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>
25 <li><a href="#linkage_private">'<tt>private</tt>' Linkage</a></li>
26 <li><a href="#linkage_linker_private">'<tt>linker_private</tt>' Linkage</a></li>
27 <li><a href="#linkage_internal">'<tt>internal</tt>' Linkage</a></li>
28 <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
29 <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
30 <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
31 <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
32 <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
33 <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
34 <li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
35 <li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
36 <li><a href="#linkage_external">'<tt>externally visible</tt>' Linkage</a></li>
37 <li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
38 <li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
41 <li><a href="#callingconv">Calling Conventions</a></li>
42 <li><a href="#namedtypes">Named Types</a></li>
43 <li><a href="#globalvars">Global Variables</a></li>
44 <li><a href="#functionstructure">Functions</a></li>
45 <li><a href="#aliasstructure">Aliases</a></li>
46 <li><a href="#namedmetadatastructure">Named Metadata</a></li>
47 <li><a href="#paramattrs">Parameter Attributes</a></li>
48 <li><a href="#fnattrs">Function Attributes</a></li>
49 <li><a href="#gc">Garbage Collector Names</a></li>
50 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
51 <li><a href="#datalayout">Data Layout</a></li>
52 <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
55 <li><a href="#typesystem">Type System</a>
57 <li><a href="#t_classifications">Type Classifications</a></li>
58 <li><a href="#t_primitive">Primitive Types</a>
60 <li><a href="#t_integer">Integer Type</a></li>
61 <li><a href="#t_floating">Floating Point Types</a></li>
62 <li><a href="#t_void">Void Type</a></li>
63 <li><a href="#t_label">Label Type</a></li>
64 <li><a href="#t_metadata">Metadata Type</a></li>
67 <li><a href="#t_derived">Derived Types</a>
69 <li><a href="#t_array">Array Type</a></li>
70 <li><a href="#t_function">Function Type</a></li>
71 <li><a href="#t_pointer">Pointer Type</a></li>
72 <li><a href="#t_struct">Structure Type</a></li>
73 <li><a href="#t_pstruct">Packed Structure Type</a></li>
74 <li><a href="#t_vector">Vector Type</a></li>
75 <li><a href="#t_opaque">Opaque Type</a></li>
78 <li><a href="#t_uprefs">Type Up-references</a></li>
81 <li><a href="#constants">Constants</a>
83 <li><a href="#simpleconstants">Simple Constants</a></li>
84 <li><a href="#complexconstants">Complex Constants</a></li>
85 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
86 <li><a href="#undefvalues">Undefined Values</a></li>
87 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
88 <li><a href="#constantexprs">Constant Expressions</a></li>
91 <li><a href="#othervalues">Other Values</a>
93 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
94 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a></li>
97 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
99 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
100 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
101 Global Variable</a></li>
102 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
103 Global Variable</a></li>
104 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
105 Global Variable</a></li>
108 <li><a href="#instref">Instruction Reference</a>
110 <li><a href="#terminators">Terminator Instructions</a>
112 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
113 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
114 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
115 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
116 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
117 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
118 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
121 <li><a href="#binaryops">Binary Operations</a>
123 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
124 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
125 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
126 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
127 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
128 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
129 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
130 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
131 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
132 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
133 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
134 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
137 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
139 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
140 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
141 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
142 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
143 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
144 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
147 <li><a href="#vectorops">Vector Operations</a>
149 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
150 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
151 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
154 <li><a href="#aggregateops">Aggregate Operations</a>
156 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
157 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
160 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
162 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
163 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
164 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
165 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
168 <li><a href="#convertops">Conversion Operations</a>
170 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
171 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
172 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
173 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
174 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
175 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
176 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
177 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
178 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
179 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
180 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
181 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
184 <li><a href="#otherops">Other Operations</a>
186 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
187 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
188 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
189 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
190 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
191 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
196 <li><a href="#intrinsics">Intrinsic Functions</a>
198 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
200 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
201 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
202 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
205 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
207 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
208 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
209 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
212 <li><a href="#int_codegen">Code Generator Intrinsics</a>
214 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
215 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
216 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
217 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
218 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
219 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
220 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
223 <li><a href="#int_libc">Standard C Library Intrinsics</a>
225 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
226 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
227 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
228 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
229 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
230 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
231 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
232 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
235 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
237 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
238 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
239 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
240 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
243 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
245 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
246 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
247 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
248 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
249 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
250 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
253 <li><a href="#int_debugger">Debugger intrinsics</a></li>
254 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
255 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
257 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
260 <li><a href="#int_atomics">Atomic intrinsics</a>
262 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
263 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
264 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
265 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
266 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
267 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
268 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
269 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
270 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
271 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
272 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
273 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
274 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
277 <li><a href="#int_memorymarkers">Memory Use Markers</a>
279 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
280 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
281 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
282 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
285 <li><a href="#int_general">General intrinsics</a>
287 <li><a href="#int_var_annotation">
288 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
289 <li><a href="#int_annotation">
290 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
291 <li><a href="#int_trap">
292 '<tt>llvm.trap</tt>' Intrinsic</a></li>
293 <li><a href="#int_stackprotector">
294 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
295 <li><a href="#int_objectsize">
296 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
303 <div class="doc_author">
304 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
305 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
308 <!-- *********************************************************************** -->
309 <div class="doc_section"> <a name="abstract">Abstract </a></div>
310 <!-- *********************************************************************** -->
312 <div class="doc_text">
314 <p>This document is a reference manual for the LLVM assembly language. LLVM is
315 a Static Single Assignment (SSA) based representation that provides type
316 safety, low-level operations, flexibility, and the capability of representing
317 'all' high-level languages cleanly. It is the common code representation
318 used throughout all phases of the LLVM compilation strategy.</p>
322 <!-- *********************************************************************** -->
323 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
324 <!-- *********************************************************************** -->
326 <div class="doc_text">
328 <p>The LLVM code representation is designed to be used in three different forms:
329 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
330 for fast loading by a Just-In-Time compiler), and as a human readable
331 assembly language representation. This allows LLVM to provide a powerful
332 intermediate representation for efficient compiler transformations and
333 analysis, while providing a natural means to debug and visualize the
334 transformations. The three different forms of LLVM are all equivalent. This
335 document describes the human readable representation and notation.</p>
337 <p>The LLVM representation aims to be light-weight and low-level while being
338 expressive, typed, and extensible at the same time. It aims to be a
339 "universal IR" of sorts, by being at a low enough level that high-level ideas
340 may be cleanly mapped to it (similar to how microprocessors are "universal
341 IR's", allowing many source languages to be mapped to them). By providing
342 type information, LLVM can be used as the target of optimizations: for
343 example, through pointer analysis, it can be proven that a C automatic
344 variable is never accessed outside of the current function, allowing it to
345 be promoted to a simple SSA value instead of a memory location.</p>
349 <!-- _______________________________________________________________________ -->
350 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
352 <div class="doc_text">
354 <p>It is important to note that this document describes 'well formed' LLVM
355 assembly language. There is a difference between what the parser accepts and
356 what is considered 'well formed'. For example, the following instruction is
357 syntactically okay, but not well formed:</p>
359 <div class="doc_code">
361 %x = <a href="#i_add">add</a> i32 1, %x
365 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
366 LLVM infrastructure provides a verification pass that may be used to verify
367 that an LLVM module is well formed. This pass is automatically run by the
368 parser after parsing input assembly and by the optimizer before it outputs
369 bitcode. The violations pointed out by the verifier pass indicate bugs in
370 transformation passes or input to the parser.</p>
374 <!-- Describe the typesetting conventions here. -->
376 <!-- *********************************************************************** -->
377 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
378 <!-- *********************************************************************** -->
380 <div class="doc_text">
382 <p>LLVM identifiers come in two basic types: global and local. Global
383 identifiers (functions, global variables) begin with the <tt>'@'</tt>
384 character. Local identifiers (register names, types) begin with
385 the <tt>'%'</tt> character. Additionally, there are three different formats
386 for identifiers, for different purposes:</p>
389 <li>Named values are represented as a string of characters with their prefix.
390 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
391 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
392 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
393 other characters in their names can be surrounded with quotes. Special
394 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
395 ASCII code for the character in hexadecimal. In this way, any character
396 can be used in a name value, even quotes themselves.</li>
398 <li>Unnamed values are represented as an unsigned numeric value with their
399 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
401 <li>Constants, which are described in a <a href="#constants">section about
402 constants</a>, below.</li>
405 <p>LLVM requires that values start with a prefix for two reasons: Compilers
406 don't need to worry about name clashes with reserved words, and the set of
407 reserved words may be expanded in the future without penalty. Additionally,
408 unnamed identifiers allow a compiler to quickly come up with a temporary
409 variable without having to avoid symbol table conflicts.</p>
411 <p>Reserved words in LLVM are very similar to reserved words in other
412 languages. There are keywords for different opcodes
413 ('<tt><a href="#i_add">add</a></tt>',
414 '<tt><a href="#i_bitcast">bitcast</a></tt>',
415 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
416 ('<tt><a href="#t_void">void</a></tt>',
417 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
418 reserved words cannot conflict with variable names, because none of them
419 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
421 <p>Here is an example of LLVM code to multiply the integer variable
422 '<tt>%X</tt>' by 8:</p>
426 <div class="doc_code">
428 %result = <a href="#i_mul">mul</a> i32 %X, 8
432 <p>After strength reduction:</p>
434 <div class="doc_code">
436 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
440 <p>And the hard way:</p>
442 <div class="doc_code">
444 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
445 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
446 %result = <a href="#i_add">add</a> i32 %1, %1
450 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
451 lexical features of LLVM:</p>
454 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
457 <li>Unnamed temporaries are created when the result of a computation is not
458 assigned to a named value.</li>
460 <li>Unnamed temporaries are numbered sequentially</li>
463 <p>It also shows a convention that we follow in this document. When
464 demonstrating instructions, we will follow an instruction with a comment that
465 defines the type and name of value produced. Comments are shown in italic
470 <!-- *********************************************************************** -->
471 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
472 <!-- *********************************************************************** -->
474 <!-- ======================================================================= -->
475 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
478 <div class="doc_text">
480 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
481 of the input programs. Each module consists of functions, global variables,
482 and symbol table entries. Modules may be combined together with the LLVM
483 linker, which merges function (and global variable) definitions, resolves
484 forward declarations, and merges symbol table entries. Here is an example of
485 the "hello world" module:</p>
487 <div class="doc_code">
489 <i>; Declare the string constant as a global constant.</i>
490 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
492 <i>; External declaration of the puts function</i>
493 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
495 <i>; Definition of main function</i>
496 define i32 @main() { <i>; i32()* </i>
497 <i>; Convert [13 x i8]* to i8 *...</i>
498 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
500 <i>; Call puts function to write out the string to stdout.</i>
501 <a href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
502 <a href="#i_ret">ret</a> i32 0<br>}
504 <i>; Named metadata</i>
505 !1 = metadata !{i32 41}
510 <p>This example is made up of a <a href="#globalvars">global variable</a> named
511 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
512 a <a href="#functionstructure">function definition</a> for
513 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
516 <p>In general, a module is made up of a list of global values, where both
517 functions and global variables are global values. Global values are
518 represented by a pointer to a memory location (in this case, a pointer to an
519 array of char, and a pointer to a function), and have one of the
520 following <a href="#linkage">linkage types</a>.</p>
524 <!-- ======================================================================= -->
525 <div class="doc_subsection">
526 <a name="linkage">Linkage Types</a>
529 <div class="doc_text">
531 <p>All Global Variables and Functions have one of the following types of
535 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
536 <dd>Global values with private linkage are only directly accessible by objects
537 in the current module. In particular, linking code into a module with an
538 private global value may cause the private to be renamed as necessary to
539 avoid collisions. Because the symbol is private to the module, all
540 references can be updated. This doesn't show up in any symbol table in the
543 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
544 <dd>Similar to private, but the symbol is passed through the assembler and
545 removed by the linker after evaluation. Note that (unlike private
546 symbols) linker_private symbols are subject to coalescing by the linker:
547 weak symbols get merged and redefinitions are rejected. However, unlike
548 normal strong symbols, they are removed by the linker from the final
549 linked image (executable or dynamic library).</dd>
551 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
552 <dd>Similar to private, but the value shows as a local symbol
553 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
554 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
556 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
557 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
558 into the object file corresponding to the LLVM module. They exist to
559 allow inlining and other optimizations to take place given knowledge of
560 the definition of the global, which is known to be somewhere outside the
561 module. Globals with <tt>available_externally</tt> linkage are allowed to
562 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
563 This linkage type is only allowed on definitions, not declarations.</dd>
565 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
566 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
567 the same name when linkage occurs. This can be used to implement
568 some forms of inline functions, templates, or other code which must be
569 generated in each translation unit that uses it, but where the body may
570 be overridden with a more definitive definition later. Unreferenced
571 <tt>linkonce</tt> globals are allowed to be discarded. Note that
572 <tt>linkonce</tt> linkage does not actually allow the optimizer to
573 inline the body of this function into callers because it doesn't know if
574 this definition of the function is the definitive definition within the
575 program or whether it will be overridden by a stronger definition.
576 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
579 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
580 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
581 <tt>linkonce</tt> linkage, except that unreferenced globals with
582 <tt>weak</tt> linkage may not be discarded. This is used for globals that
583 are declared "weak" in C source code.</dd>
585 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
586 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
587 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
589 Symbols with "<tt>common</tt>" linkage are merged in the same way as
590 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
591 <tt>common</tt> symbols may not have an explicit section,
592 must have a zero initializer, and may not be marked '<a
593 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
594 have common linkage.</dd>
597 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
598 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
599 pointer to array type. When two global variables with appending linkage
600 are linked together, the two global arrays are appended together. This is
601 the LLVM, typesafe, equivalent of having the system linker append together
602 "sections" with identical names when .o files are linked.</dd>
604 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
605 <dd>The semantics of this linkage follow the ELF object file model: the symbol
606 is weak until linked, if not linked, the symbol becomes null instead of
607 being an undefined reference.</dd>
609 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
610 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
611 <dd>Some languages allow differing globals to be merged, such as two functions
612 with different semantics. Other languages, such as <tt>C++</tt>, ensure
613 that only equivalent globals are ever merged (the "one definition rule" -
614 "ODR"). Such languages can use the <tt>linkonce_odr</tt>
615 and <tt>weak_odr</tt> linkage types to indicate that the global will only
616 be merged with equivalent globals. These linkage types are otherwise the
617 same as their non-<tt>odr</tt> versions.</dd>
619 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
620 <dd>If none of the above identifiers are used, the global is externally
621 visible, meaning that it participates in linkage and can be used to
622 resolve external symbol references.</dd>
625 <p>The next two types of linkage are targeted for Microsoft Windows platform
626 only. They are designed to support importing (exporting) symbols from (to)
627 DLLs (Dynamic Link Libraries).</p>
630 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
631 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
632 or variable via a global pointer to a pointer that is set up by the DLL
633 exporting the symbol. On Microsoft Windows targets, the pointer name is
634 formed by combining <code>__imp_</code> and the function or variable
637 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
638 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
639 pointer to a pointer in a DLL, so that it can be referenced with the
640 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
641 name is formed by combining <code>__imp_</code> and the function or
645 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
646 another module defined a "<tt>.LC0</tt>" variable and was linked with this
647 one, one of the two would be renamed, preventing a collision. Since
648 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
649 declarations), they are accessible outside of the current module.</p>
651 <p>It is illegal for a function <i>declaration</i> to have any linkage type
652 other than "externally visible", <tt>dllimport</tt>
653 or <tt>extern_weak</tt>.</p>
655 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
656 or <tt>weak_odr</tt> linkages.</p>
660 <!-- ======================================================================= -->
661 <div class="doc_subsection">
662 <a name="callingconv">Calling Conventions</a>
665 <div class="doc_text">
667 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
668 and <a href="#i_invoke">invokes</a> can all have an optional calling
669 convention specified for the call. The calling convention of any pair of
670 dynamic caller/callee must match, or the behavior of the program is
671 undefined. The following calling conventions are supported by LLVM, and more
672 may be added in the future:</p>
675 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
676 <dd>This calling convention (the default if no other calling convention is
677 specified) matches the target C calling conventions. This calling
678 convention supports varargs function calls and tolerates some mismatch in
679 the declared prototype and implemented declaration of the function (as
682 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
683 <dd>This calling convention attempts to make calls as fast as possible
684 (e.g. by passing things in registers). This calling convention allows the
685 target to use whatever tricks it wants to produce fast code for the
686 target, without having to conform to an externally specified ABI
687 (Application Binary Interface).
688 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
689 when this convention is used.</a> This calling convention does not
690 support varargs and requires the prototype of all callees to exactly match
691 the prototype of the function definition.</dd>
693 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
694 <dd>This calling convention attempts to make code in the caller as efficient
695 as possible under the assumption that the call is not commonly executed.
696 As such, these calls often preserve all registers so that the call does
697 not break any live ranges in the caller side. This calling convention
698 does not support varargs and requires the prototype of all callees to
699 exactly match the prototype of the function definition.</dd>
701 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
702 <dd>Any calling convention may be specified by number, allowing
703 target-specific calling conventions to be used. Target specific calling
704 conventions start at 64.</dd>
707 <p>More calling conventions can be added/defined on an as-needed basis, to
708 support Pascal conventions or any other well-known target-independent
713 <!-- ======================================================================= -->
714 <div class="doc_subsection">
715 <a name="visibility">Visibility Styles</a>
718 <div class="doc_text">
720 <p>All Global Variables and Functions have one of the following visibility
724 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
725 <dd>On targets that use the ELF object file format, default visibility means
726 that the declaration is visible to other modules and, in shared libraries,
727 means that the declared entity may be overridden. On Darwin, default
728 visibility means that the declaration is visible to other modules. Default
729 visibility corresponds to "external linkage" in the language.</dd>
731 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
732 <dd>Two declarations of an object with hidden visibility refer to the same
733 object if they are in the same shared object. Usually, hidden visibility
734 indicates that the symbol will not be placed into the dynamic symbol
735 table, so no other module (executable or shared library) can reference it
738 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
739 <dd>On ELF, protected visibility indicates that the symbol will be placed in
740 the dynamic symbol table, but that references within the defining module
741 will bind to the local symbol. That is, the symbol cannot be overridden by
747 <!-- ======================================================================= -->
748 <div class="doc_subsection">
749 <a name="namedtypes">Named Types</a>
752 <div class="doc_text">
754 <p>LLVM IR allows you to specify name aliases for certain types. This can make
755 it easier to read the IR and make the IR more condensed (particularly when
756 recursive types are involved). An example of a name specification is:</p>
758 <div class="doc_code">
760 %mytype = type { %mytype*, i32 }
764 <p>You may give a name to any <a href="#typesystem">type</a> except
765 "<a href="t_void">void</a>". Type name aliases may be used anywhere a type
766 is expected with the syntax "%mytype".</p>
768 <p>Note that type names are aliases for the structural type that they indicate,
769 and that you can therefore specify multiple names for the same type. This
770 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
771 uses structural typing, the name is not part of the type. When printing out
772 LLVM IR, the printer will pick <em>one name</em> to render all types of a
773 particular shape. This means that if you have code where two different
774 source types end up having the same LLVM type, that the dumper will sometimes
775 print the "wrong" or unexpected type. This is an important design point and
776 isn't going to change.</p>
780 <!-- ======================================================================= -->
781 <div class="doc_subsection">
782 <a name="globalvars">Global Variables</a>
785 <div class="doc_text">
787 <p>Global variables define regions of memory allocated at compilation time
788 instead of run-time. Global variables may optionally be initialized, may
789 have an explicit section to be placed in, and may have an optional explicit
790 alignment specified. A variable may be defined as "thread_local", which
791 means that it will not be shared by threads (each thread will have a
792 separated copy of the variable). A variable may be defined as a global
793 "constant," which indicates that the contents of the variable
794 will <b>never</b> be modified (enabling better optimization, allowing the
795 global data to be placed in the read-only section of an executable, etc).
796 Note that variables that need runtime initialization cannot be marked
797 "constant" as there is a store to the variable.</p>
799 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
800 constant, even if the final definition of the global is not. This capability
801 can be used to enable slightly better optimization of the program, but
802 requires the language definition to guarantee that optimizations based on the
803 'constantness' are valid for the translation units that do not include the
806 <p>As SSA values, global variables define pointer values that are in scope
807 (i.e. they dominate) all basic blocks in the program. Global variables
808 always define a pointer to their "content" type because they describe a
809 region of memory, and all memory objects in LLVM are accessed through
812 <p>A global variable may be declared to reside in a target-specific numbered
813 address space. For targets that support them, address spaces may affect how
814 optimizations are performed and/or what target instructions are used to
815 access the variable. The default address space is zero. The address space
816 qualifier must precede any other attributes.</p>
818 <p>LLVM allows an explicit section to be specified for globals. If the target
819 supports it, it will emit globals to the section specified.</p>
821 <p>An explicit alignment may be specified for a global. If not present, or if
822 the alignment is set to zero, the alignment of the global is set by the
823 target to whatever it feels convenient. If an explicit alignment is
824 specified, the global is forced to have at least that much alignment. All
825 alignments must be a power of 2.</p>
827 <p>For example, the following defines a global in a numbered address space with
828 an initializer, section, and alignment:</p>
830 <div class="doc_code">
832 @G = addrspace(5) constant float 1.0, section "foo", align 4
839 <!-- ======================================================================= -->
840 <div class="doc_subsection">
841 <a name="functionstructure">Functions</a>
844 <div class="doc_text">
846 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord, an
847 optional <a href="#linkage">linkage type</a>, an optional
848 <a href="#visibility">visibility style</a>, an optional
849 <a href="#callingconv">calling convention</a>, a return type, an optional
850 <a href="#paramattrs">parameter attribute</a> for the return type, a function
851 name, a (possibly empty) argument list (each with optional
852 <a href="#paramattrs">parameter attributes</a>), optional
853 <a href="#fnattrs">function attributes</a>, an optional section, an optional
854 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
855 curly brace, a list of basic blocks, and a closing curly brace.</p>
857 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
858 optional <a href="#linkage">linkage type</a>, an optional
859 <a href="#visibility">visibility style</a>, an optional
860 <a href="#callingconv">calling convention</a>, a return type, an optional
861 <a href="#paramattrs">parameter attribute</a> for the return type, a function
862 name, a possibly empty list of arguments, an optional alignment, and an
863 optional <a href="#gc">garbage collector name</a>.</p>
865 <p>A function definition contains a list of basic blocks, forming the CFG
866 (Control Flow Graph) for the function. Each basic block may optionally start
867 with a label (giving the basic block a symbol table entry), contains a list
868 of instructions, and ends with a <a href="#terminators">terminator</a>
869 instruction (such as a branch or function return).</p>
871 <p>The first basic block in a function is special in two ways: it is immediately
872 executed on entrance to the function, and it is not allowed to have
873 predecessor basic blocks (i.e. there can not be any branches to the entry
874 block of a function). Because the block can have no predecessors, it also
875 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
877 <p>LLVM allows an explicit section to be specified for functions. If the target
878 supports it, it will emit functions to the section specified.</p>
880 <p>An explicit alignment may be specified for a function. If not present, or if
881 the alignment is set to zero, the alignment of the function is set by the
882 target to whatever it feels convenient. If an explicit alignment is
883 specified, the function is forced to have at least that much alignment. All
884 alignments must be a power of 2.</p>
887 <div class="doc_code">
889 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
890 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
891 <ResultType> @<FunctionName> ([argument list])
892 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
893 [<a href="#gc">gc</a>] { ... }
899 <!-- ======================================================================= -->
900 <div class="doc_subsection">
901 <a name="aliasstructure">Aliases</a>
904 <div class="doc_text">
906 <p>Aliases act as "second name" for the aliasee value (which can be either
907 function, global variable, another alias or bitcast of global value). Aliases
908 may have an optional <a href="#linkage">linkage type</a>, and an
909 optional <a href="#visibility">visibility style</a>.</p>
912 <div class="doc_code">
914 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
920 <!-- ======================================================================= -->
921 <div class="doc_subsection">
922 <a name="namedmetadatastructure">Named Metadata</a>
925 <div class="doc_text">
927 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
928 nodes</a> (but not metadata strings) and null are the only valid operands for
929 a named metadata.</p>
932 <div class="doc_code">
934 !1 = metadata !{metadata !"one"}
941 <!-- ======================================================================= -->
942 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
944 <div class="doc_text">
946 <p>The return type and each parameter of a function type may have a set of
947 <i>parameter attributes</i> associated with them. Parameter attributes are
948 used to communicate additional information about the result or parameters of
949 a function. Parameter attributes are considered to be part of the function,
950 not of the function type, so functions with different parameter attributes
951 can have the same function type.</p>
953 <p>Parameter attributes are simple keywords that follow the type specified. If
954 multiple parameter attributes are needed, they are space separated. For
957 <div class="doc_code">
959 declare i32 @printf(i8* noalias nocapture, ...)
960 declare i32 @atoi(i8 zeroext)
961 declare signext i8 @returns_signed_char()
965 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
966 <tt>readonly</tt>) come immediately after the argument list.</p>
968 <p>Currently, only the following parameter attributes are defined:</p>
971 <dt><tt><b>zeroext</b></tt></dt>
972 <dd>This indicates to the code generator that the parameter or return value
973 should be zero-extended to a 32-bit value by the caller (for a parameter)
974 or the callee (for a return value).</dd>
976 <dt><tt><b>signext</b></tt></dt>
977 <dd>This indicates to the code generator that the parameter or return value
978 should be sign-extended to a 32-bit value by the caller (for a parameter)
979 or the callee (for a return value).</dd>
981 <dt><tt><b>inreg</b></tt></dt>
982 <dd>This indicates that this parameter or return value should be treated in a
983 special target-dependent fashion during while emitting code for a function
984 call or return (usually, by putting it in a register as opposed to memory,
985 though some targets use it to distinguish between two different kinds of
986 registers). Use of this attribute is target-specific.</dd>
988 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
989 <dd>This indicates that the pointer parameter should really be passed by value
990 to the function. The attribute implies that a hidden copy of the pointee
991 is made between the caller and the callee, so the callee is unable to
992 modify the value in the callee. This attribute is only valid on LLVM
993 pointer arguments. It is generally used to pass structs and arrays by
994 value, but is also valid on pointers to scalars. The copy is considered
995 to belong to the caller not the callee (for example,
996 <tt><a href="#readonly">readonly</a></tt> functions should not write to
997 <tt>byval</tt> parameters). This is not a valid attribute for return
998 values. The byval attribute also supports specifying an alignment with
999 the align attribute. This has a target-specific effect on the code
1000 generator that usually indicates a desired alignment for the synthesized
1003 <dt><tt><b>sret</b></tt></dt>
1004 <dd>This indicates that the pointer parameter specifies the address of a
1005 structure that is the return value of the function in the source program.
1006 This pointer must be guaranteed by the caller to be valid: loads and
1007 stores to the structure may be assumed by the callee to not to trap. This
1008 may only be applied to the first parameter. This is not a valid attribute
1009 for return values. </dd>
1011 <dt><tt><b>noalias</b></tt></dt>
1012 <dd>This indicates that the pointer does not alias any global or any other
1013 parameter. The caller is responsible for ensuring that this is the
1014 case. On a function return value, <tt>noalias</tt> additionally indicates
1015 that the pointer does not alias any other pointers visible to the
1016 caller. For further details, please see the discussion of the NoAlias
1018 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
1021 <dt><tt><b>nocapture</b></tt></dt>
1022 <dd>This indicates that the callee does not make any copies of the pointer
1023 that outlive the callee itself. This is not a valid attribute for return
1026 <dt><tt><b>nest</b></tt></dt>
1027 <dd>This indicates that the pointer parameter can be excised using the
1028 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1029 attribute for return values.</dd>
1034 <!-- ======================================================================= -->
1035 <div class="doc_subsection">
1036 <a name="gc">Garbage Collector Names</a>
1039 <div class="doc_text">
1041 <p>Each function may specify a garbage collector name, which is simply a
1044 <div class="doc_code">
1046 define void @f() gc "name" { ... }
1050 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1051 collector which will cause the compiler to alter its output in order to
1052 support the named garbage collection algorithm.</p>
1056 <!-- ======================================================================= -->
1057 <div class="doc_subsection">
1058 <a name="fnattrs">Function Attributes</a>
1061 <div class="doc_text">
1063 <p>Function attributes are set to communicate additional information about a
1064 function. Function attributes are considered to be part of the function, not
1065 of the function type, so functions with different parameter attributes can
1066 have the same function type.</p>
1068 <p>Function attributes are simple keywords that follow the type specified. If
1069 multiple attributes are needed, they are space separated. For example:</p>
1071 <div class="doc_code">
1073 define void @f() noinline { ... }
1074 define void @f() alwaysinline { ... }
1075 define void @f() alwaysinline optsize { ... }
1076 define void @f() optsize { ... }
1081 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1082 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1083 the backend should forcibly align the stack pointer. Specify the
1084 desired alignment, which must be a power of two, in parentheses.
1086 <dt><tt><b>alwaysinline</b></tt></dt>
1087 <dd>This attribute indicates that the inliner should attempt to inline this
1088 function into callers whenever possible, ignoring any active inlining size
1089 threshold for this caller.</dd>
1091 <dt><tt><b>inlinehint</b></tt></dt>
1092 <dd>This attribute indicates that the source code contained a hint that inlining
1093 this function is desirable (such as the "inline" keyword in C/C++). It
1094 is just a hint; it imposes no requirements on the inliner.</dd>
1096 <dt><tt><b>noinline</b></tt></dt>
1097 <dd>This attribute indicates that the inliner should never inline this
1098 function in any situation. This attribute may not be used together with
1099 the <tt>alwaysinline</tt> attribute.</dd>
1101 <dt><tt><b>optsize</b></tt></dt>
1102 <dd>This attribute suggests that optimization passes and code generator passes
1103 make choices that keep the code size of this function low, and otherwise
1104 do optimizations specifically to reduce code size.</dd>
1106 <dt><tt><b>noreturn</b></tt></dt>
1107 <dd>This function attribute indicates that the function never returns
1108 normally. This produces undefined behavior at runtime if the function
1109 ever does dynamically return.</dd>
1111 <dt><tt><b>nounwind</b></tt></dt>
1112 <dd>This function attribute indicates that the function never returns with an
1113 unwind or exceptional control flow. If the function does unwind, its
1114 runtime behavior is undefined.</dd>
1116 <dt><tt><b>readnone</b></tt></dt>
1117 <dd>This attribute indicates that the function computes its result (or decides
1118 to unwind an exception) based strictly on its arguments, without
1119 dereferencing any pointer arguments or otherwise accessing any mutable
1120 state (e.g. memory, control registers, etc) visible to caller functions.
1121 It does not write through any pointer arguments
1122 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1123 changes any state visible to callers. This means that it cannot unwind
1124 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1125 could use the <tt>unwind</tt> instruction.</dd>
1127 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1128 <dd>This attribute indicates that the function does not write through any
1129 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1130 arguments) or otherwise modify any state (e.g. memory, control registers,
1131 etc) visible to caller functions. It may dereference pointer arguments
1132 and read state that may be set in the caller. A readonly function always
1133 returns the same value (or unwinds an exception identically) when called
1134 with the same set of arguments and global state. It cannot unwind an
1135 exception by calling the <tt>C++</tt> exception throwing methods, but may
1136 use the <tt>unwind</tt> instruction.</dd>
1138 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1139 <dd>This attribute indicates that the function should emit a stack smashing
1140 protector. It is in the form of a "canary"—a random value placed on
1141 the stack before the local variables that's checked upon return from the
1142 function to see if it has been overwritten. A heuristic is used to
1143 determine if a function needs stack protectors or not.<br>
1145 If a function that has an <tt>ssp</tt> attribute is inlined into a
1146 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1147 function will have an <tt>ssp</tt> attribute.</dd>
1149 <dt><tt><b>sspreq</b></tt></dt>
1150 <dd>This attribute indicates that the function should <em>always</em> emit a
1151 stack smashing protector. This overrides
1152 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1154 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1155 function that doesn't have an <tt>sspreq</tt> attribute or which has
1156 an <tt>ssp</tt> attribute, then the resulting function will have
1157 an <tt>sspreq</tt> attribute.</dd>
1159 <dt><tt><b>noredzone</b></tt></dt>
1160 <dd>This attribute indicates that the code generator should not use a red
1161 zone, even if the target-specific ABI normally permits it.</dd>
1163 <dt><tt><b>noimplicitfloat</b></tt></dt>
1164 <dd>This attributes disables implicit floating point instructions.</dd>
1166 <dt><tt><b>naked</b></tt></dt>
1167 <dd>This attribute disables prologue / epilogue emission for the function.
1168 This can have very system-specific consequences.</dd>
1173 <!-- ======================================================================= -->
1174 <div class="doc_subsection">
1175 <a name="moduleasm">Module-Level Inline Assembly</a>
1178 <div class="doc_text">
1180 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1181 the GCC "file scope inline asm" blocks. These blocks are internally
1182 concatenated by LLVM and treated as a single unit, but may be separated in
1183 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1185 <div class="doc_code">
1187 module asm "inline asm code goes here"
1188 module asm "more can go here"
1192 <p>The strings can contain any character by escaping non-printable characters.
1193 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1196 <p>The inline asm code is simply printed to the machine code .s file when
1197 assembly code is generated.</p>
1201 <!-- ======================================================================= -->
1202 <div class="doc_subsection">
1203 <a name="datalayout">Data Layout</a>
1206 <div class="doc_text">
1208 <p>A module may specify a target specific data layout string that specifies how
1209 data is to be laid out in memory. The syntax for the data layout is
1212 <div class="doc_code">
1214 target datalayout = "<i>layout specification</i>"
1218 <p>The <i>layout specification</i> consists of a list of specifications
1219 separated by the minus sign character ('-'). Each specification starts with
1220 a letter and may include other information after the letter to define some
1221 aspect of the data layout. The specifications accepted are as follows:</p>
1225 <dd>Specifies that the target lays out data in big-endian form. That is, the
1226 bits with the most significance have the lowest address location.</dd>
1229 <dd>Specifies that the target lays out data in little-endian form. That is,
1230 the bits with the least significance have the lowest address
1233 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1234 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1235 <i>preferred</i> alignments. All sizes are in bits. Specifying
1236 the <i>pref</i> alignment is optional. If omitted, the
1237 preceding <tt>:</tt> should be omitted too.</dd>
1239 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1240 <dd>This specifies the alignment for an integer type of a given bit
1241 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1243 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1244 <dd>This specifies the alignment for a vector type of a given bit
1247 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1248 <dd>This specifies the alignment for a floating point type of a given bit
1249 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1252 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1253 <dd>This specifies the alignment for an aggregate type of a given bit
1256 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1257 <dd>This specifies the alignment for a stack object of a given bit
1260 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1261 <dd>This specifies a set of native integer widths for the target CPU
1262 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1263 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1264 this set are considered to support most general arithmetic
1265 operations efficiently.</dd>
1268 <p>When constructing the data layout for a given target, LLVM starts with a
1269 default set of specifications which are then (possibly) overriden by the
1270 specifications in the <tt>datalayout</tt> keyword. The default specifications
1271 are given in this list:</p>
1274 <li><tt>E</tt> - big endian</li>
1275 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1276 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1277 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1278 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1279 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1280 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1281 alignment of 64-bits</li>
1282 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1283 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1284 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1285 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1286 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1287 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1290 <p>When LLVM is determining the alignment for a given type, it uses the
1291 following rules:</p>
1294 <li>If the type sought is an exact match for one of the specifications, that
1295 specification is used.</li>
1297 <li>If no match is found, and the type sought is an integer type, then the
1298 smallest integer type that is larger than the bitwidth of the sought type
1299 is used. If none of the specifications are larger than the bitwidth then
1300 the the largest integer type is used. For example, given the default
1301 specifications above, the i7 type will use the alignment of i8 (next
1302 largest) while both i65 and i256 will use the alignment of i64 (largest
1305 <li>If no match is found, and the type sought is a vector type, then the
1306 largest vector type that is smaller than the sought vector type will be
1307 used as a fall back. This happens because <128 x double> can be
1308 implemented in terms of 64 <2 x double>, for example.</li>
1313 <!-- ======================================================================= -->
1314 <div class="doc_subsection">
1315 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1318 <div class="doc_text">
1320 <p>Any memory access must be done through a pointer value associated
1321 with an address range of the memory access, otherwise the behavior
1322 is undefined. Pointer values are associated with address ranges
1323 according to the following rules:</p>
1326 <li>A pointer value formed from a
1327 <tt><a href="#i_getelementptr">getelementptr</a></tt> instruction
1328 is associated with the addresses associated with the first operand
1329 of the <tt>getelementptr</tt>.</li>
1330 <li>An address of a global variable is associated with the address
1331 range of the variable's storage.</li>
1332 <li>The result value of an allocation instruction is associated with
1333 the address range of the allocated storage.</li>
1334 <li>A null pointer in the default address-space is associated with
1336 <li>A pointer value formed by an
1337 <tt><a href="#i_inttoptr">inttoptr</a></tt> is associated with all
1338 address ranges of all pointer values that contribute (directly or
1339 indirectly) to the computation of the pointer's value.</li>
1340 <li>The result value of a
1341 <tt><a href="#i_bitcast">bitcast</a></tt> is associated with all
1342 addresses associated with the operand of the <tt>bitcast</tt>.</li>
1343 <li>An integer constant other than zero or a pointer value returned
1344 from a function not defined within LLVM may be associated with address
1345 ranges allocated through mechanisms other than those provided by
1346 LLVM. Such ranges shall not overlap with any ranges of addresses
1347 allocated by mechanisms provided by LLVM.</li>
1350 <p>LLVM IR does not associate types with memory. The result type of a
1351 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1352 alignment of the memory from which to load, as well as the
1353 interpretation of the value. The first operand of a
1354 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1355 and alignment of the store.</p>
1357 <p>Consequently, type-based alias analysis, aka TBAA, aka
1358 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1359 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1360 additional information which specialized optimization passes may use
1361 to implement type-based alias analysis.</p>
1365 <!-- *********************************************************************** -->
1366 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1367 <!-- *********************************************************************** -->
1369 <div class="doc_text">
1371 <p>The LLVM type system is one of the most important features of the
1372 intermediate representation. Being typed enables a number of optimizations
1373 to be performed on the intermediate representation directly, without having
1374 to do extra analyses on the side before the transformation. A strong type
1375 system makes it easier to read the generated code and enables novel analyses
1376 and transformations that are not feasible to perform on normal three address
1377 code representations.</p>
1381 <!-- ======================================================================= -->
1382 <div class="doc_subsection"> <a name="t_classifications">Type
1383 Classifications</a> </div>
1385 <div class="doc_text">
1387 <p>The types fall into a few useful classifications:</p>
1389 <table border="1" cellspacing="0" cellpadding="4">
1391 <tr><th>Classification</th><th>Types</th></tr>
1393 <td><a href="#t_integer">integer</a></td>
1394 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1397 <td><a href="#t_floating">floating point</a></td>
1398 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1401 <td><a name="t_firstclass">first class</a></td>
1402 <td><a href="#t_integer">integer</a>,
1403 <a href="#t_floating">floating point</a>,
1404 <a href="#t_pointer">pointer</a>,
1405 <a href="#t_vector">vector</a>,
1406 <a href="#t_struct">structure</a>,
1407 <a href="#t_array">array</a>,
1408 <a href="#t_label">label</a>,
1409 <a href="#t_metadata">metadata</a>.
1413 <td><a href="#t_primitive">primitive</a></td>
1414 <td><a href="#t_label">label</a>,
1415 <a href="#t_void">void</a>,
1416 <a href="#t_floating">floating point</a>,
1417 <a href="#t_metadata">metadata</a>.</td>
1420 <td><a href="#t_derived">derived</a></td>
1421 <td><a href="#t_integer">integer</a>,
1422 <a href="#t_array">array</a>,
1423 <a href="#t_function">function</a>,
1424 <a href="#t_pointer">pointer</a>,
1425 <a href="#t_struct">structure</a>,
1426 <a href="#t_pstruct">packed structure</a>,
1427 <a href="#t_vector">vector</a>,
1428 <a href="#t_opaque">opaque</a>.
1434 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1435 important. Values of these types are the only ones which can be produced by
1440 <!-- ======================================================================= -->
1441 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1443 <div class="doc_text">
1445 <p>The primitive types are the fundamental building blocks of the LLVM
1450 <!-- _______________________________________________________________________ -->
1451 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1453 <div class="doc_text">
1456 <p>The integer type is a very simple type that simply specifies an arbitrary
1457 bit width for the integer type desired. Any bit width from 1 bit to
1458 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1465 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1469 <table class="layout">
1471 <td class="left"><tt>i1</tt></td>
1472 <td class="left">a single-bit integer.</td>
1475 <td class="left"><tt>i32</tt></td>
1476 <td class="left">a 32-bit integer.</td>
1479 <td class="left"><tt>i1942652</tt></td>
1480 <td class="left">a really big integer of over 1 million bits.</td>
1486 <!-- _______________________________________________________________________ -->
1487 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1489 <div class="doc_text">
1493 <tr><th>Type</th><th>Description</th></tr>
1494 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1495 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1496 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1497 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1498 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1504 <!-- _______________________________________________________________________ -->
1505 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1507 <div class="doc_text">
1510 <p>The void type does not represent any value and has no size.</p>
1519 <!-- _______________________________________________________________________ -->
1520 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1522 <div class="doc_text">
1525 <p>The label type represents code labels.</p>
1534 <!-- _______________________________________________________________________ -->
1535 <div class="doc_subsubsection"> <a name="t_metadata">Metadata Type</a> </div>
1537 <div class="doc_text">
1540 <p>The metadata type represents embedded metadata. No derived types may be
1541 created from metadata except for <a href="#t_function">function</a>
1552 <!-- ======================================================================= -->
1553 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1555 <div class="doc_text">
1557 <p>The real power in LLVM comes from the derived types in the system. This is
1558 what allows a programmer to represent arrays, functions, pointers, and other
1559 useful types. Each of these types contain one or more element types which
1560 may be a primitive type, or another derived type. For example, it is
1561 possible to have a two dimensional array, using an array as the element type
1562 of another array.</p>
1566 <!-- _______________________________________________________________________ -->
1567 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1569 <div class="doc_text">
1572 <p>The array type is a very simple derived type that arranges elements
1573 sequentially in memory. The array type requires a size (number of elements)
1574 and an underlying data type.</p>
1578 [<# elements> x <elementtype>]
1581 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1582 be any type with a size.</p>
1585 <table class="layout">
1587 <td class="left"><tt>[40 x i32]</tt></td>
1588 <td class="left">Array of 40 32-bit integer values.</td>
1591 <td class="left"><tt>[41 x i32]</tt></td>
1592 <td class="left">Array of 41 32-bit integer values.</td>
1595 <td class="left"><tt>[4 x i8]</tt></td>
1596 <td class="left">Array of 4 8-bit integer values.</td>
1599 <p>Here are some examples of multidimensional arrays:</p>
1600 <table class="layout">
1602 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1603 <td class="left">3x4 array of 32-bit integer values.</td>
1606 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1607 <td class="left">12x10 array of single precision floating point values.</td>
1610 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1611 <td class="left">2x3x4 array of 16-bit integer values.</td>
1615 <p>There is no restriction on indexing beyond the end of the array implied by
1616 a static type (though there are restrictions on indexing beyond the bounds
1617 of an allocated object in some cases). This means that single-dimension
1618 'variable sized array' addressing can be implemented in LLVM with a zero
1619 length array type. An implementation of 'pascal style arrays' in LLVM could
1620 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1624 <!-- _______________________________________________________________________ -->
1625 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1627 <div class="doc_text">
1630 <p>The function type can be thought of as a function signature. It consists of
1631 a return type and a list of formal parameter types. The return type of a
1632 function type is a scalar type, a void type, or a struct type. If the return
1633 type is a struct type then all struct elements must be of first class types,
1634 and the struct must have at least one element.</p>
1638 <returntype> (<parameter list>)
1641 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1642 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1643 which indicates that the function takes a variable number of arguments.
1644 Variable argument functions can access their arguments with
1645 the <a href="#int_varargs">variable argument handling intrinsic</a>
1646 functions. '<tt><returntype></tt>' is a any type except
1647 <a href="#t_label">label</a>.</p>
1650 <table class="layout">
1652 <td class="left"><tt>i32 (i32)</tt></td>
1653 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1655 </tr><tr class="layout">
1656 <td class="left"><tt>float (i16 signext, i32 *) *
1658 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1659 an <tt>i16</tt> that should be sign extended and a
1660 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1663 </tr><tr class="layout">
1664 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1665 <td class="left">A vararg function that takes at least one
1666 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1667 which returns an integer. This is the signature for <tt>printf</tt> in
1670 </tr><tr class="layout">
1671 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1672 <td class="left">A function taking an <tt>i32</tt>, returning a
1673 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
1680 <!-- _______________________________________________________________________ -->
1681 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1683 <div class="doc_text">
1686 <p>The structure type is used to represent a collection of data members together
1687 in memory. The packing of the field types is defined to match the ABI of the
1688 underlying processor. The elements of a structure may be any type that has a
1691 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
1692 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
1693 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
1694 Structures in registers are accessed using the
1695 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
1696 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
1699 { <type list> }
1703 <table class="layout">
1705 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1706 <td class="left">A triple of three <tt>i32</tt> values</td>
1707 </tr><tr class="layout">
1708 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1709 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1710 second element is a <a href="#t_pointer">pointer</a> to a
1711 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1712 an <tt>i32</tt>.</td>
1718 <!-- _______________________________________________________________________ -->
1719 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1722 <div class="doc_text">
1725 <p>The packed structure type is used to represent a collection of data members
1726 together in memory. There is no padding between fields. Further, the
1727 alignment of a packed structure is 1 byte. The elements of a packed
1728 structure may be any type that has a size.</p>
1730 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1731 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1732 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1736 < { <type list> } >
1740 <table class="layout">
1742 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1743 <td class="left">A triple of three <tt>i32</tt> values</td>
1744 </tr><tr class="layout">
1746 <tt>< { float, i32 (i32)* } ></tt></td>
1747 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1748 second element is a <a href="#t_pointer">pointer</a> to a
1749 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1750 an <tt>i32</tt>.</td>
1756 <!-- _______________________________________________________________________ -->
1757 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1759 <div class="doc_text">
1762 <p>As in many languages, the pointer type represents a pointer or reference to
1763 another object, which must live in memory. Pointer types may have an optional
1764 address space attribute defining the target-specific numbered address space
1765 where the pointed-to object resides. The default address space is zero.</p>
1767 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
1768 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
1776 <table class="layout">
1778 <td class="left"><tt>[4 x i32]*</tt></td>
1779 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1780 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1783 <td class="left"><tt>i32 (i32 *) *</tt></td>
1784 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1785 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1789 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1790 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1791 that resides in address space #5.</td>
1797 <!-- _______________________________________________________________________ -->
1798 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1800 <div class="doc_text">
1803 <p>A vector type is a simple derived type that represents a vector of elements.
1804 Vector types are used when multiple primitive data are operated in parallel
1805 using a single instruction (SIMD). A vector type requires a size (number of
1806 elements) and an underlying primitive data type. Vector types are considered
1807 <a href="#t_firstclass">first class</a>.</p>
1811 < <# elements> x <elementtype> >
1814 <p>The number of elements is a constant integer value; elementtype may be any
1815 integer or floating point type.</p>
1818 <table class="layout">
1820 <td class="left"><tt><4 x i32></tt></td>
1821 <td class="left">Vector of 4 32-bit integer values.</td>
1824 <td class="left"><tt><8 x float></tt></td>
1825 <td class="left">Vector of 8 32-bit floating-point values.</td>
1828 <td class="left"><tt><2 x i64></tt></td>
1829 <td class="left">Vector of 2 64-bit integer values.</td>
1835 <!-- _______________________________________________________________________ -->
1836 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1837 <div class="doc_text">
1840 <p>Opaque types are used to represent unknown types in the system. This
1841 corresponds (for example) to the C notion of a forward declared structure
1842 type. In LLVM, opaque types can eventually be resolved to any type (not just
1843 a structure type).</p>
1851 <table class="layout">
1853 <td class="left"><tt>opaque</tt></td>
1854 <td class="left">An opaque type.</td>
1860 <!-- ======================================================================= -->
1861 <div class="doc_subsection">
1862 <a name="t_uprefs">Type Up-references</a>
1865 <div class="doc_text">
1868 <p>An "up reference" allows you to refer to a lexically enclosing type without
1869 requiring it to have a name. For instance, a structure declaration may
1870 contain a pointer to any of the types it is lexically a member of. Example
1871 of up references (with their equivalent as named type declarations)
1875 { \2 * } %x = type { %x* }
1876 { \2 }* %y = type { %y }*
1880 <p>An up reference is needed by the asmprinter for printing out cyclic types
1881 when there is no declared name for a type in the cycle. Because the
1882 asmprinter does not want to print out an infinite type string, it needs a
1883 syntax to handle recursive types that have no names (all names are optional
1891 <p>The level is the count of the lexical type that is being referred to.</p>
1894 <table class="layout">
1896 <td class="left"><tt>\1*</tt></td>
1897 <td class="left">Self-referential pointer.</td>
1900 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1901 <td class="left">Recursive structure where the upref refers to the out-most
1908 <!-- *********************************************************************** -->
1909 <div class="doc_section"> <a name="constants">Constants</a> </div>
1910 <!-- *********************************************************************** -->
1912 <div class="doc_text">
1914 <p>LLVM has several different basic types of constants. This section describes
1915 them all and their syntax.</p>
1919 <!-- ======================================================================= -->
1920 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1922 <div class="doc_text">
1925 <dt><b>Boolean constants</b></dt>
1926 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1927 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
1929 <dt><b>Integer constants</b></dt>
1930 <dd>Standard integers (such as '4') are constants of
1931 the <a href="#t_integer">integer</a> type. Negative numbers may be used
1932 with integer types.</dd>
1934 <dt><b>Floating point constants</b></dt>
1935 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1936 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1937 notation (see below). The assembler requires the exact decimal value of a
1938 floating-point constant. For example, the assembler accepts 1.25 but
1939 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1940 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1942 <dt><b>Null pointer constants</b></dt>
1943 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1944 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1947 <p>The one non-intuitive notation for constants is the hexadecimal form of
1948 floating point constants. For example, the form '<tt>double
1949 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
1950 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
1951 constants are required (and the only time that they are generated by the
1952 disassembler) is when a floating point constant must be emitted but it cannot
1953 be represented as a decimal floating point number in a reasonable number of
1954 digits. For example, NaN's, infinities, and other special values are
1955 represented in their IEEE hexadecimal format so that assembly and disassembly
1956 do not cause any bits to change in the constants.</p>
1958 <p>When using the hexadecimal form, constants of types float and double are
1959 represented using the 16-digit form shown above (which matches the IEEE754
1960 representation for double); float values must, however, be exactly
1961 representable as IEE754 single precision. Hexadecimal format is always used
1962 for long double, and there are three forms of long double. The 80-bit format
1963 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
1964 The 128-bit format used by PowerPC (two adjacent doubles) is represented
1965 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
1966 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
1967 currently supported target uses this format. Long doubles will only work if
1968 they match the long double format on your target. All hexadecimal formats
1969 are big-endian (sign bit at the left).</p>
1973 <!-- ======================================================================= -->
1974 <div class="doc_subsection">
1975 <a name="aggregateconstants"></a> <!-- old anchor -->
1976 <a name="complexconstants">Complex Constants</a>
1979 <div class="doc_text">
1981 <p>Complex constants are a (potentially recursive) combination of simple
1982 constants and smaller complex constants.</p>
1985 <dt><b>Structure constants</b></dt>
1986 <dd>Structure constants are represented with notation similar to structure
1987 type definitions (a comma separated list of elements, surrounded by braces
1988 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1989 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
1990 Structure constants must have <a href="#t_struct">structure type</a>, and
1991 the number and types of elements must match those specified by the
1994 <dt><b>Array constants</b></dt>
1995 <dd>Array constants are represented with notation similar to array type
1996 definitions (a comma separated list of elements, surrounded by square
1997 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
1998 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
1999 the number and types of elements must match those specified by the
2002 <dt><b>Vector constants</b></dt>
2003 <dd>Vector constants are represented with notation similar to vector type
2004 definitions (a comma separated list of elements, surrounded by
2005 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2006 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2007 have <a href="#t_vector">vector type</a>, and the number and types of
2008 elements must match those specified by the type.</dd>
2010 <dt><b>Zero initialization</b></dt>
2011 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2012 value to zero of <em>any</em> type, including scalar and aggregate types.
2013 This is often used to avoid having to print large zero initializers
2014 (e.g. for large arrays) and is always exactly equivalent to using explicit
2015 zero initializers.</dd>
2017 <dt><b>Metadata node</b></dt>
2018 <dd>A metadata node is a structure-like constant with
2019 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2020 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2021 be interpreted as part of the instruction stream, metadata is a place to
2022 attach additional information such as debug info.</dd>
2027 <!-- ======================================================================= -->
2028 <div class="doc_subsection">
2029 <a name="globalconstants">Global Variable and Function Addresses</a>
2032 <div class="doc_text">
2034 <p>The addresses of <a href="#globalvars">global variables</a>
2035 and <a href="#functionstructure">functions</a> are always implicitly valid
2036 (link-time) constants. These constants are explicitly referenced when
2037 the <a href="#identifiers">identifier for the global</a> is used and always
2038 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2039 legal LLVM file:</p>
2041 <div class="doc_code">
2045 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2051 <!-- ======================================================================= -->
2052 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
2053 <div class="doc_text">
2055 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2056 indicates that the user of the value may receive an unspecified bit-pattern.
2057 Undefined values may be of any type (other than label or void) and be used
2058 anywhere a constant is permitted.</p>
2060 <p>Undefined values are useful because they indicate to the compiler that the
2061 program is well defined no matter what value is used. This gives the
2062 compiler more freedom to optimize. Here are some examples of (potentially
2063 surprising) transformations that are valid (in pseudo IR):</p>
2066 <div class="doc_code">
2078 <p>This is safe because all of the output bits are affected by the undef bits.
2079 Any output bit can have a zero or one depending on the input bits.</p>
2081 <div class="doc_code">
2094 <p>These logical operations have bits that are not always affected by the input.
2095 For example, if "%X" has a zero bit, then the output of the 'and' operation will
2096 always be a zero, no matter what the corresponding bit from the undef is. As
2097 such, it is unsafe to optimize or assume that the result of the and is undef.
2098 However, it is safe to assume that all bits of the undef could be 0, and
2099 optimize the and to 0. Likewise, it is safe to assume that all the bits of
2100 the undef operand to the or could be set, allowing the or to be folded to
2103 <div class="doc_code">
2105 %A = select undef, %X, %Y
2106 %B = select undef, 42, %Y
2107 %C = select %X, %Y, undef
2119 <p>This set of examples show that undefined select (and conditional branch)
2120 conditions can go "either way" but they have to come from one of the two
2121 operands. In the %A example, if %X and %Y were both known to have a clear low
2122 bit, then %A would have to have a cleared low bit. However, in the %C example,
2123 the optimizer is allowed to assume that the undef operand could be the same as
2124 %Y, allowing the whole select to be eliminated.</p>
2127 <div class="doc_code">
2129 %A = xor undef, undef
2148 <p>This example points out that two undef operands are not necessarily the same.
2149 This can be surprising to people (and also matches C semantics) where they
2150 assume that "X^X" is always zero, even if X is undef. This isn't true for a
2151 number of reasons, but the short answer is that an undef "variable" can
2152 arbitrarily change its value over its "live range". This is true because the
2153 "variable" doesn't actually <em>have a live range</em>. Instead, the value is
2154 logically read from arbitrary registers that happen to be around when needed,
2155 so the value is not necessarily consistent over time. In fact, %A and %C need
2156 to have the same semantics or the core LLVM "replace all uses with" concept
2159 <div class="doc_code">
2169 <p>These examples show the crucial difference between an <em>undefined
2170 value</em> and <em>undefined behavior</em>. An undefined value (like undef) is
2171 allowed to have an arbitrary bit-pattern. This means that the %A operation
2172 can be constant folded to undef because the undef could be an SNaN, and fdiv is
2173 not (currently) defined on SNaN's. However, in the second example, we can make
2174 a more aggressive assumption: because the undef is allowed to be an arbitrary
2175 value, we are allowed to assume that it could be zero. Since a divide by zero
2176 has <em>undefined behavior</em>, we are allowed to assume that the operation
2177 does not execute at all. This allows us to delete the divide and all code after
2178 it: since the undefined operation "can't happen", the optimizer can assume that
2179 it occurs in dead code.
2182 <div class="doc_code">
2184 a: store undef -> %X
2185 b: store %X -> undef
2192 <p>These examples reiterate the fdiv example: a store "of" an undefined value
2193 can be assumed to not have any effect: we can assume that the value is
2194 overwritten with bits that happen to match what was already there. However, a
2195 store "to" an undefined location could clobber arbitrary memory, therefore, it
2196 has undefined behavior.</p>
2200 <!-- ======================================================================= -->
2201 <div class="doc_subsection"><a name="blockaddress">Addresses of Basic
2203 <div class="doc_text">
2205 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2207 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2208 basic block in the specified function, and always has an i8* type. Taking
2209 the address of the entry block is illegal.</p>
2211 <p>This value only has defined behavior when used as an operand to the
2212 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction or for comparisons
2213 against null. Pointer equality tests between labels addresses is undefined
2214 behavior - though, again, comparison against null is ok, and no label is
2215 equal to the null pointer. This may also be passed around as an opaque
2216 pointer sized value as long as the bits are not inspected. This allows
2217 <tt>ptrtoint</tt> and arithmetic to be performed on these values so long as
2218 the original value is reconstituted before the <tt>indirectbr</tt>.</p>
2220 <p>Finally, some targets may provide defined semantics when
2221 using the value as the operand to an inline assembly, but that is target
2228 <!-- ======================================================================= -->
2229 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
2232 <div class="doc_text">
2234 <p>Constant expressions are used to allow expressions involving other constants
2235 to be used as constants. Constant expressions may be of
2236 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2237 operation that does not have side effects (e.g. load and call are not
2238 supported). The following is the syntax for constant expressions:</p>
2241 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
2242 <dd>Truncate a constant to another type. The bit size of CST must be larger
2243 than the bit size of TYPE. Both types must be integers.</dd>
2245 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
2246 <dd>Zero extend a constant to another type. The bit size of CST must be
2247 smaller or equal to the bit size of TYPE. Both types must be
2250 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
2251 <dd>Sign extend a constant to another type. The bit size of CST must be
2252 smaller or equal to the bit size of TYPE. Both types must be
2255 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
2256 <dd>Truncate a floating point constant to another floating point type. The
2257 size of CST must be larger than the size of TYPE. Both types must be
2258 floating point.</dd>
2260 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
2261 <dd>Floating point extend a constant to another type. The size of CST must be
2262 smaller or equal to the size of TYPE. Both types must be floating
2265 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
2266 <dd>Convert a floating point constant to the corresponding unsigned integer
2267 constant. TYPE must be a scalar or vector integer type. CST must be of
2268 scalar or vector floating point type. Both CST and TYPE must be scalars,
2269 or vectors of the same number of elements. If the value won't fit in the
2270 integer type, the results are undefined.</dd>
2272 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
2273 <dd>Convert a floating point constant to the corresponding signed integer
2274 constant. TYPE must be a scalar or vector integer type. CST must be of
2275 scalar or vector floating point type. Both CST and TYPE must be scalars,
2276 or vectors of the same number of elements. If the value won't fit in the
2277 integer type, the results are undefined.</dd>
2279 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
2280 <dd>Convert an unsigned integer constant to the corresponding floating point
2281 constant. TYPE must be a scalar or vector floating point type. CST must be
2282 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2283 vectors of the same number of elements. If the value won't fit in the
2284 floating point type, the results are undefined.</dd>
2286 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
2287 <dd>Convert a signed integer constant to the corresponding floating point
2288 constant. TYPE must be a scalar or vector floating point type. CST must be
2289 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2290 vectors of the same number of elements. If the value won't fit in the
2291 floating point type, the results are undefined.</dd>
2293 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
2294 <dd>Convert a pointer typed constant to the corresponding integer constant
2295 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2296 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2297 make it fit in <tt>TYPE</tt>.</dd>
2299 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
2300 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2301 type. CST must be of integer type. The CST value is zero extended,
2302 truncated, or unchanged to make it fit in a pointer size. This one is
2303 <i>really</i> dangerous!</dd>
2305 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
2306 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2307 are the same as those for the <a href="#i_bitcast">bitcast
2308 instruction</a>.</dd>
2310 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2311 <dt><b><tt>getelementptr inbounds ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2312 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2313 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2314 instruction, the index list may have zero or more indexes, which are
2315 required to make sense for the type of "CSTPTR".</dd>
2317 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
2318 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2320 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
2321 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2323 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
2324 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2326 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
2327 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2330 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
2331 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2334 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
2335 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2338 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
2339 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2340 be any of the <a href="#binaryops">binary</a>
2341 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2342 on operands are the same as those for the corresponding instruction
2343 (e.g. no bitwise operations on floating point values are allowed).</dd>
2348 <!-- *********************************************************************** -->
2349 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2350 <!-- *********************************************************************** -->
2352 <!-- ======================================================================= -->
2353 <div class="doc_subsection">
2354 <a name="inlineasm">Inline Assembler Expressions</a>
2357 <div class="doc_text">
2359 <p>LLVM supports inline assembler expressions (as opposed
2360 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2361 a special value. This value represents the inline assembler as a string
2362 (containing the instructions to emit), a list of operand constraints (stored
2363 as a string), a flag that indicates whether or not the inline asm
2364 expression has side effects, and a flag indicating whether the function
2365 containing the asm needs to align its stack conservatively. An example
2366 inline assembler expression is:</p>
2368 <div class="doc_code">
2370 i32 (i32) asm "bswap $0", "=r,r"
2374 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2375 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2378 <div class="doc_code">
2380 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2384 <p>Inline asms with side effects not visible in the constraint list must be
2385 marked as having side effects. This is done through the use of the
2386 '<tt>sideeffect</tt>' keyword, like so:</p>
2388 <div class="doc_code">
2390 call void asm sideeffect "eieio", ""()
2394 <p>In some cases inline asms will contain code that will not work unless the
2395 stack is aligned in some way, such as calls or SSE instructions on x86,
2396 yet will not contain code that does that alignment within the asm.
2397 The compiler should make conservative assumptions about what the asm might
2398 contain and should generate its usual stack alignment code in the prologue
2399 if the '<tt>alignstack</tt>' keyword is present:</p>
2401 <div class="doc_code">
2403 call void asm alignstack "eieio", ""()
2407 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2410 <p>TODO: The format of the asm and constraints string still need to be
2411 documented here. Constraints on what can be done (e.g. duplication, moving,
2412 etc need to be documented). This is probably best done by reference to
2413 another document that covers inline asm from a holistic perspective.</p>
2417 <!-- ======================================================================= -->
2418 <div class="doc_subsection"><a name="metadata">Metadata Nodes and Metadata
2422 <div class="doc_text">
2424 <p>LLVM IR allows metadata to be attached to instructions in the program that
2425 can convey extra information about the code to the optimizers and code
2426 generator. One example application of metadata is source-level debug
2427 information. There are two metadata primitives: strings and nodes. All
2428 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2429 preceding exclamation point ('<tt>!</tt>').</p>
2431 <p>A metadata string is a string surrounded by double quotes. It can contain
2432 any character by escaping non-printable characters with "\xx" where "xx" is
2433 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2435 <p>Metadata nodes are represented with notation similar to structure constants
2436 (a comma separated list of elements, surrounded by braces and preceded by an
2437 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2438 10}</tt>". Metadata nodes can have any values as their operand.</p>
2440 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2441 metadata nodes, which can be looked up in the module symbol table. For
2442 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2447 <!-- *********************************************************************** -->
2448 <div class="doc_section">
2449 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2451 <!-- *********************************************************************** -->
2453 <p>LLVM has a number of "magic" global variables that contain data that affect
2454 code generation or other IR semantics. These are documented here. All globals
2455 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2456 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2459 <!-- ======================================================================= -->
2460 <div class="doc_subsection">
2461 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2464 <div class="doc_text">
2466 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2467 href="#linkage_appending">appending linkage</a>. This array contains a list of
2468 pointers to global variables and functions which may optionally have a pointer
2469 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2475 @llvm.used = appending global [2 x i8*] [
2477 i8* bitcast (i32* @Y to i8*)
2478 ], section "llvm.metadata"
2481 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2482 compiler, assembler, and linker are required to treat the symbol as if there is
2483 a reference to the global that it cannot see. For example, if a variable has
2484 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2485 list, it cannot be deleted. This is commonly used to represent references from
2486 inline asms and other things the compiler cannot "see", and corresponds to
2487 "attribute((used))" in GNU C.</p>
2489 <p>On some targets, the code generator must emit a directive to the assembler or
2490 object file to prevent the assembler and linker from molesting the symbol.</p>
2494 <!-- ======================================================================= -->
2495 <div class="doc_subsection">
2496 <a name="intg_compiler_used">The '<tt>llvm.compiler.used</tt>' Global Variable</a>
2499 <div class="doc_text">
2501 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2502 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2503 touching the symbol. On targets that support it, this allows an intelligent
2504 linker to optimize references to the symbol without being impeded as it would be
2505 by <tt>@llvm.used</tt>.</p>
2507 <p>This is a rare construct that should only be used in rare circumstances, and
2508 should not be exposed to source languages.</p>
2512 <!-- ======================================================================= -->
2513 <div class="doc_subsection">
2514 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2517 <div class="doc_text">
2519 <p>TODO: Describe this.</p>
2523 <!-- ======================================================================= -->
2524 <div class="doc_subsection">
2525 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
2528 <div class="doc_text">
2530 <p>TODO: Describe this.</p>
2535 <!-- *********************************************************************** -->
2536 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2537 <!-- *********************************************************************** -->
2539 <div class="doc_text">
2541 <p>The LLVM instruction set consists of several different classifications of
2542 instructions: <a href="#terminators">terminator
2543 instructions</a>, <a href="#binaryops">binary instructions</a>,
2544 <a href="#bitwiseops">bitwise binary instructions</a>,
2545 <a href="#memoryops">memory instructions</a>, and
2546 <a href="#otherops">other instructions</a>.</p>
2550 <!-- ======================================================================= -->
2551 <div class="doc_subsection"> <a name="terminators">Terminator
2552 Instructions</a> </div>
2554 <div class="doc_text">
2556 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
2557 in a program ends with a "Terminator" instruction, which indicates which
2558 block should be executed after the current block is finished. These
2559 terminator instructions typically yield a '<tt>void</tt>' value: they produce
2560 control flow, not values (the one exception being the
2561 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2563 <p>There are six different terminator instructions: the
2564 '<a href="#i_ret"><tt>ret</tt></a>' instruction, the
2565 '<a href="#i_br"><tt>br</tt></a>' instruction, the
2566 '<a href="#i_switch"><tt>switch</tt></a>' instruction, the
2567 '<a href="#i_indirectbr">'<tt>indirectbr</tt></a>' Instruction, the
2568 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the
2569 '<a href="#i_unwind"><tt>unwind</tt></a>' instruction, and the
2570 '<a href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2574 <!-- _______________________________________________________________________ -->
2575 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2576 Instruction</a> </div>
2578 <div class="doc_text">
2582 ret <type> <value> <i>; Return a value from a non-void function</i>
2583 ret void <i>; Return from void function</i>
2587 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
2588 a value) from a function back to the caller.</p>
2590 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
2591 value and then causes control flow, and one that just causes control flow to
2595 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
2596 return value. The type of the return value must be a
2597 '<a href="#t_firstclass">first class</a>' type.</p>
2599 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
2600 non-void return type and contains a '<tt>ret</tt>' instruction with no return
2601 value or a return value with a type that does not match its type, or if it
2602 has a void return type and contains a '<tt>ret</tt>' instruction with a
2606 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
2607 the calling function's context. If the caller is a
2608 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
2609 instruction after the call. If the caller was an
2610 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
2611 the beginning of the "normal" destination block. If the instruction returns
2612 a value, that value shall set the call or invoke instruction's return
2617 ret i32 5 <i>; Return an integer value of 5</i>
2618 ret void <i>; Return from a void function</i>
2619 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
2623 <!-- _______________________________________________________________________ -->
2624 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2626 <div class="doc_text">
2630 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2634 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
2635 different basic block in the current function. There are two forms of this
2636 instruction, corresponding to a conditional branch and an unconditional
2640 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
2641 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
2642 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
2646 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2647 argument is evaluated. If the value is <tt>true</tt>, control flows to the
2648 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2649 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2654 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
2655 br i1 %cond, label %IfEqual, label %IfUnequal
2657 <a href="#i_ret">ret</a> i32 1
2659 <a href="#i_ret">ret</a> i32 0
2664 <!-- _______________________________________________________________________ -->
2665 <div class="doc_subsubsection">
2666 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2669 <div class="doc_text">
2673 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2677 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2678 several different places. It is a generalization of the '<tt>br</tt>'
2679 instruction, allowing a branch to occur to one of many possible
2683 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2684 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
2685 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
2686 The table is not allowed to contain duplicate constant entries.</p>
2689 <p>The <tt>switch</tt> instruction specifies a table of values and
2690 destinations. When the '<tt>switch</tt>' instruction is executed, this table
2691 is searched for the given value. If the value is found, control flow is
2692 transferred to the corresponding destination; otherwise, control flow is
2693 transferred to the default destination.</p>
2695 <h5>Implementation:</h5>
2696 <p>Depending on properties of the target machine and the particular
2697 <tt>switch</tt> instruction, this instruction may be code generated in
2698 different ways. For example, it could be generated as a series of chained
2699 conditional branches or with a lookup table.</p>
2703 <i>; Emulate a conditional br instruction</i>
2704 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2705 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2707 <i>; Emulate an unconditional br instruction</i>
2708 switch i32 0, label %dest [ ]
2710 <i>; Implement a jump table:</i>
2711 switch i32 %val, label %otherwise [ i32 0, label %onzero
2713 i32 2, label %ontwo ]
2719 <!-- _______________________________________________________________________ -->
2720 <div class="doc_subsubsection">
2721 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
2724 <div class="doc_text">
2728 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
2733 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
2734 within the current function, whose address is specified by
2735 "<tt>address</tt>". Address must be derived from a <a
2736 href="#blockaddress">blockaddress</a> constant.</p>
2740 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
2741 rest of the arguments indicate the full set of possible destinations that the
2742 address may point to. Blocks are allowed to occur multiple times in the
2743 destination list, though this isn't particularly useful.</p>
2745 <p>This destination list is required so that dataflow analysis has an accurate
2746 understanding of the CFG.</p>
2750 <p>Control transfers to the block specified in the address argument. All
2751 possible destination blocks must be listed in the label list, otherwise this
2752 instruction has undefined behavior. This implies that jumps to labels
2753 defined in other functions have undefined behavior as well.</p>
2755 <h5>Implementation:</h5>
2757 <p>This is typically implemented with a jump through a register.</p>
2761 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
2767 <!-- _______________________________________________________________________ -->
2768 <div class="doc_subsubsection">
2769 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2772 <div class="doc_text">
2776 <result> = invoke [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] <ptr to function ty> <function ptr val>(<function args>) [<a href="#fnattrs">fn attrs</a>]
2777 to label <normal label> unwind label <exception label>
2781 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2782 function, with the possibility of control flow transfer to either the
2783 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
2784 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
2785 control flow will return to the "normal" label. If the callee (or any
2786 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
2787 instruction, control is interrupted and continued at the dynamically nearest
2788 "exception" label.</p>
2791 <p>This instruction requires several arguments:</p>
2794 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
2795 convention</a> the call should use. If none is specified, the call
2796 defaults to using C calling conventions.</li>
2798 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2799 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
2800 '<tt>inreg</tt>' attributes are valid here.</li>
2802 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2803 function value being invoked. In most cases, this is a direct function
2804 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
2805 off an arbitrary pointer to function value.</li>
2807 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2808 function to be invoked. </li>
2810 <li>'<tt>function args</tt>': argument list whose types match the function
2811 signature argument types. If the function signature indicates the
2812 function accepts a variable number of arguments, the extra arguments can
2815 <li>'<tt>normal label</tt>': the label reached when the called function
2816 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2818 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2819 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2821 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2822 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2823 '<tt>readnone</tt>' attributes are valid here.</li>
2827 <p>This instruction is designed to operate as a standard
2828 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
2829 primary difference is that it establishes an association with a label, which
2830 is used by the runtime library to unwind the stack.</p>
2832 <p>This instruction is used in languages with destructors to ensure that proper
2833 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2834 exception. Additionally, this is important for implementation of
2835 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2837 <p>For the purposes of the SSA form, the definition of the value returned by the
2838 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
2839 block to the "normal" label. If the callee unwinds then no return value is
2842 <p>Note that the code generator does not yet completely support unwind, and
2843 that the invoke/unwind semantics are likely to change in future versions.</p>
2847 %retval = invoke i32 @Test(i32 15) to label %Continue
2848 unwind label %TestCleanup <i>; {i32}:retval set</i>
2849 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2850 unwind label %TestCleanup <i>; {i32}:retval set</i>
2855 <!-- _______________________________________________________________________ -->
2857 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2858 Instruction</a> </div>
2860 <div class="doc_text">
2868 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2869 at the first callee in the dynamic call stack which used
2870 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
2871 This is primarily used to implement exception handling.</p>
2874 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2875 immediately halt. The dynamic call stack is then searched for the
2876 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
2877 Once found, execution continues at the "exceptional" destination block
2878 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
2879 instruction in the dynamic call chain, undefined behavior results.</p>
2881 <p>Note that the code generator does not yet completely support unwind, and
2882 that the invoke/unwind semantics are likely to change in future versions.</p>
2886 <!-- _______________________________________________________________________ -->
2888 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2889 Instruction</a> </div>
2891 <div class="doc_text">
2899 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2900 instruction is used to inform the optimizer that a particular portion of the
2901 code is not reachable. This can be used to indicate that the code after a
2902 no-return function cannot be reached, and other facts.</p>
2905 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2909 <!-- ======================================================================= -->
2910 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2912 <div class="doc_text">
2914 <p>Binary operators are used to do most of the computation in a program. They
2915 require two operands of the same type, execute an operation on them, and
2916 produce a single value. The operands might represent multiple data, as is
2917 the case with the <a href="#t_vector">vector</a> data type. The result value
2918 has the same type as its operands.</p>
2920 <p>There are several different binary operators:</p>
2924 <!-- _______________________________________________________________________ -->
2925 <div class="doc_subsubsection">
2926 <a name="i_add">'<tt>add</tt>' Instruction</a>
2929 <div class="doc_text">
2933 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2934 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2935 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2936 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2940 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2943 <p>The two arguments to the '<tt>add</tt>' instruction must
2944 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
2945 integer values. Both arguments must have identical types.</p>
2948 <p>The value produced is the integer sum of the two operands.</p>
2950 <p>If the sum has unsigned overflow, the result returned is the mathematical
2951 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
2953 <p>Because LLVM integers use a two's complement representation, this instruction
2954 is appropriate for both signed and unsigned integers.</p>
2956 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
2957 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
2958 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
2959 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
2963 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2968 <!-- _______________________________________________________________________ -->
2969 <div class="doc_subsubsection">
2970 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
2973 <div class="doc_text">
2977 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2981 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
2984 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
2985 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2986 floating point values. Both arguments must have identical types.</p>
2989 <p>The value produced is the floating point sum of the two operands.</p>
2993 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
2998 <!-- _______________________________________________________________________ -->
2999 <div class="doc_subsubsection">
3000 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3003 <div class="doc_text">
3007 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3008 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3009 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3010 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3014 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3017 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3018 '<tt>neg</tt>' instruction present in most other intermediate
3019 representations.</p>
3022 <p>The two arguments to the '<tt>sub</tt>' instruction must
3023 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3024 integer values. Both arguments must have identical types.</p>
3027 <p>The value produced is the integer difference of the two operands.</p>
3029 <p>If the difference has unsigned overflow, the result returned is the
3030 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3033 <p>Because LLVM integers use a two's complement representation, this instruction
3034 is appropriate for both signed and unsigned integers.</p>
3036 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3037 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3038 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3039 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
3043 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3044 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3049 <!-- _______________________________________________________________________ -->
3050 <div class="doc_subsubsection">
3051 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3054 <div class="doc_text">
3058 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3062 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3065 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3066 '<tt>fneg</tt>' instruction present in most other intermediate
3067 representations.</p>
3070 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3071 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3072 floating point values. Both arguments must have identical types.</p>
3075 <p>The value produced is the floating point difference of the two operands.</p>
3079 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3080 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3085 <!-- _______________________________________________________________________ -->
3086 <div class="doc_subsubsection">
3087 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3090 <div class="doc_text">
3094 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3095 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3096 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3097 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3101 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3104 <p>The two arguments to the '<tt>mul</tt>' instruction must
3105 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3106 integer values. Both arguments must have identical types.</p>
3109 <p>The value produced is the integer product of the two operands.</p>
3111 <p>If the result of the multiplication has unsigned overflow, the result
3112 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3113 width of the result.</p>
3115 <p>Because LLVM integers use a two's complement representation, and the result
3116 is the same width as the operands, this instruction returns the correct
3117 result for both signed and unsigned integers. If a full product
3118 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3119 be sign-extended or zero-extended as appropriate to the width of the full
3122 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3123 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3124 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3125 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
3129 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3134 <!-- _______________________________________________________________________ -->
3135 <div class="doc_subsubsection">
3136 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3139 <div class="doc_text">
3143 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3147 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3150 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3151 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3152 floating point values. Both arguments must have identical types.</p>
3155 <p>The value produced is the floating point product of the two operands.</p>
3159 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3164 <!-- _______________________________________________________________________ -->
3165 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
3168 <div class="doc_text">
3172 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3176 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3179 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3180 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3181 values. Both arguments must have identical types.</p>
3184 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3186 <p>Note that unsigned integer division and signed integer division are distinct
3187 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3189 <p>Division by zero leads to undefined behavior.</p>
3193 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3198 <!-- _______________________________________________________________________ -->
3199 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
3202 <div class="doc_text">
3206 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3207 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3211 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3214 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3215 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3216 values. Both arguments must have identical types.</p>
3219 <p>The value produced is the signed integer quotient of the two operands rounded
3222 <p>Note that signed integer division and unsigned integer division are distinct
3223 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3225 <p>Division by zero leads to undefined behavior. Overflow also leads to
3226 undefined behavior; this is a rare case, but can occur, for example, by doing
3227 a 32-bit division of -2147483648 by -1.</p>
3229 <p>If the <tt>exact</tt> keyword is present, the result value of the
3230 <tt>sdiv</tt> is undefined if the result would be rounded or if overflow
3235 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3240 <!-- _______________________________________________________________________ -->
3241 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
3242 Instruction</a> </div>
3244 <div class="doc_text">
3248 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3252 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3255 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3256 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3257 floating point values. Both arguments must have identical types.</p>
3260 <p>The value produced is the floating point quotient of the two operands.</p>
3264 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3269 <!-- _______________________________________________________________________ -->
3270 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3273 <div class="doc_text">
3277 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3281 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3282 division of its two arguments.</p>
3285 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3286 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3287 values. Both arguments must have identical types.</p>
3290 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3291 This instruction always performs an unsigned division to get the
3294 <p>Note that unsigned integer remainder and signed integer remainder are
3295 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3297 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3301 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3306 <!-- _______________________________________________________________________ -->
3307 <div class="doc_subsubsection">
3308 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3311 <div class="doc_text">
3315 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3319 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3320 division of its two operands. This instruction can also take
3321 <a href="#t_vector">vector</a> versions of the values in which case the
3322 elements must be integers.</p>
3325 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3326 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3327 values. Both arguments must have identical types.</p>
3330 <p>This instruction returns the <i>remainder</i> of a division (where the result
3331 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
3332 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
3333 a value. For more information about the difference,
3334 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3335 Math Forum</a>. For a table of how this is implemented in various languages,
3336 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3337 Wikipedia: modulo operation</a>.</p>
3339 <p>Note that signed integer remainder and unsigned integer remainder are
3340 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3342 <p>Taking the remainder of a division by zero leads to undefined behavior.
3343 Overflow also leads to undefined behavior; this is a rare case, but can
3344 occur, for example, by taking the remainder of a 32-bit division of
3345 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3346 lets srem be implemented using instructions that return both the result of
3347 the division and the remainder.)</p>
3351 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3356 <!-- _______________________________________________________________________ -->
3357 <div class="doc_subsubsection">
3358 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
3360 <div class="doc_text">
3364 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3368 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3369 its two operands.</p>
3372 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3373 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3374 floating point values. Both arguments must have identical types.</p>
3377 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3378 has the same sign as the dividend.</p>
3382 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3387 <!-- ======================================================================= -->
3388 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
3389 Operations</a> </div>
3391 <div class="doc_text">
3393 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3394 program. They are generally very efficient instructions and can commonly be
3395 strength reduced from other instructions. They require two operands of the
3396 same type, execute an operation on them, and produce a single value. The
3397 resulting value is the same type as its operands.</p>
3401 <!-- _______________________________________________________________________ -->
3402 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
3403 Instruction</a> </div>
3405 <div class="doc_text">
3409 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3413 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3414 a specified number of bits.</p>
3417 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3418 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3419 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3422 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3423 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3424 is (statically or dynamically) negative or equal to or larger than the number
3425 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3426 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3427 shift amount in <tt>op2</tt>.</p>
3431 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3432 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3433 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3434 <result> = shl i32 1, 32 <i>; undefined</i>
3435 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3440 <!-- _______________________________________________________________________ -->
3441 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
3442 Instruction</a> </div>
3444 <div class="doc_text">
3448 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3452 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
3453 operand shifted to the right a specified number of bits with zero fill.</p>
3456 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
3457 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3458 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3461 <p>This instruction always performs a logical shift right operation. The most
3462 significant bits of the result will be filled with zero bits after the shift.
3463 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
3464 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3465 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3466 shift amount in <tt>op2</tt>.</p>
3470 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
3471 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
3472 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
3473 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
3474 <result> = lshr i32 1, 32 <i>; undefined</i>
3475 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3480 <!-- _______________________________________________________________________ -->
3481 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
3482 Instruction</a> </div>
3483 <div class="doc_text">
3487 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3491 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
3492 operand shifted to the right a specified number of bits with sign
3496 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
3497 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3498 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3501 <p>This instruction always performs an arithmetic shift right operation, The
3502 most significant bits of the result will be filled with the sign bit
3503 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
3504 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
3505 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
3506 the corresponding shift amount in <tt>op2</tt>.</p>
3510 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
3511 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
3512 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
3513 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
3514 <result> = ashr i32 1, 32 <i>; undefined</i>
3515 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3520 <!-- _______________________________________________________________________ -->
3521 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
3522 Instruction</a> </div>
3524 <div class="doc_text">
3528 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3532 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
3536 <p>The two arguments to the '<tt>and</tt>' instruction must be
3537 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3538 values. Both arguments must have identical types.</p>
3541 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
3543 <table border="1" cellspacing="0" cellpadding="4">
3575 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
3576 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
3577 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
3580 <!-- _______________________________________________________________________ -->
3581 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
3583 <div class="doc_text">
3587 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3591 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
3595 <p>The two arguments to the '<tt>or</tt>' instruction must be
3596 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3597 values. Both arguments must have identical types.</p>
3600 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
3602 <table border="1" cellspacing="0" cellpadding="4">
3634 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
3635 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
3636 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3641 <!-- _______________________________________________________________________ -->
3642 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3643 Instruction</a> </div>
3645 <div class="doc_text">
3649 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3653 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
3654 its two operands. The <tt>xor</tt> is used to implement the "one's
3655 complement" operation, which is the "~" operator in C.</p>
3658 <p>The two arguments to the '<tt>xor</tt>' instruction must be
3659 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3660 values. Both arguments must have identical types.</p>
3663 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3665 <table border="1" cellspacing="0" cellpadding="4">
3697 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3698 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3699 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3700 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3705 <!-- ======================================================================= -->
3706 <div class="doc_subsection">
3707 <a name="vectorops">Vector Operations</a>
3710 <div class="doc_text">
3712 <p>LLVM supports several instructions to represent vector operations in a
3713 target-independent manner. These instructions cover the element-access and
3714 vector-specific operations needed to process vectors effectively. While LLVM
3715 does directly support these vector operations, many sophisticated algorithms
3716 will want to use target-specific intrinsics to take full advantage of a
3717 specific target.</p>
3721 <!-- _______________________________________________________________________ -->
3722 <div class="doc_subsubsection">
3723 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3726 <div class="doc_text">
3730 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3734 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
3735 from a vector at a specified index.</p>
3739 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
3740 of <a href="#t_vector">vector</a> type. The second operand is an index
3741 indicating the position from which to extract the element. The index may be
3745 <p>The result is a scalar of the same type as the element type of
3746 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3747 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3748 results are undefined.</p>
3752 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
3757 <!-- _______________________________________________________________________ -->
3758 <div class="doc_subsubsection">
3759 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3762 <div class="doc_text">
3766 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3770 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
3771 vector at a specified index.</p>
3774 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
3775 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
3776 whose type must equal the element type of the first operand. The third
3777 operand is an index indicating the position at which to insert the value.
3778 The index may be a variable.</p>
3781 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
3782 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
3783 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3784 results are undefined.</p>
3788 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3793 <!-- _______________________________________________________________________ -->
3794 <div class="doc_subsubsection">
3795 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3798 <div class="doc_text">
3802 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3806 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
3807 from two input vectors, returning a vector with the same element type as the
3808 input and length that is the same as the shuffle mask.</p>
3811 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3812 with types that match each other. The third argument is a shuffle mask whose
3813 element type is always 'i32'. The result of the instruction is a vector
3814 whose length is the same as the shuffle mask and whose element type is the
3815 same as the element type of the first two operands.</p>
3817 <p>The shuffle mask operand is required to be a constant vector with either
3818 constant integer or undef values.</p>
3821 <p>The elements of the two input vectors are numbered from left to right across
3822 both of the vectors. The shuffle mask operand specifies, for each element of
3823 the result vector, which element of the two input vectors the result element
3824 gets. The element selector may be undef (meaning "don't care") and the
3825 second operand may be undef if performing a shuffle from only one vector.</p>
3829 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3830 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
3831 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
3832 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3833 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
3834 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
3835 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3836 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > <i>; yields <8 x i32></i>
3841 <!-- ======================================================================= -->
3842 <div class="doc_subsection">
3843 <a name="aggregateops">Aggregate Operations</a>
3846 <div class="doc_text">
3848 <p>LLVM supports several instructions for working with aggregate values.</p>
3852 <!-- _______________________________________________________________________ -->
3853 <div class="doc_subsubsection">
3854 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3857 <div class="doc_text">
3861 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3865 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3866 or array element from an aggregate value.</p>
3869 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
3870 of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type. The
3871 operands are constant indices to specify which value to extract in a similar
3872 manner as indices in a
3873 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
3876 <p>The result is the value at the position in the aggregate specified by the
3881 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3886 <!-- _______________________________________________________________________ -->
3887 <div class="doc_subsubsection">
3888 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3891 <div class="doc_text">
3895 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx> <i>; yields <aggregate type></i>
3899 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a struct field or
3900 array element in an aggregate.</p>
3904 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
3905 of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type. The
3906 second operand is a first-class value to insert. The following operands are
3907 constant indices indicating the position at which to insert the value in a
3908 similar manner as indices in a
3909 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction. The
3910 value to insert must have the same type as the value identified by the
3914 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
3915 that of <tt>val</tt> except that the value at the position specified by the
3916 indices is that of <tt>elt</tt>.</p>
3920 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
3921 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
3927 <!-- ======================================================================= -->
3928 <div class="doc_subsection">
3929 <a name="memoryops">Memory Access and Addressing Operations</a>
3932 <div class="doc_text">
3934 <p>A key design point of an SSA-based representation is how it represents
3935 memory. In LLVM, no memory locations are in SSA form, which makes things
3936 very simple. This section describes how to read, write, and allocate
3941 <!-- _______________________________________________________________________ -->
3942 <div class="doc_subsubsection">
3943 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3946 <div class="doc_text">
3950 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3954 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3955 currently executing function, to be automatically released when this function
3956 returns to its caller. The object is always allocated in the generic address
3957 space (address space zero).</p>
3960 <p>The '<tt>alloca</tt>' instruction
3961 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
3962 runtime stack, returning a pointer of the appropriate type to the program.
3963 If "NumElements" is specified, it is the number of elements allocated,
3964 otherwise "NumElements" is defaulted to be one. If a constant alignment is
3965 specified, the value result of the allocation is guaranteed to be aligned to
3966 at least that boundary. If not specified, or if zero, the target can choose
3967 to align the allocation on any convenient boundary compatible with the
3970 <p>'<tt>type</tt>' may be any sized type.</p>
3973 <p>Memory is allocated; a pointer is returned. The operation is undefined if
3974 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3975 memory is automatically released when the function returns. The
3976 '<tt>alloca</tt>' instruction is commonly used to represent automatic
3977 variables that must have an address available. When the function returns
3978 (either with the <tt><a href="#i_ret">ret</a></tt>
3979 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
3980 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
3984 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3985 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3986 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3987 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3992 <!-- _______________________________________________________________________ -->
3993 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3994 Instruction</a> </div>
3996 <div class="doc_text">
4000 <result> = load <ty>* <pointer>[, align <alignment>]
4001 <result> = volatile load <ty>* <pointer>[, align <alignment>]
4005 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4008 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4009 from which to load. The pointer must point to
4010 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4011 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4012 number or order of execution of this <tt>load</tt> with other
4013 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
4016 <p>The optional constant "align" argument specifies the alignment of the
4017 operation (that is, the alignment of the memory address). A value of 0 or an
4018 omitted "align" argument means that the operation has the preferential
4019 alignment for the target. It is the responsibility of the code emitter to
4020 ensure that the alignment information is correct. Overestimating the
4021 alignment results in an undefined behavior. Underestimating the alignment may
4022 produce less efficient code. An alignment of 1 is always safe.</p>
4025 <p>The location of memory pointed to is loaded. If the value being loaded is of
4026 scalar type then the number of bytes read does not exceed the minimum number
4027 of bytes needed to hold all bits of the type. For example, loading an
4028 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4029 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4030 is undefined if the value was not originally written using a store of the
4035 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4036 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4037 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4042 <!-- _______________________________________________________________________ -->
4043 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
4044 Instruction</a> </div>
4046 <div class="doc_text">
4050 store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
4051 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
4055 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4058 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4059 and an address at which to store it. The type of the
4060 '<tt><pointer></tt>' operand must be a pointer to
4061 the <a href="#t_firstclass">first class</a> type of the
4062 '<tt><value></tt>' operand. If the <tt>store</tt> is marked
4063 as <tt>volatile</tt>, then the optimizer is not allowed to modify the number
4064 or order of execution of this <tt>store</tt> with other
4065 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
4068 <p>The optional constant "align" argument specifies the alignment of the
4069 operation (that is, the alignment of the memory address). A value of 0 or an
4070 omitted "align" argument means that the operation has the preferential
4071 alignment for the target. It is the responsibility of the code emitter to
4072 ensure that the alignment information is correct. Overestimating the
4073 alignment results in an undefined behavior. Underestimating the alignment may
4074 produce less efficient code. An alignment of 1 is always safe.</p>
4077 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4078 location specified by the '<tt><pointer></tt>' operand. If
4079 '<tt><value></tt>' is of scalar type then the number of bytes written
4080 does not exceed the minimum number of bytes needed to hold all bits of the
4081 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4082 writing a value of a type like <tt>i20</tt> with a size that is not an
4083 integral number of bytes, it is unspecified what happens to the extra bits
4084 that do not belong to the type, but they will typically be overwritten.</p>
4088 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4089 store i32 3, i32* %ptr <i>; yields {void}</i>
4090 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4095 <!-- _______________________________________________________________________ -->
4096 <div class="doc_subsubsection">
4097 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
4100 <div class="doc_text">
4104 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4105 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4109 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
4110 subelement of an aggregate data structure. It performs address calculation
4111 only and does not access memory.</p>
4114 <p>The first argument is always a pointer, and forms the basis of the
4115 calculation. The remaining arguments are indices that indicate which of the
4116 elements of the aggregate object are indexed. The interpretation of each
4117 index is dependent on the type being indexed into. The first index always
4118 indexes the pointer value given as the first argument, the second index
4119 indexes a value of the type pointed to (not necessarily the value directly
4120 pointed to, since the first index can be non-zero), etc. The first type
4121 indexed into must be a pointer value, subsequent types can be arrays, vectors
4122 and structs. Note that subsequent types being indexed into can never be
4123 pointers, since that would require loading the pointer before continuing
4126 <p>The type of each index argument depends on the type it is indexing into.
4127 When indexing into a (optionally packed) structure, only <tt>i32</tt> integer
4128 <b>constants</b> are allowed. When indexing into an array, pointer or
4129 vector, integers of any width are allowed, and they are not required to be
4132 <p>For example, let's consider a C code fragment and how it gets compiled to
4135 <div class="doc_code">
4148 int *foo(struct ST *s) {
4149 return &s[1].Z.B[5][13];
4154 <p>The LLVM code generated by the GCC frontend is:</p>
4156 <div class="doc_code">
4158 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
4159 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
4161 define i32* @foo(%ST* %s) {
4163 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
4170 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
4171 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
4172 }</tt>' type, a structure. The second index indexes into the third element
4173 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
4174 i8 }</tt>' type, another structure. The third index indexes into the second
4175 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
4176 array. The two dimensions of the array are subscripted into, yielding an
4177 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
4178 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
4180 <p>Note that it is perfectly legal to index partially through a structure,
4181 returning a pointer to an inner element. Because of this, the LLVM code for
4182 the given testcase is equivalent to:</p>
4185 define i32* @foo(%ST* %s) {
4186 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
4187 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
4188 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
4189 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
4190 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
4195 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
4196 <tt>getelementptr</tt> is undefined if the base pointer is not an
4197 <i>in bounds</i> address of an allocated object, or if any of the addresses
4198 that would be formed by successive addition of the offsets implied by the
4199 indices to the base address with infinitely precise arithmetic are not an
4200 <i>in bounds</i> address of that allocated object.
4201 The <i>in bounds</i> addresses for an allocated object are all the addresses
4202 that point into the object, plus the address one byte past the end.</p>
4204 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
4205 the base address with silently-wrapping two's complement arithmetic, and
4206 the result value of the <tt>getelementptr</tt> may be outside the object
4207 pointed to by the base pointer. The result value may not necessarily be
4208 used to access memory though, even if it happens to point into allocated
4209 storage. See the <a href="#pointeraliasing">Pointer Aliasing Rules</a>
4210 section for more information.</p>
4212 <p>The getelementptr instruction is often confusing. For some more insight into
4213 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
4217 <i>; yields [12 x i8]*:aptr</i>
4218 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4219 <i>; yields i8*:vptr</i>
4220 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4221 <i>; yields i8*:eptr</i>
4222 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4223 <i>; yields i32*:iptr</i>
4224 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4229 <!-- ======================================================================= -->
4230 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
4233 <div class="doc_text">
4235 <p>The instructions in this category are the conversion instructions (casting)
4236 which all take a single operand and a type. They perform various bit
4237 conversions on the operand.</p>
4241 <!-- _______________________________________________________________________ -->
4242 <div class="doc_subsubsection">
4243 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
4245 <div class="doc_text">
4249 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
4253 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
4254 type <tt>ty2</tt>.</p>
4257 <p>The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
4258 be an <a href="#t_integer">integer</a> type, and a type that specifies the
4259 size and type of the result, which must be
4260 an <a href="#t_integer">integer</a> type. The bit size of <tt>value</tt> must
4261 be larger than the bit size of <tt>ty2</tt>. Equal sized types are not
4265 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
4266 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
4267 source size must be larger than the destination size, <tt>trunc</tt> cannot
4268 be a <i>no-op cast</i>. It will always truncate bits.</p>
4272 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
4273 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
4274 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
4279 <!-- _______________________________________________________________________ -->
4280 <div class="doc_subsubsection">
4281 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
4283 <div class="doc_text">
4287 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
4291 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
4296 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
4297 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4298 also be of <a href="#t_integer">integer</a> type. The bit size of the
4299 <tt>value</tt> must be smaller than the bit size of the destination type,
4303 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
4304 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
4306 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
4310 %X = zext i32 257 to i64 <i>; yields i64:257</i>
4311 %Y = zext i1 true to i32 <i>; yields i32:1</i>
4316 <!-- _______________________________________________________________________ -->
4317 <div class="doc_subsubsection">
4318 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
4320 <div class="doc_text">
4324 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
4328 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
4331 <p>The '<tt>sext</tt>' instruction takes a value to cast, which must be of
4332 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4333 also be of <a href="#t_integer">integer</a> type. The bit size of the
4334 <tt>value</tt> must be smaller than the bit size of the destination type,
4338 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
4339 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
4340 of the type <tt>ty2</tt>.</p>
4342 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
4346 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
4347 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
4352 <!-- _______________________________________________________________________ -->
4353 <div class="doc_subsubsection">
4354 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
4357 <div class="doc_text">
4361 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
4365 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
4369 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
4370 point</a> value to cast and a <a href="#t_floating">floating point</a> type
4371 to cast it to. The size of <tt>value</tt> must be larger than the size of
4372 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
4373 <i>no-op cast</i>.</p>
4376 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
4377 <a href="#t_floating">floating point</a> type to a smaller
4378 <a href="#t_floating">floating point</a> type. If the value cannot fit
4379 within the destination type, <tt>ty2</tt>, then the results are
4384 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
4385 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
4390 <!-- _______________________________________________________________________ -->
4391 <div class="doc_subsubsection">
4392 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
4394 <div class="doc_text">
4398 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
4402 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
4403 floating point value.</p>
4406 <p>The '<tt>fpext</tt>' instruction takes a
4407 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
4408 a <a href="#t_floating">floating point</a> type to cast it to. The source
4409 type must be smaller than the destination type.</p>
4412 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
4413 <a href="#t_floating">floating point</a> type to a larger
4414 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
4415 used to make a <i>no-op cast</i> because it always changes bits. Use
4416 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
4420 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
4421 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
4426 <!-- _______________________________________________________________________ -->
4427 <div class="doc_subsubsection">
4428 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
4430 <div class="doc_text">
4434 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
4438 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
4439 unsigned integer equivalent of type <tt>ty2</tt>.</p>
4442 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
4443 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4444 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4445 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4446 vector integer type with the same number of elements as <tt>ty</tt></p>
4449 <p>The '<tt>fptoui</tt>' instruction converts its
4450 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4451 towards zero) unsigned integer value. If the value cannot fit
4452 in <tt>ty2</tt>, the results are undefined.</p>
4456 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
4457 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
4458 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
4463 <!-- _______________________________________________________________________ -->
4464 <div class="doc_subsubsection">
4465 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
4467 <div class="doc_text">
4471 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
4475 <p>The '<tt>fptosi</tt>' instruction converts
4476 <a href="#t_floating">floating point</a> <tt>value</tt> to
4477 type <tt>ty2</tt>.</p>
4480 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
4481 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4482 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4483 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4484 vector integer type with the same number of elements as <tt>ty</tt></p>
4487 <p>The '<tt>fptosi</tt>' instruction converts its
4488 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4489 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
4490 the results are undefined.</p>
4494 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
4495 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
4496 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
4501 <!-- _______________________________________________________________________ -->
4502 <div class="doc_subsubsection">
4503 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
4505 <div class="doc_text">
4509 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4513 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
4514 integer and converts that value to the <tt>ty2</tt> type.</p>
4517 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
4518 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4519 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4520 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4521 floating point type with the same number of elements as <tt>ty</tt></p>
4524 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4525 integer quantity and converts it to the corresponding floating point
4526 value. If the value cannot fit in the floating point value, the results are
4531 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
4532 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
4537 <!-- _______________________________________________________________________ -->
4538 <div class="doc_subsubsection">
4539 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4541 <div class="doc_text">
4545 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4549 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
4550 and converts that value to the <tt>ty2</tt> type.</p>
4553 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4554 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4555 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4556 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4557 floating point type with the same number of elements as <tt>ty</tt></p>
4560 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
4561 quantity and converts it to the corresponding floating point value. If the
4562 value cannot fit in the floating point value, the results are undefined.</p>
4566 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
4567 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
4572 <!-- _______________________________________________________________________ -->
4573 <div class="doc_subsubsection">
4574 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4576 <div class="doc_text">
4580 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
4584 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4585 the integer type <tt>ty2</tt>.</p>
4588 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4589 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
4590 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
4593 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4594 <tt>ty2</tt> by interpreting the pointer value as an integer and either
4595 truncating or zero extending that value to the size of the integer type. If
4596 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4597 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4598 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4603 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4604 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4609 <!-- _______________________________________________________________________ -->
4610 <div class="doc_subsubsection">
4611 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4613 <div class="doc_text">
4617 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4621 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
4622 pointer type, <tt>ty2</tt>.</p>
4625 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4626 value to cast, and a type to cast it to, which must be a
4627 <a href="#t_pointer">pointer</a> type.</p>
4630 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4631 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4632 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4633 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
4634 than the size of a pointer then a zero extension is done. If they are the
4635 same size, nothing is done (<i>no-op cast</i>).</p>
4639 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4640 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4641 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4646 <!-- _______________________________________________________________________ -->
4647 <div class="doc_subsubsection">
4648 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4650 <div class="doc_text">
4654 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4658 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4659 <tt>ty2</tt> without changing any bits.</p>
4662 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
4663 non-aggregate first class value, and a type to cast it to, which must also be
4664 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
4665 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
4666 identical. If the source type is a pointer, the destination type must also be
4667 a pointer. This instruction supports bitwise conversion of vectors to
4668 integers and to vectors of other types (as long as they have the same
4672 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4673 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4674 this conversion. The conversion is done as if the <tt>value</tt> had been
4675 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
4676 be converted to other pointer types with this instruction. To convert
4677 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
4678 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4682 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4683 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4684 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4689 <!-- ======================================================================= -->
4690 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4692 <div class="doc_text">
4694 <p>The instructions in this category are the "miscellaneous" instructions, which
4695 defy better classification.</p>
4699 <!-- _______________________________________________________________________ -->
4700 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4703 <div class="doc_text">
4707 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4711 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
4712 boolean values based on comparison of its two integer, integer vector, or
4713 pointer operands.</p>
4716 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4717 the condition code indicating the kind of comparison to perform. It is not a
4718 value, just a keyword. The possible condition code are:</p>
4721 <li><tt>eq</tt>: equal</li>
4722 <li><tt>ne</tt>: not equal </li>
4723 <li><tt>ugt</tt>: unsigned greater than</li>
4724 <li><tt>uge</tt>: unsigned greater or equal</li>
4725 <li><tt>ult</tt>: unsigned less than</li>
4726 <li><tt>ule</tt>: unsigned less or equal</li>
4727 <li><tt>sgt</tt>: signed greater than</li>
4728 <li><tt>sge</tt>: signed greater or equal</li>
4729 <li><tt>slt</tt>: signed less than</li>
4730 <li><tt>sle</tt>: signed less or equal</li>
4733 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4734 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
4735 typed. They must also be identical types.</p>
4738 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
4739 condition code given as <tt>cond</tt>. The comparison performed always yields
4740 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
4741 result, as follows:</p>
4744 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4745 <tt>false</tt> otherwise. No sign interpretation is necessary or
4748 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4749 <tt>false</tt> otherwise. No sign interpretation is necessary or
4752 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4753 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4755 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4756 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4757 to <tt>op2</tt>.</li>
4759 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4760 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4762 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
4763 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4765 <li><tt>sgt</tt>: interprets the operands as signed values and yields
4766 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4768 <li><tt>sge</tt>: interprets the operands as signed values and yields
4769 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4770 to <tt>op2</tt>.</li>
4772 <li><tt>slt</tt>: interprets the operands as signed values and yields
4773 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4775 <li><tt>sle</tt>: interprets the operands as signed values and yields
4776 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4779 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4780 values are compared as if they were integers.</p>
4782 <p>If the operands are integer vectors, then they are compared element by
4783 element. The result is an <tt>i1</tt> vector with the same number of elements
4784 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
4788 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4789 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4790 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4791 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4792 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4793 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4796 <p>Note that the code generator does not yet support vector types with
4797 the <tt>icmp</tt> instruction.</p>
4801 <!-- _______________________________________________________________________ -->
4802 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4805 <div class="doc_text">
4809 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4813 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
4814 values based on comparison of its operands.</p>
4816 <p>If the operands are floating point scalars, then the result type is a boolean
4817 (<a href="#t_integer"><tt>i1</tt></a>).</p>
4819 <p>If the operands are floating point vectors, then the result type is a vector
4820 of boolean with the same number of elements as the operands being
4824 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4825 the condition code indicating the kind of comparison to perform. It is not a
4826 value, just a keyword. The possible condition code are:</p>
4829 <li><tt>false</tt>: no comparison, always returns false</li>
4830 <li><tt>oeq</tt>: ordered and equal</li>
4831 <li><tt>ogt</tt>: ordered and greater than </li>
4832 <li><tt>oge</tt>: ordered and greater than or equal</li>
4833 <li><tt>olt</tt>: ordered and less than </li>
4834 <li><tt>ole</tt>: ordered and less than or equal</li>
4835 <li><tt>one</tt>: ordered and not equal</li>
4836 <li><tt>ord</tt>: ordered (no nans)</li>
4837 <li><tt>ueq</tt>: unordered or equal</li>
4838 <li><tt>ugt</tt>: unordered or greater than </li>
4839 <li><tt>uge</tt>: unordered or greater than or equal</li>
4840 <li><tt>ult</tt>: unordered or less than </li>
4841 <li><tt>ule</tt>: unordered or less than or equal</li>
4842 <li><tt>une</tt>: unordered or not equal</li>
4843 <li><tt>uno</tt>: unordered (either nans)</li>
4844 <li><tt>true</tt>: no comparison, always returns true</li>
4847 <p><i>Ordered</i> means that neither operand is a QNAN while
4848 <i>unordered</i> means that either operand may be a QNAN.</p>
4850 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
4851 a <a href="#t_floating">floating point</a> type or
4852 a <a href="#t_vector">vector</a> of floating point type. They must have
4853 identical types.</p>
4856 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4857 according to the condition code given as <tt>cond</tt>. If the operands are
4858 vectors, then the vectors are compared element by element. Each comparison
4859 performed always yields an <a href="#t_integer">i1</a> result, as
4863 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4865 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4866 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4868 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4869 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4871 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4872 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4874 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4875 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4877 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4878 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4880 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4881 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4883 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4885 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4886 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4888 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4889 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4891 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4892 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4894 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4895 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4897 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4898 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4900 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4901 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4903 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4905 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4910 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4911 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4912 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4913 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4916 <p>Note that the code generator does not yet support vector types with
4917 the <tt>fcmp</tt> instruction.</p>
4921 <!-- _______________________________________________________________________ -->
4922 <div class="doc_subsubsection">
4923 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4926 <div class="doc_text">
4930 <result> = phi <ty> [ <val0>, <label0>], ...
4934 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
4935 SSA graph representing the function.</p>
4938 <p>The type of the incoming values is specified with the first type field. After
4939 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
4940 one pair for each predecessor basic block of the current block. Only values
4941 of <a href="#t_firstclass">first class</a> type may be used as the value
4942 arguments to the PHI node. Only labels may be used as the label
4945 <p>There must be no non-phi instructions between the start of a basic block and
4946 the PHI instructions: i.e. PHI instructions must be first in a basic
4949 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
4950 occur on the edge from the corresponding predecessor block to the current
4951 block (but after any definition of an '<tt>invoke</tt>' instruction's return
4952 value on the same edge).</p>
4955 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4956 specified by the pair corresponding to the predecessor basic block that
4957 executed just prior to the current block.</p>
4961 Loop: ; Infinite loop that counts from 0 on up...
4962 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4963 %nextindvar = add i32 %indvar, 1
4969 <!-- _______________________________________________________________________ -->
4970 <div class="doc_subsubsection">
4971 <a name="i_select">'<tt>select</tt>' Instruction</a>
4974 <div class="doc_text">
4978 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4980 <i>selty</i> is either i1 or {<N x i1>}
4984 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
4985 condition, without branching.</p>
4989 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
4990 values indicating the condition, and two values of the
4991 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
4992 vectors and the condition is a scalar, then entire vectors are selected, not
4993 individual elements.</p>
4996 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
4997 first value argument; otherwise, it returns the second value argument.</p>
4999 <p>If the condition is a vector of i1, then the value arguments must be vectors
5000 of the same size, and the selection is done element by element.</p>
5004 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
5007 <p>Note that the code generator does not yet support conditions
5008 with vector type.</p>
5012 <!-- _______________________________________________________________________ -->
5013 <div class="doc_subsubsection">
5014 <a name="i_call">'<tt>call</tt>' Instruction</a>
5017 <div class="doc_text">
5021 <result> = [tail] call [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] <ty> [<fnty>*] <fnptrval>(<function args>) [<a href="#fnattrs">fn attrs</a>]
5025 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
5028 <p>This instruction requires several arguments:</p>
5031 <li>The optional "tail" marker indicates that the callee function does not
5032 access any allocas or varargs in the caller. Note that calls may be
5033 marked "tail" even if they do not occur before
5034 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
5035 present, the function call is eligible for tail call optimization,
5036 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
5037 optimized into a jump</a>. As of this writing, the extra requirements for
5038 a call to actually be optimized are:
5040 <li>Caller and callee both have the calling
5041 convention <tt>fastcc</tt>.</li>
5042 <li>The call is in tail position (ret immediately follows call and ret
5043 uses value of call or is void).</li>
5044 <li>Option <tt>-tailcallopt</tt> is enabled,
5045 or <code>llvm::PerformTailCallOpt</code> is <code>true</code>.</li>
5046 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
5047 constraints are met.</a></li>
5051 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
5052 convention</a> the call should use. If none is specified, the call
5053 defaults to using C calling conventions. The calling convention of the
5054 call must match the calling convention of the target function, or else the
5055 behavior is undefined.</li>
5057 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
5058 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
5059 '<tt>inreg</tt>' attributes are valid here.</li>
5061 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
5062 type of the return value. Functions that return no value are marked
5063 <tt><a href="#t_void">void</a></tt>.</li>
5065 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5066 being invoked. The argument types must match the types implied by this
5067 signature. This type can be omitted if the function is not varargs and if
5068 the function type does not return a pointer to a function.</li>
5070 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5071 be invoked. In most cases, this is a direct function invocation, but
5072 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5073 to function value.</li>
5075 <li>'<tt>function args</tt>': argument list whose types match the function
5076 signature argument types. All arguments must be of
5077 <a href="#t_firstclass">first class</a> type. If the function signature
5078 indicates the function accepts a variable number of arguments, the extra
5079 arguments can be specified.</li>
5081 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
5082 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
5083 '<tt>readnone</tt>' attributes are valid here.</li>
5087 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
5088 a specified function, with its incoming arguments bound to the specified
5089 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
5090 function, control flow continues with the instruction after the function
5091 call, and the return value of the function is bound to the result
5096 %retval = call i32 @test(i32 %argc)
5097 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
5098 %X = tail call i32 @foo() <i>; yields i32</i>
5099 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
5100 call void %foo(i8 97 signext)
5102 %struct.A = type { i32, i8 }
5103 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
5104 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
5105 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
5106 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
5107 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
5110 <p>llvm treats calls to some functions with names and arguments that match the
5111 standard C99 library as being the C99 library functions, and may perform
5112 optimizations or generate code for them under that assumption. This is
5113 something we'd like to change in the future to provide better support for
5114 freestanding environments and non-C-based langauges.</p>
5118 <!-- _______________________________________________________________________ -->
5119 <div class="doc_subsubsection">
5120 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
5123 <div class="doc_text">
5127 <resultval> = va_arg <va_list*> <arglist>, <argty>
5131 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
5132 the "variable argument" area of a function call. It is used to implement the
5133 <tt>va_arg</tt> macro in C.</p>
5136 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
5137 argument. It returns a value of the specified argument type and increments
5138 the <tt>va_list</tt> to point to the next argument. The actual type
5139 of <tt>va_list</tt> is target specific.</p>
5142 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
5143 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
5144 to the next argument. For more information, see the variable argument
5145 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
5147 <p>It is legal for this instruction to be called in a function which does not
5148 take a variable number of arguments, for example, the <tt>vfprintf</tt>
5151 <p><tt>va_arg</tt> is an LLVM instruction instead of
5152 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
5156 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
5158 <p>Note that the code generator does not yet fully support va_arg on many
5159 targets. Also, it does not currently support va_arg with aggregate types on
5164 <!-- *********************************************************************** -->
5165 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
5166 <!-- *********************************************************************** -->
5168 <div class="doc_text">
5170 <p>LLVM supports the notion of an "intrinsic function". These functions have
5171 well known names and semantics and are required to follow certain
5172 restrictions. Overall, these intrinsics represent an extension mechanism for
5173 the LLVM language that does not require changing all of the transformations
5174 in LLVM when adding to the language (or the bitcode reader/writer, the
5175 parser, etc...).</p>
5177 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
5178 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5179 begin with this prefix. Intrinsic functions must always be external
5180 functions: you cannot define the body of intrinsic functions. Intrinsic
5181 functions may only be used in call or invoke instructions: it is illegal to
5182 take the address of an intrinsic function. Additionally, because intrinsic
5183 functions are part of the LLVM language, it is required if any are added that
5184 they be documented here.</p>
5186 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5187 family of functions that perform the same operation but on different data
5188 types. Because LLVM can represent over 8 million different integer types,
5189 overloading is used commonly to allow an intrinsic function to operate on any
5190 integer type. One or more of the argument types or the result type can be
5191 overloaded to accept any integer type. Argument types may also be defined as
5192 exactly matching a previous argument's type or the result type. This allows
5193 an intrinsic function which accepts multiple arguments, but needs all of them
5194 to be of the same type, to only be overloaded with respect to a single
5195 argument or the result.</p>
5197 <p>Overloaded intrinsics will have the names of its overloaded argument types
5198 encoded into its function name, each preceded by a period. Only those types
5199 which are overloaded result in a name suffix. Arguments whose type is matched
5200 against another type do not. For example, the <tt>llvm.ctpop</tt> function
5201 can take an integer of any width and returns an integer of exactly the same
5202 integer width. This leads to a family of functions such as
5203 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
5204 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5205 suffix is required. Because the argument's type is matched against the return
5206 type, it does not require its own name suffix.</p>
5208 <p>To learn how to add an intrinsic function, please see the
5209 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
5213 <!-- ======================================================================= -->
5214 <div class="doc_subsection">
5215 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
5218 <div class="doc_text">
5220 <p>Variable argument support is defined in LLVM with
5221 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
5222 intrinsic functions. These functions are related to the similarly named
5223 macros defined in the <tt><stdarg.h></tt> header file.</p>
5225 <p>All of these functions operate on arguments that use a target-specific value
5226 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5227 not define what this type is, so all transformations should be prepared to
5228 handle these functions regardless of the type used.</p>
5230 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
5231 instruction and the variable argument handling intrinsic functions are
5234 <div class="doc_code">
5236 define i32 @test(i32 %X, ...) {
5237 ; Initialize variable argument processing
5239 %ap2 = bitcast i8** %ap to i8*
5240 call void @llvm.va_start(i8* %ap2)
5242 ; Read a single integer argument
5243 %tmp = va_arg i8** %ap, i32
5245 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5247 %aq2 = bitcast i8** %aq to i8*
5248 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5249 call void @llvm.va_end(i8* %aq2)
5251 ; Stop processing of arguments.
5252 call void @llvm.va_end(i8* %ap2)
5256 declare void @llvm.va_start(i8*)
5257 declare void @llvm.va_copy(i8*, i8*)
5258 declare void @llvm.va_end(i8*)
5264 <!-- _______________________________________________________________________ -->
5265 <div class="doc_subsubsection">
5266 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
5270 <div class="doc_text">
5274 declare void %llvm.va_start(i8* <arglist>)
5278 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
5279 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
5282 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
5285 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
5286 macro available in C. In a target-dependent way, it initializes
5287 the <tt>va_list</tt> element to which the argument points, so that the next
5288 call to <tt>va_arg</tt> will produce the first variable argument passed to
5289 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
5290 need to know the last argument of the function as the compiler can figure
5295 <!-- _______________________________________________________________________ -->
5296 <div class="doc_subsubsection">
5297 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
5300 <div class="doc_text">
5304 declare void @llvm.va_end(i8* <arglist>)
5308 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
5309 which has been initialized previously
5310 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
5311 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
5314 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
5317 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
5318 macro available in C. In a target-dependent way, it destroys
5319 the <tt>va_list</tt> element to which the argument points. Calls
5320 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
5321 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
5322 with calls to <tt>llvm.va_end</tt>.</p>
5326 <!-- _______________________________________________________________________ -->
5327 <div class="doc_subsubsection">
5328 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
5331 <div class="doc_text">
5335 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
5339 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
5340 from the source argument list to the destination argument list.</p>
5343 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
5344 The second argument is a pointer to a <tt>va_list</tt> element to copy
5348 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
5349 macro available in C. In a target-dependent way, it copies the
5350 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
5351 element. This intrinsic is necessary because
5352 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
5353 arbitrarily complex and require, for example, memory allocation.</p>
5357 <!-- ======================================================================= -->
5358 <div class="doc_subsection">
5359 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
5362 <div class="doc_text">
5364 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
5365 Collection</a> (GC) requires the implementation and generation of these
5366 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
5367 roots on the stack</a>, as well as garbage collector implementations that
5368 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
5369 barriers. Front-ends for type-safe garbage collected languages should generate
5370 these intrinsics to make use of the LLVM garbage collectors. For more details,
5371 see <a href="GarbageCollection.html">Accurate Garbage Collection with
5374 <p>The garbage collection intrinsics only operate on objects in the generic
5375 address space (address space zero).</p>
5379 <!-- _______________________________________________________________________ -->
5380 <div class="doc_subsubsection">
5381 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
5384 <div class="doc_text">
5388 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5392 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
5393 the code generator, and allows some metadata to be associated with it.</p>
5396 <p>The first argument specifies the address of a stack object that contains the
5397 root pointer. The second pointer (which must be either a constant or a
5398 global value address) contains the meta-data to be associated with the
5402 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
5403 location. At compile-time, the code generator generates information to allow
5404 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
5405 intrinsic may only be used in a function which <a href="#gc">specifies a GC
5410 <!-- _______________________________________________________________________ -->
5411 <div class="doc_subsubsection">
5412 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
5415 <div class="doc_text">
5419 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5423 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
5424 locations, allowing garbage collector implementations that require read
5428 <p>The second argument is the address to read from, which should be an address
5429 allocated from the garbage collector. The first object is a pointer to the
5430 start of the referenced object, if needed by the language runtime (otherwise
5434 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
5435 instruction, but may be replaced with substantially more complex code by the
5436 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
5437 may only be used in a function which <a href="#gc">specifies a GC
5442 <!-- _______________________________________________________________________ -->
5443 <div class="doc_subsubsection">
5444 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
5447 <div class="doc_text">
5451 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5455 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5456 locations, allowing garbage collector implementations that require write
5457 barriers (such as generational or reference counting collectors).</p>
5460 <p>The first argument is the reference to store, the second is the start of the
5461 object to store it to, and the third is the address of the field of Obj to
5462 store to. If the runtime does not require a pointer to the object, Obj may
5466 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5467 instruction, but may be replaced with substantially more complex code by the
5468 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5469 may only be used in a function which <a href="#gc">specifies a GC
5474 <!-- ======================================================================= -->
5475 <div class="doc_subsection">
5476 <a name="int_codegen">Code Generator Intrinsics</a>
5479 <div class="doc_text">
5481 <p>These intrinsics are provided by LLVM to expose special features that may
5482 only be implemented with code generator support.</p>
5486 <!-- _______________________________________________________________________ -->
5487 <div class="doc_subsubsection">
5488 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5491 <div class="doc_text">
5495 declare i8 *@llvm.returnaddress(i32 <level>)
5499 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5500 target-specific value indicating the return address of the current function
5501 or one of its callers.</p>
5504 <p>The argument to this intrinsic indicates which function to return the address
5505 for. Zero indicates the calling function, one indicates its caller, etc.
5506 The argument is <b>required</b> to be a constant integer value.</p>
5509 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
5510 indicating the return address of the specified call frame, or zero if it
5511 cannot be identified. The value returned by this intrinsic is likely to be
5512 incorrect or 0 for arguments other than zero, so it should only be used for
5513 debugging purposes.</p>
5515 <p>Note that calling this intrinsic does not prevent function inlining or other
5516 aggressive transformations, so the value returned may not be that of the
5517 obvious source-language caller.</p>
5521 <!-- _______________________________________________________________________ -->
5522 <div class="doc_subsubsection">
5523 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5526 <div class="doc_text">
5530 declare i8 *@llvm.frameaddress(i32 <level>)
5534 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5535 target-specific frame pointer value for the specified stack frame.</p>
5538 <p>The argument to this intrinsic indicates which function to return the frame
5539 pointer for. Zero indicates the calling function, one indicates its caller,
5540 etc. The argument is <b>required</b> to be a constant integer value.</p>
5543 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
5544 indicating the frame address of the specified call frame, or zero if it
5545 cannot be identified. The value returned by this intrinsic is likely to be
5546 incorrect or 0 for arguments other than zero, so it should only be used for
5547 debugging purposes.</p>
5549 <p>Note that calling this intrinsic does not prevent function inlining or other
5550 aggressive transformations, so the value returned may not be that of the
5551 obvious source-language caller.</p>
5555 <!-- _______________________________________________________________________ -->
5556 <div class="doc_subsubsection">
5557 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5560 <div class="doc_text">
5564 declare i8 *@llvm.stacksave()
5568 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
5569 of the function stack, for use
5570 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
5571 useful for implementing language features like scoped automatic variable
5572 sized arrays in C99.</p>
5575 <p>This intrinsic returns a opaque pointer value that can be passed
5576 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
5577 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
5578 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
5579 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
5580 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
5581 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
5585 <!-- _______________________________________________________________________ -->
5586 <div class="doc_subsubsection">
5587 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5590 <div class="doc_text">
5594 declare void @llvm.stackrestore(i8 * %ptr)
5598 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5599 the function stack to the state it was in when the
5600 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
5601 executed. This is useful for implementing language features like scoped
5602 automatic variable sized arrays in C99.</p>
5605 <p>See the description
5606 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
5610 <!-- _______________________________________________________________________ -->
5611 <div class="doc_subsubsection">
5612 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5615 <div class="doc_text">
5619 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5623 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
5624 insert a prefetch instruction if supported; otherwise, it is a noop.
5625 Prefetches have no effect on the behavior of the program but can change its
5626 performance characteristics.</p>
5629 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
5630 specifier determining if the fetch should be for a read (0) or write (1),
5631 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5632 locality, to (3) - extremely local keep in cache. The <tt>rw</tt>
5633 and <tt>locality</tt> arguments must be constant integers.</p>
5636 <p>This intrinsic does not modify the behavior of the program. In particular,
5637 prefetches cannot trap and do not produce a value. On targets that support
5638 this intrinsic, the prefetch can provide hints to the processor cache for
5639 better performance.</p>
5643 <!-- _______________________________________________________________________ -->
5644 <div class="doc_subsubsection">
5645 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5648 <div class="doc_text">
5652 declare void @llvm.pcmarker(i32 <id>)
5656 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
5657 Counter (PC) in a region of code to simulators and other tools. The method
5658 is target specific, but it is expected that the marker will use exported
5659 symbols to transmit the PC of the marker. The marker makes no guarantees
5660 that it will remain with any specific instruction after optimizations. It is
5661 possible that the presence of a marker will inhibit optimizations. The
5662 intended use is to be inserted after optimizations to allow correlations of
5663 simulation runs.</p>
5666 <p><tt>id</tt> is a numerical id identifying the marker.</p>
5669 <p>This intrinsic does not modify the behavior of the program. Backends that do
5670 not support this intrinisic may ignore it.</p>
5674 <!-- _______________________________________________________________________ -->
5675 <div class="doc_subsubsection">
5676 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5679 <div class="doc_text">
5683 declare i64 @llvm.readcyclecounter( )
5687 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5688 counter register (or similar low latency, high accuracy clocks) on those
5689 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5690 should map to RPCC. As the backing counters overflow quickly (on the order
5691 of 9 seconds on alpha), this should only be used for small timings.</p>
5694 <p>When directly supported, reading the cycle counter should not modify any
5695 memory. Implementations are allowed to either return a application specific
5696 value or a system wide value. On backends without support, this is lowered
5697 to a constant 0.</p>
5701 <!-- ======================================================================= -->
5702 <div class="doc_subsection">
5703 <a name="int_libc">Standard C Library Intrinsics</a>
5706 <div class="doc_text">
5708 <p>LLVM provides intrinsics for a few important standard C library functions.
5709 These intrinsics allow source-language front-ends to pass information about
5710 the alignment of the pointer arguments to the code generator, providing
5711 opportunity for more efficient code generation.</p>
5715 <!-- _______________________________________________________________________ -->
5716 <div class="doc_subsubsection">
5717 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5720 <div class="doc_text">
5723 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
5724 integer bit width. Not all targets support all bit widths however.</p>
5727 declare void @llvm.memcpy.i8(i8 * <dest>, i8 * <src>,
5728 i8 <len>, i32 <align>)
5729 declare void @llvm.memcpy.i16(i8 * <dest>, i8 * <src>,
5730 i16 <len>, i32 <align>)
5731 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5732 i32 <len>, i32 <align>)
5733 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5734 i64 <len>, i32 <align>)
5738 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5739 source location to the destination location.</p>
5741 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5742 intrinsics do not return a value, and takes an extra alignment argument.</p>
5745 <p>The first argument is a pointer to the destination, the second is a pointer
5746 to the source. The third argument is an integer argument specifying the
5747 number of bytes to copy, and the fourth argument is the alignment of the
5748 source and destination locations.</p>
5750 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5751 then the caller guarantees that both the source and destination pointers are
5752 aligned to that boundary.</p>
5755 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5756 source location to the destination location, which are not allowed to
5757 overlap. It copies "len" bytes of memory over. If the argument is known to
5758 be aligned to some boundary, this can be specified as the fourth argument,
5759 otherwise it should be set to 0 or 1.</p>
5763 <!-- _______________________________________________________________________ -->
5764 <div class="doc_subsubsection">
5765 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5768 <div class="doc_text">
5771 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5772 width. Not all targets support all bit widths however.</p>
5775 declare void @llvm.memmove.i8(i8 * <dest>, i8 * <src>,
5776 i8 <len>, i32 <align>)
5777 declare void @llvm.memmove.i16(i8 * <dest>, i8 * <src>,
5778 i16 <len>, i32 <align>)
5779 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5780 i32 <len>, i32 <align>)
5781 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5782 i64 <len>, i32 <align>)
5786 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
5787 source location to the destination location. It is similar to the
5788 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
5791 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5792 intrinsics do not return a value, and takes an extra alignment argument.</p>
5795 <p>The first argument is a pointer to the destination, the second is a pointer
5796 to the source. The third argument is an integer argument specifying the
5797 number of bytes to copy, and the fourth argument is the alignment of the
5798 source and destination locations.</p>
5800 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5801 then the caller guarantees that the source and destination pointers are
5802 aligned to that boundary.</p>
5805 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
5806 source location to the destination location, which may overlap. It copies
5807 "len" bytes of memory over. If the argument is known to be aligned to some
5808 boundary, this can be specified as the fourth argument, otherwise it should
5809 be set to 0 or 1.</p>
5813 <!-- _______________________________________________________________________ -->
5814 <div class="doc_subsubsection">
5815 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5818 <div class="doc_text">
5821 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5822 width. Not all targets support all bit widths however.</p>
5825 declare void @llvm.memset.i8(i8 * <dest>, i8 <val>,
5826 i8 <len>, i32 <align>)
5827 declare void @llvm.memset.i16(i8 * <dest>, i8 <val>,
5828 i16 <len>, i32 <align>)
5829 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5830 i32 <len>, i32 <align>)
5831 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5832 i64 <len>, i32 <align>)
5836 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
5837 particular byte value.</p>
5839 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
5840 intrinsic does not return a value, and takes an extra alignment argument.</p>
5843 <p>The first argument is a pointer to the destination to fill, the second is the
5844 byte value to fill it with, the third argument is an integer argument
5845 specifying the number of bytes to fill, and the fourth argument is the known
5846 alignment of destination location.</p>
5848 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5849 then the caller guarantees that the destination pointer is aligned to that
5853 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
5854 at the destination location. If the argument is known to be aligned to some
5855 boundary, this can be specified as the fourth argument, otherwise it should
5856 be set to 0 or 1.</p>
5860 <!-- _______________________________________________________________________ -->
5861 <div class="doc_subsubsection">
5862 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5865 <div class="doc_text">
5868 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5869 floating point or vector of floating point type. Not all targets support all
5873 declare float @llvm.sqrt.f32(float %Val)
5874 declare double @llvm.sqrt.f64(double %Val)
5875 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5876 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5877 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5881 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5882 returning the same value as the libm '<tt>sqrt</tt>' functions would.
5883 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
5884 behavior for negative numbers other than -0.0 (which allows for better
5885 optimization, because there is no need to worry about errno being
5886 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
5889 <p>The argument and return value are floating point numbers of the same
5893 <p>This function returns the sqrt of the specified operand if it is a
5894 nonnegative floating point number.</p>
5898 <!-- _______________________________________________________________________ -->
5899 <div class="doc_subsubsection">
5900 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5903 <div class="doc_text">
5906 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5907 floating point or vector of floating point type. Not all targets support all
5911 declare float @llvm.powi.f32(float %Val, i32 %power)
5912 declare double @llvm.powi.f64(double %Val, i32 %power)
5913 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5914 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5915 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5919 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5920 specified (positive or negative) power. The order of evaluation of
5921 multiplications is not defined. When a vector of floating point type is
5922 used, the second argument remains a scalar integer value.</p>
5925 <p>The second argument is an integer power, and the first is a value to raise to
5929 <p>This function returns the first value raised to the second power with an
5930 unspecified sequence of rounding operations.</p>
5934 <!-- _______________________________________________________________________ -->
5935 <div class="doc_subsubsection">
5936 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5939 <div class="doc_text">
5942 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5943 floating point or vector of floating point type. Not all targets support all
5947 declare float @llvm.sin.f32(float %Val)
5948 declare double @llvm.sin.f64(double %Val)
5949 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5950 declare fp128 @llvm.sin.f128(fp128 %Val)
5951 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5955 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
5958 <p>The argument and return value are floating point numbers of the same
5962 <p>This function returns the sine of the specified operand, returning the same
5963 values as the libm <tt>sin</tt> functions would, and handles error conditions
5964 in the same way.</p>
5968 <!-- _______________________________________________________________________ -->
5969 <div class="doc_subsubsection">
5970 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5973 <div class="doc_text">
5976 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5977 floating point or vector of floating point type. Not all targets support all
5981 declare float @llvm.cos.f32(float %Val)
5982 declare double @llvm.cos.f64(double %Val)
5983 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5984 declare fp128 @llvm.cos.f128(fp128 %Val)
5985 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5989 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
5992 <p>The argument and return value are floating point numbers of the same
5996 <p>This function returns the cosine of the specified operand, returning the same
5997 values as the libm <tt>cos</tt> functions would, and handles error conditions
5998 in the same way.</p>
6002 <!-- _______________________________________________________________________ -->
6003 <div class="doc_subsubsection">
6004 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
6007 <div class="doc_text">
6010 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
6011 floating point or vector of floating point type. Not all targets support all
6015 declare float @llvm.pow.f32(float %Val, float %Power)
6016 declare double @llvm.pow.f64(double %Val, double %Power)
6017 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6018 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6019 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6023 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
6024 specified (positive or negative) power.</p>
6027 <p>The second argument is a floating point power, and the first is a value to
6028 raise to that power.</p>
6031 <p>This function returns the first value raised to the second power, returning
6032 the same values as the libm <tt>pow</tt> functions would, and handles error
6033 conditions in the same way.</p>
6037 <!-- ======================================================================= -->
6038 <div class="doc_subsection">
6039 <a name="int_manip">Bit Manipulation Intrinsics</a>
6042 <div class="doc_text">
6044 <p>LLVM provides intrinsics for a few important bit manipulation operations.
6045 These allow efficient code generation for some algorithms.</p>
6049 <!-- _______________________________________________________________________ -->
6050 <div class="doc_subsubsection">
6051 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
6054 <div class="doc_text">
6057 <p>This is an overloaded intrinsic function. You can use bswap on any integer
6058 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
6061 declare i16 @llvm.bswap.i16(i16 <id>)
6062 declare i32 @llvm.bswap.i32(i32 <id>)
6063 declare i64 @llvm.bswap.i64(i64 <id>)
6067 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
6068 values with an even number of bytes (positive multiple of 16 bits). These
6069 are useful for performing operations on data that is not in the target's
6070 native byte order.</p>
6073 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
6074 and low byte of the input i16 swapped. Similarly,
6075 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
6076 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
6077 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
6078 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
6079 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
6080 more, respectively).</p>
6084 <!-- _______________________________________________________________________ -->
6085 <div class="doc_subsubsection">
6086 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
6089 <div class="doc_text">
6092 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
6093 width. Not all targets support all bit widths however.</p>
6096 declare i8 @llvm.ctpop.i8(i8 <src>)
6097 declare i16 @llvm.ctpop.i16(i16 <src>)
6098 declare i32 @llvm.ctpop.i32(i32 <src>)
6099 declare i64 @llvm.ctpop.i64(i64 <src>)
6100 declare i256 @llvm.ctpop.i256(i256 <src>)
6104 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
6108 <p>The only argument is the value to be counted. The argument may be of any
6109 integer type. The return type must match the argument type.</p>
6112 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.</p>
6116 <!-- _______________________________________________________________________ -->
6117 <div class="doc_subsubsection">
6118 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
6121 <div class="doc_text">
6124 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
6125 integer bit width. Not all targets support all bit widths however.</p>
6128 declare i8 @llvm.ctlz.i8 (i8 <src>)
6129 declare i16 @llvm.ctlz.i16(i16 <src>)
6130 declare i32 @llvm.ctlz.i32(i32 <src>)
6131 declare i64 @llvm.ctlz.i64(i64 <src>)
6132 declare i256 @llvm.ctlz.i256(i256 <src>)
6136 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
6137 leading zeros in a variable.</p>
6140 <p>The only argument is the value to be counted. The argument may be of any
6141 integer type. The return type must match the argument type.</p>
6144 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
6145 zeros in a variable. If the src == 0 then the result is the size in bits of
6146 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
6150 <!-- _______________________________________________________________________ -->
6151 <div class="doc_subsubsection">
6152 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
6155 <div class="doc_text">
6158 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
6159 integer bit width. Not all targets support all bit widths however.</p>
6162 declare i8 @llvm.cttz.i8 (i8 <src>)
6163 declare i16 @llvm.cttz.i16(i16 <src>)
6164 declare i32 @llvm.cttz.i32(i32 <src>)
6165 declare i64 @llvm.cttz.i64(i64 <src>)
6166 declare i256 @llvm.cttz.i256(i256 <src>)
6170 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
6174 <p>The only argument is the value to be counted. The argument may be of any
6175 integer type. The return type must match the argument type.</p>
6178 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
6179 zeros in a variable. If the src == 0 then the result is the size in bits of
6180 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
6184 <!-- ======================================================================= -->
6185 <div class="doc_subsection">
6186 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
6189 <div class="doc_text">
6191 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
6195 <!-- _______________________________________________________________________ -->
6196 <div class="doc_subsubsection">
6197 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
6200 <div class="doc_text">
6203 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
6204 on any integer bit width.</p>
6207 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6208 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6209 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6213 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6214 a signed addition of the two arguments, and indicate whether an overflow
6215 occurred during the signed summation.</p>
6218 <p>The arguments (%a and %b) and the first element of the result structure may
6219 be of integer types of any bit width, but they must have the same bit
6220 width. The second element of the result structure must be of
6221 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6222 undergo signed addition.</p>
6225 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6226 a signed addition of the two variables. They return a structure — the
6227 first element of which is the signed summation, and the second element of
6228 which is a bit specifying if the signed summation resulted in an
6233 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6234 %sum = extractvalue {i32, i1} %res, 0
6235 %obit = extractvalue {i32, i1} %res, 1
6236 br i1 %obit, label %overflow, label %normal
6241 <!-- _______________________________________________________________________ -->
6242 <div class="doc_subsubsection">
6243 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
6246 <div class="doc_text">
6249 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
6250 on any integer bit width.</p>
6253 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6254 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6255 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6259 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6260 an unsigned addition of the two arguments, and indicate whether a carry
6261 occurred during the unsigned summation.</p>
6264 <p>The arguments (%a and %b) and the first element of the result structure may
6265 be of integer types of any bit width, but they must have the same bit
6266 width. The second element of the result structure must be of
6267 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6268 undergo unsigned addition.</p>
6271 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6272 an unsigned addition of the two arguments. They return a structure —
6273 the first element of which is the sum, and the second element of which is a
6274 bit specifying if the unsigned summation resulted in a carry.</p>
6278 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6279 %sum = extractvalue {i32, i1} %res, 0
6280 %obit = extractvalue {i32, i1} %res, 1
6281 br i1 %obit, label %carry, label %normal
6286 <!-- _______________________________________________________________________ -->
6287 <div class="doc_subsubsection">
6288 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
6291 <div class="doc_text">
6294 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
6295 on any integer bit width.</p>
6298 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6299 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6300 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6304 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6305 a signed subtraction of the two arguments, and indicate whether an overflow
6306 occurred during the signed subtraction.</p>
6309 <p>The arguments (%a and %b) and the first element of the result structure may
6310 be of integer types of any bit width, but they must have the same bit
6311 width. The second element of the result structure must be of
6312 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6313 undergo signed subtraction.</p>
6316 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6317 a signed subtraction of the two arguments. They return a structure —
6318 the first element of which is the subtraction, and the second element of
6319 which is a bit specifying if the signed subtraction resulted in an
6324 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6325 %sum = extractvalue {i32, i1} %res, 0
6326 %obit = extractvalue {i32, i1} %res, 1
6327 br i1 %obit, label %overflow, label %normal
6332 <!-- _______________________________________________________________________ -->
6333 <div class="doc_subsubsection">
6334 <a name="int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6337 <div class="doc_text">
6340 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6341 on any integer bit width.</p>
6344 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6345 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6346 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6350 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6351 an unsigned subtraction of the two arguments, and indicate whether an
6352 overflow occurred during the unsigned subtraction.</p>
6355 <p>The arguments (%a and %b) and the first element of the result structure may
6356 be of integer types of any bit width, but they must have the same bit
6357 width. The second element of the result structure must be of
6358 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6359 undergo unsigned subtraction.</p>
6362 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6363 an unsigned subtraction of the two arguments. They return a structure —
6364 the first element of which is the subtraction, and the second element of
6365 which is a bit specifying if the unsigned subtraction resulted in an
6370 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6371 %sum = extractvalue {i32, i1} %res, 0
6372 %obit = extractvalue {i32, i1} %res, 1
6373 br i1 %obit, label %overflow, label %normal
6378 <!-- _______________________________________________________________________ -->
6379 <div class="doc_subsubsection">
6380 <a name="int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6383 <div class="doc_text">
6386 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6387 on any integer bit width.</p>
6390 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6391 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6392 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6397 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6398 a signed multiplication of the two arguments, and indicate whether an
6399 overflow occurred during the signed multiplication.</p>
6402 <p>The arguments (%a and %b) and the first element of the result structure may
6403 be of integer types of any bit width, but they must have the same bit
6404 width. The second element of the result structure must be of
6405 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6406 undergo signed multiplication.</p>
6409 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6410 a signed multiplication of the two arguments. They return a structure —
6411 the first element of which is the multiplication, and the second element of
6412 which is a bit specifying if the signed multiplication resulted in an
6417 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6418 %sum = extractvalue {i32, i1} %res, 0
6419 %obit = extractvalue {i32, i1} %res, 1
6420 br i1 %obit, label %overflow, label %normal
6425 <!-- _______________________________________________________________________ -->
6426 <div class="doc_subsubsection">
6427 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6430 <div class="doc_text">
6433 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6434 on any integer bit width.</p>
6437 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6438 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6439 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6443 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6444 a unsigned multiplication of the two arguments, and indicate whether an
6445 overflow occurred during the unsigned multiplication.</p>
6448 <p>The arguments (%a and %b) and the first element of the result structure may
6449 be of integer types of any bit width, but they must have the same bit
6450 width. The second element of the result structure must be of
6451 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6452 undergo unsigned multiplication.</p>
6455 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6456 an unsigned multiplication of the two arguments. They return a structure
6457 — the first element of which is the multiplication, and the second
6458 element of which is a bit specifying if the unsigned multiplication resulted
6463 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6464 %sum = extractvalue {i32, i1} %res, 0
6465 %obit = extractvalue {i32, i1} %res, 1
6466 br i1 %obit, label %overflow, label %normal
6471 <!-- ======================================================================= -->
6472 <div class="doc_subsection">
6473 <a name="int_debugger">Debugger Intrinsics</a>
6476 <div class="doc_text">
6478 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
6479 prefix), are described in
6480 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
6481 Level Debugging</a> document.</p>
6485 <!-- ======================================================================= -->
6486 <div class="doc_subsection">
6487 <a name="int_eh">Exception Handling Intrinsics</a>
6490 <div class="doc_text">
6492 <p>The LLVM exception handling intrinsics (which all start with
6493 <tt>llvm.eh.</tt> prefix), are described in
6494 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6495 Handling</a> document.</p>
6499 <!-- ======================================================================= -->
6500 <div class="doc_subsection">
6501 <a name="int_trampoline">Trampoline Intrinsic</a>
6504 <div class="doc_text">
6506 <p>This intrinsic makes it possible to excise one parameter, marked with
6507 the <tt>nest</tt> attribute, from a function. The result is a callable
6508 function pointer lacking the nest parameter - the caller does not need to
6509 provide a value for it. Instead, the value to use is stored in advance in a
6510 "trampoline", a block of memory usually allocated on the stack, which also
6511 contains code to splice the nest value into the argument list. This is used
6512 to implement the GCC nested function address extension.</p>
6514 <p>For example, if the function is
6515 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6516 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
6519 <div class="doc_code">
6521 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6522 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6523 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6524 %fp = bitcast i8* %p to i32 (i32, i32)*
6528 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6529 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
6533 <!-- _______________________________________________________________________ -->
6534 <div class="doc_subsubsection">
6535 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6538 <div class="doc_text">
6542 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6546 <p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
6547 function pointer suitable for executing it.</p>
6550 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6551 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
6552 sufficiently aligned block of memory; this memory is written to by the
6553 intrinsic. Note that the size and the alignment are target-specific - LLVM
6554 currently provides no portable way of determining them, so a front-end that
6555 generates this intrinsic needs to have some target-specific knowledge.
6556 The <tt>func</tt> argument must hold a function bitcast to
6557 an <tt>i8*</tt>.</p>
6560 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
6561 dependent code, turning it into a function. A pointer to this function is
6562 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
6563 function pointer type</a> before being called. The new function's signature
6564 is the same as that of <tt>func</tt> with any arguments marked with
6565 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
6566 is allowed, and it must be of pointer type. Calling the new function is
6567 equivalent to calling <tt>func</tt> with the same argument list, but
6568 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
6569 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
6570 by <tt>tramp</tt> is modified, then the effect of any later call to the
6571 returned function pointer is undefined.</p>
6575 <!-- ======================================================================= -->
6576 <div class="doc_subsection">
6577 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6580 <div class="doc_text">
6582 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
6583 hardware constructs for atomic operations and memory synchronization. This
6584 provides an interface to the hardware, not an interface to the programmer. It
6585 is aimed at a low enough level to allow any programming models or APIs
6586 (Application Programming Interfaces) which need atomic behaviors to map
6587 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6588 hardware provides a "universal IR" for source languages, it also provides a
6589 starting point for developing a "universal" atomic operation and
6590 synchronization IR.</p>
6592 <p>These do <em>not</em> form an API such as high-level threading libraries,
6593 software transaction memory systems, atomic primitives, and intrinsic
6594 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6595 application libraries. The hardware interface provided by LLVM should allow
6596 a clean implementation of all of these APIs and parallel programming models.
6597 No one model or paradigm should be selected above others unless the hardware
6598 itself ubiquitously does so.</p>
6602 <!-- _______________________________________________________________________ -->
6603 <div class="doc_subsubsection">
6604 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6606 <div class="doc_text">
6609 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device> )
6613 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6614 specific pairs of memory access types.</p>
6617 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6618 The first four arguments enables a specific barrier as listed below. The
6619 fith argument specifies that the barrier applies to io or device or uncached
6623 <li><tt>ll</tt>: load-load barrier</li>
6624 <li><tt>ls</tt>: load-store barrier</li>
6625 <li><tt>sl</tt>: store-load barrier</li>
6626 <li><tt>ss</tt>: store-store barrier</li>
6627 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6631 <p>This intrinsic causes the system to enforce some ordering constraints upon
6632 the loads and stores of the program. This barrier does not
6633 indicate <em>when</em> any events will occur, it only enforces
6634 an <em>order</em> in which they occur. For any of the specified pairs of load
6635 and store operations (f.ex. load-load, or store-load), all of the first
6636 operations preceding the barrier will complete before any of the second
6637 operations succeeding the barrier begin. Specifically the semantics for each
6638 pairing is as follows:</p>
6641 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6642 after the barrier begins.</li>
6643 <li><tt>ls</tt>: All loads before the barrier must complete before any
6644 store after the barrier begins.</li>
6645 <li><tt>ss</tt>: All stores before the barrier must complete before any
6646 store after the barrier begins.</li>
6647 <li><tt>sl</tt>: All stores before the barrier must complete before any
6648 load after the barrier begins.</li>
6651 <p>These semantics are applied with a logical "and" behavior when more than one
6652 is enabled in a single memory barrier intrinsic.</p>
6654 <p>Backends may implement stronger barriers than those requested when they do
6655 not support as fine grained a barrier as requested. Some architectures do
6656 not need all types of barriers and on such architectures, these become
6661 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6662 %ptr = bitcast i8* %mallocP to i32*
6665 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6666 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6667 <i>; guarantee the above finishes</i>
6668 store i32 8, %ptr <i>; before this begins</i>
6673 <!-- _______________________________________________________________________ -->
6674 <div class="doc_subsubsection">
6675 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
6678 <div class="doc_text">
6681 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6682 any integer bit width and for different address spaces. Not all targets
6683 support all bit widths however.</p>
6686 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
6687 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6688 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6689 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6693 <p>This loads a value in memory and compares it to a given value. If they are
6694 equal, it stores a new value into the memory.</p>
6697 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
6698 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
6699 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
6700 this integer type. While any bit width integer may be used, targets may only
6701 lower representations they support in hardware.</p>
6704 <p>This entire intrinsic must be executed atomically. It first loads the value
6705 in memory pointed to by <tt>ptr</tt> and compares it with the
6706 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
6707 memory. The loaded value is yielded in all cases. This provides the
6708 equivalent of an atomic compare-and-swap operation within the SSA
6713 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6714 %ptr = bitcast i8* %mallocP to i32*
6717 %val1 = add i32 4, 4
6718 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6719 <i>; yields {i32}:result1 = 4</i>
6720 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6721 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6723 %val2 = add i32 1, 1
6724 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6725 <i>; yields {i32}:result2 = 8</i>
6726 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6728 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6733 <!-- _______________________________________________________________________ -->
6734 <div class="doc_subsubsection">
6735 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6737 <div class="doc_text">
6740 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6741 integer bit width. Not all targets support all bit widths however.</p>
6744 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6745 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6746 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6747 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6751 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6752 the value from memory. It then stores the value in <tt>val</tt> in the memory
6753 at <tt>ptr</tt>.</p>
6756 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
6757 the <tt>val</tt> argument and the result must be integers of the same bit
6758 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6759 integer type. The targets may only lower integer representations they
6763 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6764 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6765 equivalent of an atomic swap operation within the SSA framework.</p>
6769 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6770 %ptr = bitcast i8* %mallocP to i32*
6773 %val1 = add i32 4, 4
6774 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6775 <i>; yields {i32}:result1 = 4</i>
6776 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6777 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6779 %val2 = add i32 1, 1
6780 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6781 <i>; yields {i32}:result2 = 8</i>
6783 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6784 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6789 <!-- _______________________________________________________________________ -->
6790 <div class="doc_subsubsection">
6791 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6795 <div class="doc_text">
6798 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
6799 any integer bit width. Not all targets support all bit widths however.</p>
6802 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6803 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6804 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6805 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6809 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
6810 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
6813 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6814 and the second an integer value. The result is also an integer value. These
6815 integer types can have any bit width, but they must all have the same bit
6816 width. The targets may only lower integer representations they support.</p>
6819 <p>This intrinsic does a series of operations atomically. It first loads the
6820 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6821 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
6825 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6826 %ptr = bitcast i8* %mallocP to i32*
6828 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6829 <i>; yields {i32}:result1 = 4</i>
6830 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6831 <i>; yields {i32}:result2 = 8</i>
6832 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6833 <i>; yields {i32}:result3 = 10</i>
6834 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6839 <!-- _______________________________________________________________________ -->
6840 <div class="doc_subsubsection">
6841 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6845 <div class="doc_text">
6848 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6849 any integer bit width and for different address spaces. Not all targets
6850 support all bit widths however.</p>
6853 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6854 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6855 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6856 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6860 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6861 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
6864 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6865 and the second an integer value. The result is also an integer value. These
6866 integer types can have any bit width, but they must all have the same bit
6867 width. The targets may only lower integer representations they support.</p>
6870 <p>This intrinsic does a series of operations atomically. It first loads the
6871 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6872 result to <tt>ptr</tt>. It yields the original value stored
6873 at <tt>ptr</tt>.</p>
6877 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6878 %ptr = bitcast i8* %mallocP to i32*
6880 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6881 <i>; yields {i32}:result1 = 8</i>
6882 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6883 <i>; yields {i32}:result2 = 4</i>
6884 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6885 <i>; yields {i32}:result3 = 2</i>
6886 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6891 <!-- _______________________________________________________________________ -->
6892 <div class="doc_subsubsection">
6893 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6894 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6895 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6896 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6899 <div class="doc_text">
6902 <p>These are overloaded intrinsics. You can
6903 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
6904 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
6905 bit width and for different address spaces. Not all targets support all bit
6909 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6910 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6911 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6912 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6916 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6917 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6918 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6919 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6923 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6924 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6925 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6926 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6930 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6931 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6932 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6933 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6937 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6938 the value stored in memory at <tt>ptr</tt>. It yields the original value
6939 at <tt>ptr</tt>.</p>
6942 <p>These intrinsics take two arguments, the first a pointer to an integer value
6943 and the second an integer value. The result is also an integer value. These
6944 integer types can have any bit width, but they must all have the same bit
6945 width. The targets may only lower integer representations they support.</p>
6948 <p>These intrinsics does a series of operations atomically. They first load the
6949 value stored at <tt>ptr</tt>. They then do the bitwise
6950 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
6951 original value stored at <tt>ptr</tt>.</p>
6955 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6956 %ptr = bitcast i8* %mallocP to i32*
6957 store i32 0x0F0F, %ptr
6958 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6959 <i>; yields {i32}:result0 = 0x0F0F</i>
6960 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6961 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6962 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6963 <i>; yields {i32}:result2 = 0xF0</i>
6964 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6965 <i>; yields {i32}:result3 = FF</i>
6966 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6971 <!-- _______________________________________________________________________ -->
6972 <div class="doc_subsubsection">
6973 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6974 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6975 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6976 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6979 <div class="doc_text">
6982 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6983 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6984 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6985 address spaces. Not all targets support all bit widths however.</p>
6988 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6989 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6990 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6991 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6995 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6996 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6997 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6998 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
7002 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
7003 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
7004 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
7005 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
7009 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
7010 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
7011 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
7012 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
7016 <p>These intrinsics takes the signed or unsigned minimum or maximum of
7017 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
7018 original value at <tt>ptr</tt>.</p>
7021 <p>These intrinsics take two arguments, the first a pointer to an integer value
7022 and the second an integer value. The result is also an integer value. These
7023 integer types can have any bit width, but they must all have the same bit
7024 width. The targets may only lower integer representations they support.</p>
7027 <p>These intrinsics does a series of operations atomically. They first load the
7028 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
7029 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
7030 yield the original value stored at <tt>ptr</tt>.</p>
7034 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7035 %ptr = bitcast i8* %mallocP to i32*
7037 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
7038 <i>; yields {i32}:result0 = 7</i>
7039 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
7040 <i>; yields {i32}:result1 = -2</i>
7041 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
7042 <i>; yields {i32}:result2 = 8</i>
7043 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
7044 <i>; yields {i32}:result3 = 8</i>
7045 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
7051 <!-- ======================================================================= -->
7052 <div class="doc_subsection">
7053 <a name="int_memorymarkers">Memory Use Markers</a>
7056 <div class="doc_text">
7058 <p>This class of intrinsics exists to information about the lifetime of memory
7059 objects and ranges where variables are immutable.</p>
7063 <!-- _______________________________________________________________________ -->
7064 <div class="doc_subsubsection">
7065 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
7068 <div class="doc_text">
7072 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
7076 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
7077 object's lifetime.</p>
7080 <p>The first argument is a constant integer representing the size of the
7081 object, or -1 if it is variable sized. The second argument is a pointer to
7085 <p>This intrinsic indicates that before this point in the code, the value of the
7086 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7087 never be used and has an undefined value. A load from the pointer that
7088 precedes this intrinsic can be replaced with
7089 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
7093 <!-- _______________________________________________________________________ -->
7094 <div class="doc_subsubsection">
7095 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
7098 <div class="doc_text">
7102 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
7106 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
7107 object's lifetime.</p>
7110 <p>The first argument is a constant integer representing the size of the
7111 object, or -1 if it is variable sized. The second argument is a pointer to
7115 <p>This intrinsic indicates that after this point in the code, the value of the
7116 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7117 never be used and has an undefined value. Any stores into the memory object
7118 following this intrinsic may be removed as dead.
7122 <!-- _______________________________________________________________________ -->
7123 <div class="doc_subsubsection">
7124 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
7127 <div class="doc_text">
7131 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>) readonly
7135 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
7136 a memory object will not change.</p>
7139 <p>The first argument is a constant integer representing the size of the
7140 object, or -1 if it is variable sized. The second argument is a pointer to
7144 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
7145 the return value, the referenced memory location is constant and
7150 <!-- _______________________________________________________________________ -->
7151 <div class="doc_subsubsection">
7152 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
7155 <div class="doc_text">
7159 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7163 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
7164 a memory object are mutable.</p>
7167 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
7168 The second argument is a constant integer representing the size of the
7169 object, or -1 if it is variable sized and the third argument is a pointer
7173 <p>This intrinsic indicates that the memory is mutable again.</p>
7177 <!-- ======================================================================= -->
7178 <div class="doc_subsection">
7179 <a name="int_general">General Intrinsics</a>
7182 <div class="doc_text">
7184 <p>This class of intrinsics is designed to be generic and has no specific
7189 <!-- _______________________________________________________________________ -->
7190 <div class="doc_subsubsection">
7191 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
7194 <div class="doc_text">
7198 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
7202 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
7205 <p>The first argument is a pointer to a value, the second is a pointer to a
7206 global string, the third is a pointer to a global string which is the source
7207 file name, and the last argument is the line number.</p>
7210 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7211 This can be useful for special purpose optimizations that want to look for
7212 these annotations. These have no other defined use, they are ignored by code
7213 generation and optimization.</p>
7217 <!-- _______________________________________________________________________ -->
7218 <div class="doc_subsubsection">
7219 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
7222 <div class="doc_text">
7225 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
7226 any integer bit width.</p>
7229 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
7230 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
7231 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
7232 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
7233 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
7237 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
7240 <p>The first argument is an integer value (result of some expression), the
7241 second is a pointer to a global string, the third is a pointer to a global
7242 string which is the source file name, and the last argument is the line
7243 number. It returns the value of the first argument.</p>
7246 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7247 arbitrary strings. This can be useful for special purpose optimizations that
7248 want to look for these annotations. These have no other defined use, they
7249 are ignored by code generation and optimization.</p>
7253 <!-- _______________________________________________________________________ -->
7254 <div class="doc_subsubsection">
7255 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
7258 <div class="doc_text">
7262 declare void @llvm.trap()
7266 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
7272 <p>This intrinsics is lowered to the target dependent trap instruction. If the
7273 target does not have a trap instruction, this intrinsic will be lowered to
7274 the call of the <tt>abort()</tt> function.</p>
7278 <!-- _______________________________________________________________________ -->
7279 <div class="doc_subsubsection">
7280 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
7283 <div class="doc_text">
7287 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
7291 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
7292 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
7293 ensure that it is placed on the stack before local variables.</p>
7296 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
7297 arguments. The first argument is the value loaded from the stack
7298 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
7299 that has enough space to hold the value of the guard.</p>
7302 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
7303 the <tt>AllocaInst</tt> stack slot to be before local variables on the
7304 stack. This is to ensure that if a local variable on the stack is
7305 overwritten, it will destroy the value of the guard. When the function exits,
7306 the guard on the stack is checked against the original guard. If they're
7307 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
7312 <!-- _______________________________________________________________________ -->
7313 <div class="doc_subsubsection">
7314 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
7317 <div class="doc_text">
7321 declare i32 @llvm.objectsize.i32( i8* <object>, i1 <type> )
7322 declare i64 @llvm.objectsize.i64( i8* <object>, i1 <type> )
7326 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information
7327 to the optimizers to discover at compile time either a) when an
7328 operation like memcpy will either overflow a buffer that corresponds to
7329 an object, or b) to determine that a runtime check for overflow isn't
7330 necessary. An object in this context means an allocation of a
7331 specific class, structure, array, or other object.</p>
7334 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
7335 argument is a pointer to or into the <tt>object</tt>. The second argument
7336 is a boolean 0 or 1. This argument determines whether you want the
7337 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
7338 1, variables are not allowed.</p>
7341 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
7342 representing the size of the object concerned or <tt>i32/i64 -1 or 0</tt>
7343 (depending on the <tt>type</tt> argument if the size cannot be determined
7344 at compile time.</p>
7348 <!-- *********************************************************************** -->
7351 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
7352 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
7353 <a href="http://validator.w3.org/check/referer"><img
7354 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
7356 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
7357 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
7358 Last modified: $Date$