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 a Static Single Assignment (SSA) based representation that provides
259 type safety, low-level operations, flexibility, and the capability of
260 representing 'all' high-level languages cleanly. It is the common code
261 representation used throughout all phases of the LLVM compilation
265 <!-- *********************************************************************** -->
266 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
267 <!-- *********************************************************************** -->
269 <div class="doc_text">
271 <p>The LLVM code representation is designed to be used in three
272 different forms: as an in-memory compiler IR, as an on-disk bitcode
273 representation (suitable for fast loading by a Just-In-Time compiler),
274 and as a human readable assembly language representation. This allows
275 LLVM to provide a powerful intermediate representation for efficient
276 compiler transformations and analysis, while providing a natural means
277 to debug and visualize the transformations. The three different forms
278 of LLVM are all equivalent. This document describes the human readable
279 representation and notation.</p>
281 <p>The LLVM representation aims to be light-weight and low-level
282 while being expressive, typed, and extensible at the same time. It
283 aims to be a "universal IR" of sorts, by being at a low enough level
284 that high-level ideas may be cleanly mapped to it (similar to how
285 microprocessors are "universal IR's", allowing many source languages to
286 be mapped to them). By providing type information, LLVM can be used as
287 the target of optimizations: for example, through pointer analysis, it
288 can be proven that a C automatic variable is never accessed outside of
289 the current function... allowing it to be promoted to a simple SSA
290 value instead of a memory location.</p>
294 <!-- _______________________________________________________________________ -->
295 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
297 <div class="doc_text">
299 <p>It is important to note that this document describes 'well formed'
300 LLVM assembly language. There is a difference between what the parser
301 accepts and what is considered 'well formed'. For example, the
302 following instruction is syntactically okay, but not well formed:</p>
304 <div class="doc_code">
306 %x = <a href="#i_add">add</a> i32 1, %x
310 <p>...because the definition of <tt>%x</tt> does not dominate all of
311 its uses. The LLVM infrastructure provides a verification pass that may
312 be used to verify that an LLVM module is well formed. This pass is
313 automatically run by the parser after parsing input assembly and by
314 the optimizer before it outputs bitcode. The violations pointed out
315 by the verifier pass indicate bugs in transformation passes or input to
319 <!-- Describe the typesetting conventions here. -->
321 <!-- *********************************************************************** -->
322 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
323 <!-- *********************************************************************** -->
325 <div class="doc_text">
327 <p>LLVM identifiers come in two basic types: global and local. Global
328 identifiers (functions, global variables) begin with the @ character. Local
329 identifiers (register names, types) begin with the % character. Additionally,
330 there are three different formats for identifiers, for different purposes:
333 <li>Named values are represented as a string of characters with their prefix.
334 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
335 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
336 Identifiers which require other characters in their names can be surrounded
337 with quotes. In this way, anything except a <tt>"</tt> character can
338 be used in a named value.</li>
340 <li>Unnamed values are represented as an unsigned numeric value with their
341 prefix. For example, %12, @2, %44.</li>
343 <li>Constants, which are described in a <a href="#constants">section about
344 constants</a>, below.</li>
347 <p>LLVM requires that values start with a prefix for two reasons: Compilers
348 don't need to worry about name clashes with reserved words, and the set of
349 reserved words may be expanded in the future without penalty. Additionally,
350 unnamed identifiers allow a compiler to quickly come up with a temporary
351 variable without having to avoid symbol table conflicts.</p>
353 <p>Reserved words in LLVM are very similar to reserved words in other
354 languages. There are keywords for different opcodes
355 ('<tt><a href="#i_add">add</a></tt>',
356 '<tt><a href="#i_bitcast">bitcast</a></tt>',
357 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
358 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
359 and others. These reserved words cannot conflict with variable names, because
360 none of them start with a prefix character ('%' or '@').</p>
362 <p>Here is an example of LLVM code to multiply the integer variable
363 '<tt>%X</tt>' by 8:</p>
367 <div class="doc_code">
369 %result = <a href="#i_mul">mul</a> i32 %X, 8
373 <p>After strength reduction:</p>
375 <div class="doc_code">
377 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
381 <p>And the hard way:</p>
383 <div class="doc_code">
385 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
386 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
387 %result = <a href="#i_add">add</a> i32 %1, %1
391 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
392 important lexical features of LLVM:</p>
396 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
399 <li>Unnamed temporaries are created when the result of a computation is not
400 assigned to a named value.</li>
402 <li>Unnamed temporaries are numbered sequentially</li>
406 <p>...and it also shows a convention that we follow in this document. When
407 demonstrating instructions, we will follow an instruction with a comment that
408 defines the type and name of value produced. Comments are shown in italic
413 <!-- *********************************************************************** -->
414 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
415 <!-- *********************************************************************** -->
417 <!-- ======================================================================= -->
418 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
421 <div class="doc_text">
423 <p>LLVM programs are composed of "Module"s, each of which is a
424 translation unit of the input programs. Each module consists of
425 functions, global variables, and symbol table entries. Modules may be
426 combined together with the LLVM linker, which merges function (and
427 global variable) definitions, resolves forward declarations, and merges
428 symbol table entries. Here is an example of the "hello world" module:</p>
430 <div class="doc_code">
431 <pre><i>; Declare the string constant as a global constant...</i>
432 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
433 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
435 <i>; External declaration of the puts function</i>
436 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
438 <i>; Definition of main function</i>
439 define i32 @main() { <i>; i32()* </i>
440 <i>; Convert [13x i8 ]* to i8 *...</i>
442 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
444 <i>; Call puts function to write out the string to stdout...</i>
446 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
448 href="#i_ret">ret</a> i32 0<br>}<br>
452 <p>This example is made up of a <a href="#globalvars">global variable</a>
453 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
454 function, and a <a href="#functionstructure">function definition</a>
455 for "<tt>main</tt>".</p>
457 <p>In general, a module is made up of a list of global values,
458 where both functions and global variables are global values. Global values are
459 represented by a pointer to a memory location (in this case, a pointer to an
460 array of char, and a pointer to a function), and have one of the following <a
461 href="#linkage">linkage types</a>.</p>
465 <!-- ======================================================================= -->
466 <div class="doc_subsection">
467 <a name="linkage">Linkage Types</a>
470 <div class="doc_text">
473 All Global Variables and Functions have one of the following types of linkage:
478 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
480 <dd>Global values with internal linkage are only directly accessible by
481 objects in the current module. In particular, linking code into a module with
482 an internal global value may cause the internal to be renamed as necessary to
483 avoid collisions. Because the symbol is internal to the module, all
484 references can be updated. This corresponds to the notion of the
485 '<tt>static</tt>' keyword in C.
488 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
490 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
491 the same name when linkage occurs. This is typically used to implement
492 inline functions, templates, or other code which must be generated in each
493 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
494 allowed to be discarded.
497 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
499 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
500 linkage, except that unreferenced <tt>common</tt> globals may not be
501 discarded. This is used for globals that may be emitted in multiple
502 translation units, but that are not guaranteed to be emitted into every
503 translation unit that uses them. One example of this is tentative
504 definitions in C, such as "<tt>int X;</tt>" at global scope.
507 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
509 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
510 that some targets may choose to emit different assembly sequences for them
511 for target-dependent reasons. This is used for globals that are declared
512 "weak" in C source code.
515 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
517 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
518 pointer to array type. When two global variables with appending linkage are
519 linked together, the two global arrays are appended together. This is the
520 LLVM, typesafe, equivalent of having the system linker append together
521 "sections" with identical names when .o files are linked.
524 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
525 <dd>The semantics of this linkage follow the ELF object file model: the
526 symbol is weak until linked, if not linked, the symbol becomes null instead
527 of being an undefined reference.
530 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
532 <dd>If none of the above identifiers are used, the global is externally
533 visible, meaning that it participates in linkage and can be used to resolve
534 external symbol references.
539 The next two types of linkage are targeted for Microsoft Windows platform
540 only. They are designed to support importing (exporting) symbols from (to)
541 DLLs (Dynamic Link Libraries).
545 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
547 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
548 or variable via a global pointer to a pointer that is set up by the DLL
549 exporting the symbol. On Microsoft Windows targets, the pointer name is
550 formed by combining <code>_imp__</code> and the function or variable name.
553 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
555 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
556 pointer to a pointer in a DLL, so that it can be referenced with the
557 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
558 name is formed by combining <code>_imp__</code> and the function or variable
564 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
565 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
566 variable and was linked with this one, one of the two would be renamed,
567 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
568 external (i.e., lacking any linkage declarations), they are accessible
569 outside of the current module.</p>
570 <p>It is illegal for a function <i>declaration</i>
571 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
572 or <tt>extern_weak</tt>.</p>
573 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
577 <!-- ======================================================================= -->
578 <div class="doc_subsection">
579 <a name="callingconv">Calling Conventions</a>
582 <div class="doc_text">
584 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
585 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
586 specified for the call. The calling convention of any pair of dynamic
587 caller/callee must match, or the behavior of the program is undefined. The
588 following calling conventions are supported by LLVM, and more may be added in
592 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
594 <dd>This calling convention (the default if no other calling convention is
595 specified) matches the target C calling conventions. This calling convention
596 supports varargs function calls and tolerates some mismatch in the declared
597 prototype and implemented declaration of the function (as does normal C).
600 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
602 <dd>This calling convention attempts to make calls as fast as possible
603 (e.g. by passing things in registers). This calling convention allows the
604 target to use whatever tricks it wants to produce fast code for the target,
605 without having to conform to an externally specified ABI (Application Binary
606 Interface). Implementations of this convention should allow arbitrary
607 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
608 supported. This calling convention does not support varargs and requires the
609 prototype of all callees to exactly match the prototype of the function
613 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
615 <dd>This calling convention attempts to make code in the caller as efficient
616 as possible under the assumption that the call is not commonly executed. As
617 such, these calls often preserve all registers so that the call does not break
618 any live ranges in the caller side. This calling convention does not support
619 varargs and requires the prototype of all callees to exactly match the
620 prototype of the function definition.
623 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
625 <dd>Any calling convention may be specified by number, allowing
626 target-specific calling conventions to be used. Target specific calling
627 conventions start at 64.
631 <p>More calling conventions can be added/defined on an as-needed basis, to
632 support pascal conventions or any other well-known target-independent
637 <!-- ======================================================================= -->
638 <div class="doc_subsection">
639 <a name="visibility">Visibility Styles</a>
642 <div class="doc_text">
645 All Global Variables and Functions have one of the following visibility styles:
649 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
651 <dd>On targets that use the ELF object file format, default visibility means
652 that the declaration is visible to other
653 modules and, in shared libraries, means that the declared entity may be
654 overridden. On Darwin, default visibility means that the declaration is
655 visible to other modules. Default visibility corresponds to "external
656 linkage" in the language.
659 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
661 <dd>Two declarations of an object with hidden visibility refer to the same
662 object if they are in the same shared object. Usually, hidden visibility
663 indicates that the symbol will not be placed into the dynamic symbol table,
664 so no other module (executable or shared library) can reference it
668 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
670 <dd>On ELF, protected visibility indicates that the symbol will be placed in
671 the dynamic symbol table, but that references within the defining module will
672 bind to the local symbol. That is, the symbol cannot be overridden by another
679 <!-- ======================================================================= -->
680 <div class="doc_subsection">
681 <a name="globalvars">Global Variables</a>
684 <div class="doc_text">
686 <p>Global variables define regions of memory allocated at compilation time
687 instead of run-time. Global variables may optionally be initialized, may have
688 an explicit section to be placed in, and may have an optional explicit alignment
689 specified. A variable may be defined as "thread_local", which means that it
690 will not be shared by threads (each thread will have a separated copy of the
691 variable). A variable may be defined as a global "constant," which indicates
692 that the contents of the variable will <b>never</b> be modified (enabling better
693 optimization, allowing the global data to be placed in the read-only section of
694 an executable, etc). Note that variables that need runtime initialization
695 cannot be marked "constant" as there is a store to the variable.</p>
698 LLVM explicitly allows <em>declarations</em> of global variables to be marked
699 constant, even if the final definition of the global is not. This capability
700 can be used to enable slightly better optimization of the program, but requires
701 the language definition to guarantee that optimizations based on the
702 'constantness' are valid for the translation units that do not include the
706 <p>As SSA values, global variables define pointer values that are in
707 scope (i.e. they dominate) all basic blocks in the program. Global
708 variables always define a pointer to their "content" type because they
709 describe a region of memory, and all memory objects in LLVM are
710 accessed through pointers.</p>
712 <p>A global variable may be declared to reside in a target-specifc numbered
713 address space. For targets that support them, address spaces may affect how
714 optimizations are performed and/or what target instructions are used to access
715 the variable. The default address space is zero. The address space qualifier
716 must precede any other attributes.</p>
718 <p>LLVM allows an explicit section to be specified for globals. If the target
719 supports it, it will emit globals to the section specified.</p>
721 <p>An explicit alignment may be specified for a global. If not present, or if
722 the alignment is set to zero, the alignment of the global is set by the target
723 to whatever it feels convenient. If an explicit alignment is specified, the
724 global is forced to have at least that much alignment. All alignments must be
727 <p>For example, the following defines a global in a numbered address space with
728 an initializer, section, and alignment:</p>
730 <div class="doc_code">
732 @G = constant float 1.0 addrspace(5), section "foo", align 4
739 <!-- ======================================================================= -->
740 <div class="doc_subsection">
741 <a name="functionstructure">Functions</a>
744 <div class="doc_text">
746 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
747 an optional <a href="#linkage">linkage type</a>, an optional
748 <a href="#visibility">visibility style</a>, an optional
749 <a href="#callingconv">calling convention</a>, a return type, an optional
750 <a href="#paramattrs">parameter attribute</a> for the return type, a function
751 name, a (possibly empty) argument list (each with optional
752 <a href="#paramattrs">parameter attributes</a>), an optional section, an
753 optional alignment, an optional <a href="#gc">garbage collector name</a>, an
754 opening curly brace, a list of basic blocks, and a closing curly brace.
756 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
757 optional <a href="#linkage">linkage type</a>, an optional
758 <a href="#visibility">visibility style</a>, an optional
759 <a href="#callingconv">calling convention</a>, a return type, an optional
760 <a href="#paramattrs">parameter attribute</a> for the return type, a function
761 name, a possibly empty list of arguments, an optional alignment, and an optional
762 <a href="#gc">garbage collector name</a>.</p>
764 <p>A function definition contains a list of basic blocks, forming the CFG
765 (Control Flow Graph) for
766 the function. Each basic block may optionally start with a label (giving the
767 basic block a symbol table entry), contains a list of instructions, and ends
768 with a <a href="#terminators">terminator</a> instruction (such as a branch or
769 function return).</p>
771 <p>The first basic block in a function is special in two ways: it is immediately
772 executed on entrance to the function, and it is not allowed to have predecessor
773 basic blocks (i.e. there can not be any branches to the entry block of a
774 function). Because the block can have no predecessors, it also cannot have any
775 <a href="#i_phi">PHI nodes</a>.</p>
777 <p>LLVM allows an explicit section to be specified for functions. If the target
778 supports it, it will emit functions to the section specified.</p>
780 <p>An explicit alignment may be specified for a function. If not present, or if
781 the alignment is set to zero, the alignment of the function is set by the target
782 to whatever it feels convenient. If an explicit alignment is specified, the
783 function is forced to have at least that much alignment. All alignments must be
789 <!-- ======================================================================= -->
790 <div class="doc_subsection">
791 <a name="aliasstructure">Aliases</a>
793 <div class="doc_text">
794 <p>Aliases act as "second name" for the aliasee value (which can be either
795 function, global variable, another alias or bitcast of global value). Aliases
796 may have an optional <a href="#linkage">linkage type</a>, and an
797 optional <a href="#visibility">visibility style</a>.</p>
801 <div class="doc_code">
803 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
811 <!-- ======================================================================= -->
812 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
813 <div class="doc_text">
814 <p>The return type and each parameter of a function type may have a set of
815 <i>parameter attributes</i> associated with them. Parameter attributes are
816 used to communicate additional information about the result or parameters of
817 a function. Parameter attributes are considered to be part of the function,
818 not of the function type, so functions with different parameter attributes
819 can have the same function type.</p>
821 <p>Parameter attributes are simple keywords that follow the type specified. If
822 multiple parameter attributes are needed, they are space separated. For
825 <div class="doc_code">
827 declare i32 @printf(i8* noalias , ...) nounwind
828 declare i32 @atoi(i8*) nounwind readonly
832 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
833 <tt>readonly</tt>) come immediately after the argument list.</p>
835 <p>Currently, only the following parameter attributes are defined:</p>
837 <dt><tt>zeroext</tt></dt>
838 <dd>This indicates that the parameter should be zero extended just before
839 a call to this function.</dd>
841 <dt><tt>signext</tt></dt>
842 <dd>This indicates that the parameter should be sign extended just before
843 a call to this function.</dd>
845 <dt><tt>inreg</tt></dt>
846 <dd>This indicates that the parameter should be placed in register (if
847 possible) during assembling function call. Support for this attribute is
850 <dt><tt>byval</tt></dt>
851 <dd>This indicates that the pointer parameter should really be passed by
852 value to the function. The attribute implies that a hidden copy of the
853 pointee is made between the caller and the callee, so the callee is unable
854 to modify the value in the callee. This attribute is only valid on LLVM
855 pointer arguments. It is generally used to pass structs and arrays by
856 value, but is also valid on scalars (even though this is silly).</dd>
858 <dt><tt>sret</tt></dt>
859 <dd>This indicates that the pointer parameter specifies the address of a
860 structure that is the return value of the function in the source program.
861 Loads and stores to the structure are assumed not to trap.
862 May only be applied to the first parameter.</dd>
864 <dt><tt>noalias</tt></dt>
865 <dd>This indicates that the parameter does not alias any global or any other
866 parameter. The caller is responsible for ensuring that this is the case,
867 usually by placing the value in a stack allocation.</dd>
869 <dt><tt>noreturn</tt></dt>
870 <dd>This function attribute indicates that the function never returns. This
871 indicates to LLVM that every call to this function should be treated as if
872 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
874 <dt><tt>nounwind</tt></dt>
875 <dd>This function attribute indicates that no exceptions unwind out of the
876 function. Usually this is because the function makes no use of exceptions,
877 but it may also be that the function catches any exceptions thrown when
880 <dt><tt>nest</tt></dt>
881 <dd>This indicates that the pointer parameter can be excised using the
882 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
883 <dt><tt>readonly</tt></dt>
884 <dd>This function attribute indicates that the function has no side-effects
885 except for producing a return value or throwing an exception. The value
886 returned must only depend on the function arguments and/or global variables.
887 It may use values obtained by dereferencing pointers.</dd>
888 <dt><tt>readnone</tt></dt>
889 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
890 function, but in addition it is not allowed to dereference any pointer arguments
896 <!-- ======================================================================= -->
897 <div class="doc_subsection">
898 <a name="gc">Garbage Collector Names</a>
901 <div class="doc_text">
902 <p>Each function may specify a garbage collector name, which is simply a
905 <div class="doc_code"><pre
906 >define void @f() gc "name" { ...</pre></div>
908 <p>The compiler declares the supported values of <i>name</i>. Specifying a
909 collector which will cause the compiler to alter its output in order to support
910 the named garbage collection algorithm.</p>
913 <!-- ======================================================================= -->
914 <div class="doc_subsection">
915 <a name="moduleasm">Module-Level Inline Assembly</a>
918 <div class="doc_text">
920 Modules may contain "module-level inline asm" blocks, which corresponds to the
921 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
922 LLVM and treated as a single unit, but may be separated in the .ll file if
923 desired. The syntax is very simple:
926 <div class="doc_code">
928 module asm "inline asm code goes here"
929 module asm "more can go here"
933 <p>The strings can contain any character by escaping non-printable characters.
934 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
939 The inline asm code is simply printed to the machine code .s file when
940 assembly code is generated.
944 <!-- ======================================================================= -->
945 <div class="doc_subsection">
946 <a name="datalayout">Data Layout</a>
949 <div class="doc_text">
950 <p>A module may specify a target specific data layout string that specifies how
951 data is to be laid out in memory. The syntax for the data layout is simply:</p>
952 <pre> target datalayout = "<i>layout specification</i>"</pre>
953 <p>The <i>layout specification</i> consists of a list of specifications
954 separated by the minus sign character ('-'). Each specification starts with a
955 letter and may include other information after the letter to define some
956 aspect of the data layout. The specifications accepted are as follows: </p>
959 <dd>Specifies that the target lays out data in big-endian form. That is, the
960 bits with the most significance have the lowest address location.</dd>
962 <dd>Specifies that the target lays out data in little-endian form. That is,
963 the bits with the least significance have the lowest address location.</dd>
964 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
965 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
966 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
967 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
969 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
970 <dd>This specifies the alignment for an integer type of a given bit
971 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
972 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
973 <dd>This specifies the alignment for a vector type of a given bit
975 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
976 <dd>This specifies the alignment for a floating point type of a given bit
977 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
979 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
980 <dd>This specifies the alignment for an aggregate type of a given bit
983 <p>When constructing the data layout for a given target, LLVM starts with a
984 default set of specifications which are then (possibly) overriden by the
985 specifications in the <tt>datalayout</tt> keyword. The default specifications
986 are given in this list:</p>
988 <li><tt>E</tt> - big endian</li>
989 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
990 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
991 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
992 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
993 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
994 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
995 alignment of 64-bits</li>
996 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
997 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
998 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
999 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1000 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1002 <p>When LLVM is determining the alignment for a given type, it uses the
1005 <li>If the type sought is an exact match for one of the specifications, that
1006 specification is used.</li>
1007 <li>If no match is found, and the type sought is an integer type, then the
1008 smallest integer type that is larger than the bitwidth of the sought type is
1009 used. If none of the specifications are larger than the bitwidth then the the
1010 largest integer type is used. For example, given the default specifications
1011 above, the i7 type will use the alignment of i8 (next largest) while both
1012 i65 and i256 will use the alignment of i64 (largest specified).</li>
1013 <li>If no match is found, and the type sought is a vector type, then the
1014 largest vector type that is smaller than the sought vector type will be used
1015 as a fall back. This happens because <128 x double> can be implemented in
1016 terms of 64 <2 x double>, for example.</li>
1020 <!-- *********************************************************************** -->
1021 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1022 <!-- *********************************************************************** -->
1024 <div class="doc_text">
1026 <p>The LLVM type system is one of the most important features of the
1027 intermediate representation. Being typed enables a number of
1028 optimizations to be performed on the intermediate representation directly,
1029 without having to do
1030 extra analyses on the side before the transformation. A strong type
1031 system makes it easier to read the generated code and enables novel
1032 analyses and transformations that are not feasible to perform on normal
1033 three address code representations.</p>
1037 <!-- ======================================================================= -->
1038 <div class="doc_subsection"> <a name="t_classifications">Type
1039 Classifications</a> </div>
1040 <div class="doc_text">
1041 <p>The types fall into a few useful
1042 classifications:</p>
1044 <table border="1" cellspacing="0" cellpadding="4">
1046 <tr><th>Classification</th><th>Types</th></tr>
1048 <td><a href="#t_integer">integer</a></td>
1049 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1052 <td><a href="#t_floating">floating point</a></td>
1053 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1056 <td><a name="t_firstclass">first class</a></td>
1057 <td><a href="#t_integer">integer</a>,
1058 <a href="#t_floating">floating point</a>,
1059 <a href="#t_pointer">pointer</a>,
1060 <a href="#t_vector">vector</a>,
1061 <a href="#t_struct">structure</a>,
1062 <a href="#t_array">array</a>,
1063 <a href="#t_label">label</a>.
1067 <td><a href="#t_primitive">primitive</a></td>
1068 <td><a href="#t_label">label</a>,
1069 <a href="#t_void">void</a>,
1070 <a href="#t_floating">floating point</a>.</td>
1073 <td><a href="#t_derived">derived</a></td>
1074 <td><a href="#t_integer">integer</a>,
1075 <a href="#t_array">array</a>,
1076 <a href="#t_function">function</a>,
1077 <a href="#t_pointer">pointer</a>,
1078 <a href="#t_struct">structure</a>,
1079 <a href="#t_pstruct">packed structure</a>,
1080 <a href="#t_vector">vector</a>,
1081 <a href="#t_opaque">opaque</a>.
1086 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1087 most important. Values of these types are the only ones which can be
1088 produced by instructions, passed as arguments, or used as operands to
1092 <!-- ======================================================================= -->
1093 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1095 <div class="doc_text">
1096 <p>The primitive types are the fundamental building blocks of the LLVM
1101 <!-- _______________________________________________________________________ -->
1102 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1104 <div class="doc_text">
1107 <tr><th>Type</th><th>Description</th></tr>
1108 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1109 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1110 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1111 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1112 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1117 <!-- _______________________________________________________________________ -->
1118 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1120 <div class="doc_text">
1122 <p>The void type does not represent any value and has no size.</p>
1131 <!-- _______________________________________________________________________ -->
1132 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1134 <div class="doc_text">
1136 <p>The label type represents code labels.</p>
1146 <!-- ======================================================================= -->
1147 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1149 <div class="doc_text">
1151 <p>The real power in LLVM comes from the derived types in the system.
1152 This is what allows a programmer to represent arrays, functions,
1153 pointers, and other useful types. Note that these derived types may be
1154 recursive: For example, it is possible to have a two dimensional array.</p>
1158 <!-- _______________________________________________________________________ -->
1159 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1161 <div class="doc_text">
1164 <p>The integer type is a very simple derived type that simply specifies an
1165 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1166 2^23-1 (about 8 million) can be specified.</p>
1174 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1178 <table class="layout">
1181 <td><tt>i1</tt></td>
1182 <td>a single-bit integer.</td>
1184 <td><tt>i32</tt></td>
1185 <td>a 32-bit integer.</td>
1187 <td><tt>i1942652</tt></td>
1188 <td>a really big integer of over 1 million bits.</td>
1194 <!-- _______________________________________________________________________ -->
1195 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1197 <div class="doc_text">
1201 <p>The array type is a very simple derived type that arranges elements
1202 sequentially in memory. The array type requires a size (number of
1203 elements) and an underlying data type.</p>
1208 [<# elements> x <elementtype>]
1211 <p>The number of elements is a constant integer value; elementtype may
1212 be any type with a size.</p>
1215 <table class="layout">
1217 <td class="left"><tt>[40 x i32]</tt></td>
1218 <td class="left">Array of 40 32-bit integer values.</td>
1221 <td class="left"><tt>[41 x i32]</tt></td>
1222 <td class="left">Array of 41 32-bit integer values.</td>
1225 <td class="left"><tt>[4 x i8]</tt></td>
1226 <td class="left">Array of 4 8-bit integer values.</td>
1229 <p>Here are some examples of multidimensional arrays:</p>
1230 <table class="layout">
1232 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1233 <td class="left">3x4 array of 32-bit integer values.</td>
1236 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1237 <td class="left">12x10 array of single precision floating point values.</td>
1240 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1241 <td class="left">2x3x4 array of 16-bit integer values.</td>
1245 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1246 length array. Normally, accesses past the end of an array are undefined in
1247 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1248 As a special case, however, zero length arrays are recognized to be variable
1249 length. This allows implementation of 'pascal style arrays' with the LLVM
1250 type "{ i32, [0 x float]}", for example.</p>
1254 <!-- _______________________________________________________________________ -->
1255 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1256 <div class="doc_text">
1260 <p>The function type can be thought of as a function signature. It
1261 consists of a return type and a list of formal parameter types. The
1262 return type of a function type is a scalar type, a void type, or a struct type.
1263 If the return type is a struct type then all struct elements must be of first
1264 class types, and the struct must have at least one element.</p>
1269 <returntype list> (<parameter list>)
1272 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1273 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1274 which indicates that the function takes a variable number of arguments.
1275 Variable argument functions can access their arguments with the <a
1276 href="#int_varargs">variable argument handling intrinsic</a> functions.
1277 '<tt><returntype list></tt>' is a comma-separated list of
1278 <a href="#t_firstclass">first class</a> type specifiers.</p>
1281 <table class="layout">
1283 <td class="left"><tt>i32 (i32)</tt></td>
1284 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1286 </tr><tr class="layout">
1287 <td class="left"><tt>float (i16 signext, i32 *) *
1289 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1290 an <tt>i16</tt> that should be sign extended and a
1291 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1294 </tr><tr class="layout">
1295 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1296 <td class="left">A vararg function that takes at least one
1297 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1298 which returns an integer. This is the signature for <tt>printf</tt> in
1301 </tr><tr class="layout">
1302 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1303 <td class="left">A function taking an <tt>i32></tt>, returning two
1304 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1310 <!-- _______________________________________________________________________ -->
1311 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1312 <div class="doc_text">
1314 <p>The structure type is used to represent a collection of data members
1315 together in memory. The packing of the field types is defined to match
1316 the ABI of the underlying processor. The elements of a structure may
1317 be any type that has a size.</p>
1318 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1319 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1320 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1323 <pre> { <type list> }<br></pre>
1325 <table class="layout">
1327 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1328 <td class="left">A triple of three <tt>i32</tt> values</td>
1329 </tr><tr class="layout">
1330 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1331 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1332 second element is a <a href="#t_pointer">pointer</a> to a
1333 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1334 an <tt>i32</tt>.</td>
1339 <!-- _______________________________________________________________________ -->
1340 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1342 <div class="doc_text">
1344 <p>The packed structure type is used to represent a collection of data members
1345 together in memory. There is no padding between fields. Further, the alignment
1346 of a packed structure is 1 byte. The elements of a packed structure may
1347 be any type that has a size.</p>
1348 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1349 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1350 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1353 <pre> < { <type list> } > <br></pre>
1355 <table class="layout">
1357 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1358 <td class="left">A triple of three <tt>i32</tt> values</td>
1359 </tr><tr class="layout">
1360 <td class="left"><tt>< { float, i32 (i32)* } ></tt></td>
1361 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1362 second element is a <a href="#t_pointer">pointer</a> to a
1363 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1364 an <tt>i32</tt>.</td>
1369 <!-- _______________________________________________________________________ -->
1370 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1371 <div class="doc_text">
1373 <p>As in many languages, the pointer type represents a pointer or
1374 reference to another object, which must live in memory. Pointer types may have
1375 an optional address space attribute defining the target-specific numbered
1376 address space where the pointed-to object resides. The default address space is
1379 <pre> <type> *<br></pre>
1381 <table class="layout">
1383 <td class="left"><tt>[4x i32]*</tt></td>
1384 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1385 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1388 <td class="left"><tt>i32 (i32 *) *</tt></td>
1389 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1390 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1394 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1395 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1396 that resides in address space #5.</td>
1401 <!-- _______________________________________________________________________ -->
1402 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1403 <div class="doc_text">
1407 <p>A vector type is a simple derived type that represents a vector
1408 of elements. Vector types are used when multiple primitive data
1409 are operated in parallel using a single instruction (SIMD).
1410 A vector type requires a size (number of
1411 elements) and an underlying primitive data type. Vectors must have a power
1412 of two length (1, 2, 4, 8, 16 ...). Vector types are
1413 considered <a href="#t_firstclass">first class</a>.</p>
1418 < <# elements> x <elementtype> >
1421 <p>The number of elements is a constant integer value; elementtype may
1422 be any integer or floating point type.</p>
1426 <table class="layout">
1428 <td class="left"><tt><4 x i32></tt></td>
1429 <td class="left">Vector of 4 32-bit integer values.</td>
1432 <td class="left"><tt><8 x float></tt></td>
1433 <td class="left">Vector of 8 32-bit floating-point values.</td>
1436 <td class="left"><tt><2 x i64></tt></td>
1437 <td class="left">Vector of 2 64-bit integer values.</td>
1442 <!-- _______________________________________________________________________ -->
1443 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1444 <div class="doc_text">
1448 <p>Opaque types are used to represent unknown types in the system. This
1449 corresponds (for example) to the C notion of a forward declared structure type.
1450 In LLVM, opaque types can eventually be resolved to any type (not just a
1451 structure type).</p>
1461 <table class="layout">
1463 <td class="left"><tt>opaque</tt></td>
1464 <td class="left">An opaque type.</td>
1470 <!-- *********************************************************************** -->
1471 <div class="doc_section"> <a name="constants">Constants</a> </div>
1472 <!-- *********************************************************************** -->
1474 <div class="doc_text">
1476 <p>LLVM has several different basic types of constants. This section describes
1477 them all and their syntax.</p>
1481 <!-- ======================================================================= -->
1482 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1484 <div class="doc_text">
1487 <dt><b>Boolean constants</b></dt>
1489 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1490 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1493 <dt><b>Integer constants</b></dt>
1495 <dd>Standard integers (such as '4') are constants of the <a
1496 href="#t_integer">integer</a> type. Negative numbers may be used with
1500 <dt><b>Floating point constants</b></dt>
1502 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1503 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1504 notation (see below). The assembler requires the exact decimal value of
1505 a floating-point constant. For example, the assembler accepts 1.25 but
1506 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1507 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1509 <dt><b>Null pointer constants</b></dt>
1511 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1512 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1516 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1517 of floating point constants. For example, the form '<tt>double
1518 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1519 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1520 (and the only time that they are generated by the disassembler) is when a
1521 floating point constant must be emitted but it cannot be represented as a
1522 decimal floating point number. For example, NaN's, infinities, and other
1523 special values are represented in their IEEE hexadecimal format so that
1524 assembly and disassembly do not cause any bits to change in the constants.</p>
1528 <!-- ======================================================================= -->
1529 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1532 <div class="doc_text">
1533 <p>Aggregate constants arise from aggregation of simple constants
1534 and smaller aggregate constants.</p>
1537 <dt><b>Structure constants</b></dt>
1539 <dd>Structure constants are represented with notation similar to structure
1540 type definitions (a comma separated list of elements, surrounded by braces
1541 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1542 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1543 must have <a href="#t_struct">structure type</a>, and the number and
1544 types of elements must match those specified by the type.
1547 <dt><b>Array constants</b></dt>
1549 <dd>Array constants are represented with notation similar to array type
1550 definitions (a comma separated list of elements, surrounded by square brackets
1551 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1552 constants must have <a href="#t_array">array type</a>, and the number and
1553 types of elements must match those specified by the type.
1556 <dt><b>Vector constants</b></dt>
1558 <dd>Vector constants are represented with notation similar to vector type
1559 definitions (a comma separated list of elements, surrounded by
1560 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1561 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1562 href="#t_vector">vector type</a>, and the number and types of elements must
1563 match those specified by the type.
1566 <dt><b>Zero initialization</b></dt>
1568 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1569 value to zero of <em>any</em> type, including scalar and aggregate types.
1570 This is often used to avoid having to print large zero initializers (e.g. for
1571 large arrays) and is always exactly equivalent to using explicit zero
1578 <!-- ======================================================================= -->
1579 <div class="doc_subsection">
1580 <a name="globalconstants">Global Variable and Function Addresses</a>
1583 <div class="doc_text">
1585 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1586 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1587 constants. These constants are explicitly referenced when the <a
1588 href="#identifiers">identifier for the global</a> is used and always have <a
1589 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1592 <div class="doc_code">
1596 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1602 <!-- ======================================================================= -->
1603 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1604 <div class="doc_text">
1605 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1606 no specific value. Undefined values may be of any type and be used anywhere
1607 a constant is permitted.</p>
1609 <p>Undefined values indicate to the compiler that the program is well defined
1610 no matter what value is used, giving the compiler more freedom to optimize.
1614 <!-- ======================================================================= -->
1615 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1618 <div class="doc_text">
1620 <p>Constant expressions are used to allow expressions involving other constants
1621 to be used as constants. Constant expressions may be of any <a
1622 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1623 that does not have side effects (e.g. load and call are not supported). The
1624 following is the syntax for constant expressions:</p>
1627 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1628 <dd>Truncate a constant to another type. The bit size of CST must be larger
1629 than the bit size of TYPE. Both types must be integers.</dd>
1631 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1632 <dd>Zero extend a constant to another type. The bit size of CST must be
1633 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1635 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1636 <dd>Sign extend a constant to another type. The bit size of CST must be
1637 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1639 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1640 <dd>Truncate a floating point constant to another floating point type. The
1641 size of CST must be larger than the size of TYPE. Both types must be
1642 floating point.</dd>
1644 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1645 <dd>Floating point extend a constant to another type. The size of CST must be
1646 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1648 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1649 <dd>Convert a floating point constant to the corresponding unsigned integer
1650 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1651 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1652 of the same number of elements. If the value won't fit in the integer type,
1653 the results are undefined.</dd>
1655 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1656 <dd>Convert a floating point constant to the corresponding signed integer
1657 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1658 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1659 of the same number of elements. If the value won't fit in the integer type,
1660 the results are undefined.</dd>
1662 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1663 <dd>Convert an unsigned integer constant to the corresponding floating point
1664 constant. TYPE must be a scalar or vector floating point type. CST must be of
1665 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1666 of the same number of elements. If the value won't fit in the floating point
1667 type, the results are undefined.</dd>
1669 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1670 <dd>Convert a signed integer constant to the corresponding floating point
1671 constant. TYPE must be a scalar or vector floating point type. CST must be of
1672 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1673 of the same number of elements. If the value won't fit in the floating point
1674 type, the results are undefined.</dd>
1676 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1677 <dd>Convert a pointer typed constant to the corresponding integer constant
1678 TYPE must be an integer type. CST must be of pointer type. The CST value is
1679 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1681 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1682 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1683 pointer type. CST must be of integer type. The CST value is zero extended,
1684 truncated, or unchanged to make it fit in a pointer size. This one is
1685 <i>really</i> dangerous!</dd>
1687 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1688 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1689 identical (same number of bits). The conversion is done as if the CST value
1690 was stored to memory and read back as TYPE. In other words, no bits change
1691 with this operator, just the type. This can be used for conversion of
1692 vector types to any other type, as long as they have the same bit width. For
1693 pointers it is only valid to cast to another pointer type.
1696 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1698 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1699 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1700 instruction, the index list may have zero or more indexes, which are required
1701 to make sense for the type of "CSTPTR".</dd>
1703 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1705 <dd>Perform the <a href="#i_select">select operation</a> on
1708 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1709 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1711 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1712 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1714 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1715 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1717 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1718 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1720 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1722 <dd>Perform the <a href="#i_extractelement">extractelement
1723 operation</a> on constants.
1725 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1727 <dd>Perform the <a href="#i_insertelement">insertelement
1728 operation</a> on constants.</dd>
1731 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1733 <dd>Perform the <a href="#i_shufflevector">shufflevector
1734 operation</a> on constants.</dd>
1736 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1738 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1739 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1740 binary</a> operations. The constraints on operands are the same as those for
1741 the corresponding instruction (e.g. no bitwise operations on floating point
1742 values are allowed).</dd>
1746 <!-- *********************************************************************** -->
1747 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1748 <!-- *********************************************************************** -->
1750 <!-- ======================================================================= -->
1751 <div class="doc_subsection">
1752 <a name="inlineasm">Inline Assembler Expressions</a>
1755 <div class="doc_text">
1758 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1759 Module-Level Inline Assembly</a>) through the use of a special value. This
1760 value represents the inline assembler as a string (containing the instructions
1761 to emit), a list of operand constraints (stored as a string), and a flag that
1762 indicates whether or not the inline asm expression has side effects. An example
1763 inline assembler expression is:
1766 <div class="doc_code">
1768 i32 (i32) asm "bswap $0", "=r,r"
1773 Inline assembler expressions may <b>only</b> be used as the callee operand of
1774 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1777 <div class="doc_code">
1779 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1784 Inline asms with side effects not visible in the constraint list must be marked
1785 as having side effects. This is done through the use of the
1786 '<tt>sideeffect</tt>' keyword, like so:
1789 <div class="doc_code">
1791 call void asm sideeffect "eieio", ""()
1795 <p>TODO: The format of the asm and constraints string still need to be
1796 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1797 need to be documented).
1802 <!-- *********************************************************************** -->
1803 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1804 <!-- *********************************************************************** -->
1806 <div class="doc_text">
1808 <p>The LLVM instruction set consists of several different
1809 classifications of instructions: <a href="#terminators">terminator
1810 instructions</a>, <a href="#binaryops">binary instructions</a>,
1811 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1812 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1813 instructions</a>.</p>
1817 <!-- ======================================================================= -->
1818 <div class="doc_subsection"> <a name="terminators">Terminator
1819 Instructions</a> </div>
1821 <div class="doc_text">
1823 <p>As mentioned <a href="#functionstructure">previously</a>, every
1824 basic block in a program ends with a "Terminator" instruction, which
1825 indicates which block should be executed after the current block is
1826 finished. These terminator instructions typically yield a '<tt>void</tt>'
1827 value: they produce control flow, not values (the one exception being
1828 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1829 <p>There are six different terminator instructions: the '<a
1830 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1831 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1832 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1833 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1834 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1838 <!-- _______________________________________________________________________ -->
1839 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1840 Instruction</a> </div>
1841 <div class="doc_text">
1843 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1844 ret void <i>; Return from void function</i>
1845 ret <type> <value>, <type> <value> <i>; Return two values from a non-void function </i>
1850 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1851 value) from a function back to the caller.</p>
1852 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1853 returns value(s) and then causes control flow, and one that just causes
1854 control flow to occur.</p>
1858 <p>The '<tt>ret</tt>' instruction may return zero, one or multiple values.
1859 The type of each return value must be a '<a href="#t_firstclass">first
1860 class</a>' type. Note that a function is not <a href="#wellformed">well
1861 formed</a> if there exists a '<tt>ret</tt>' instruction inside of the
1862 function that returns values that do not match the return type of the
1867 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1868 returns back to the calling function's context. If the caller is a "<a
1869 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1870 the instruction after the call. If the caller was an "<a
1871 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1872 at the beginning of the "normal" destination block. If the instruction
1873 returns a value, that value shall set the call or invoke instruction's
1874 return value. If the instruction returns multiple values then these
1875 values can only be accessed through a '<a href="#i_getresult"><tt>getresult</tt>
1876 </a>' instruction.</p>
1881 ret i32 5 <i>; Return an integer value of 5</i>
1882 ret void <i>; Return from a void function</i>
1883 ret i32 4, i8 2 <i>; Return two values 4 and 2 </i>
1886 <!-- _______________________________________________________________________ -->
1887 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1888 <div class="doc_text">
1890 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1893 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1894 transfer to a different basic block in the current function. There are
1895 two forms of this instruction, corresponding to a conditional branch
1896 and an unconditional branch.</p>
1898 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1899 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1900 unconditional form of the '<tt>br</tt>' instruction takes a single
1901 '<tt>label</tt>' value as a target.</p>
1903 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1904 argument is evaluated. If the value is <tt>true</tt>, control flows
1905 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1906 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1908 <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
1909 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1911 <!-- _______________________________________________________________________ -->
1912 <div class="doc_subsubsection">
1913 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1916 <div class="doc_text">
1920 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1925 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1926 several different places. It is a generalization of the '<tt>br</tt>'
1927 instruction, allowing a branch to occur to one of many possible
1933 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1934 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1935 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1936 table is not allowed to contain duplicate constant entries.</p>
1940 <p>The <tt>switch</tt> instruction specifies a table of values and
1941 destinations. When the '<tt>switch</tt>' instruction is executed, this
1942 table is searched for the given value. If the value is found, control flow is
1943 transfered to the corresponding destination; otherwise, control flow is
1944 transfered to the default destination.</p>
1946 <h5>Implementation:</h5>
1948 <p>Depending on properties of the target machine and the particular
1949 <tt>switch</tt> instruction, this instruction may be code generated in different
1950 ways. For example, it could be generated as a series of chained conditional
1951 branches or with a lookup table.</p>
1956 <i>; Emulate a conditional br instruction</i>
1957 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1958 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1960 <i>; Emulate an unconditional br instruction</i>
1961 switch i32 0, label %dest [ ]
1963 <i>; Implement a jump table:</i>
1964 switch i32 %val, label %otherwise [ i32 0, label %onzero
1966 i32 2, label %ontwo ]
1970 <!-- _______________________________________________________________________ -->
1971 <div class="doc_subsubsection">
1972 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1975 <div class="doc_text">
1980 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> <function ptr val>(<function args>)
1981 to label <normal label> unwind label <exception label>
1986 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1987 function, with the possibility of control flow transfer to either the
1988 '<tt>normal</tt>' label or the
1989 '<tt>exception</tt>' label. If the callee function returns with the
1990 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1991 "normal" label. If the callee (or any indirect callees) returns with the "<a
1992 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1993 continued at the dynamically nearest "exception" label. If the callee function
1994 returns multiple values then individual return values are only accessible through
1995 a '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
1999 <p>This instruction requires several arguments:</p>
2003 The optional "cconv" marker indicates which <a href="#callingconv">calling
2004 convention</a> the call should use. If none is specified, the call defaults
2005 to using C calling conventions.
2007 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2008 function value being invoked. In most cases, this is a direct function
2009 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2010 an arbitrary pointer to function value.
2013 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2014 function to be invoked. </li>
2016 <li>'<tt>function args</tt>': argument list whose types match the function
2017 signature argument types. If the function signature indicates the function
2018 accepts a variable number of arguments, the extra arguments can be
2021 <li>'<tt>normal label</tt>': the label reached when the called function
2022 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2024 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2025 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2031 <p>This instruction is designed to operate as a standard '<tt><a
2032 href="#i_call">call</a></tt>' instruction in most regards. The primary
2033 difference is that it establishes an association with a label, which is used by
2034 the runtime library to unwind the stack.</p>
2036 <p>This instruction is used in languages with destructors to ensure that proper
2037 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2038 exception. Additionally, this is important for implementation of
2039 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2043 %retval = invoke i32 @Test(i32 15) to label %Continue
2044 unwind label %TestCleanup <i>; {i32}:retval set</i>
2045 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2046 unwind label %TestCleanup <i>; {i32}:retval set</i>
2051 <!-- _______________________________________________________________________ -->
2053 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2054 Instruction</a> </div>
2056 <div class="doc_text">
2065 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2066 at the first callee in the dynamic call stack which used an <a
2067 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2068 primarily used to implement exception handling.</p>
2072 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2073 immediately halt. The dynamic call stack is then searched for the first <a
2074 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2075 execution continues at the "exceptional" destination block specified by the
2076 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2077 dynamic call chain, undefined behavior results.</p>
2080 <!-- _______________________________________________________________________ -->
2082 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2083 Instruction</a> </div>
2085 <div class="doc_text">
2094 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2095 instruction is used to inform the optimizer that a particular portion of the
2096 code is not reachable. This can be used to indicate that the code after a
2097 no-return function cannot be reached, and other facts.</p>
2101 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2106 <!-- ======================================================================= -->
2107 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2108 <div class="doc_text">
2109 <p>Binary operators are used to do most of the computation in a
2110 program. They require two operands of the same type, execute an operation on them, and
2111 produce a single value. The operands might represent
2112 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2113 The result value has the same type as its operands.</p>
2114 <p>There are several different binary operators:</p>
2116 <!-- _______________________________________________________________________ -->
2117 <div class="doc_subsubsection">
2118 <a name="i_add">'<tt>add</tt>' Instruction</a>
2121 <div class="doc_text">
2126 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2131 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2135 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2136 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2137 <a href="#t_vector">vector</a> values. Both arguments must have identical
2142 <p>The value produced is the integer or floating point sum of the two
2145 <p>If an integer sum has unsigned overflow, the result returned is the
2146 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2149 <p>Because LLVM integers use a two's complement representation, this
2150 instruction is appropriate for both signed and unsigned integers.</p>
2155 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2158 <!-- _______________________________________________________________________ -->
2159 <div class="doc_subsubsection">
2160 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2163 <div class="doc_text">
2168 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2173 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2176 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2177 '<tt>neg</tt>' instruction present in most other intermediate
2178 representations.</p>
2182 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2183 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2184 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2189 <p>The value produced is the integer or floating point difference of
2190 the two operands.</p>
2192 <p>If an integer difference has unsigned overflow, the result returned is the
2193 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2196 <p>Because LLVM integers use a two's complement representation, this
2197 instruction is appropriate for both signed and unsigned integers.</p>
2201 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2202 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2206 <!-- _______________________________________________________________________ -->
2207 <div class="doc_subsubsection">
2208 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2211 <div class="doc_text">
2214 <pre> <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2217 <p>The '<tt>mul</tt>' instruction returns the product of its two
2222 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2223 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2224 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2229 <p>The value produced is the integer or floating point product of the
2232 <p>If the result of an integer multiplication has unsigned overflow,
2233 the result returned is the mathematical result modulo
2234 2<sup>n</sup>, where n is the bit width of the result.</p>
2235 <p>Because LLVM integers use a two's complement representation, and the
2236 result is the same width as the operands, this instruction returns the
2237 correct result for both signed and unsigned integers. If a full product
2238 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2239 should be sign-extended or zero-extended as appropriate to the
2240 width of the full product.</p>
2242 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2246 <!-- _______________________________________________________________________ -->
2247 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2249 <div class="doc_text">
2251 <pre> <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2254 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2259 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2260 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2261 values. Both arguments must have identical types.</p>
2265 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2266 <p>Note that unsigned integer division and signed integer division are distinct
2267 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2268 <p>Division by zero leads to undefined behavior.</p>
2270 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2273 <!-- _______________________________________________________________________ -->
2274 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2276 <div class="doc_text">
2279 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2284 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2289 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2290 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2291 values. Both arguments must have identical types.</p>
2294 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2295 <p>Note that signed integer division and unsigned integer division are distinct
2296 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2297 <p>Division by zero leads to undefined behavior. Overflow also leads to
2298 undefined behavior; this is a rare case, but can occur, for example,
2299 by doing a 32-bit division of -2147483648 by -1.</p>
2301 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2304 <!-- _______________________________________________________________________ -->
2305 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2306 Instruction</a> </div>
2307 <div class="doc_text">
2310 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2314 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2319 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2320 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2321 of floating point values. Both arguments must have identical types.</p>
2325 <p>The value produced is the floating point quotient of the two operands.</p>
2330 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2334 <!-- _______________________________________________________________________ -->
2335 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2337 <div class="doc_text">
2339 <pre> <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2342 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2343 unsigned division of its two arguments.</p>
2345 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2346 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2347 values. Both arguments must have identical types.</p>
2349 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2350 This instruction always performs an unsigned division to get the remainder.</p>
2351 <p>Note that unsigned integer remainder and signed integer remainder are
2352 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2353 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2355 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2359 <!-- _______________________________________________________________________ -->
2360 <div class="doc_subsubsection">
2361 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2364 <div class="doc_text">
2369 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2374 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2375 signed division of its two operands. This instruction can also take
2376 <a href="#t_vector">vector</a> versions of the values in which case
2377 the elements must be integers.</p>
2381 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2382 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2383 values. Both arguments must have identical types.</p>
2387 <p>This instruction returns the <i>remainder</i> of a division (where the result
2388 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2389 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
2390 a value. For more information about the difference, see <a
2391 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2392 Math Forum</a>. For a table of how this is implemented in various languages,
2393 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2394 Wikipedia: modulo operation</a>.</p>
2395 <p>Note that signed integer remainder and unsigned integer remainder are
2396 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2397 <p>Taking the remainder of a division by zero leads to undefined behavior.
2398 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2399 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2400 (The remainder doesn't actually overflow, but this rule lets srem be
2401 implemented using instructions that return both the result of the division
2402 and the remainder.)</p>
2404 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2408 <!-- _______________________________________________________________________ -->
2409 <div class="doc_subsubsection">
2410 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2412 <div class="doc_text">
2415 <pre> <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2418 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2419 division of its two operands.</p>
2421 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2422 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2423 of floating point values. Both arguments must have identical types.</p>
2427 <p>This instruction returns the <i>remainder</i> of a division.
2428 The remainder has the same sign as the dividend.</p>
2433 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2437 <!-- ======================================================================= -->
2438 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2439 Operations</a> </div>
2440 <div class="doc_text">
2441 <p>Bitwise binary operators are used to do various forms of
2442 bit-twiddling in a program. They are generally very efficient
2443 instructions and can commonly be strength reduced from other
2444 instructions. They require two operands of the same type, execute an operation on them,
2445 and produce a single value. The resulting value is the same type as its operands.</p>
2448 <!-- _______________________________________________________________________ -->
2449 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2450 Instruction</a> </div>
2451 <div class="doc_text">
2453 <pre> <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2458 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2459 the left a specified number of bits.</p>
2463 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2464 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2465 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2469 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 2<sup>n</sup>,
2470 where n is the width of the result. If <tt>op2</tt> is (statically or dynamically) negative or
2471 equal to or larger than the number of bits in <tt>op1</tt>, the result is undefined.</p>
2473 <h5>Example:</h5><pre>
2474 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2475 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2476 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2477 <result> = shl i32 1, 32 <i>; undefined</i>
2480 <!-- _______________________________________________________________________ -->
2481 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2482 Instruction</a> </div>
2483 <div class="doc_text">
2485 <pre> <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2489 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2490 operand shifted to the right a specified number of bits with zero fill.</p>
2493 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2494 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2495 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2499 <p>This instruction always performs a logical shift right operation. The most
2500 significant bits of the result will be filled with zero bits after the
2501 shift. If <tt>op2</tt> is (statically or dynamically) equal to or larger than
2502 the number of bits in <tt>op1</tt>, the result is undefined.</p>
2506 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2507 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2508 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2509 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2510 <result> = lshr i32 1, 32 <i>; undefined</i>
2514 <!-- _______________________________________________________________________ -->
2515 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2516 Instruction</a> </div>
2517 <div class="doc_text">
2520 <pre> <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2524 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2525 operand shifted to the right a specified number of bits with sign extension.</p>
2528 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2529 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2530 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2533 <p>This instruction always performs an arithmetic shift right operation,
2534 The most significant bits of the result will be filled with the sign bit
2535 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
2536 larger than the number of bits in <tt>op1</tt>, the result is undefined.
2541 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2542 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2543 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2544 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2545 <result> = ashr i32 1, 32 <i>; undefined</i>
2549 <!-- _______________________________________________________________________ -->
2550 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2551 Instruction</a> </div>
2553 <div class="doc_text">
2558 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2563 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2564 its two operands.</p>
2568 <p>The two arguments to the '<tt>and</tt>' instruction must be
2569 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2570 values. Both arguments must have identical types.</p>
2573 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2575 <div style="align: center">
2576 <table border="1" cellspacing="0" cellpadding="4">
2608 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2609 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2610 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2613 <!-- _______________________________________________________________________ -->
2614 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2615 <div class="doc_text">
2617 <pre> <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2620 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2621 or of its two operands.</p>
2624 <p>The two arguments to the '<tt>or</tt>' instruction must be
2625 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2626 values. Both arguments must have identical types.</p>
2628 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2630 <div style="align: center">
2631 <table border="1" cellspacing="0" cellpadding="4">
2662 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2663 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2664 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2667 <!-- _______________________________________________________________________ -->
2668 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2669 Instruction</a> </div>
2670 <div class="doc_text">
2672 <pre> <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2675 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2676 or of its two operands. The <tt>xor</tt> is used to implement the
2677 "one's complement" operation, which is the "~" operator in C.</p>
2679 <p>The two arguments to the '<tt>xor</tt>' instruction must be
2680 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2681 values. Both arguments must have identical types.</p>
2685 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2687 <div style="align: center">
2688 <table border="1" cellspacing="0" cellpadding="4">
2720 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2721 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2722 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2723 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2727 <!-- ======================================================================= -->
2728 <div class="doc_subsection">
2729 <a name="vectorops">Vector Operations</a>
2732 <div class="doc_text">
2734 <p>LLVM supports several instructions to represent vector operations in a
2735 target-independent manner. These instructions cover the element-access and
2736 vector-specific operations needed to process vectors effectively. While LLVM
2737 does directly support these vector operations, many sophisticated algorithms
2738 will want to use target-specific intrinsics to take full advantage of a specific
2743 <!-- _______________________________________________________________________ -->
2744 <div class="doc_subsubsection">
2745 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2748 <div class="doc_text">
2753 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2759 The '<tt>extractelement</tt>' instruction extracts a single scalar
2760 element from a vector at a specified index.
2767 The first operand of an '<tt>extractelement</tt>' instruction is a
2768 value of <a href="#t_vector">vector</a> type. The second operand is
2769 an index indicating the position from which to extract the element.
2770 The index may be a variable.</p>
2775 The result is a scalar of the same type as the element type of
2776 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2777 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2778 results are undefined.
2784 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2789 <!-- _______________________________________________________________________ -->
2790 <div class="doc_subsubsection">
2791 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2794 <div class="doc_text">
2799 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2805 The '<tt>insertelement</tt>' instruction inserts a scalar
2806 element into a vector at a specified index.
2813 The first operand of an '<tt>insertelement</tt>' instruction is a
2814 value of <a href="#t_vector">vector</a> type. The second operand is a
2815 scalar value whose type must equal the element type of the first
2816 operand. The third operand is an index indicating the position at
2817 which to insert the value. The index may be a variable.</p>
2822 The result is a vector of the same type as <tt>val</tt>. Its
2823 element values are those of <tt>val</tt> except at position
2824 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2825 exceeds the length of <tt>val</tt>, the results are undefined.
2831 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2835 <!-- _______________________________________________________________________ -->
2836 <div class="doc_subsubsection">
2837 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2840 <div class="doc_text">
2845 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2851 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2852 from two input vectors, returning a vector of the same type.
2858 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2859 with types that match each other and types that match the result of the
2860 instruction. The third argument is a shuffle mask, which has the same number
2861 of elements as the other vector type, but whose element type is always 'i32'.
2865 The shuffle mask operand is required to be a constant vector with either
2866 constant integer or undef values.
2872 The elements of the two input vectors are numbered from left to right across
2873 both of the vectors. The shuffle mask operand specifies, for each element of
2874 the result vector, which element of the two input registers the result element
2875 gets. The element selector may be undef (meaning "don't care") and the second
2876 operand may be undef if performing a shuffle from only one vector.
2882 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2883 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2884 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2885 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2890 <!-- ======================================================================= -->
2891 <div class="doc_subsection">
2892 <a name="aggregateops">Aggregate Operations</a>
2895 <div class="doc_text">
2897 <p>LLVM supports several instructions for working with aggregate values.
2902 <!-- _______________________________________________________________________ -->
2903 <div class="doc_subsubsection">
2904 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
2907 <div class="doc_text">
2912 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
2918 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
2919 or array element from an aggregate value.
2926 The first operand of an '<tt>extractvalue</tt>' instruction is a
2927 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
2928 type. The operands are constant indices to specify which value to extract
2929 in a similar manner as indices in a
2930 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2936 The result is the value at the position in the aggregate specified by
2943 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
2948 <!-- _______________________________________________________________________ -->
2949 <div class="doc_subsubsection">
2950 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
2953 <div class="doc_text">
2958 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
2964 The '<tt>insertvalue</tt>' instruction inserts a value
2965 into a struct field or array element in an aggregate.
2972 The first operand of an '<tt>insertvalue</tt>' instruction is a
2973 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
2974 The second operand is a first-class value to insert.
2975 The following operands are constant indices
2976 indicating the position at which to insert the value in a similar manner as
2978 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2979 The value to insert must have the same type as the value identified
2985 The result is an aggregate of the same type as <tt>val</tt>. Its
2986 value is that of <tt>val</tt> except that the value at the position
2987 specified by the indices is that of <tt>elt</tt>.
2993 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
2998 <!-- ======================================================================= -->
2999 <div class="doc_subsection">
3000 <a name="memoryops">Memory Access and Addressing Operations</a>
3003 <div class="doc_text">
3005 <p>A key design point of an SSA-based representation is how it
3006 represents memory. In LLVM, no memory locations are in SSA form, which
3007 makes things very simple. This section describes how to read, write,
3008 allocate, and free memory in LLVM.</p>
3012 <!-- _______________________________________________________________________ -->
3013 <div class="doc_subsubsection">
3014 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3017 <div class="doc_text">
3022 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3027 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3028 heap and returns a pointer to it. The object is always allocated in the generic
3029 address space (address space zero).</p>
3033 <p>The '<tt>malloc</tt>' instruction allocates
3034 <tt>sizeof(<type>)*NumElements</tt>
3035 bytes of memory from the operating system and returns a pointer of the
3036 appropriate type to the program. If "NumElements" is specified, it is the
3037 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3038 If a constant alignment is specified, the value result of the allocation is guaranteed to
3039 be aligned to at least that boundary. If not specified, or if zero, the target can
3040 choose to align the allocation on any convenient boundary.</p>
3042 <p>'<tt>type</tt>' must be a sized type.</p>
3046 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3047 a pointer is returned. The result of a zero byte allocattion is undefined. The
3048 result is null if there is insufficient memory available.</p>
3053 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
3055 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3056 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3057 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3058 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3059 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3063 <!-- _______________________________________________________________________ -->
3064 <div class="doc_subsubsection">
3065 <a name="i_free">'<tt>free</tt>' Instruction</a>
3068 <div class="doc_text">
3073 free <type> <value> <i>; yields {void}</i>
3078 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3079 memory heap to be reallocated in the future.</p>
3083 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3084 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3089 <p>Access to the memory pointed to by the pointer is no longer defined
3090 after this instruction executes. If the pointer is null, the operation
3096 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3097 free [4 x i8]* %array
3101 <!-- _______________________________________________________________________ -->
3102 <div class="doc_subsubsection">
3103 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3106 <div class="doc_text">
3111 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3116 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3117 currently executing function, to be automatically released when this function
3118 returns to its caller. The object is always allocated in the generic address
3119 space (address space zero).</p>
3123 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3124 bytes of memory on the runtime stack, returning a pointer of the
3125 appropriate type to the program. If "NumElements" is specified, it is the
3126 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3127 If a constant alignment is specified, the value result of the allocation is guaranteed
3128 to be aligned to at least that boundary. If not specified, or if zero, the target
3129 can choose to align the allocation on any convenient boundary.</p>
3131 <p>'<tt>type</tt>' may be any sized type.</p>
3135 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3136 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3137 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3138 instruction is commonly used to represent automatic variables that must
3139 have an address available. When the function returns (either with the <tt><a
3140 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3141 instructions), the memory is reclaimed. Allocating zero bytes
3142 is legal, but the result is undefined.</p>
3147 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3148 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3149 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3150 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3154 <!-- _______________________________________________________________________ -->
3155 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3156 Instruction</a> </div>
3157 <div class="doc_text">
3159 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3161 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3163 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3164 address from which to load. The pointer must point to a <a
3165 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3166 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3167 the number or order of execution of this <tt>load</tt> with other
3168 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3171 The optional constant "align" argument specifies the alignment of the operation
3172 (that is, the alignment of the memory address). A value of 0 or an
3173 omitted "align" argument means that the operation has the preferential
3174 alignment for the target. It is the responsibility of the code emitter
3175 to ensure that the alignment information is correct. Overestimating
3176 the alignment results in an undefined behavior. Underestimating the
3177 alignment may produce less efficient code. An alignment of 1 is always
3181 <p>The location of memory pointed to is loaded.</p>
3183 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3185 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3186 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3189 <!-- _______________________________________________________________________ -->
3190 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3191 Instruction</a> </div>
3192 <div class="doc_text">
3194 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3195 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3198 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3200 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3201 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3202 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3203 of the '<tt><value></tt>'
3204 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3205 optimizer is not allowed to modify the number or order of execution of
3206 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3207 href="#i_store">store</a></tt> instructions.</p>
3209 The optional constant "align" argument specifies the alignment of the operation
3210 (that is, the alignment of the memory address). A value of 0 or an
3211 omitted "align" argument means that the operation has the preferential
3212 alignment for the target. It is the responsibility of the code emitter
3213 to ensure that the alignment information is correct. Overestimating
3214 the alignment results in an undefined behavior. Underestimating the
3215 alignment may produce less efficient code. An alignment of 1 is always
3219 <p>The contents of memory are updated to contain '<tt><value></tt>'
3220 at the location specified by the '<tt><pointer></tt>' operand.</p>
3222 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3223 store i32 3, i32* %ptr <i>; yields {void}</i>
3224 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3228 <!-- _______________________________________________________________________ -->
3229 <div class="doc_subsubsection">
3230 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3233 <div class="doc_text">
3236 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
3242 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3243 subelement of an aggregate data structure.</p>
3247 <p>This instruction takes a list of integer operands that indicate what
3248 elements of the aggregate object to index to. The actual types of the arguments
3249 provided depend on the type of the first pointer argument. The
3250 '<tt>getelementptr</tt>' instruction is used to index down through the type
3251 levels of a structure or to a specific index in an array. When indexing into a
3252 structure, only <tt>i32</tt> integer constants are allowed. When indexing
3253 into an array or pointer, only integers of 32 or 64 bits are allowed; 32-bit
3254 values will be sign extended to 64-bits if required.</p>
3256 <p>For example, let's consider a C code fragment and how it gets
3257 compiled to LLVM:</p>
3259 <div class="doc_code">
3272 int *foo(struct ST *s) {
3273 return &s[1].Z.B[5][13];
3278 <p>The LLVM code generated by the GCC frontend is:</p>
3280 <div class="doc_code">
3282 %RT = type { i8 , [10 x [20 x i32]], i8 }
3283 %ST = type { i32, double, %RT }
3285 define i32* %foo(%ST* %s) {
3287 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3295 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3296 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3297 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3298 <a href="#t_integer">integer</a> type but the value will always be sign extended
3299 to 64-bits. <a href="#t_struct">Structure</a> and <a href="#t_pstruct">packed
3300 structure</a> types require <tt>i32</tt> <b>constants</b>.</p>
3302 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3303 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3304 }</tt>' type, a structure. The second index indexes into the third element of
3305 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3306 i8 }</tt>' type, another structure. The third index indexes into the second
3307 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3308 array. The two dimensions of the array are subscripted into, yielding an
3309 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3310 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3312 <p>Note that it is perfectly legal to index partially through a
3313 structure, returning a pointer to an inner element. Because of this,
3314 the LLVM code for the given testcase is equivalent to:</p>
3317 define i32* %foo(%ST* %s) {
3318 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3319 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3320 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3321 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3322 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3327 <p>Note that it is undefined to access an array out of bounds: array and
3328 pointer indexes must always be within the defined bounds of the array type.
3329 The one exception for this rule is zero length arrays. These arrays are
3330 defined to be accessible as variable length arrays, which requires access
3331 beyond the zero'th element.</p>
3333 <p>The getelementptr instruction is often confusing. For some more insight
3334 into how it works, see <a href="GetElementPtr.html">the getelementptr
3340 <i>; yields [12 x i8]*:aptr</i>
3341 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3345 <!-- ======================================================================= -->
3346 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3348 <div class="doc_text">
3349 <p>The instructions in this category are the conversion instructions (casting)
3350 which all take a single operand and a type. They perform various bit conversions
3354 <!-- _______________________________________________________________________ -->
3355 <div class="doc_subsubsection">
3356 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3358 <div class="doc_text">
3362 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3367 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3372 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3373 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3374 and type of the result, which must be an <a href="#t_integer">integer</a>
3375 type. The bit size of <tt>value</tt> must be larger than the bit size of
3376 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3380 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3381 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3382 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3383 It will always truncate bits.</p>
3387 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3388 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3389 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3393 <!-- _______________________________________________________________________ -->
3394 <div class="doc_subsubsection">
3395 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3397 <div class="doc_text">
3401 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3405 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3410 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3411 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3412 also be of <a href="#t_integer">integer</a> type. The bit size of the
3413 <tt>value</tt> must be smaller than the bit size of the destination type,
3417 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3418 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3420 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3424 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3425 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3429 <!-- _______________________________________________________________________ -->
3430 <div class="doc_subsubsection">
3431 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3433 <div class="doc_text">
3437 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3441 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3445 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3446 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3447 also be of <a href="#t_integer">integer</a> type. The bit size of the
3448 <tt>value</tt> must be smaller than the bit size of the destination type,
3453 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3454 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3455 the type <tt>ty2</tt>.</p>
3457 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3461 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3462 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3466 <!-- _______________________________________________________________________ -->
3467 <div class="doc_subsubsection">
3468 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3471 <div class="doc_text">
3476 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3480 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3485 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3486 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3487 cast it to. The size of <tt>value</tt> must be larger than the size of
3488 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3489 <i>no-op cast</i>.</p>
3492 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3493 <a href="#t_floating">floating point</a> type to a smaller
3494 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3495 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3499 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3500 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3504 <!-- _______________________________________________________________________ -->
3505 <div class="doc_subsubsection">
3506 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3508 <div class="doc_text">
3512 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3516 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3517 floating point value.</p>
3520 <p>The '<tt>fpext</tt>' instruction takes a
3521 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3522 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3523 type must be smaller than the destination type.</p>
3526 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3527 <a href="#t_floating">floating point</a> type to a larger
3528 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3529 used to make a <i>no-op cast</i> because it always changes bits. Use
3530 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3534 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3535 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3539 <!-- _______________________________________________________________________ -->
3540 <div class="doc_subsubsection">
3541 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3543 <div class="doc_text">
3547 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3551 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3552 unsigned integer equivalent of type <tt>ty2</tt>.
3556 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3557 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3558 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3559 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3560 vector integer type with the same number of elements as <tt>ty</tt></p>
3563 <p> The '<tt>fptoui</tt>' instruction converts its
3564 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3565 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3566 the results are undefined.</p>
3570 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3571 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3572 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3576 <!-- _______________________________________________________________________ -->
3577 <div class="doc_subsubsection">
3578 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3580 <div class="doc_text">
3584 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3588 <p>The '<tt>fptosi</tt>' instruction converts
3589 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3593 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3594 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3595 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3596 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3597 vector integer type with the same number of elements as <tt>ty</tt></p>
3600 <p>The '<tt>fptosi</tt>' instruction converts its
3601 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3602 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3603 the results are undefined.</p>
3607 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3608 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3609 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3613 <!-- _______________________________________________________________________ -->
3614 <div class="doc_subsubsection">
3615 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3617 <div class="doc_text">
3621 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3625 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3626 integer and converts that value to the <tt>ty2</tt> type.</p>
3629 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3630 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3631 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3632 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3633 floating point type with the same number of elements as <tt>ty</tt></p>
3636 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3637 integer quantity and converts it to the corresponding floating point value. If
3638 the value cannot fit in the floating point value, the results are undefined.</p>
3642 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3643 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3647 <!-- _______________________________________________________________________ -->
3648 <div class="doc_subsubsection">
3649 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3651 <div class="doc_text">
3655 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3659 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3660 integer and converts that value to the <tt>ty2</tt> type.</p>
3663 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3664 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3665 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3666 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3667 floating point type with the same number of elements as <tt>ty</tt></p>
3670 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3671 integer quantity and converts it to the corresponding floating point value. If
3672 the value cannot fit in the floating point value, the results are undefined.</p>
3676 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3677 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3681 <!-- _______________________________________________________________________ -->
3682 <div class="doc_subsubsection">
3683 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3685 <div class="doc_text">
3689 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3693 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3694 the integer type <tt>ty2</tt>.</p>
3697 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3698 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3699 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3702 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3703 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3704 truncating or zero extending that value to the size of the integer type. If
3705 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3706 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3707 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3712 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3713 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3717 <!-- _______________________________________________________________________ -->
3718 <div class="doc_subsubsection">
3719 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3721 <div class="doc_text">
3725 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3729 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3730 a pointer type, <tt>ty2</tt>.</p>
3733 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3734 value to cast, and a type to cast it to, which must be a
3735 <a href="#t_pointer">pointer</a> type.
3738 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3739 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3740 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3741 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3742 the size of a pointer then a zero extension is done. If they are the same size,
3743 nothing is done (<i>no-op cast</i>).</p>
3747 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3748 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3749 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3753 <!-- _______________________________________________________________________ -->
3754 <div class="doc_subsubsection">
3755 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3757 <div class="doc_text">
3761 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3766 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3767 <tt>ty2</tt> without changing any bits.</p>
3771 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3772 a first class value, and a type to cast it to, which must also be a <a
3773 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3774 and the destination type, <tt>ty2</tt>, must be identical. If the source
3775 type is a pointer, the destination type must also be a pointer. This
3776 instruction supports bitwise conversion of vectors to integers and to vectors
3777 of other types (as long as they have the same size).</p>
3780 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3781 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3782 this conversion. The conversion is done as if the <tt>value</tt> had been
3783 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3784 converted to other pointer types with this instruction. To convert pointers to
3785 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3786 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3790 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3791 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3792 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3796 <!-- ======================================================================= -->
3797 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3798 <div class="doc_text">
3799 <p>The instructions in this category are the "miscellaneous"
3800 instructions, which defy better classification.</p>
3803 <!-- _______________________________________________________________________ -->
3804 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3806 <div class="doc_text">
3808 <pre> <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1}:result</i>
3811 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3812 of its two integer or pointer operands.</p>
3814 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3815 the condition code indicating the kind of comparison to perform. It is not
3816 a value, just a keyword. The possible condition code are:
3818 <li><tt>eq</tt>: equal</li>
3819 <li><tt>ne</tt>: not equal </li>
3820 <li><tt>ugt</tt>: unsigned greater than</li>
3821 <li><tt>uge</tt>: unsigned greater or equal</li>
3822 <li><tt>ult</tt>: unsigned less than</li>
3823 <li><tt>ule</tt>: unsigned less or equal</li>
3824 <li><tt>sgt</tt>: signed greater than</li>
3825 <li><tt>sge</tt>: signed greater or equal</li>
3826 <li><tt>slt</tt>: signed less than</li>
3827 <li><tt>sle</tt>: signed less or equal</li>
3829 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3830 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3832 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to
3833 the condition code given as <tt>cond</tt>. The comparison performed always
3834 yields a <a href="#t_primitive">i1</a> result, as follows:
3836 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3837 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3839 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3840 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3841 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3842 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3843 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3844 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3845 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3846 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
3847 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3848 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3849 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3850 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3851 <li><tt>sge</tt>: interprets the operands as signed values and yields
3852 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3853 <li><tt>slt</tt>: interprets the operands as signed values and yields
3854 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
3855 <li><tt>sle</tt>: interprets the operands as signed values and yields
3856 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3858 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3859 values are compared as if they were integers.</p>
3862 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3863 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3864 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3865 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3866 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3867 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3871 <!-- _______________________________________________________________________ -->
3872 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3874 <div class="doc_text">
3876 <pre> <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1}:result</i>
3879 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3880 of its floating point operands.</p>
3882 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3883 the condition code indicating the kind of comparison to perform. It is not
3884 a value, just a keyword. The possible condition code are:
3886 <li><tt>false</tt>: no comparison, always returns false</li>
3887 <li><tt>oeq</tt>: ordered and equal</li>
3888 <li><tt>ogt</tt>: ordered and greater than </li>
3889 <li><tt>oge</tt>: ordered and greater than or equal</li>
3890 <li><tt>olt</tt>: ordered and less than </li>
3891 <li><tt>ole</tt>: ordered and less than or equal</li>
3892 <li><tt>one</tt>: ordered and not equal</li>
3893 <li><tt>ord</tt>: ordered (no nans)</li>
3894 <li><tt>ueq</tt>: unordered or equal</li>
3895 <li><tt>ugt</tt>: unordered or greater than </li>
3896 <li><tt>uge</tt>: unordered or greater than or equal</li>
3897 <li><tt>ult</tt>: unordered or less than </li>
3898 <li><tt>ule</tt>: unordered or less than or equal</li>
3899 <li><tt>une</tt>: unordered or not equal</li>
3900 <li><tt>uno</tt>: unordered (either nans)</li>
3901 <li><tt>true</tt>: no comparison, always returns true</li>
3903 <p><i>Ordered</i> means that neither operand is a QNAN while
3904 <i>unordered</i> means that either operand may be a QNAN.</p>
3905 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3906 <a href="#t_floating">floating point</a> typed. They must have identical
3909 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
3910 according to the condition code given as <tt>cond</tt>. The comparison performed
3911 always yields a <a href="#t_primitive">i1</a> result, as follows:
3913 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3914 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3915 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
3916 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3917 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
3918 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3919 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3920 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3921 <tt>op1</tt> is less than <tt>op2</tt>.</li>
3922 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3923 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3924 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3925 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
3926 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3927 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3928 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
3929 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3930 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3931 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3932 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3933 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3934 <tt>op1</tt> is less than <tt>op2</tt>.</li>
3935 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3936 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3937 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3938 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
3939 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3940 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3944 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3945 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3946 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3947 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3951 <!-- _______________________________________________________________________ -->
3952 <div class="doc_subsubsection">
3953 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
3955 <div class="doc_text">
3957 <pre> <result> = vicmp <cond> <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3960 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
3961 element-wise comparison of its two integer vector operands.</p>
3963 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
3964 the condition code indicating the kind of comparison to perform. It is not
3965 a value, just a keyword. The possible condition code are:
3967 <li><tt>eq</tt>: equal</li>
3968 <li><tt>ne</tt>: not equal </li>
3969 <li><tt>ugt</tt>: unsigned greater than</li>
3970 <li><tt>uge</tt>: unsigned greater or equal</li>
3971 <li><tt>ult</tt>: unsigned less than</li>
3972 <li><tt>ule</tt>: unsigned less or equal</li>
3973 <li><tt>sgt</tt>: signed greater than</li>
3974 <li><tt>sge</tt>: signed greater or equal</li>
3975 <li><tt>slt</tt>: signed less than</li>
3976 <li><tt>sle</tt>: signed less or equal</li>
3978 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
3979 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
3981 <p>The '<tt>vicmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
3982 according to the condition code given as <tt>cond</tt>. The comparison yields a
3983 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
3984 identical type as the values being compared. The most significant bit in each
3985 element is 1 if the element-wise comparison evaluates to true, and is 0
3986 otherwise. All other bits of the result are undefined. The condition codes
3987 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
3992 <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>
3993 <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>
3997 <!-- _______________________________________________________________________ -->
3998 <div class="doc_subsubsection">
3999 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
4001 <div class="doc_text">
4003 <pre> <result> = vfcmp <cond> <ty> <op1>, <op2></pre>
4005 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4006 element-wise comparison of its two floating point vector operands. The output
4007 elements have the same width as the input elements.</p>
4009 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4010 the condition code indicating the kind of comparison to perform. It is not
4011 a value, just a keyword. The possible condition code are:
4013 <li><tt>false</tt>: no comparison, always returns false</li>
4014 <li><tt>oeq</tt>: ordered and equal</li>
4015 <li><tt>ogt</tt>: ordered and greater than </li>
4016 <li><tt>oge</tt>: ordered and greater than or equal</li>
4017 <li><tt>olt</tt>: ordered and less than </li>
4018 <li><tt>ole</tt>: ordered and less than or equal</li>
4019 <li><tt>one</tt>: ordered and not equal</li>
4020 <li><tt>ord</tt>: ordered (no nans)</li>
4021 <li><tt>ueq</tt>: unordered or equal</li>
4022 <li><tt>ugt</tt>: unordered or greater than </li>
4023 <li><tt>uge</tt>: unordered or greater than or equal</li>
4024 <li><tt>ult</tt>: unordered or less than </li>
4025 <li><tt>ule</tt>: unordered or less than or equal</li>
4026 <li><tt>une</tt>: unordered or not equal</li>
4027 <li><tt>uno</tt>: unordered (either nans)</li>
4028 <li><tt>true</tt>: no comparison, always returns true</li>
4030 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4031 <a href="#t_floating">floating point</a> typed. They must also be identical
4034 <p>The '<tt>vfcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4035 according to the condition code given as <tt>cond</tt>. The comparison yields a
4036 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4037 an identical number of elements as the values being compared, and each element
4038 having identical with to the width of the floating point elements. The most
4039 significant bit in each element is 1 if the element-wise comparison evaluates to
4040 true, and is 0 otherwise. All other bits of the result are undefined. The
4041 condition codes are evaluated identically to the
4042 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.
4046 <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>
4047 <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>
4051 <!-- _______________________________________________________________________ -->
4052 <div class="doc_subsubsection">
4053 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4056 <div class="doc_text">
4060 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4062 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4063 the SSA graph representing the function.</p>
4066 <p>The type of the incoming values is specified with the first type
4067 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4068 as arguments, with one pair for each predecessor basic block of the
4069 current block. Only values of <a href="#t_firstclass">first class</a>
4070 type may be used as the value arguments to the PHI node. Only labels
4071 may be used as the label arguments.</p>
4073 <p>There must be no non-phi instructions between the start of a basic
4074 block and the PHI instructions: i.e. PHI instructions must be first in
4079 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4080 specified by the pair corresponding to the predecessor basic block that executed
4081 just prior to the current block.</p>
4085 Loop: ; Infinite loop that counts from 0 on up...
4086 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4087 %nextindvar = add i32 %indvar, 1
4092 <!-- _______________________________________________________________________ -->
4093 <div class="doc_subsubsection">
4094 <a name="i_select">'<tt>select</tt>' Instruction</a>
4097 <div class="doc_text">
4102 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4108 The '<tt>select</tt>' instruction is used to choose one value based on a
4109 condition, without branching.
4116 The '<tt>select</tt>' instruction requires an 'i1' value indicating the
4117 condition, and two values of the same <a href="#t_firstclass">first class</a>
4118 type. If the val1/val2 are vectors, the entire vectors are selected, not
4119 individual elements.
4125 If the i1 condition evaluates is 1, the instruction returns the first
4126 value argument; otherwise, it returns the second value argument.
4132 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4137 <!-- _______________________________________________________________________ -->
4138 <div class="doc_subsubsection">
4139 <a name="i_call">'<tt>call</tt>' Instruction</a>
4142 <div class="doc_text">
4146 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
4151 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4155 <p>This instruction requires several arguments:</p>
4159 <p>The optional "tail" marker indicates whether the callee function accesses
4160 any allocas or varargs in the caller. If the "tail" marker is present, the
4161 function call is eligible for tail call optimization. Note that calls may
4162 be marked "tail" even if they do not occur before a <a
4163 href="#i_ret"><tt>ret</tt></a> instruction.
4166 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4167 convention</a> the call should use. If none is specified, the call defaults
4168 to using C calling conventions.
4171 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4172 the type of the return value. Functions that return no value are marked
4173 <tt><a href="#t_void">void</a></tt>.</p>
4176 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4177 value being invoked. The argument types must match the types implied by
4178 this signature. This type can be omitted if the function is not varargs
4179 and if the function type does not return a pointer to a function.</p>
4182 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4183 be invoked. In most cases, this is a direct function invocation, but
4184 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4185 to function value.</p>
4188 <p>'<tt>function args</tt>': argument list whose types match the
4189 function signature argument types. All arguments must be of
4190 <a href="#t_firstclass">first class</a> type. If the function signature
4191 indicates the function accepts a variable number of arguments, the extra
4192 arguments can be specified.</p>
4198 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4199 transfer to a specified function, with its incoming arguments bound to
4200 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4201 instruction in the called function, control flow continues with the
4202 instruction after the function call, and the return value of the
4203 function is bound to the result argument. If the callee returns multiple
4204 values then the return values of the function are only accessible through
4205 the '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
4210 %retval = call i32 @test(i32 %argc)
4211 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4212 %X = tail call i32 @foo() <i>; yields i32</i>
4213 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4214 call void %foo(i8 97 signext)
4216 %struct.A = type { i32, i8 }
4217 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4218 %gr = getresult %struct.A %r, 0 <i>; yields i32</i>
4219 %gr1 = getresult %struct.A %r, 1 <i>; yields i8</i>
4224 <!-- _______________________________________________________________________ -->
4225 <div class="doc_subsubsection">
4226 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4229 <div class="doc_text">
4234 <resultval> = va_arg <va_list*> <arglist>, <argty>
4239 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4240 the "variable argument" area of a function call. It is used to implement the
4241 <tt>va_arg</tt> macro in C.</p>
4245 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4246 the argument. It returns a value of the specified argument type and
4247 increments the <tt>va_list</tt> to point to the next argument. The
4248 actual type of <tt>va_list</tt> is target specific.</p>
4252 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4253 type from the specified <tt>va_list</tt> and causes the
4254 <tt>va_list</tt> to point to the next argument. For more information,
4255 see the variable argument handling <a href="#int_varargs">Intrinsic
4258 <p>It is legal for this instruction to be called in a function which does not
4259 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4262 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4263 href="#intrinsics">intrinsic function</a> because it takes a type as an
4268 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4272 <!-- _______________________________________________________________________ -->
4273 <div class="doc_subsubsection">
4274 <a name="i_getresult">'<tt>getresult</tt>' Instruction</a>
4277 <div class="doc_text">
4281 <resultval> = getresult <type> <retval>, <index>
4286 <p> The '<tt>getresult</tt>' instruction is used to extract individual values
4287 from a '<tt><a href="#i_call">call</a></tt>'
4288 or '<tt><a href="#i_invoke">invoke</a></tt>' instruction that returns multiple
4293 <p>The '<tt>getresult</tt>' instruction takes a call or invoke value as its
4294 first argument, or an undef value. The value must have <a
4295 href="#t_struct">structure type</a>. The second argument is a constant
4296 unsigned index value which must be in range for the number of values returned
4301 <p>The '<tt>getresult</tt>' instruction extracts the element identified by
4302 '<tt>index</tt>' from the aggregate value.</p>
4307 %struct.A = type { i32, i8 }
4309 %r = call %struct.A @foo()
4310 %gr = getresult %struct.A %r, 0 <i>; yields i32:%gr</i>
4311 %gr1 = getresult %struct.A %r, 1 <i>; yields i8:%gr1</i>
4318 <!-- *********************************************************************** -->
4319 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4320 <!-- *********************************************************************** -->
4322 <div class="doc_text">
4324 <p>LLVM supports the notion of an "intrinsic function". These functions have
4325 well known names and semantics and are required to follow certain restrictions.
4326 Overall, these intrinsics represent an extension mechanism for the LLVM
4327 language that does not require changing all of the transformations in LLVM when
4328 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4330 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4331 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4332 begin with this prefix. Intrinsic functions must always be external functions:
4333 you cannot define the body of intrinsic functions. Intrinsic functions may
4334 only be used in call or invoke instructions: it is illegal to take the address
4335 of an intrinsic function. Additionally, because intrinsic functions are part
4336 of the LLVM language, it is required if any are added that they be documented
4339 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4340 a family of functions that perform the same operation but on different data
4341 types. Because LLVM can represent over 8 million different integer types,
4342 overloading is used commonly to allow an intrinsic function to operate on any
4343 integer type. One or more of the argument types or the result type can be
4344 overloaded to accept any integer type. Argument types may also be defined as
4345 exactly matching a previous argument's type or the result type. This allows an
4346 intrinsic function which accepts multiple arguments, but needs all of them to
4347 be of the same type, to only be overloaded with respect to a single argument or
4350 <p>Overloaded intrinsics will have the names of its overloaded argument types
4351 encoded into its function name, each preceded by a period. Only those types
4352 which are overloaded result in a name suffix. Arguments whose type is matched
4353 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4354 take an integer of any width and returns an integer of exactly the same integer
4355 width. This leads to a family of functions such as
4356 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4357 Only one type, the return type, is overloaded, and only one type suffix is
4358 required. Because the argument's type is matched against the return type, it
4359 does not require its own name suffix.</p>
4361 <p>To learn how to add an intrinsic function, please see the
4362 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4367 <!-- ======================================================================= -->
4368 <div class="doc_subsection">
4369 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4372 <div class="doc_text">
4374 <p>Variable argument support is defined in LLVM with the <a
4375 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4376 intrinsic functions. These functions are related to the similarly
4377 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4379 <p>All of these functions operate on arguments that use a
4380 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4381 language reference manual does not define what this type is, so all
4382 transformations should be prepared to handle these functions regardless of
4385 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4386 instruction and the variable argument handling intrinsic functions are
4389 <div class="doc_code">
4391 define i32 @test(i32 %X, ...) {
4392 ; Initialize variable argument processing
4394 %ap2 = bitcast i8** %ap to i8*
4395 call void @llvm.va_start(i8* %ap2)
4397 ; Read a single integer argument
4398 %tmp = va_arg i8** %ap, i32
4400 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4402 %aq2 = bitcast i8** %aq to i8*
4403 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4404 call void @llvm.va_end(i8* %aq2)
4406 ; Stop processing of arguments.
4407 call void @llvm.va_end(i8* %ap2)
4411 declare void @llvm.va_start(i8*)
4412 declare void @llvm.va_copy(i8*, i8*)
4413 declare void @llvm.va_end(i8*)
4419 <!-- _______________________________________________________________________ -->
4420 <div class="doc_subsubsection">
4421 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4425 <div class="doc_text">
4427 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4429 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
4430 <tt>*<arglist></tt> for subsequent use by <tt><a
4431 href="#i_va_arg">va_arg</a></tt>.</p>
4435 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4439 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4440 macro available in C. In a target-dependent way, it initializes the
4441 <tt>va_list</tt> element to which the argument points, so that the next call to
4442 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4443 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4444 last argument of the function as the compiler can figure that out.</p>
4448 <!-- _______________________________________________________________________ -->
4449 <div class="doc_subsubsection">
4450 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4453 <div class="doc_text">
4455 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4458 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4459 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4460 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4464 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4468 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4469 macro available in C. In a target-dependent way, it destroys the
4470 <tt>va_list</tt> element to which the argument points. Calls to <a
4471 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4472 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4473 <tt>llvm.va_end</tt>.</p>
4477 <!-- _______________________________________________________________________ -->
4478 <div class="doc_subsubsection">
4479 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4482 <div class="doc_text">
4487 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4492 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4493 from the source argument list to the destination argument list.</p>
4497 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4498 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4503 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4504 macro available in C. In a target-dependent way, it copies the source
4505 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4506 intrinsic is necessary because the <tt><a href="#int_va_start">
4507 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4508 example, memory allocation.</p>
4512 <!-- ======================================================================= -->
4513 <div class="doc_subsection">
4514 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4517 <div class="doc_text">
4520 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4521 Collection</a> (GC) requires the implementation and generation of these
4523 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4524 stack</a>, as well as garbage collector implementations that require <a
4525 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4526 Front-ends for type-safe garbage collected languages should generate these
4527 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4528 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4531 <p>The garbage collection intrinsics only operate on objects in the generic
4532 address space (address space zero).</p>
4536 <!-- _______________________________________________________________________ -->
4537 <div class="doc_subsubsection">
4538 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4541 <div class="doc_text">
4546 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4551 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4552 the code generator, and allows some metadata to be associated with it.</p>
4556 <p>The first argument specifies the address of a stack object that contains the
4557 root pointer. The second pointer (which must be either a constant or a global
4558 value address) contains the meta-data to be associated with the root.</p>
4562 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4563 location. At compile-time, the code generator generates information to allow
4564 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4565 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4571 <!-- _______________________________________________________________________ -->
4572 <div class="doc_subsubsection">
4573 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4576 <div class="doc_text">
4581 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4586 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4587 locations, allowing garbage collector implementations that require read
4592 <p>The second argument is the address to read from, which should be an address
4593 allocated from the garbage collector. The first object is a pointer to the
4594 start of the referenced object, if needed by the language runtime (otherwise
4599 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4600 instruction, but may be replaced with substantially more complex code by the
4601 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4602 may only be used in a function which <a href="#gc">specifies a GC
4608 <!-- _______________________________________________________________________ -->
4609 <div class="doc_subsubsection">
4610 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4613 <div class="doc_text">
4618 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4623 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4624 locations, allowing garbage collector implementations that require write
4625 barriers (such as generational or reference counting collectors).</p>
4629 <p>The first argument is the reference to store, the second is the start of the
4630 object to store it to, and the third is the address of the field of Obj to
4631 store to. If the runtime does not require a pointer to the object, Obj may be
4636 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4637 instruction, but may be replaced with substantially more complex code by the
4638 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4639 may only be used in a function which <a href="#gc">specifies a GC
4646 <!-- ======================================================================= -->
4647 <div class="doc_subsection">
4648 <a name="int_codegen">Code Generator Intrinsics</a>
4651 <div class="doc_text">
4653 These intrinsics are provided by LLVM to expose special features that may only
4654 be implemented with code generator support.
4659 <!-- _______________________________________________________________________ -->
4660 <div class="doc_subsubsection">
4661 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4664 <div class="doc_text">
4668 declare i8 *@llvm.returnaddress(i32 <level>)
4674 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4675 target-specific value indicating the return address of the current function
4676 or one of its callers.
4682 The argument to this intrinsic indicates which function to return the address
4683 for. Zero indicates the calling function, one indicates its caller, etc. The
4684 argument is <b>required</b> to be a constant integer value.
4690 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4691 the return address of the specified call frame, or zero if it cannot be
4692 identified. The value returned by this intrinsic is likely to be incorrect or 0
4693 for arguments other than zero, so it should only be used for debugging purposes.
4697 Note that calling this intrinsic does not prevent function inlining or other
4698 aggressive transformations, so the value returned may not be that of the obvious
4699 source-language caller.
4704 <!-- _______________________________________________________________________ -->
4705 <div class="doc_subsubsection">
4706 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4709 <div class="doc_text">
4713 declare i8 *@llvm.frameaddress(i32 <level>)
4719 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4720 target-specific frame pointer value for the specified stack frame.
4726 The argument to this intrinsic indicates which function to return the frame
4727 pointer for. Zero indicates the calling function, one indicates its caller,
4728 etc. The argument is <b>required</b> to be a constant integer value.
4734 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4735 the frame address of the specified call frame, or zero if it cannot be
4736 identified. The value returned by this intrinsic is likely to be incorrect or 0
4737 for arguments other than zero, so it should only be used for debugging purposes.
4741 Note that calling this intrinsic does not prevent function inlining or other
4742 aggressive transformations, so the value returned may not be that of the obvious
4743 source-language caller.
4747 <!-- _______________________________________________________________________ -->
4748 <div class="doc_subsubsection">
4749 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4752 <div class="doc_text">
4756 declare i8 *@llvm.stacksave()
4762 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4763 the function stack, for use with <a href="#int_stackrestore">
4764 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4765 features like scoped automatic variable sized arrays in C99.
4771 This intrinsic returns a opaque pointer value that can be passed to <a
4772 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4773 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4774 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4775 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4776 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4777 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4782 <!-- _______________________________________________________________________ -->
4783 <div class="doc_subsubsection">
4784 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4787 <div class="doc_text">
4791 declare void @llvm.stackrestore(i8 * %ptr)
4797 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4798 the function stack to the state it was in when the corresponding <a
4799 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4800 useful for implementing language features like scoped automatic variable sized
4807 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4813 <!-- _______________________________________________________________________ -->
4814 <div class="doc_subsubsection">
4815 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4818 <div class="doc_text">
4822 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4829 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4830 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4832 effect on the behavior of the program but can change its performance
4839 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4840 determining if the fetch should be for a read (0) or write (1), and
4841 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4842 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4843 <tt>locality</tt> arguments must be constant integers.
4849 This intrinsic does not modify the behavior of the program. In particular,
4850 prefetches cannot trap and do not produce a value. On targets that support this
4851 intrinsic, the prefetch can provide hints to the processor cache for better
4857 <!-- _______________________________________________________________________ -->
4858 <div class="doc_subsubsection">
4859 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4862 <div class="doc_text">
4866 declare void @llvm.pcmarker(i32 <id>)
4873 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4875 code to simulators and other tools. The method is target specific, but it is
4876 expected that the marker will use exported symbols to transmit the PC of the
4878 The marker makes no guarantees that it will remain with any specific instruction
4879 after optimizations. It is possible that the presence of a marker will inhibit
4880 optimizations. The intended use is to be inserted after optimizations to allow
4881 correlations of simulation runs.
4887 <tt>id</tt> is a numerical id identifying the marker.
4893 This intrinsic does not modify the behavior of the program. Backends that do not
4894 support this intrinisic may ignore it.
4899 <!-- _______________________________________________________________________ -->
4900 <div class="doc_subsubsection">
4901 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4904 <div class="doc_text">
4908 declare i64 @llvm.readcyclecounter( )
4915 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4916 counter register (or similar low latency, high accuracy clocks) on those targets
4917 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4918 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4919 should only be used for small timings.
4925 When directly supported, reading the cycle counter should not modify any memory.
4926 Implementations are allowed to either return a application specific value or a
4927 system wide value. On backends without support, this is lowered to a constant 0.
4932 <!-- ======================================================================= -->
4933 <div class="doc_subsection">
4934 <a name="int_libc">Standard C Library Intrinsics</a>
4937 <div class="doc_text">
4939 LLVM provides intrinsics for a few important standard C library functions.
4940 These intrinsics allow source-language front-ends to pass information about the
4941 alignment of the pointer arguments to the code generator, providing opportunity
4942 for more efficient code generation.
4947 <!-- _______________________________________________________________________ -->
4948 <div class="doc_subsubsection">
4949 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4952 <div class="doc_text">
4956 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4957 i32 <len>, i32 <align>)
4958 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4959 i64 <len>, i32 <align>)
4965 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4966 location to the destination location.
4970 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4971 intrinsics do not return a value, and takes an extra alignment argument.
4977 The first argument is a pointer to the destination, the second is a pointer to
4978 the source. The third argument is an integer argument
4979 specifying the number of bytes to copy, and the fourth argument is the alignment
4980 of the source and destination locations.
4984 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4985 the caller guarantees that both the source and destination pointers are aligned
4992 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4993 location to the destination location, which are not allowed to overlap. It
4994 copies "len" bytes of memory over. If the argument is known to be aligned to
4995 some boundary, this can be specified as the fourth argument, otherwise it should
5001 <!-- _______________________________________________________________________ -->
5002 <div class="doc_subsubsection">
5003 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5006 <div class="doc_text">
5010 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5011 i32 <len>, i32 <align>)
5012 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5013 i64 <len>, i32 <align>)
5019 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5020 location to the destination location. It is similar to the
5021 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5025 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5026 intrinsics do not return a value, and takes an extra alignment argument.
5032 The first argument is a pointer to the destination, the second is a pointer to
5033 the source. The third argument is an integer argument
5034 specifying the number of bytes to copy, and the fourth argument is the alignment
5035 of the source and destination locations.
5039 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5040 the caller guarantees that the source and destination pointers are aligned to
5047 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5048 location to the destination location, which may overlap. It
5049 copies "len" bytes of memory over. If the argument is known to be aligned to
5050 some boundary, this can be specified as the fourth argument, otherwise it should
5056 <!-- _______________________________________________________________________ -->
5057 <div class="doc_subsubsection">
5058 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5061 <div class="doc_text">
5065 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5066 i32 <len>, i32 <align>)
5067 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5068 i64 <len>, i32 <align>)
5074 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5079 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5080 does not return a value, and takes an extra alignment argument.
5086 The first argument is a pointer to the destination to fill, the second is the
5087 byte value to fill it with, the third argument is an integer
5088 argument specifying the number of bytes to fill, and the fourth argument is the
5089 known alignment of destination location.
5093 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5094 the caller guarantees that the destination pointer is aligned to that boundary.
5100 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5102 destination location. If the argument is known to be aligned to some boundary,
5103 this can be specified as the fourth argument, otherwise it should be set to 0 or
5109 <!-- _______________________________________________________________________ -->
5110 <div class="doc_subsubsection">
5111 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5114 <div class="doc_text">
5117 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5118 floating point or vector of floating point type. Not all targets support all
5121 declare float @llvm.sqrt.f32(float %Val)
5122 declare double @llvm.sqrt.f64(double %Val)
5123 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5124 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5125 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5131 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5132 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5133 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5134 negative numbers other than -0.0 (which allows for better optimization, because
5135 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5136 defined to return -0.0 like IEEE sqrt.
5142 The argument and return value are floating point numbers of the same type.
5148 This function returns the sqrt of the specified operand if it is a nonnegative
5149 floating point number.
5153 <!-- _______________________________________________________________________ -->
5154 <div class="doc_subsubsection">
5155 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5158 <div class="doc_text">
5161 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5162 floating point or vector of floating point type. Not all targets support all
5165 declare float @llvm.powi.f32(float %Val, i32 %power)
5166 declare double @llvm.powi.f64(double %Val, i32 %power)
5167 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5168 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5169 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5175 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5176 specified (positive or negative) power. The order of evaluation of
5177 multiplications is not defined. When a vector of floating point type is
5178 used, the second argument remains a scalar integer value.
5184 The second argument is an integer power, and the first is a value to raise to
5191 This function returns the first value raised to the second power with an
5192 unspecified sequence of rounding operations.</p>
5195 <!-- _______________________________________________________________________ -->
5196 <div class="doc_subsubsection">
5197 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5200 <div class="doc_text">
5203 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5204 floating point or vector of floating point type. Not all targets support all
5207 declare float @llvm.sin.f32(float %Val)
5208 declare double @llvm.sin.f64(double %Val)
5209 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5210 declare fp128 @llvm.sin.f128(fp128 %Val)
5211 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5217 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5223 The argument and return value are floating point numbers of the same type.
5229 This function returns the sine of the specified operand, returning the
5230 same values as the libm <tt>sin</tt> functions would, and handles error
5231 conditions in the same way.</p>
5234 <!-- _______________________________________________________________________ -->
5235 <div class="doc_subsubsection">
5236 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5239 <div class="doc_text">
5242 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5243 floating point or vector of floating point type. Not all targets support all
5246 declare float @llvm.cos.f32(float %Val)
5247 declare double @llvm.cos.f64(double %Val)
5248 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5249 declare fp128 @llvm.cos.f128(fp128 %Val)
5250 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5256 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5262 The argument and return value are floating point numbers of the same type.
5268 This function returns the cosine of the specified operand, returning the
5269 same values as the libm <tt>cos</tt> functions would, and handles error
5270 conditions in the same way.</p>
5273 <!-- _______________________________________________________________________ -->
5274 <div class="doc_subsubsection">
5275 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5278 <div class="doc_text">
5281 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5282 floating point or vector of floating point type. Not all targets support all
5285 declare float @llvm.pow.f32(float %Val, float %Power)
5286 declare double @llvm.pow.f64(double %Val, double %Power)
5287 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5288 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5289 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5295 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5296 specified (positive or negative) power.
5302 The second argument is a floating point power, and the first is a value to
5303 raise to that power.
5309 This function returns the first value raised to the second power,
5311 same values as the libm <tt>pow</tt> functions would, and handles error
5312 conditions in the same way.</p>
5316 <!-- ======================================================================= -->
5317 <div class="doc_subsection">
5318 <a name="int_manip">Bit Manipulation Intrinsics</a>
5321 <div class="doc_text">
5323 LLVM provides intrinsics for a few important bit manipulation operations.
5324 These allow efficient code generation for some algorithms.
5329 <!-- _______________________________________________________________________ -->
5330 <div class="doc_subsubsection">
5331 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5334 <div class="doc_text">
5337 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5338 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
5340 declare i16 @llvm.bswap.i16(i16 <id>)
5341 declare i32 @llvm.bswap.i32(i32 <id>)
5342 declare i64 @llvm.bswap.i64(i64 <id>)
5348 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5349 values with an even number of bytes (positive multiple of 16 bits). These are
5350 useful for performing operations on data that is not in the target's native
5357 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5358 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5359 intrinsic returns an i32 value that has the four bytes of the input i32
5360 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5361 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5362 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5363 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5368 <!-- _______________________________________________________________________ -->
5369 <div class="doc_subsubsection">
5370 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5373 <div class="doc_text">
5376 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5377 width. Not all targets support all bit widths however.
5379 declare i8 @llvm.ctpop.i8 (i8 <src>)
5380 declare i16 @llvm.ctpop.i16(i16 <src>)
5381 declare i32 @llvm.ctpop.i32(i32 <src>)
5382 declare i64 @llvm.ctpop.i64(i64 <src>)
5383 declare i256 @llvm.ctpop.i256(i256 <src>)
5389 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5396 The only argument is the value to be counted. The argument may be of any
5397 integer type. The return type must match the argument type.
5403 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5407 <!-- _______________________________________________________________________ -->
5408 <div class="doc_subsubsection">
5409 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5412 <div class="doc_text">
5415 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5416 integer bit width. Not all targets support all bit widths however.
5418 declare i8 @llvm.ctlz.i8 (i8 <src>)
5419 declare i16 @llvm.ctlz.i16(i16 <src>)
5420 declare i32 @llvm.ctlz.i32(i32 <src>)
5421 declare i64 @llvm.ctlz.i64(i64 <src>)
5422 declare i256 @llvm.ctlz.i256(i256 <src>)
5428 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5429 leading zeros in a variable.
5435 The only argument is the value to be counted. The argument may be of any
5436 integer type. The return type must match the argument type.
5442 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5443 in a variable. If the src == 0 then the result is the size in bits of the type
5444 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5450 <!-- _______________________________________________________________________ -->
5451 <div class="doc_subsubsection">
5452 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5455 <div class="doc_text">
5458 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5459 integer bit width. Not all targets support all bit widths however.
5461 declare i8 @llvm.cttz.i8 (i8 <src>)
5462 declare i16 @llvm.cttz.i16(i16 <src>)
5463 declare i32 @llvm.cttz.i32(i32 <src>)
5464 declare i64 @llvm.cttz.i64(i64 <src>)
5465 declare i256 @llvm.cttz.i256(i256 <src>)
5471 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5478 The only argument is the value to be counted. The argument may be of any
5479 integer type. The return type must match the argument type.
5485 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5486 in a variable. If the src == 0 then the result is the size in bits of the type
5487 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5491 <!-- _______________________________________________________________________ -->
5492 <div class="doc_subsubsection">
5493 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5496 <div class="doc_text">
5499 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5500 on any integer bit width.
5502 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5503 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5507 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5508 range of bits from an integer value and returns them in the same bit width as
5509 the original value.</p>
5512 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5513 any bit width but they must have the same bit width. The second and third
5514 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5517 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5518 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5519 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5520 operates in forward mode.</p>
5521 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5522 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5523 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5525 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5526 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5527 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5528 to determine the number of bits to retain.</li>
5529 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5530 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5532 <p>In reverse mode, a similar computation is made except that the bits are
5533 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5534 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5535 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5536 <tt>i16 0x0026 (000000100110)</tt>.</p>
5539 <div class="doc_subsubsection">
5540 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5543 <div class="doc_text">
5546 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5547 on any integer bit width.
5549 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5550 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5554 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5555 of bits in an integer value with another integer value. It returns the integer
5556 with the replaced bits.</p>
5559 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5560 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5561 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5562 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5563 type since they specify only a bit index.</p>
5566 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5567 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5568 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5569 operates in forward mode.</p>
5570 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5571 truncating it down to the size of the replacement area or zero extending it
5572 up to that size.</p>
5573 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5574 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5575 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5576 to the <tt>%hi</tt>th bit.
5577 <p>In reverse mode, a similar computation is made except that the bits are
5578 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5579 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5582 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5583 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5584 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5585 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5586 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5590 <!-- ======================================================================= -->
5591 <div class="doc_subsection">
5592 <a name="int_debugger">Debugger Intrinsics</a>
5595 <div class="doc_text">
5597 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5598 are described in the <a
5599 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5600 Debugging</a> document.
5605 <!-- ======================================================================= -->
5606 <div class="doc_subsection">
5607 <a name="int_eh">Exception Handling Intrinsics</a>
5610 <div class="doc_text">
5611 <p> The LLVM exception handling intrinsics (which all start with
5612 <tt>llvm.eh.</tt> prefix), are described in the <a
5613 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5614 Handling</a> document. </p>
5617 <!-- ======================================================================= -->
5618 <div class="doc_subsection">
5619 <a name="int_trampoline">Trampoline Intrinsic</a>
5622 <div class="doc_text">
5624 This intrinsic makes it possible to excise one parameter, marked with
5625 the <tt>nest</tt> attribute, from a function. The result is a callable
5626 function pointer lacking the nest parameter - the caller does not need
5627 to provide a value for it. Instead, the value to use is stored in
5628 advance in a "trampoline", a block of memory usually allocated
5629 on the stack, which also contains code to splice the nest value into the
5630 argument list. This is used to implement the GCC nested function address
5634 For example, if the function is
5635 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5636 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5638 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5639 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5640 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5641 %fp = bitcast i8* %p to i32 (i32, i32)*
5643 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5644 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5647 <!-- _______________________________________________________________________ -->
5648 <div class="doc_subsubsection">
5649 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5651 <div class="doc_text">
5654 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5658 This fills the memory pointed to by <tt>tramp</tt> with code
5659 and returns a function pointer suitable for executing it.
5663 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5664 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5665 and sufficiently aligned block of memory; this memory is written to by the
5666 intrinsic. Note that the size and the alignment are target-specific - LLVM
5667 currently provides no portable way of determining them, so a front-end that
5668 generates this intrinsic needs to have some target-specific knowledge.
5669 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5673 The block of memory pointed to by <tt>tramp</tt> is filled with target
5674 dependent code, turning it into a function. A pointer to this function is
5675 returned, but needs to be bitcast to an
5676 <a href="#int_trampoline">appropriate function pointer type</a>
5677 before being called. The new function's signature is the same as that of
5678 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5679 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5680 of pointer type. Calling the new function is equivalent to calling
5681 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5682 missing <tt>nest</tt> argument. If, after calling
5683 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5684 modified, then the effect of any later call to the returned function pointer is
5689 <!-- ======================================================================= -->
5690 <div class="doc_subsection">
5691 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5694 <div class="doc_text">
5696 These intrinsic functions expand the "universal IR" of LLVM to represent
5697 hardware constructs for atomic operations and memory synchronization. This
5698 provides an interface to the hardware, not an interface to the programmer. It
5699 is aimed at a low enough level to allow any programming models or APIs
5700 (Application Programming Interfaces) which
5701 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5702 hardware behavior. Just as hardware provides a "universal IR" for source
5703 languages, it also provides a starting point for developing a "universal"
5704 atomic operation and synchronization IR.
5707 These do <em>not</em> form an API such as high-level threading libraries,
5708 software transaction memory systems, atomic primitives, and intrinsic
5709 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5710 application libraries. The hardware interface provided by LLVM should allow
5711 a clean implementation of all of these APIs and parallel programming models.
5712 No one model or paradigm should be selected above others unless the hardware
5713 itself ubiquitously does so.
5718 <!-- _______________________________________________________________________ -->
5719 <div class="doc_subsubsection">
5720 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5722 <div class="doc_text">
5725 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5731 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5732 specific pairs of memory access types.
5736 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5737 The first four arguments enables a specific barrier as listed below. The fith
5738 argument specifies that the barrier applies to io or device or uncached memory.
5742 <li><tt>ll</tt>: load-load barrier</li>
5743 <li><tt>ls</tt>: load-store barrier</li>
5744 <li><tt>sl</tt>: store-load barrier</li>
5745 <li><tt>ss</tt>: store-store barrier</li>
5746 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5750 This intrinsic causes the system to enforce some ordering constraints upon
5751 the loads and stores of the program. This barrier does not indicate
5752 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5753 which they occur. For any of the specified pairs of load and store operations
5754 (f.ex. load-load, or store-load), all of the first operations preceding the
5755 barrier will complete before any of the second operations succeeding the
5756 barrier begin. Specifically the semantics for each pairing is as follows:
5759 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5760 after the barrier begins.</li>
5762 <li><tt>ls</tt>: All loads before the barrier must complete before any
5763 store after the barrier begins.</li>
5764 <li><tt>ss</tt>: All stores before the barrier must complete before any
5765 store after the barrier begins.</li>
5766 <li><tt>sl</tt>: All stores before the barrier must complete before any
5767 load after the barrier begins.</li>
5770 These semantics are applied with a logical "and" behavior when more than one
5771 is enabled in a single memory barrier intrinsic.
5774 Backends may implement stronger barriers than those requested when they do not
5775 support as fine grained a barrier as requested. Some architectures do not
5776 need all types of barriers and on such architectures, these become noops.
5783 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5784 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5785 <i>; guarantee the above finishes</i>
5786 store i32 8, %ptr <i>; before this begins</i>
5790 <!-- _______________________________________________________________________ -->
5791 <div class="doc_subsubsection">
5792 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
5794 <div class="doc_text">
5797 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
5798 any integer bit width and for different address spaces. Not all targets
5799 support all bit widths however.</p>
5802 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5803 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5804 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5805 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5810 This loads a value in memory and compares it to a given value. If they are
5811 equal, it stores a new value into the memory.
5815 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
5816 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5817 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5818 this integer type. While any bit width integer may be used, targets may only
5819 lower representations they support in hardware.
5824 This entire intrinsic must be executed atomically. It first loads the value
5825 in memory pointed to by <tt>ptr</tt> and compares it with the value
5826 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5827 loaded value is yielded in all cases. This provides the equivalent of an
5828 atomic compare-and-swap operation within the SSA framework.
5836 %val1 = add i32 4, 4
5837 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
5838 <i>; yields {i32}:result1 = 4</i>
5839 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5840 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5842 %val2 = add i32 1, 1
5843 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
5844 <i>; yields {i32}:result2 = 8</i>
5845 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5847 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5851 <!-- _______________________________________________________________________ -->
5852 <div class="doc_subsubsection">
5853 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5855 <div class="doc_text">
5859 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5860 integer bit width. Not all targets support all bit widths however.</p>
5862 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
5863 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
5864 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
5865 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
5870 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5871 the value from memory. It then stores the value in <tt>val</tt> in the memory
5877 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
5878 <tt>val</tt> argument and the result must be integers of the same bit width.
5879 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5880 integer type. The targets may only lower integer representations they
5885 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5886 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5887 equivalent of an atomic swap operation within the SSA framework.
5895 %val1 = add i32 4, 4
5896 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
5897 <i>; yields {i32}:result1 = 4</i>
5898 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5899 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5901 %val2 = add i32 1, 1
5902 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
5903 <i>; yields {i32}:result2 = 8</i>
5905 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5906 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5910 <!-- _______________________________________________________________________ -->
5911 <div class="doc_subsubsection">
5912 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
5915 <div class="doc_text">
5918 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on any
5919 integer bit width. Not all targets support all bit widths however.</p>
5921 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
5922 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
5923 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
5924 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
5929 This intrinsic adds <tt>delta</tt> to the value stored in memory at
5930 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5935 The intrinsic takes two arguments, the first a pointer to an integer value
5936 and the second an integer value. The result is also an integer value. These
5937 integer types can have any bit width, but they must all have the same bit
5938 width. The targets may only lower integer representations they support.
5942 This intrinsic does a series of operations atomically. It first loads the
5943 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
5944 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5951 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
5952 <i>; yields {i32}:result1 = 4</i>
5953 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
5954 <i>; yields {i32}:result2 = 8</i>
5955 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
5956 <i>; yields {i32}:result3 = 10</i>
5957 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
5961 <!-- _______________________________________________________________________ -->
5962 <div class="doc_subsubsection">
5963 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
5966 <div class="doc_text">
5969 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
5970 any integer bit width and for different address spaces. Not all targets
5971 support all bit widths however.</p>
5973 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
5974 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
5975 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
5976 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
5981 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
5982 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5987 The intrinsic takes two arguments, the first a pointer to an integer value
5988 and the second an integer value. The result is also an integer value. These
5989 integer types can have any bit width, but they must all have the same bit
5990 width. The targets may only lower integer representations they support.
5994 This intrinsic does a series of operations atomically. It first loads the
5995 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
5996 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6003 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6004 <i>; yields {i32}:result1 = 8</i>
6005 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6006 <i>; yields {i32}:result2 = 4</i>
6007 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6008 <i>; yields {i32}:result3 = 2</i>
6009 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6013 <!-- _______________________________________________________________________ -->
6014 <div class="doc_subsubsection">
6015 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6016 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6017 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6018 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6021 <div class="doc_text">
6024 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6025 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
6026 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6027 address spaces. Not all targets support all bit widths however.</p>
6029 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6030 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6031 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6032 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6037 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6038 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6039 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6040 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6045 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6046 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6047 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6048 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6053 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6054 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6055 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6056 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6061 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6062 the value stored in memory at <tt>ptr</tt>. It yields the original value
6068 These intrinsics take two arguments, the first a pointer to an integer value
6069 and the second an integer value. The result is also an integer value. These
6070 integer types can have any bit width, but they must all have the same bit
6071 width. The targets may only lower integer representations they support.
6075 These intrinsics does a series of operations atomically. They first load the
6076 value stored at <tt>ptr</tt>. They then do the bitwise operation
6077 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6078 value stored at <tt>ptr</tt>.
6084 store i32 0x0F0F, %ptr
6085 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6086 <i>; yields {i32}:result0 = 0x0F0F</i>
6087 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6088 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6089 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6090 <i>; yields {i32}:result2 = 0xF0</i>
6091 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6092 <i>; yields {i32}:result3 = FF</i>
6093 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6098 <!-- _______________________________________________________________________ -->
6099 <div class="doc_subsubsection">
6100 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6101 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6102 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6103 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6106 <div class="doc_text">
6109 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6110 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6111 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6112 address spaces. Not all targets
6113 support all bit widths however.</p>
6115 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6116 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6117 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6118 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6123 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6124 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6125 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6126 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6131 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6132 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6133 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6134 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6139 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6140 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6141 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6142 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6147 These intrinsics takes the signed or unsigned minimum or maximum of
6148 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6149 original value at <tt>ptr</tt>.
6154 These intrinsics take two arguments, the first a pointer to an integer value
6155 and the second an integer value. The result is also an integer value. These
6156 integer types can have any bit width, but they must all have the same bit
6157 width. The targets may only lower integer representations they support.
6161 These intrinsics does a series of operations atomically. They first load the
6162 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
6163 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
6164 the original value stored at <tt>ptr</tt>.
6171 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6172 <i>; yields {i32}:result0 = 7</i>
6173 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6174 <i>; yields {i32}:result1 = -2</i>
6175 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6176 <i>; yields {i32}:result2 = 8</i>
6177 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6178 <i>; yields {i32}:result3 = 8</i>
6179 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6183 <!-- ======================================================================= -->
6184 <div class="doc_subsection">
6185 <a name="int_general">General Intrinsics</a>
6188 <div class="doc_text">
6189 <p> This class of intrinsics is designed to be generic and has
6190 no specific purpose. </p>
6193 <!-- _______________________________________________________________________ -->
6194 <div class="doc_subsubsection">
6195 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6198 <div class="doc_text">
6202 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6208 The '<tt>llvm.var.annotation</tt>' intrinsic
6214 The first argument is a pointer to a value, the second is a pointer to a
6215 global string, the third is a pointer to a global string which is the source
6216 file name, and the last argument is the line number.
6222 This intrinsic allows annotation of local variables with arbitrary strings.
6223 This can be useful for special purpose optimizations that want to look for these
6224 annotations. These have no other defined use, they are ignored by code
6225 generation and optimization.
6229 <!-- _______________________________________________________________________ -->
6230 <div class="doc_subsubsection">
6231 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6234 <div class="doc_text">
6237 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6238 any integer bit width.
6241 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6242 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6243 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6244 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6245 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6251 The '<tt>llvm.annotation</tt>' intrinsic.
6257 The first argument is an integer value (result of some expression),
6258 the second is a pointer to a global string, the third is a pointer to a global
6259 string which is the source file name, and the last argument is the line number.
6260 It returns the value of the first argument.
6266 This intrinsic allows annotations to be put on arbitrary expressions
6267 with arbitrary strings. This can be useful for special purpose optimizations
6268 that want to look for these annotations. These have no other defined use, they
6269 are ignored by code generation and optimization.
6272 <!-- _______________________________________________________________________ -->
6273 <div class="doc_subsubsection">
6274 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6277 <div class="doc_text">
6281 declare void @llvm.trap()
6287 The '<tt>llvm.trap</tt>' intrinsic
6299 This intrinsics is lowered to the target dependent trap instruction. If the
6300 target does not have a trap instruction, this intrinsic will be lowered to the
6301 call of the abort() function.
6305 <!-- *********************************************************************** -->
6308 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6309 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
6310 <a href="http://validator.w3.org/check/referer"><img
6311 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
6313 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6314 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6315 Last modified: $Date$