vector type</a>, and the number and types of elements must
1560 match those specified by the type.
1563 <dt><b>Zero initialization</b></dt>
1565 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1566 value to zero of <em>any</em> type, including scalar and aggregate types.
1567 This is often used to avoid having to print large zero initializers (e.g. for
1568 large arrays) and is always exactly equivalent to using explicit zero
1575 <!-- ======================================================================= -->
1576 <div class="doc_subsection">
1577 <a name="globalconstants">Global Variable and Function Addresses</a>
1580 <div class="doc_text">
1582 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1583 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1584 constants. These constants are explicitly referenced when the <a
1585 href="#identifiers">identifier for the global</a> is used and always have <a
1586 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1589 <div class="doc_code">
1593 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1599 <!-- ======================================================================= -->
1600 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1601 <div class="doc_text">
1602 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1603 no specific value. Undefined values may be of any type and be used anywhere
1604 a constant is permitted.</p>
1606 <p>Undefined values indicate to the compiler that the program is well defined
1607 no matter what value is used, giving the compiler more freedom to optimize.
1611 <!-- ======================================================================= -->
1612 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1615 <div class="doc_text">
1617 <p>Constant expressions are used to allow expressions involving other constants
1618 to be used as constants. Constant expressions may be of any <a
1619 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1620 that does not have side effects (e.g. load and call are not supported). The
1621 following is the syntax for constant expressions:</p>
1624 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1625 <dd>Truncate a constant to another type. The bit size of CST must be larger
1626 than the bit size of TYPE. Both types must be integers.</dd>
1628 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1629 <dd>Zero extend a constant to another type. The bit size of CST must be
1630 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1632 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1633 <dd>Sign extend a constant to another type. The bit size of CST must be
1634 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1636 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1637 <dd>Truncate a floating point constant to another floating point type. The
1638 size of CST must be larger than the size of TYPE. Both types must be
1639 floating point.</dd>
1641 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1642 <dd>Floating point extend a constant to another type. The size of CST must be
1643 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1645 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1646 <dd>Convert a floating point constant to the corresponding unsigned integer
1647 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1648 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1649 of the same number of elements. If the value won't fit in the integer type,
1650 the results are undefined.</dd>
1652 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1653 <dd>Convert a floating point constant to the corresponding signed integer
1654 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1655 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1656 of the same number of elements. If the value won't fit in the integer type,
1657 the results are undefined.</dd>
1659 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1660 <dd>Convert an unsigned integer constant to the corresponding floating point
1661 constant. TYPE must be a scalar or vector floating point type. CST must be of
1662 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1663 of the same number of elements. If the value won't fit in the floating point
1664 type, the results are undefined.</dd>
1666 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1667 <dd>Convert a signed integer constant to the corresponding floating point
1668 constant. TYPE must be a scalar or vector floating point type. CST must be of
1669 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1670 of the same number of elements. If the value won't fit in the floating point
1671 type, the results are undefined.</dd>
1673 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1674 <dd>Convert a pointer typed constant to the corresponding integer constant
1675 TYPE must be an integer type. CST must be of pointer type. The CST value is
1676 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1678 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1679 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1680 pointer type. CST must be of integer type. The CST value is zero extended,
1681 truncated, or unchanged to make it fit in a pointer size. This one is
1682 <i>really</i> dangerous!</dd>
1684 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1685 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1686 identical (same number of bits). The conversion is done as if the CST value
1687 was stored to memory and read back as TYPE. In other words, no bits change
1688 with this operator, just the type. This can be used for conversion of
1689 vector types to any other type, as long as they have the same bit width. For
1690 pointers it is only valid to cast to another pointer type.
1693 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1695 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1696 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1697 instruction, the index list may have zero or more indexes, which are required
1698 to make sense for the type of "CSTPTR".</dd>
1700 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1702 <dd>Perform the <a href="#i_select">select operation</a> on
1705 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1706 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1708 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1709 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1711 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1712 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1714 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1715 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1717 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1719 <dd>Perform the <a href="#i_extractelement">extractelement
1720 operation</a> on constants.
1722 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1724 <dd>Perform the <a href="#i_insertelement">insertelement
1725 operation</a> on constants.</dd>
1728 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1730 <dd>Perform the <a href="#i_shufflevector">shufflevector
1731 operation</a> on constants.</dd>
1733 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1735 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1736 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1737 binary</a> operations. The constraints on operands are the same as those for
1738 the corresponding instruction (e.g. no bitwise operations on floating point
1739 values are allowed).</dd>
1743 <!-- *********************************************************************** -->
1744 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1745 <!-- *********************************************************************** -->
1747 <!-- ======================================================================= -->
1748 <div class="doc_subsection">
1749 <a name="inlineasm">Inline Assembler Expressions</a>
1752 <div class="doc_text">
1755 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1756 Module-Level Inline Assembly</a>) through the use of a special value. This
1757 value represents the inline assembler as a string (containing the instructions
1758 to emit), a list of operand constraints (stored as a string), and a flag that
1759 indicates whether or not the inline asm expression has side effects. An example
1760 inline assembler expression is:
1763 <div class="doc_code">
1765 i32 (i32) asm "bswap $0", "=r,r"
1770 Inline assembler expressions may <b>only</b> be used as the callee operand of
1771 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1774 <div class="doc_code">
1776 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1781 Inline asms with side effects not visible in the constraint list must be marked
1782 as having side effects. This is done through the use of the
1783 '<tt>sideeffect</tt>' keyword, like so:
1786 <div class="doc_code">
1788 call void asm sideeffect "eieio", ""()
1792 <p>TODO: The format of the asm and constraints string still need to be
1793 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1794 need to be documented).
1799 <!-- *********************************************************************** -->
1800 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1801 <!-- *********************************************************************** -->
1803 <div class="doc_text">
1805 <p>The LLVM instruction set consists of several different
1806 classifications of instructions: <a href="#terminators">terminator
1807 instructions</a>, <a href="#binaryops">binary instructions</a>,
1808 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1809 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1810 instructions</a>.</p>
1814 <!-- ======================================================================= -->
1815 <div class="doc_subsection"> <a name="terminators">Terminator
1816 Instructions</a> </div>
1818 <div class="doc_text">
1820 <p>As mentioned <a href="#functionstructure">previously</a>, every
1821 basic block in a program ends with a "Terminator" instruction, which
1822 indicates which block should be executed after the current block is
1823 finished. These terminator instructions typically yield a '<tt>void</tt>'
1824 value: they produce control flow, not values (the one exception being
1825 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1826 <p>There are six different terminator instructions: the '<a
1827 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1828 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1829 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1830 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1831 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1835 <!-- _______________________________________________________________________ -->
1836 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1837 Instruction</a> </div>
1838 <div class="doc_text">
1840 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1841 ret void <i>; Return from void function</i>
1842 ret <type> <value>, <type> <value> <i>; Return two values from a non-void function </i>
1847 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1848 value) from a function back to the caller.</p>
1849 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1850 returns value(s) and then causes control flow, and one that just causes
1851 control flow to occur.</p>
1855 <p>The '<tt>ret</tt>' instruction may return zero, one or multiple values.
1856 The type of each return value must be a '<a href="#t_firstclass">first
1857 class</a>' type. Note that a function is not <a href="#wellformed">well
1858 formed</a> if there exists a '<tt>ret</tt>' instruction inside of the
1859 function that returns values that do not match the return type of the
1864 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1865 returns back to the calling function's context. If the caller is a "<a
1866 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1867 the instruction after the call. If the caller was an "<a
1868 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1869 at the beginning of the "normal" destination block. If the instruction
1870 returns a value, that value shall set the call or invoke instruction's
1871 return value. If the instruction returns multiple values then these
1872 values can only be accessed through a '<a href="#i_getresult"><tt>getresult</tt>
1873 </a>' instruction.</p>
1878 ret i32 5 <i>; Return an integer value of 5</i>
1879 ret void <i>; Return from a void function</i>
1880 ret i32 4, i8 2 <i>; Return two values 4 and 2 </i>
1883 <!-- _______________________________________________________________________ -->
1884 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1885 <div class="doc_text">
1887 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1890 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1891 transfer to a different basic block in the current function. There are
1892 two forms of this instruction, corresponding to a conditional branch
1893 and an unconditional branch.</p>
1895 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1896 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1897 unconditional form of the '<tt>br</tt>' instruction takes a single
1898 '<tt>label</tt>' value as a target.</p>
1900 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1901 argument is evaluated. If the value is <tt>true</tt>, control flows
1902 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1903 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1905 <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
1906 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1908 <!-- _______________________________________________________________________ -->
1909 <div class="doc_subsubsection">
1910 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1913 <div class="doc_text">
1917 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1922 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1923 several different places. It is a generalization of the '<tt>br</tt>'
1924 instruction, allowing a branch to occur to one of many possible
1930 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1931 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1932 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1933 table is not allowed to contain duplicate constant entries.</p>
1937 <p>The <tt>switch</tt> instruction specifies a table of values and
1938 destinations. When the '<tt>switch</tt>' instruction is executed, this
1939 table is searched for the given value. If the value is found, control flow is
1940 transfered to the corresponding destination; otherwise, control flow is
1941 transfered to the default destination.</p>
1943 <h5>Implementation:</h5>
1945 <p>Depending on properties of the target machine and the particular
1946 <tt>switch</tt> instruction, this instruction may be code generated in different
1947 ways. For example, it could be generated as a series of chained conditional
1948 branches or with a lookup table.</p>
1953 <i>; Emulate a conditional br instruction</i>
1954 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1955 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1957 <i>; Emulate an unconditional br instruction</i>
1958 switch i32 0, label %dest [ ]
1960 <i>; Implement a jump table:</i>
1961 switch i32 %val, label %otherwise [ i32 0, label %onzero
1963 i32 2, label %ontwo ]
1967 <!-- _______________________________________________________________________ -->
1968 <div class="doc_subsubsection">
1969 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1972 <div class="doc_text">
1977 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> <function ptr val>(<function args>)
1978 to label <normal label> unwind label <exception label>
1983 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1984 function, with the possibility of control flow transfer to either the
1985 '<tt>normal</tt>' label or the
1986 '<tt>exception</tt>' label. If the callee function returns with the
1987 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1988 "normal" label. If the callee (or any indirect callees) returns with the "<a
1989 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1990 continued at the dynamically nearest "exception" label. If the callee function
1991 returns multiple values then individual return values are only accessible through
1992 a '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
1996 <p>This instruction requires several arguments:</p>
2000 The optional "cconv" marker indicates which <a href="#callingconv">calling
2001 convention</a> the call should use. If none is specified, the call defaults
2002 to using C calling conventions.
2004 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2005 function value being invoked. In most cases, this is a direct function
2006 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2007 an arbitrary pointer to function value.
2010 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2011 function to be invoked. </li>
2013 <li>'<tt>function args</tt>': argument list whose types match the function
2014 signature argument types. If the function signature indicates the function
2015 accepts a variable number of arguments, the extra arguments can be
2018 <li>'<tt>normal label</tt>': the label reached when the called function
2019 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2021 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2022 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2028 <p>This instruction is designed to operate as a standard '<tt><a
2029 href="#i_call">call</a></tt>' instruction in most regards. The primary
2030 difference is that it establishes an association with a label, which is used by
2031 the runtime library to unwind the stack.</p>
2033 <p>This instruction is used in languages with destructors to ensure that proper
2034 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2035 exception. Additionally, this is important for implementation of
2036 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2040 %retval = invoke i32 @Test(i32 15) to label %Continue
2041 unwind label %TestCleanup <i>; {i32}:retval set</i>
2042 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2043 unwind label %TestCleanup <i>; {i32}:retval set</i>
2048 <!-- _______________________________________________________________________ -->
2050 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2051 Instruction</a> </div>
2053 <div class="doc_text">
2062 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2063 at the first callee in the dynamic call stack which used an <a
2064 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2065 primarily used to implement exception handling.</p>
2069 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2070 immediately halt. The dynamic call stack is then searched for the first <a
2071 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2072 execution continues at the "exceptional" destination block specified by the
2073 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2074 dynamic call chain, undefined behavior results.</p>
2077 <!-- _______________________________________________________________________ -->
2079 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2080 Instruction</a> </div>
2082 <div class="doc_text">
2091 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2092 instruction is used to inform the optimizer that a particular portion of the
2093 code is not reachable. This can be used to indicate that the code after a
2094 no-return function cannot be reached, and other facts.</p>
2098 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2103 <!-- ======================================================================= -->
2104 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2105 <div class="doc_text">
2106 <p>Binary operators are used to do most of the computation in a
2107 program. They require two operands of the same type, execute an operation on them, and
2108 produce a single value. The operands might represent
2109 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2110 The result value has the same type as its operands.</p>
2111 <p>There are several different binary operators:</p>
2113 <!-- _______________________________________________________________________ -->
2114 <div class="doc_subsubsection">
2115 <a name="i_add">'<tt>add</tt>' Instruction</a>
2118 <div class="doc_text">
2123 <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2128 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2132 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2133 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2134 <a href="#t_vector">vector</a> values. Both arguments must have identical
2139 <p>The value produced is the integer or floating point sum of the two
2142 <p>If an integer sum has unsigned overflow, the result returned is the
2143 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2146 <p>Because LLVM integers use a two's complement representation, this
2147 instruction is appropriate for both signed and unsigned integers.</p>
2152 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2155 <!-- _______________________________________________________________________ -->
2156 <div class="doc_subsubsection">
2157 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2160 <div class="doc_text">
2165 <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2170 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2173 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2174 '<tt>neg</tt>' instruction present in most other intermediate
2175 representations.</p>
2179 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2180 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2181 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2186 <p>The value produced is the integer or floating point difference of
2187 the two operands.</p>
2189 <p>If an integer difference has unsigned overflow, the result returned is the
2190 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2193 <p>Because LLVM integers use a two's complement representation, this
2194 instruction is appropriate for both signed and unsigned integers.</p>
2198 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2199 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2203 <!-- _______________________________________________________________________ -->
2204 <div class="doc_subsubsection">
2205 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2208 <div class="doc_text">
2211 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2214 <p>The '<tt>mul</tt>' instruction returns the product of its two
2219 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2220 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2221 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2226 <p>The value produced is the integer or floating point product of the
2229 <p>If the result of an integer multiplication has unsigned overflow,
2230 the result returned is the mathematical result modulo
2231 2<sup>n</sup>, where n is the bit width of the result.</p>
2232 <p>Because LLVM integers use a two's complement representation, and the
2233 result is the same width as the operands, this instruction returns the
2234 correct result for both signed and unsigned integers. If a full product
2235 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2236 should be sign-extended or zero-extended as appropriate to the
2237 width of the full product.</p>
2239 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2243 <!-- _______________________________________________________________________ -->
2244 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2246 <div class="doc_text">
2248 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2251 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2256 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2257 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2258 values. Both arguments must have identical types.</p>
2262 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2263 <p>Note that unsigned integer division and signed integer division are distinct
2264 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2265 <p>Division by zero leads to undefined behavior.</p>
2267 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2270 <!-- _______________________________________________________________________ -->
2271 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2273 <div class="doc_text">
2276 <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2281 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2286 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2287 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2288 values. Both arguments must have identical types.</p>
2291 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2292 <p>Note that signed integer division and unsigned integer division are distinct
2293 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2294 <p>Division by zero leads to undefined behavior. Overflow also leads to
2295 undefined behavior; this is a rare case, but can occur, for example,
2296 by doing a 32-bit division of -2147483648 by -1.</p>
2298 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2301 <!-- _______________________________________________________________________ -->
2302 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2303 Instruction</a> </div>
2304 <div class="doc_text">
2307 <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2311 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2316 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2317 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2318 of floating point values. Both arguments must have identical types.</p>
2322 <p>The value produced is the floating point quotient of the two operands.</p>
2327 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2331 <!-- _______________________________________________________________________ -->
2332 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2334 <div class="doc_text">
2336 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2339 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2340 unsigned division of its two arguments.</p>
2342 <p>The two arguments to the '<tt>urem</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>
2346 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2347 This instruction always performs an unsigned division to get the remainder.</p>
2348 <p>Note that unsigned integer remainder and signed integer remainder are
2349 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2350 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2352 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2356 <!-- _______________________________________________________________________ -->
2357 <div class="doc_subsubsection">
2358 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2361 <div class="doc_text">
2366 <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2371 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2372 signed division of its two operands. This instruction can also take
2373 <a href="#t_vector">vector</a> versions of the values in which case
2374 the elements must be integers.</p>
2378 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2379 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2380 values. Both arguments must have identical types.</p>
2384 <p>This instruction returns the <i>remainder</i> of a division (where the result
2385 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2386 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2387 a value. For more information about the difference, see <a
2388 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2389 Math Forum</a>. For a table of how this is implemented in various languages,
2390 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2391 Wikipedia: modulo operation</a>.</p>
2392 <p>Note that signed integer remainder and unsigned integer remainder are
2393 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2394 <p>Taking the remainder of a division by zero leads to undefined behavior.
2395 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2396 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2397 (The remainder doesn't actually overflow, but this rule lets srem be
2398 implemented using instructions that return both the result of the division
2399 and the remainder.)</p>
2401 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2405 <!-- _______________________________________________________________________ -->
2406 <div class="doc_subsubsection">
2407 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2409 <div class="doc_text">
2412 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2415 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2416 division of its two operands.</p>
2418 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2419 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2420 of floating point values. Both arguments must have identical types.</p>
2424 <p>This instruction returns the <i>remainder</i> of a division.
2425 The remainder has the same sign as the dividend.</p>
2430 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2434 <!-- ======================================================================= -->
2435 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2436 Operations</a> </div>
2437 <div class="doc_text">
2438 <p>Bitwise binary operators are used to do various forms of
2439 bit-twiddling in a program. They are generally very efficient
2440 instructions and can commonly be strength reduced from other
2441 instructions. They require two operands of the same type, execute an operation on them,
2442 and produce a single value. The resulting value is the same type as its operands.</p>
2445 <!-- _______________________________________________________________________ -->
2446 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2447 Instruction</a> </div>
2448 <div class="doc_text">
2450 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2455 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2456 the left a specified number of bits.</p>
2460 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2461 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2462 type. '<tt>var2</tt>' is treated as an unsigned value.</p>
2466 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup> mod 2<sup>n</sup>,
2467 where n is the width of the result. If <tt>var2</tt> is (statically or dynamically) negative or
2468 equal to or larger than the number of bits in <tt>var1</tt>, the result is undefined.</p>
2470 <h5>Example:</h5><pre>
2471 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2472 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2473 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2474 <result> = shl i32 1, 32 <i>; undefined</i>
2477 <!-- _______________________________________________________________________ -->
2478 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2479 Instruction</a> </div>
2480 <div class="doc_text">
2482 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2486 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2487 operand shifted to the right a specified number of bits with zero fill.</p>
2490 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2491 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2492 type. '<tt>var2</tt>' is treated as an unsigned value.</p>
2496 <p>This instruction always performs a logical shift right operation. The most
2497 significant bits of the result will be filled with zero bits after the
2498 shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2499 the number of bits in <tt>var1</tt>, the result is undefined.</p>
2503 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2504 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2505 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2506 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2507 <result> = lshr i32 1, 32 <i>; undefined</i>
2511 <!-- _______________________________________________________________________ -->
2512 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2513 Instruction</a> </div>
2514 <div class="doc_text">
2517 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2521 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2522 operand shifted to the right a specified number of bits with sign extension.</p>
2525 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2526 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2527 type. '<tt>var2</tt>' is treated as an unsigned value.</p>
2530 <p>This instruction always performs an arithmetic shift right operation,
2531 The most significant bits of the result will be filled with the sign bit
2532 of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2533 larger than the number of bits in <tt>var1</tt>, the result is undefined.
2538 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2539 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2540 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2541 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2542 <result> = ashr i32 1, 32 <i>; undefined</i>
2546 <!-- _______________________________________________________________________ -->
2547 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2548 Instruction</a> </div>
2550 <div class="doc_text">
2555 <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2560 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2561 its two operands.</p>
2565 <p>The two arguments to the '<tt>and</tt>' instruction must be
2566 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2567 values. Both arguments must have identical types.</p>
2570 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2572 <div style="align: center">
2573 <table border="1" cellspacing="0" cellpadding="4">
2605 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2606 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2607 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2610 <!-- _______________________________________________________________________ -->
2611 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2612 <div class="doc_text">
2614 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2617 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2618 or of its two operands.</p>
2621 <p>The two arguments to the '<tt>or</tt>' instruction must be
2622 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2623 values. Both arguments must have identical types.</p>
2625 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2627 <div style="align: center">
2628 <table border="1" cellspacing="0" cellpadding="4">
2659 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2660 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2661 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2664 <!-- _______________________________________________________________________ -->
2665 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2666 Instruction</a> </div>
2667 <div class="doc_text">
2669 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2672 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2673 or of its two operands. The <tt>xor</tt> is used to implement the
2674 "one's complement" operation, which is the "~" operator in C.</p>
2676 <p>The two arguments to the '<tt>xor</tt>' instruction must be
2677 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2678 values. Both arguments must have identical types.</p>
2682 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2684 <div style="align: center">
2685 <table border="1" cellspacing="0" cellpadding="4">
2717 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2718 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2719 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2720 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2724 <!-- ======================================================================= -->
2725 <div class="doc_subsection">
2726 <a name="vectorops">Vector Operations</a>
2729 <div class="doc_text">
2731 <p>LLVM supports several instructions to represent vector operations in a
2732 target-independent manner. These instructions cover the element-access and
2733 vector-specific operations needed to process vectors effectively. While LLVM
2734 does directly support these vector operations, many sophisticated algorithms
2735 will want to use target-specific intrinsics to take full advantage of a specific
2740 <!-- _______________________________________________________________________ -->
2741 <div class="doc_subsubsection">
2742 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2745 <div class="doc_text">
2750 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2756 The '<tt>extractelement</tt>' instruction extracts a single scalar
2757 element from a vector at a specified index.
2764 The first operand of an '<tt>extractelement</tt>' instruction is a
2765 value of <a href="#t_vector">vector</a> type. The second operand is
2766 an index indicating the position from which to extract the element.
2767 The index may be a variable.</p>
2772 The result is a scalar of the same type as the element type of
2773 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2774 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2775 results are undefined.
2781 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2786 <!-- _______________________________________________________________________ -->
2787 <div class="doc_subsubsection">
2788 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2791 <div class="doc_text">
2796 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2802 The '<tt>insertelement</tt>' instruction inserts a scalar
2803 element into a vector at a specified index.
2810 The first operand of an '<tt>insertelement</tt>' instruction is a
2811 value of <a href="#t_vector">vector</a> type. The second operand is a
2812 scalar value whose type must equal the element type of the first
2813 operand. The third operand is an index indicating the position at
2814 which to insert the value. The index may be a variable.</p>
2819 The result is a vector of the same type as <tt>val</tt>. Its
2820 element values are those of <tt>val</tt> except at position
2821 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2822 exceeds the length of <tt>val</tt>, the results are undefined.
2828 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2832 <!-- _______________________________________________________________________ -->
2833 <div class="doc_subsubsection">
2834 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2837 <div class="doc_text">
2842 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2848 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2849 from two input vectors, returning a vector of the same type.
2855 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2856 with types that match each other and types that match the result of the
2857 instruction. The third argument is a shuffle mask, which has the same number
2858 of elements as the other vector type, but whose element type is always 'i32'.
2862 The shuffle mask operand is required to be a constant vector with either
2863 constant integer or undef values.
2869 The elements of the two input vectors are numbered from left to right across
2870 both of the vectors. The shuffle mask operand specifies, for each element of
2871 the result vector, which element of the two input registers the result element
2872 gets. The element selector may be undef (meaning "don't care") and the second
2873 operand may be undef if performing a shuffle from only one vector.
2879 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2880 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2881 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2882 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2887 <!-- ======================================================================= -->
2888 <div class="doc_subsection">
2889 <a name="aggregateops">Aggregate Operations</a>
2892 <div class="doc_text">
2894 <p>LLVM supports several instructions for working with aggregate values.
2899 <!-- _______________________________________________________________________ -->
2900 <div class="doc_subsubsection">
2901 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
2904 <div class="doc_text">
2909 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
2915 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
2916 or array element from an aggregate value.
2923 The first operand of an '<tt>extractvalue</tt>' instruction is a
2924 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
2925 type. The operands are constant indices to specify which value to extract
2926 in a similar manner as indices in a
2927 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2933 The result is the value at the position in the aggregate specified by
2940 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
2945 <!-- _______________________________________________________________________ -->
2946 <div class="doc_subsubsection">
2947 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
2950 <div class="doc_text">
2955 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
2961 The '<tt>insertvalue</tt>' instruction inserts a value
2962 into a struct field or array element in an aggregate.
2969 The first operand of an '<tt>insertvalue</tt>' instruction is a
2970 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
2971 The second operand is a first-class value to insert.
2972 The following operands are constant indices
2973 indicating the position at which to insert the value in a similar manner as
2975 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2976 The value to insert must have the same type as the value identified
2982 The result is an aggregate of the same type as <tt>val</tt>. Its
2983 value is that of <tt>val</tt> except that the value at the position
2984 specified by the indices is that of <tt>elt</tt>.
2990 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
2995 <!-- ======================================================================= -->
2996 <div class="doc_subsection">
2997 <a name="memoryops">Memory Access and Addressing Operations</a>
3000 <div class="doc_text">
3002 <p>A key design point of an SSA-based representation is how it
3003 represents memory. In LLVM, no memory locations are in SSA form, which
3004 makes things very simple. This section describes how to read, write,
3005 allocate, and free memory in LLVM.</p>
3009 <!-- _______________________________________________________________________ -->
3010 <div class="doc_subsubsection">
3011 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3014 <div class="doc_text">
3019 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3024 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3025 heap and returns a pointer to it. The object is always allocated in the generic
3026 address space (address space zero).</p>
3030 <p>The '<tt>malloc</tt>' instruction allocates
3031 <tt>sizeof(<type>)*NumElements</tt>
3032 bytes of memory from the operating system and returns a pointer of the
3033 appropriate type to the program. If "NumElements" is specified, it is the
3034 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3035 If a constant alignment is specified, the value result of the allocation is guaranteed to
3036 be aligned to at least that boundary. If not specified, or if zero, the target can
3037 choose to align the allocation on any convenient boundary.</p>
3039 <p>'<tt>type</tt>' must be a sized type.</p>
3043 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3044 a pointer is returned. The result of a zero byte allocattion is undefined. The
3045 result is null if there is insufficient memory available.</p>
3050 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
3052 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3053 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3054 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3055 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3056 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3060 <!-- _______________________________________________________________________ -->
3061 <div class="doc_subsubsection">
3062 <a name="i_free">'<tt>free</tt>' Instruction</a>
3065 <div class="doc_text">
3070 free <type> <value> <i>; yields {void}</i>
3075 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3076 memory heap to be reallocated in the future.</p>
3080 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3081 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3086 <p>Access to the memory pointed to by the pointer is no longer defined
3087 after this instruction executes. If the pointer is null, the operation
3093 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3094 free [4 x i8]* %array
3098 <!-- _______________________________________________________________________ -->
3099 <div class="doc_subsubsection">
3100 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3103 <div class="doc_text">
3108 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3113 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3114 currently executing function, to be automatically released when this function
3115 returns to its caller. The object is always allocated in the generic address
3116 space (address space zero).</p>
3120 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3121 bytes of memory on the runtime stack, returning a pointer of the
3122 appropriate type to the program. If "NumElements" is specified, it is the
3123 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3124 If a constant alignment is specified, the value result of the allocation is guaranteed
3125 to be aligned to at least that boundary. If not specified, or if zero, the target
3126 can choose to align the allocation on any convenient boundary.</p>
3128 <p>'<tt>type</tt>' may be any sized type.</p>
3132 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3133 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3134 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3135 instruction is commonly used to represent automatic variables that must
3136 have an address available. When the function returns (either with the <tt><a
3137 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3138 instructions), the memory is reclaimed. Allocating zero bytes
3139 is legal, but the result is undefined.</p>
3144 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3145 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3146 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3147 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3151 <!-- _______________________________________________________________________ -->
3152 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3153 Instruction</a> </div>
3154 <div class="doc_text">
3156 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3158 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3160 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3161 address from which to load. The pointer must point to a <a
3162 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3163 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3164 the number or order of execution of this <tt>load</tt> with other
3165 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3168 The optional constant "align" argument specifies the alignment of the operation
3169 (that is, the alignment of the memory address). A value of 0 or an
3170 omitted "align" argument means that the operation has the preferential
3171 alignment for the target. It is the responsibility of the code emitter
3172 to ensure that the alignment information is correct. Overestimating
3173 the alignment results in an undefined behavior. Underestimating the
3174 alignment may produce less efficient code. An alignment of 1 is always
3178 <p>The location of memory pointed to is loaded.</p>
3180 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3182 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3183 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3186 <!-- _______________________________________________________________________ -->
3187 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3188 Instruction</a> </div>
3189 <div class="doc_text">
3191 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3192 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3195 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3197 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3198 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3199 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3200 of the '<tt><value></tt>'
3201 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3202 optimizer is not allowed to modify the number or order of execution of
3203 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3204 href="#i_store">store</a></tt> instructions.</p>
3206 The optional constant "align" argument specifies the alignment of the operation
3207 (that is, the alignment of the memory address). A value of 0 or an
3208 omitted "align" argument means that the operation has the preferential
3209 alignment for the target. It is the responsibility of the code emitter
3210 to ensure that the alignment information is correct. Overestimating
3211 the alignment results in an undefined behavior. Underestimating the
3212 alignment may produce less efficient code. An alignment of 1 is always
3216 <p>The contents of memory are updated to contain '<tt><value></tt>'
3217 at the location specified by the '<tt><pointer></tt>' operand.</p>
3219 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3220 store i32 3, i32* %ptr <i>; yields {void}</i>
3221 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3225 <!-- _______________________________________________________________________ -->
3226 <div class="doc_subsubsection">
3227 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3230 <div class="doc_text">
3233 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
3239 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3240 subelement of an aggregate data structure.</p>
3244 <p>This instruction takes a list of integer operands that indicate what
3245 elements of the aggregate object to index to. The actual types of the arguments
3246 provided depend on the type of the first pointer argument. The
3247 '<tt>getelementptr</tt>' instruction is used to index down through the type
3248 levels of a structure or to a specific index in an array. When indexing into a
3249 structure, only <tt>i32</tt> integer constants are allowed. When indexing
3250 into an array or pointer, only integers of 32 or 64 bits are allowed; 32-bit
3251 values will be sign extended to 64-bits if required.</p>
3253 <p>For example, let's consider a C code fragment and how it gets
3254 compiled to LLVM:</p>
3256 <div class="doc_code">
3269 int *foo(struct ST *s) {
3270 return &s[1].Z.B[5][13];
3275 <p>The LLVM code generated by the GCC frontend is:</p>
3277 <div class="doc_code">
3279 %RT = type { i8 , [10 x [20 x i32]], i8 }
3280 %ST = type { i32, double, %RT }
3282 define i32* %foo(%ST* %s) {
3284 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3292 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3293 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3294 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3295 <a href="#t_integer">integer</a> type but the value will always be sign extended
3296 to 64-bits. <a href="#t_struct">Structure</a> and <a href="#t_pstruct">packed
3297 structure</a> types require <tt>i32</tt> <b>constants</b>.</p>
3299 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3300 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3301 }</tt>' type, a structure. The second index indexes into the third element of
3302 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3303 i8 }</tt>' type, another structure. The third index indexes into the second
3304 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3305 array. The two dimensions of the array are subscripted into, yielding an
3306 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3307 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3309 <p>Note that it is perfectly legal to index partially through a
3310 structure, returning a pointer to an inner element. Because of this,
3311 the LLVM code for the given testcase is equivalent to:</p>
3314 define i32* %foo(%ST* %s) {
3315 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3316 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3317 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3318 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3319 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3324 <p>Note that it is undefined to access an array out of bounds: array and
3325 pointer indexes must always be within the defined bounds of the array type.
3326 The one exception for this rule is zero length arrays. These arrays are
3327 defined to be accessible as variable length arrays, which requires access
3328 beyond the zero'th element.</p>
3330 <p>The getelementptr instruction is often confusing. For some more insight
3331 into how it works, see <a href="GetElementPtr.html">the getelementptr
3337 <i>; yields [12 x i8]*:aptr</i>
3338 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3342 <!-- ======================================================================= -->
3343 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3345 <div class="doc_text">
3346 <p>The instructions in this category are the conversion instructions (casting)
3347 which all take a single operand and a type. They perform various bit conversions
3351 <!-- _______________________________________________________________________ -->
3352 <div class="doc_subsubsection">
3353 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3355 <div class="doc_text">
3359 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3364 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3369 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3370 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3371 and type of the result, which must be an <a href="#t_integer">integer</a>
3372 type. The bit size of <tt>value</tt> must be larger than the bit size of
3373 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3377 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3378 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3379 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3380 It will always truncate bits.</p>
3384 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3385 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3386 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3390 <!-- _______________________________________________________________________ -->
3391 <div class="doc_subsubsection">
3392 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3394 <div class="doc_text">
3398 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3402 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3407 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3408 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3409 also be of <a href="#t_integer">integer</a> type. The bit size of the
3410 <tt>value</tt> must be smaller than the bit size of the destination type,
3414 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3415 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3417 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3421 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3422 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3426 <!-- _______________________________________________________________________ -->
3427 <div class="doc_subsubsection">
3428 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3430 <div class="doc_text">
3434 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3438 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3442 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3443 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3444 also be of <a href="#t_integer">integer</a> type. The bit size of the
3445 <tt>value</tt> must be smaller than the bit size of the destination type,
3450 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3451 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3452 the type <tt>ty2</tt>.</p>
3454 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3458 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3459 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3463 <!-- _______________________________________________________________________ -->
3464 <div class="doc_subsubsection">
3465 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3468 <div class="doc_text">
3473 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3477 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3482 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3483 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3484 cast it to. The size of <tt>value</tt> must be larger than the size of
3485 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3486 <i>no-op cast</i>.</p>
3489 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3490 <a href="#t_floating">floating point</a> type to a smaller
3491 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3492 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3496 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3497 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3501 <!-- _______________________________________________________________________ -->
3502 <div class="doc_subsubsection">
3503 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3505 <div class="doc_text">
3509 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3513 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3514 floating point value.</p>
3517 <p>The '<tt>fpext</tt>' instruction takes a
3518 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3519 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3520 type must be smaller than the destination type.</p>
3523 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3524 <a href="#t_floating">floating point</a> type to a larger
3525 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3526 used to make a <i>no-op cast</i> because it always changes bits. Use
3527 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3531 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3532 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3536 <!-- _______________________________________________________________________ -->
3537 <div class="doc_subsubsection">
3538 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3540 <div class="doc_text">
3544 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3548 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3549 unsigned integer equivalent of type <tt>ty2</tt>.
3553 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3554 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3555 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3556 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3557 vector integer type with the same number of elements as <tt>ty</tt></p>
3560 <p> The '<tt>fptoui</tt>' instruction converts its
3561 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3562 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3563 the results are undefined.</p>
3567 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3568 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3569 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3573 <!-- _______________________________________________________________________ -->
3574 <div class="doc_subsubsection">
3575 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3577 <div class="doc_text">
3581 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3585 <p>The '<tt>fptosi</tt>' instruction converts
3586 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3590 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3591 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3592 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3593 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3594 vector integer type with the same number of elements as <tt>ty</tt></p>
3597 <p>The '<tt>fptosi</tt>' instruction converts its
3598 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3599 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3600 the results are undefined.</p>
3604 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3605 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3606 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3610 <!-- _______________________________________________________________________ -->
3611 <div class="doc_subsubsection">
3612 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3614 <div class="doc_text">
3618 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3622 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3623 integer and converts that value to the <tt>ty2</tt> type.</p>
3626 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3627 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3628 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3629 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3630 floating point type with the same number of elements as <tt>ty</tt></p>
3633 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3634 integer quantity and converts it to the corresponding floating point value. If
3635 the value cannot fit in the floating point value, the results are undefined.</p>
3639 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3640 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3644 <!-- _______________________________________________________________________ -->
3645 <div class="doc_subsubsection">
3646 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3648 <div class="doc_text">
3652 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3656 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3657 integer and converts that value to the <tt>ty2</tt> type.</p>
3660 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3661 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3662 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3663 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3664 floating point type with the same number of elements as <tt>ty</tt></p>
3667 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3668 integer quantity and converts it to the corresponding floating point value. If
3669 the value cannot fit in the floating point value, the results are undefined.</p>
3673 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3674 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3678 <!-- _______________________________________________________________________ -->
3679 <div class="doc_subsubsection">
3680 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3682 <div class="doc_text">
3686 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3690 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3691 the integer type <tt>ty2</tt>.</p>
3694 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3695 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3696 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3699 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3700 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3701 truncating or zero extending that value to the size of the integer type. If
3702 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3703 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3704 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3709 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3710 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3714 <!-- _______________________________________________________________________ -->
3715 <div class="doc_subsubsection">
3716 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3718 <div class="doc_text">
3722 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3726 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3727 a pointer type, <tt>ty2</tt>.</p>
3730 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3731 value to cast, and a type to cast it to, which must be a
3732 <a href="#t_pointer">pointer</a> type.
3735 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3736 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3737 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3738 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3739 the size of a pointer then a zero extension is done. If they are the same size,
3740 nothing is done (<i>no-op cast</i>).</p>
3744 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3745 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3746 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3750 <!-- _______________________________________________________________________ -->
3751 <div class="doc_subsubsection">
3752 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3754 <div class="doc_text">
3758 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3763 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3764 <tt>ty2</tt> without changing any bits.</p>
3768 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3769 a first class value, and a type to cast it to, which must also be a <a
3770 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3771 and the destination type, <tt>ty2</tt>, must be identical. If the source
3772 type is a pointer, the destination type must also be a pointer. This
3773 instruction supports bitwise conversion of vectors to integers and to vectors
3774 of other types (as long as they have the same size).</p>
3777 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3778 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3779 this conversion. The conversion is done as if the <tt>value</tt> had been
3780 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3781 converted to other pointer types with this instruction. To convert pointers to
3782 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3783 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3787 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3788 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3789 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3793 <!-- ======================================================================= -->
3794 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3795 <div class="doc_text">
3796 <p>The instructions in this category are the "miscellaneous"
3797 instructions, which defy better classification.</p>
3800 <!-- _______________________________________________________________________ -->
3801 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3803 <div class="doc_text">
3805 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3808 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3809 of its two integer or pointer operands.</p>
3811 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3812 the condition code indicating the kind of comparison to perform. It is not
3813 a value, just a keyword. The possible condition code are:
3815 <li><tt>eq</tt>: equal</li>
3816 <li><tt>ne</tt>: not equal </li>
3817 <li><tt>ugt</tt>: unsigned greater than</li>
3818 <li><tt>uge</tt>: unsigned greater or equal</li>
3819 <li><tt>ult</tt>: unsigned less than</li>
3820 <li><tt>ule</tt>: unsigned less or equal</li>
3821 <li><tt>sgt</tt>: signed greater than</li>
3822 <li><tt>sge</tt>: signed greater or equal</li>
3823 <li><tt>slt</tt>: signed less than</li>
3824 <li><tt>sle</tt>: signed less or equal</li>
3826 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3827 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3829 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3830 the condition code given as <tt>cond</tt>. The comparison performed always
3831 yields a <a href="#t_primitive">i1</a> result, as follows:
3833 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3834 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3836 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3837 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3838 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3839 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3840 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3841 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3842 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3843 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3844 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3845 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3846 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3847 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3848 <li><tt>sge</tt>: interprets the operands as signed values and yields
3849 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3850 <li><tt>slt</tt>: interprets the operands as signed values and yields
3851 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3852 <li><tt>sle</tt>: interprets the operands as signed values and yields
3853 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3855 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3856 values are compared as if they were integers.</p>
3859 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3860 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3861 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3862 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3863 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3864 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3868 <!-- _______________________________________________________________________ -->
3869 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3871 <div class="doc_text">
3873 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3876 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3877 of its floating point operands.</p>
3879 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3880 the condition code indicating the kind of comparison to perform. It is not
3881 a value, just a keyword. The possible condition code are:
3883 <li><tt>false</tt>: no comparison, always returns false</li>
3884 <li><tt>oeq</tt>: ordered and equal</li>
3885 <li><tt>ogt</tt>: ordered and greater than </li>
3886 <li><tt>oge</tt>: ordered and greater than or equal</li>
3887 <li><tt>olt</tt>: ordered and less than </li>
3888 <li><tt>ole</tt>: ordered and less than or equal</li>
3889 <li><tt>one</tt>: ordered and not equal</li>
3890 <li><tt>ord</tt>: ordered (no nans)</li>
3891 <li><tt>ueq</tt>: unordered or equal</li>
3892 <li><tt>ugt</tt>: unordered or greater than </li>
3893 <li><tt>uge</tt>: unordered or greater than or equal</li>
3894 <li><tt>ult</tt>: unordered or less than </li>
3895 <li><tt>ule</tt>: unordered or less than or equal</li>
3896 <li><tt>une</tt>: unordered or not equal</li>
3897 <li><tt>uno</tt>: unordered (either nans)</li>
3898 <li><tt>true</tt>: no comparison, always returns true</li>
3900 <p><i>Ordered</i> means that neither operand is a QNAN while
3901 <i>unordered</i> means that either operand may be a QNAN.</p>
3902 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3903 <a href="#t_floating">floating point</a> typed. They must have identical
3906 <p>The '<tt>fcmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3907 according to the condition code given as <tt>cond</tt>. The comparison performed
3908 always yields a <a href="#t_primitive">i1</a> result, as follows:
3910 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3911 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3912 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3913 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3914 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3915 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3916 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3917 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3918 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3919 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3920 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3921 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3922 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3923 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3924 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3925 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3926 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3927 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3928 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3929 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3930 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3931 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3932 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3933 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3934 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3935 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3936 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3937 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3941 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3942 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3943 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3944 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3948 <!-- _______________________________________________________________________ -->
3949 <div class="doc_subsubsection">
3950 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
3952 <div class="doc_text">
3954 <pre> <result> = vicmp <cond> <ty> <var1>, <var2> <i>; yields {ty}:result</i>
3957 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
3958 element-wise comparison of its two integer vector operands.</p>
3960 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
3961 the condition code indicating the kind of comparison to perform. It is not
3962 a value, just a keyword. The possible condition code are:
3964 <li><tt>eq</tt>: equal</li>
3965 <li><tt>ne</tt>: not equal </li>
3966 <li><tt>ugt</tt>: unsigned greater than</li>
3967 <li><tt>uge</tt>: unsigned greater or equal</li>
3968 <li><tt>ult</tt>: unsigned less than</li>
3969 <li><tt>ule</tt>: unsigned less or equal</li>
3970 <li><tt>sgt</tt>: signed greater than</li>
3971 <li><tt>sge</tt>: signed greater or equal</li>
3972 <li><tt>slt</tt>: signed less than</li>
3973 <li><tt>sle</tt>: signed less or equal</li>
3975 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
3976 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
3978 <p>The '<tt>vicmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3979 according to the condition code given as <tt>cond</tt>. The comparison yields a
3980 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
3981 identical type as the values being compared. The most significant bit in each
3982 element is 1 if the element-wise comparison evaluates to true, and is 0
3983 otherwise. All other bits of the result are undefined. The condition codes
3984 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
3989 <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>
3990 <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>
3994 <!-- _______________________________________________________________________ -->
3995 <div class="doc_subsubsection">
3996 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
3998 <div class="doc_text">
4000 <pre> <result> = vfcmp <cond> <ty> <var1>, <var2></pre>
4002 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4003 element-wise comparison of its two floating point vector operands. The output
4004 elements have the same width as the input elements.</p>
4006 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4007 the condition code indicating the kind of comparison to perform. It is not
4008 a value, just a keyword. The possible condition code are:
4010 <li><tt>false</tt>: no comparison, always returns false</li>
4011 <li><tt>oeq</tt>: ordered and equal</li>
4012 <li><tt>ogt</tt>: ordered and greater than </li>
4013 <li><tt>oge</tt>: ordered and greater than or equal</li>
4014 <li><tt>olt</tt>: ordered and less than </li>
4015 <li><tt>ole</tt>: ordered and less than or equal</li>
4016 <li><tt>one</tt>: ordered and not equal</li>
4017 <li><tt>ord</tt>: ordered (no nans)</li>
4018 <li><tt>ueq</tt>: unordered or equal</li>
4019 <li><tt>ugt</tt>: unordered or greater than </li>
4020 <li><tt>uge</tt>: unordered or greater than or equal</li>
4021 <li><tt>ult</tt>: unordered or less than </li>
4022 <li><tt>ule</tt>: unordered or less than or equal</li>
4023 <li><tt>une</tt>: unordered or not equal</li>
4024 <li><tt>uno</tt>: unordered (either nans)</li>
4025 <li><tt>true</tt>: no comparison, always returns true</li>
4027 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4028 <a href="#t_floating">floating point</a> typed. They must also be identical
4031 <p>The '<tt>vfcmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
4032 according to the condition code given as <tt>cond</tt>. The comparison yields a
4033 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4034 an identical number of elements as the values being compared, and each element
4035 having identical with to the width of the floating point elements. The most
4036 significant bit in each element is 1 if the element-wise comparison evaluates to
4037 true, and is 0 otherwise. All other bits of the result are undefined. The
4038 condition codes are evaluated identically to the
4039 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.
4043 <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>
4044 <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>
4048 <!-- _______________________________________________________________________ -->
4049 <div class="doc_subsubsection">
4050 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4053 <div class="doc_text">
4057 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4059 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4060 the SSA graph representing the function.</p>
4063 <p>The type of the incoming values is specified with the first type
4064 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4065 as arguments, with one pair for each predecessor basic block of the
4066 current block. Only values of <a href="#t_firstclass">first class</a>
4067 type may be used as the value arguments to the PHI node. Only labels
4068 may be used as the label arguments.</p>
4070 <p>There must be no non-phi instructions between the start of a basic
4071 block and the PHI instructions: i.e. PHI instructions must be first in
4076 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4077 specified by the pair corresponding to the predecessor basic block that executed
4078 just prior to the current block.</p>
4082 Loop: ; Infinite loop that counts from 0 on up...
4083 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4084 %nextindvar = add i32 %indvar, 1
4089 <!-- _______________________________________________________________________ -->
4090 <div class="doc_subsubsection">
4091 <a name="i_select">'<tt>select</tt>' Instruction</a>
4094 <div class="doc_text">
4099 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4105 The '<tt>select</tt>' instruction is used to choose one value based on a
4106 condition, without branching.
4113 The '<tt>select</tt>' instruction requires an 'i1' value indicating the
4114 condition, and two values of the same <a href="#t_firstclass">first class</a>
4115 type. If the val1/val2 are vectors, the entire vectors are selected, not
4116 individual elements.
4122 If the i1 condition evaluates is 1, the instruction returns the first
4123 value argument; otherwise, it returns the second value argument.
4129 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4134 <!-- _______________________________________________________________________ -->
4135 <div class="doc_subsubsection">
4136 <a name="i_call">'<tt>call</tt>' Instruction</a>
4139 <div class="doc_text">
4143 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
4148 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4152 <p>This instruction requires several arguments:</p>
4156 <p>The optional "tail" marker indicates whether the callee function accesses
4157 any allocas or varargs in the caller. If the "tail" marker is present, the
4158 function call is eligible for tail call optimization. Note that calls may
4159 be marked "tail" even if they do not occur before a <a
4160 href="#i_ret"><tt>ret</tt></a> instruction.
4163 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4164 convention</a> the call should use. If none is specified, the call defaults
4165 to using C calling conventions.
4168 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4169 the type of the return value. Functions that return no value are marked
4170 <tt><a href="#t_void">void</a></tt>.</p>
4173 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4174 value being invoked. The argument types must match the types implied by
4175 this signature. This type can be omitted if the function is not varargs
4176 and if the function type does not return a pointer to a function.</p>
4179 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4180 be invoked. In most cases, this is a direct function invocation, but
4181 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4182 to function value.</p>
4185 <p>'<tt>function args</tt>': argument list whose types match the
4186 function signature argument types. All arguments must be of
4187 <a href="#t_firstclass">first class</a> type. If the function signature
4188 indicates the function accepts a variable number of arguments, the extra
4189 arguments can be specified.</p>
4195 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4196 transfer to a specified function, with its incoming arguments bound to
4197 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4198 instruction in the called function, control flow continues with the
4199 instruction after the function call, and the return value of the
4200 function is bound to the result argument. If the callee returns multiple
4201 values then the return values of the function are only accessible through
4202 the '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
4207 %retval = call i32 @test(i32 %argc)
4208 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4209 %X = tail call i32 @foo() <i>; yields i32</i>
4210 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4211 call void %foo(i8 97 signext)
4213 %struct.A = type { i32, i8 }
4214 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4215 %gr = getresult %struct.A %r, 0 <i>; yields i32</i>
4216 %gr1 = getresult %struct.A %r, 1 <i>; yields i8</i>
4221 <!-- _______________________________________________________________________ -->
4222 <div class="doc_subsubsection">
4223 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4226 <div class="doc_text">
4231 <resultval> = va_arg <va_list*> <arglist>, <argty>
4236 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4237 the "variable argument" area of a function call. It is used to implement the
4238 <tt>va_arg</tt> macro in C.</p>
4242 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4243 the argument. It returns a value of the specified argument type and
4244 increments the <tt>va_list</tt> to point to the next argument. The
4245 actual type of <tt>va_list</tt> is target specific.</p>
4249 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4250 type from the specified <tt>va_list</tt> and causes the
4251 <tt>va_list</tt> to point to the next argument. For more information,
4252 see the variable argument handling <a href="#int_varargs">Intrinsic
4255 <p>It is legal for this instruction to be called in a function which does not
4256 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4259 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4260 href="#intrinsics">intrinsic function</a> because it takes a type as an
4265 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4269 <!-- _______________________________________________________________________ -->
4270 <div class="doc_subsubsection">
4271 <a name="i_getresult">'<tt>getresult</tt>' Instruction</a>
4274 <div class="doc_text">
4278 <resultval> = getresult <type> <retval>, <index>
4283 <p> The '<tt>getresult</tt>' instruction is used to extract individual values
4284 from a '<tt><a href="#i_call">call</a></tt>'
4285 or '<tt><a href="#i_invoke">invoke</a></tt>' instruction that returns multiple
4290 <p>The '<tt>getresult</tt>' instruction takes a call or invoke value as its
4291 first argument, or an undef value. The value must have <a
4292 href="#t_struct">structure type</a>. The second argument is a constant
4293 unsigned index value which must be in range for the number of values returned
4298 <p>The '<tt>getresult</tt>' instruction extracts the element identified by
4299 '<tt>index</tt>' from the aggregate value.</p>
4304 %struct.A = type { i32, i8 }
4306 %r = call %struct.A @foo()
4307 %gr = getresult %struct.A %r, 0 <i>; yields i32:%gr</i>
4308 %gr1 = getresult %struct.A %r, 1 <i>; yields i8:%gr1</i>
4315 <!-- *********************************************************************** -->
4316 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4317 <!-- *********************************************************************** -->
4319 <div class="doc_text">
4321 <p>LLVM supports the notion of an "intrinsic function". These functions have
4322 well known names and semantics and are required to follow certain restrictions.
4323 Overall, these intrinsics represent an extension mechanism for the LLVM
4324 language that does not require changing all of the transformations in LLVM when
4325 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4327 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4328 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4329 begin with this prefix. Intrinsic functions must always be external functions:
4330 you cannot define the body of intrinsic functions. Intrinsic functions may
4331 only be used in call or invoke instructions: it is illegal to take the address
4332 of an intrinsic function. Additionally, because intrinsic functions are part
4333 of the LLVM language, it is required if any are added that they be documented
4336 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4337 a family of functions that perform the same operation but on different data
4338 types. Because LLVM can represent over 8 million different integer types,
4339 overloading is used commonly to allow an intrinsic function to operate on any
4340 integer type. One or more of the argument types or the result type can be
4341 overloaded to accept any integer type. Argument types may also be defined as
4342 exactly matching a previous argument's type or the result type. This allows an
4343 intrinsic function which accepts multiple arguments, but needs all of them to
4344 be of the same type, to only be overloaded with respect to a single argument or
4347 <p>Overloaded intrinsics will have the names of its overloaded argument types
4348 encoded into its function name, each preceded by a period. Only those types
4349 which are overloaded result in a name suffix. Arguments whose type is matched
4350 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4351 take an integer of any width and returns an integer of exactly the same integer
4352 width. This leads to a family of functions such as
4353 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4354 Only one type, the return type, is overloaded, and only one type suffix is
4355 required. Because the argument's type is matched against the return type, it
4356 does not require its own name suffix.</p>
4358 <p>To learn how to add an intrinsic function, please see the
4359 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4364 <!-- ======================================================================= -->
4365 <div class="doc_subsection">
4366 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4369 <div class="doc_text">
4371 <p>Variable argument support is defined in LLVM with the <a
4372 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4373 intrinsic functions. These functions are related to the similarly
4374 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4376 <p>All of these functions operate on arguments that use a
4377 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4378 language reference manual does not define what this type is, so all
4379 transformations should be prepared to handle these functions regardless of
4382 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4383 instruction and the variable argument handling intrinsic functions are
4386 <div class="doc_code">
4388 define i32 @test(i32 %X, ...) {
4389 ; Initialize variable argument processing
4391 %ap2 = bitcast i8** %ap to i8*
4392 call void @llvm.va_start(i8* %ap2)
4394 ; Read a single integer argument
4395 %tmp = va_arg i8** %ap, i32
4397 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4399 %aq2 = bitcast i8** %aq to i8*
4400 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4401 call void @llvm.va_end(i8* %aq2)
4403 ; Stop processing of arguments.
4404 call void @llvm.va_end(i8* %ap2)
4408 declare void @llvm.va_start(i8*)
4409 declare void @llvm.va_copy(i8*, i8*)
4410 declare void @llvm.va_end(i8*)
4416 <!-- _______________________________________________________________________ -->
4417 <div class="doc_subsubsection">
4418 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4422 <div class="doc_text">
4424 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4426 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
4427 <tt>*<arglist></tt> for subsequent use by <tt><a
4428 href="#i_va_arg">va_arg</a></tt>.</p>
4432 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4436 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4437 macro available in C. In a target-dependent way, it initializes the
4438 <tt>va_list</tt> element to which the argument points, so that the next call to
4439 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4440 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4441 last argument of the function as the compiler can figure that out.</p>
4445 <!-- _______________________________________________________________________ -->
4446 <div class="doc_subsubsection">
4447 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4450 <div class="doc_text">
4452 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4455 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4456 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4457 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4461 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4465 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4466 macro available in C. In a target-dependent way, it destroys the
4467 <tt>va_list</tt> element to which the argument points. Calls to <a
4468 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4469 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4470 <tt>llvm.va_end</tt>.</p>
4474 <!-- _______________________________________________________________________ -->
4475 <div class="doc_subsubsection">
4476 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4479 <div class="doc_text">
4484 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4489 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4490 from the source argument list to the destination argument list.</p>
4494 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4495 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4500 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4501 macro available in C. In a target-dependent way, it copies the source
4502 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4503 intrinsic is necessary because the <tt><a href="#int_va_start">
4504 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4505 example, memory allocation.</p>
4509 <!-- ======================================================================= -->
4510 <div class="doc_subsection">
4511 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4514 <div class="doc_text">
4517 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4518 Collection</a> requires the implementation and generation of these intrinsics.
4519 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4520 stack</a>, as well as garbage collector implementations that require <a
4521 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4522 Front-ends for type-safe garbage collected languages should generate these
4523 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4524 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4527 <p>The garbage collection intrinsics only operate on objects in the generic
4528 address space (address space zero).</p>
4532 <!-- _______________________________________________________________________ -->
4533 <div class="doc_subsubsection">
4534 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4537 <div class="doc_text">
4542 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4547 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4548 the code generator, and allows some metadata to be associated with it.</p>
4552 <p>The first argument specifies the address of a stack object that contains the
4553 root pointer. The second pointer (which must be either a constant or a global
4554 value address) contains the meta-data to be associated with the root.</p>
4558 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4559 location. At compile-time, the code generator generates information to allow
4560 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4561 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4567 <!-- _______________________________________________________________________ -->
4568 <div class="doc_subsubsection">
4569 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4572 <div class="doc_text">
4577 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4582 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4583 locations, allowing garbage collector implementations that require read
4588 <p>The second argument is the address to read from, which should be an address
4589 allocated from the garbage collector. The first object is a pointer to the
4590 start of the referenced object, if needed by the language runtime (otherwise
4595 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4596 instruction, but may be replaced with substantially more complex code by the
4597 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4598 may only be used in a function which <a href="#gc">specifies a GC
4604 <!-- _______________________________________________________________________ -->
4605 <div class="doc_subsubsection">
4606 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4609 <div class="doc_text">
4614 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4619 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4620 locations, allowing garbage collector implementations that require write
4621 barriers (such as generational or reference counting collectors).</p>
4625 <p>The first argument is the reference to store, the second is the start of the
4626 object to store it to, and the third is the address of the field of Obj to
4627 store to. If the runtime does not require a pointer to the object, Obj may be
4632 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4633 instruction, but may be replaced with substantially more complex code by the
4634 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4635 may only be used in a function which <a href="#gc">specifies a GC
4642 <!-- ======================================================================= -->
4643 <div class="doc_subsection">
4644 <a name="int_codegen">Code Generator Intrinsics</a>
4647 <div class="doc_text">
4649 These intrinsics are provided by LLVM to expose special features that may only
4650 be implemented with code generator support.
4655 <!-- _______________________________________________________________________ -->
4656 <div class="doc_subsubsection">
4657 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4660 <div class="doc_text">
4664 declare i8 *@llvm.returnaddress(i32 <level>)
4670 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4671 target-specific value indicating the return address of the current function
4672 or one of its callers.
4678 The argument to this intrinsic indicates which function to return the address
4679 for. Zero indicates the calling function, one indicates its caller, etc. The
4680 argument is <b>required</b> to be a constant integer value.
4686 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4687 the return address of the specified call frame, or zero if it cannot be
4688 identified. The value returned by this intrinsic is likely to be incorrect or 0
4689 for arguments other than zero, so it should only be used for debugging purposes.
4693 Note that calling this intrinsic does not prevent function inlining or other
4694 aggressive transformations, so the value returned may not be that of the obvious
4695 source-language caller.
4700 <!-- _______________________________________________________________________ -->
4701 <div class="doc_subsubsection">
4702 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4705 <div class="doc_text">
4709 declare i8 *@llvm.frameaddress(i32 <level>)
4715 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4716 target-specific frame pointer value for the specified stack frame.
4722 The argument to this intrinsic indicates which function to return the frame
4723 pointer for. Zero indicates the calling function, one indicates its caller,
4724 etc. The argument is <b>required</b> to be a constant integer value.
4730 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4731 the frame address of the specified call frame, or zero if it cannot be
4732 identified. The value returned by this intrinsic is likely to be incorrect or 0
4733 for arguments other than zero, so it should only be used for debugging purposes.
4737 Note that calling this intrinsic does not prevent function inlining or other
4738 aggressive transformations, so the value returned may not be that of the obvious
4739 source-language caller.
4743 <!-- _______________________________________________________________________ -->
4744 <div class="doc_subsubsection">
4745 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4748 <div class="doc_text">
4752 declare i8 *@llvm.stacksave()
4758 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4759 the function stack, for use with <a href="#int_stackrestore">
4760 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4761 features like scoped automatic variable sized arrays in C99.
4767 This intrinsic returns a opaque pointer value that can be passed to <a
4768 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4769 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4770 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4771 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4772 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4773 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4778 <!-- _______________________________________________________________________ -->
4779 <div class="doc_subsubsection">
4780 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4783 <div class="doc_text">
4787 declare void @llvm.stackrestore(i8 * %ptr)
4793 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4794 the function stack to the state it was in when the corresponding <a
4795 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4796 useful for implementing language features like scoped automatic variable sized
4803 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4809 <!-- _______________________________________________________________________ -->
4810 <div class="doc_subsubsection">
4811 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4814 <div class="doc_text">
4818 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4825 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4826 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4828 effect on the behavior of the program but can change its performance
4835 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4836 determining if the fetch should be for a read (0) or write (1), and
4837 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4838 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4839 <tt>locality</tt> arguments must be constant integers.
4845 This intrinsic does not modify the behavior of the program. In particular,
4846 prefetches cannot trap and do not produce a value. On targets that support this
4847 intrinsic, the prefetch can provide hints to the processor cache for better
4853 <!-- _______________________________________________________________________ -->
4854 <div class="doc_subsubsection">
4855 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4858 <div class="doc_text">
4862 declare void @llvm.pcmarker(i32 <id>)
4869 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4871 code to simulators and other tools. The method is target specific, but it is
4872 expected that the marker will use exported symbols to transmit the PC of the marker.
4873 The marker makes no guarantees that it will remain with any specific instruction
4874 after optimizations. It is possible that the presence of a marker will inhibit
4875 optimizations. The intended use is to be inserted after optimizations to allow
4876 correlations of simulation runs.
4882 <tt>id</tt> is a numerical id identifying the marker.
4888 This intrinsic does not modify the behavior of the program. Backends that do not
4889 support this intrinisic may ignore it.
4894 <!-- _______________________________________________________________________ -->
4895 <div class="doc_subsubsection">
4896 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4899 <div class="doc_text">
4903 declare i64 @llvm.readcyclecounter( )
4910 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4911 counter register (or similar low latency, high accuracy clocks) on those targets
4912 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4913 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4914 should only be used for small timings.
4920 When directly supported, reading the cycle counter should not modify any memory.
4921 Implementations are allowed to either return a application specific value or a
4922 system wide value. On backends without support, this is lowered to a constant 0.
4927 <!-- ======================================================================= -->
4928 <div class="doc_subsection">
4929 <a name="int_libc">Standard C Library Intrinsics</a>
4932 <div class="doc_text">
4934 LLVM provides intrinsics for a few important standard C library functions.
4935 These intrinsics allow source-language front-ends to pass information about the
4936 alignment of the pointer arguments to the code generator, providing opportunity
4937 for more efficient code generation.
4942 <!-- _______________________________________________________________________ -->
4943 <div class="doc_subsubsection">
4944 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4947 <div class="doc_text">
4951 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4952 i32 <len>, i32 <align>)
4953 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4954 i64 <len>, i32 <align>)
4960 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4961 location to the destination location.
4965 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4966 intrinsics do not return a value, and takes an extra alignment argument.
4972 The first argument is a pointer to the destination, the second is a pointer to
4973 the source. The third argument is an integer argument
4974 specifying the number of bytes to copy, and the fourth argument is the alignment
4975 of the source and destination locations.
4979 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4980 the caller guarantees that both the source and destination pointers are aligned
4987 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4988 location to the destination location, which are not allowed to overlap. It
4989 copies "len" bytes of memory over. If the argument is known to be aligned to
4990 some boundary, this can be specified as the fourth argument, otherwise it should
4996 <!-- _______________________________________________________________________ -->
4997 <div class="doc_subsubsection">
4998 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5001 <div class="doc_text">
5005 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5006 i32 <len>, i32 <align>)
5007 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5008 i64 <len>, i32 <align>)
5014 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5015 location to the destination location. It is similar to the
5016 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5020 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5021 intrinsics do not return a value, and takes an extra alignment argument.
5027 The first argument is a pointer to the destination, the second is a pointer to
5028 the source. The third argument is an integer argument
5029 specifying the number of bytes to copy, and the fourth argument is the alignment
5030 of the source and destination locations.
5034 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5035 the caller guarantees that the source and destination pointers are aligned to
5042 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5043 location to the destination location, which may overlap. It
5044 copies "len" bytes of memory over. If the argument is known to be aligned to
5045 some boundary, this can be specified as the fourth argument, otherwise it should
5051 <!-- _______________________________________________________________________ -->
5052 <div class="doc_subsubsection">
5053 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5056 <div class="doc_text">
5060 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5061 i32 <len>, i32 <align>)
5062 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5063 i64 <len>, i32 <align>)
5069 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5074 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5075 does not return a value, and takes an extra alignment argument.
5081 The first argument is a pointer to the destination to fill, the second is the
5082 byte value to fill it with, the third argument is an integer
5083 argument specifying the number of bytes to fill, and the fourth argument is the
5084 known alignment of destination location.
5088 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5089 the caller guarantees that the destination pointer is aligned to that boundary.
5095 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5097 destination location. If the argument is known to be aligned to some boundary,
5098 this can be specified as the fourth argument, otherwise it should be set to 0 or
5104 <!-- _______________________________________________________________________ -->
5105 <div class="doc_subsubsection">
5106 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5109 <div class="doc_text">
5112 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5113 floating point or vector of floating point type. Not all targets support all
5116 declare float @llvm.sqrt.f32(float %Val)
5117 declare double @llvm.sqrt.f64(double %Val)
5118 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5119 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5120 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5126 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5127 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5128 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5129 negative numbers other than -0.0 (which allows for better optimization, because
5130 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5131 defined to return -0.0 like IEEE sqrt.
5137 The argument and return value are floating point numbers of the same type.
5143 This function returns the sqrt of the specified operand if it is a nonnegative
5144 floating point number.
5148 <!-- _______________________________________________________________________ -->
5149 <div class="doc_subsubsection">
5150 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5153 <div class="doc_text">
5156 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5157 floating point or vector of floating point type. Not all targets support all
5160 declare float @llvm.powi.f32(float %Val, i32 %power)
5161 declare double @llvm.powi.f64(double %Val, i32 %power)
5162 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5163 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5164 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5170 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5171 specified (positive or negative) power. The order of evaluation of
5172 multiplications is not defined. When a vector of floating point type is
5173 used, the second argument remains a scalar integer value.
5179 The second argument is an integer power, and the first is a value to raise to
5186 This function returns the first value raised to the second power with an
5187 unspecified sequence of rounding operations.</p>
5190 <!-- _______________________________________________________________________ -->
5191 <div class="doc_subsubsection">
5192 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5195 <div class="doc_text">
5198 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5199 floating point or vector of floating point type. Not all targets support all
5202 declare float @llvm.sin.f32(float %Val)
5203 declare double @llvm.sin.f64(double %Val)
5204 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5205 declare fp128 @llvm.sin.f128(fp128 %Val)
5206 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5212 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5218 The argument and return value are floating point numbers of the same type.
5224 This function returns the sine of the specified operand, returning the
5225 same values as the libm <tt>sin</tt> functions would, and handles error
5226 conditions in the same way.</p>
5229 <!-- _______________________________________________________________________ -->
5230 <div class="doc_subsubsection">
5231 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5234 <div class="doc_text">
5237 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5238 floating point or vector of floating point type. Not all targets support all
5241 declare float @llvm.cos.f32(float %Val)
5242 declare double @llvm.cos.f64(double %Val)
5243 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5244 declare fp128 @llvm.cos.f128(fp128 %Val)
5245 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5251 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5257 The argument and return value are floating point numbers of the same type.
5263 This function returns the cosine of the specified operand, returning the
5264 same values as the libm <tt>cos</tt> functions would, and handles error
5265 conditions in the same way.</p>
5268 <!-- _______________________________________________________________________ -->
5269 <div class="doc_subsubsection">
5270 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5273 <div class="doc_text">
5276 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5277 floating point or vector of floating point type. Not all targets support all
5280 declare float @llvm.pow.f32(float %Val, float %Power)
5281 declare double @llvm.pow.f64(double %Val, double %Power)
5282 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5283 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5284 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5290 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5291 specified (positive or negative) power.
5297 The second argument is a floating point power, and the first is a value to
5298 raise to that power.
5304 This function returns the first value raised to the second power,
5306 same values as the libm <tt>pow</tt> functions would, and handles error
5307 conditions in the same way.</p>
5311 <!-- ======================================================================= -->
5312 <div class="doc_subsection">
5313 <a name="int_manip">Bit Manipulation Intrinsics</a>
5316 <div class="doc_text">
5318 LLVM provides intrinsics for a few important bit manipulation operations.
5319 These allow efficient code generation for some algorithms.
5324 <!-- _______________________________________________________________________ -->
5325 <div class="doc_subsubsection">
5326 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5329 <div class="doc_text">
5332 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5333 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
5335 declare i16 @llvm.bswap.i16(i16 <id>)
5336 declare i32 @llvm.bswap.i32(i32 <id>)
5337 declare i64 @llvm.bswap.i64(i64 <id>)
5343 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5344 values with an even number of bytes (positive multiple of 16 bits). These are
5345 useful for performing operations on data that is not in the target's native
5352 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5353 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5354 intrinsic returns an i32 value that has the four bytes of the input i32
5355 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5356 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5357 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5358 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5363 <!-- _______________________________________________________________________ -->
5364 <div class="doc_subsubsection">
5365 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5368 <div class="doc_text">
5371 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5372 width. Not all targets support all bit widths however.
5374 declare i8 @llvm.ctpop.i8 (i8 <src>)
5375 declare i16 @llvm.ctpop.i16(i16 <src>)
5376 declare i32 @llvm.ctpop.i32(i32 <src>)
5377 declare i64 @llvm.ctpop.i64(i64 <src>)
5378 declare i256 @llvm.ctpop.i256(i256 <src>)
5384 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5391 The only argument is the value to be counted. The argument may be of any
5392 integer type. The return type must match the argument type.
5398 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5402 <!-- _______________________________________________________________________ -->
5403 <div class="doc_subsubsection">
5404 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5407 <div class="doc_text">
5410 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5411 integer bit width. Not all targets support all bit widths however.
5413 declare i8 @llvm.ctlz.i8 (i8 <src>)
5414 declare i16 @llvm.ctlz.i16(i16 <src>)
5415 declare i32 @llvm.ctlz.i32(i32 <src>)
5416 declare i64 @llvm.ctlz.i64(i64 <src>)
5417 declare i256 @llvm.ctlz.i256(i256 <src>)
5423 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5424 leading zeros in a variable.
5430 The only argument is the value to be counted. The argument may be of any
5431 integer type. The return type must match the argument type.
5437 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5438 in a variable. If the src == 0 then the result is the size in bits of the type
5439 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5445 <!-- _______________________________________________________________________ -->
5446 <div class="doc_subsubsection">
5447 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5450 <div class="doc_text">
5453 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5454 integer bit width. Not all targets support all bit widths however.
5456 declare i8 @llvm.cttz.i8 (i8 <src>)
5457 declare i16 @llvm.cttz.i16(i16 <src>)
5458 declare i32 @llvm.cttz.i32(i32 <src>)
5459 declare i64 @llvm.cttz.i64(i64 <src>)
5460 declare i256 @llvm.cttz.i256(i256 <src>)
5466 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5473 The only argument is the value to be counted. The argument may be of any
5474 integer type. The return type must match the argument type.
5480 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5481 in a variable. If the src == 0 then the result is the size in bits of the type
5482 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5486 <!-- _______________________________________________________________________ -->
5487 <div class="doc_subsubsection">
5488 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5491 <div class="doc_text">
5494 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5495 on any integer bit width.
5497 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5498 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5502 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5503 range of bits from an integer value and returns them in the same bit width as
5504 the original value.</p>
5507 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5508 any bit width but they must have the same bit width. The second and third
5509 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5512 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5513 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5514 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5515 operates in forward mode.</p>
5516 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5517 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5518 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5520 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5521 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5522 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5523 to determine the number of bits to retain.</li>
5524 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5525 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5527 <p>In reverse mode, a similar computation is made except that the bits are
5528 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5529 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5530 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5531 <tt>i16 0x0026 (000000100110)</tt>.</p>
5534 <div class="doc_subsubsection">
5535 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5538 <div class="doc_text">
5541 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5542 on any integer bit width.
5544 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5545 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5549 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5550 of bits in an integer value with another integer value. It returns the integer
5551 with the replaced bits.</p>
5554 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5555 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5556 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5557 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5558 type since they specify only a bit index.</p>
5561 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5562 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5563 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5564 operates in forward mode.</p>
5565 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5566 truncating it down to the size of the replacement area or zero extending it
5567 up to that size.</p>
5568 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5569 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5570 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5571 to the <tt>%hi</tt>th bit.
5572 <p>In reverse mode, a similar computation is made except that the bits are
5573 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5574 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5577 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5578 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5579 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5580 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5581 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5585 <!-- ======================================================================= -->
5586 <div class="doc_subsection">
5587 <a name="int_debugger">Debugger Intrinsics</a>
5590 <div class="doc_text">
5592 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5593 are described in the <a
5594 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5595 Debugging</a> document.
5600 <!-- ======================================================================= -->
5601 <div class="doc_subsection">
5602 <a name="int_eh">Exception Handling Intrinsics</a>
5605 <div class="doc_text">
5606 <p> The LLVM exception handling intrinsics (which all start with
5607 <tt>llvm.eh.</tt> prefix), are described in the <a
5608 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5609 Handling</a> document. </p>
5612 <!-- ======================================================================= -->
5613 <div class="doc_subsection">
5614 <a name="int_trampoline">Trampoline Intrinsic</a>
5617 <div class="doc_text">
5619 This intrinsic makes it possible to excise one parameter, marked with
5620 the <tt>nest</tt> attribute, from a function. The result is a callable
5621 function pointer lacking the nest parameter - the caller does not need
5622 to provide a value for it. Instead, the value to use is stored in
5623 advance in a "trampoline", a block of memory usually allocated
5624 on the stack, which also contains code to splice the nest value into the
5625 argument list. This is used to implement the GCC nested function address
5629 For example, if the function is
5630 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5631 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5633 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5634 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5635 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5636 %fp = bitcast i8* %p to i32 (i32, i32)*
5638 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5639 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5642 <!-- _______________________________________________________________________ -->
5643 <div class="doc_subsubsection">
5644 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5646 <div class="doc_text">
5649 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5653 This fills the memory pointed to by <tt>tramp</tt> with code
5654 and returns a function pointer suitable for executing it.
5658 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5659 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5660 and sufficiently aligned block of memory; this memory is written to by the
5661 intrinsic. Note that the size and the alignment are target-specific - LLVM
5662 currently provides no portable way of determining them, so a front-end that
5663 generates this intrinsic needs to have some target-specific knowledge.
5664 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5668 The block of memory pointed to by <tt>tramp</tt> is filled with target
5669 dependent code, turning it into a function. A pointer to this function is
5670 returned, but needs to be bitcast to an
5671 <a href="#int_trampoline">appropriate function pointer type</a>
5672 before being called. The new function's signature is the same as that of
5673 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5674 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5675 of pointer type. Calling the new function is equivalent to calling
5676 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5677 missing <tt>nest</tt> argument. If, after calling
5678 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5679 modified, then the effect of any later call to the returned function pointer is
5684 <!-- ======================================================================= -->
5685 <div class="doc_subsection">
5686 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5689 <div class="doc_text">
5691 These intrinsic functions expand the "universal IR" of LLVM to represent
5692 hardware constructs for atomic operations and memory synchronization. This
5693 provides an interface to the hardware, not an interface to the programmer. It
5694 is aimed at a low enough level to allow any programming models or APIs which
5695 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5696 hardware behavior. Just as hardware provides a "universal IR" for source
5697 languages, it also provides a starting point for developing a "universal"
5698 atomic operation and synchronization IR.
5701 These do <em>not</em> form an API such as high-level threading libraries,
5702 software transaction memory systems, atomic primitives, and intrinsic
5703 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5704 application libraries. The hardware interface provided by LLVM should allow
5705 a clean implementation of all of these APIs and parallel programming models.
5706 No one model or paradigm should be selected above others unless the hardware
5707 itself ubiquitously does so.
5712 <!-- _______________________________________________________________________ -->
5713 <div class="doc_subsubsection">
5714 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5716 <div class="doc_text">
5719 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5725 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5726 specific pairs of memory access types.
5730 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5731 The first four arguments enables a specific barrier as listed below. The fith
5732 argument specifies that the barrier applies to io or device or uncached memory.
5736 <li><tt>ll</tt>: load-load barrier</li>
5737 <li><tt>ls</tt>: load-store barrier</li>
5738 <li><tt>sl</tt>: store-load barrier</li>
5739 <li><tt>ss</tt>: store-store barrier</li>
5740 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5744 This intrinsic causes the system to enforce some ordering constraints upon
5745 the loads and stores of the program. This barrier does not indicate
5746 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5747 which they occur. For any of the specified pairs of load and store operations
5748 (f.ex. load-load, or store-load), all of the first operations preceding the
5749 barrier will complete before any of the second operations succeeding the
5750 barrier begin. Specifically the semantics for each pairing is as follows:
5753 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5754 after the barrier begins.</li>
5756 <li><tt>ls</tt>: All loads before the barrier must complete before any
5757 store after the barrier begins.</li>
5758 <li><tt>ss</tt>: All stores before the barrier must complete before any
5759 store after the barrier begins.</li>
5760 <li><tt>sl</tt>: All stores before the barrier must complete before any
5761 load after the barrier begins.</li>
5764 These semantics are applied with a logical "and" behavior when more than one
5765 is enabled in a single memory barrier intrinsic.
5768 Backends may implement stronger barriers than those requested when they do not
5769 support as fine grained a barrier as requested. Some architectures do not
5770 need all types of barriers and on such architectures, these become noops.
5777 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5778 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5779 <i>; guarantee the above finishes</i>
5780 store i32 8, %ptr <i>; before this begins</i>
5784 <!-- _______________________________________________________________________ -->
5785 <div class="doc_subsubsection">
5786 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
5788 <div class="doc_text">
5791 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
5792 any integer bit width and for different address spaces. Not all targets
5793 support all bit widths however.</p>
5796 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5797 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5798 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5799 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5804 This loads a value in memory and compares it to a given value. If they are
5805 equal, it stores a new value into the memory.
5809 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
5810 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5811 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5812 this integer type. While any bit width integer may be used, targets may only
5813 lower representations they support in hardware.
5818 This entire intrinsic must be executed atomically. It first loads the value
5819 in memory pointed to by <tt>ptr</tt> and compares it with the value
5820 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5821 loaded value is yielded in all cases. This provides the equivalent of an
5822 atomic compare-and-swap operation within the SSA framework.
5830 %val1 = add i32 4, 4
5831 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
5832 <i>; yields {i32}:result1 = 4</i>
5833 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5834 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5836 %val2 = add i32 1, 1
5837 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
5838 <i>; yields {i32}:result2 = 8</i>
5839 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5841 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5845 <!-- _______________________________________________________________________ -->
5846 <div class="doc_subsubsection">
5847 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5849 <div class="doc_text">
5853 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5854 integer bit width. Not all targets support all bit widths however.</p>
5856 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
5857 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
5858 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
5859 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
5864 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5865 the value from memory. It then stores the value in <tt>val</tt> in the memory
5871 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
5872 <tt>val</tt> argument and the result must be integers of the same bit width.
5873 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5874 integer type. The targets may only lower integer representations they
5879 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5880 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5881 equivalent of an atomic swap operation within the SSA framework.
5889 %val1 = add i32 4, 4
5890 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
5891 <i>; yields {i32}:result1 = 4</i>
5892 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5893 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5895 %val2 = add i32 1, 1
5896 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
5897 <i>; yields {i32}:result2 = 8</i>
5899 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5900 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5904 <!-- _______________________________________________________________________ -->
5905 <div class="doc_subsubsection">
5906 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
5909 <div class="doc_text">
5912 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on any
5913 integer bit width. Not all targets support all bit widths however.</p>
5915 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
5916 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
5917 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
5918 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
5923 This intrinsic adds <tt>delta</tt> to the value stored in memory at
5924 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5929 The intrinsic takes two arguments, the first a pointer to an integer value
5930 and the second an integer value. The result is also an integer value. These
5931 integer types can have any bit width, but they must all have the same bit
5932 width. The targets may only lower integer representations they support.
5936 This intrinsic does a series of operations atomically. It first loads the
5937 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
5938 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5945 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
5946 <i>; yields {i32}:result1 = 4</i>
5947 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
5948 <i>; yields {i32}:result2 = 8</i>
5949 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
5950 <i>; yields {i32}:result3 = 10</i>
5951 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
5955 <!-- _______________________________________________________________________ -->
5956 <div class="doc_subsubsection">
5957 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
5960 <div class="doc_text">
5963 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
5964 any integer bit width and for different address spaces. Not all targets
5965 support all bit widths however.</p>
5967 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
5968 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
5969 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
5970 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
5975 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
5976 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5981 The intrinsic takes two arguments, the first a pointer to an integer value
5982 and the second an integer value. The result is also an integer value. These
5983 integer types can have any bit width, but they must all have the same bit
5984 width. The targets may only lower integer representations they support.
5988 This intrinsic does a series of operations atomically. It first loads the
5989 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
5990 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5997 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
5998 <i>; yields {i32}:result1 = 8</i>
5999 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6000 <i>; yields {i32}:result2 = 4</i>
6001 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6002 <i>; yields {i32}:result3 = 2</i>
6003 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6007 <!-- _______________________________________________________________________ -->
6008 <div class="doc_subsubsection">
6009 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6010 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6011 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6012 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6015 <div class="doc_text">
6018 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6019 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
6020 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6021 address spaces. Not all targets support all bit widths however.</p>
6023 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6024 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6025 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6026 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6031 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6032 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6033 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6034 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6039 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6040 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6041 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6042 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6047 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6048 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6049 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6050 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6055 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6056 the value stored in memory at <tt>ptr</tt>. It yields the original value
6062 These intrinsics take two arguments, the first a pointer to an integer value
6063 and the second an integer value. The result is also an integer value. These
6064 integer types can have any bit width, but they must all have the same bit
6065 width. The targets may only lower integer representations they support.
6069 These intrinsics does a series of operations atomically. They first load the
6070 value stored at <tt>ptr</tt>. They then do the bitwise operation
6071 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6072 value stored at <tt>ptr</tt>.
6078 store i32 0x0F0F, %ptr
6079 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6080 <i>; yields {i32}:result0 = 0x0F0F</i>
6081 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6082 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6083 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6084 <i>; yields {i32}:result2 = 0xF0</i>
6085 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6086 <i>; yields {i32}:result3 = FF</i>
6087 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6092 <!-- _______________________________________________________________________ -->
6093 <div class="doc_subsubsection">
6094 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6095 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6096 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6097 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6100 <div class="doc_text">
6103 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6104 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6105 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6106 address spaces. Not all targets
6107 support all bit widths however.</p>
6109 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6110 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6111 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6112 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6117 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6118 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6119 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6120 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6125 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6126 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6127 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6128 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6133 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6134 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6135 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6136 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6141 These intrinsics takes the signed or unsigned minimum or maximum of
6142 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6143 original value at <tt>ptr</tt>.
6148 These intrinsics take two arguments, the first a pointer to an integer value
6149 and the second an integer value. The result is also an integer value. These
6150 integer types can have any bit width, but they must all have the same bit
6151 width. The targets may only lower integer representations they support.
6155 These intrinsics does a series of operations atomically. They first load the
6156 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
6157 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
6158 the original value stored at <tt>ptr</tt>.
6165 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6166 <i>; yields {i32}:result0 = 7</i>
6167 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6168 <i>; yields {i32}:result1 = -2</i>
6169 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6170 <i>; yields {i32}:result2 = 8</i>
6171 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6172 <i>; yields {i32}:result3 = 8</i>
6173 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6177 <!-- ======================================================================= -->
6178 <div class="doc_subsection">
6179 <a name="int_general">General Intrinsics</a>
6182 <div class="doc_text">
6183 <p> This class of intrinsics is designed to be generic and has
6184 no specific purpose. </p>
6187 <!-- _______________________________________________________________________ -->
6188 <div class="doc_subsubsection">
6189 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6192 <div class="doc_text">
6196 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6202 The '<tt>llvm.var.annotation</tt>' intrinsic
6208 The first argument is a pointer to a value, the second is a pointer to a
6209 global string, the third is a pointer to a global string which is the source
6210 file name, and the last argument is the line number.
6216 This intrinsic allows annotation of local variables with arbitrary strings.
6217 This can be useful for special purpose optimizations that want to look for these
6218 annotations. These have no other defined use, they are ignored by code
6219 generation and optimization.
6223 <!-- _______________________________________________________________________ -->
6224 <div class="doc_subsubsection">
6225 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6228 <div class="doc_text">
6231 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6232 any integer bit width.
6235 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6236 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6237 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6238 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6239 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6245 The '<tt>llvm.annotation</tt>' intrinsic.
6251 The first argument is an integer value (result of some expression),
6252 the second is a pointer to a global string, the third is a pointer to a global
6253 string which is the source file name, and the last argument is the line number.
6254 It returns the value of the first argument.
6260 This intrinsic allows annotations to be put on arbitrary expressions
6261 with arbitrary strings. This can be useful for special purpose optimizations
6262 that want to look for these annotations. These have no other defined use, they
6263 are ignored by code generation and optimization.
6266 <!-- _______________________________________________________________________ -->
6267 <div class="doc_subsubsection">
6268 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6271 <div class="doc_text">
6275 declare void @llvm.trap()
6281 The '<tt>llvm.trap</tt>' intrinsic
6293 This intrinsics is lowered to the target dependent trap instruction. If the
6294 target does not have a trap instruction, this intrinsic will be lowered to the
6295 call of the abort() function.
6299 <!-- *********************************************************************** -->
6302 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6303 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
6304 <a href="http://validator.w3.org/check/referer"><img
6305 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
6307 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6308 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6309 Last modified: $Date$