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 <h1>LLVM Language Reference Manual</h1>
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_linker_private_weak">'<tt>linker_private_weak</tt>' Linkage</a></li>
28 <li><a href="#linkage_linker_private_weak_def_auto">'<tt>linker_private_weak_def_auto</tt>' Linkage</a></li>
29 <li><a href="#linkage_internal">'<tt>internal</tt>' Linkage</a></li>
30 <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
31 <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
32 <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
33 <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
34 <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
35 <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
36 <li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
37 <li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
38 <li><a href="#linkage_external">'<tt>external</tt>' Linkage</a></li>
39 <li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
40 <li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
43 <li><a href="#callingconv">Calling Conventions</a></li>
44 <li><a href="#namedtypes">Named Types</a></li>
45 <li><a href="#globalvars">Global Variables</a></li>
46 <li><a href="#functionstructure">Functions</a></li>
47 <li><a href="#aliasstructure">Aliases</a></li>
48 <li><a href="#namedmetadatastructure">Named Metadata</a></li>
49 <li><a href="#paramattrs">Parameter Attributes</a></li>
50 <li><a href="#fnattrs">Function Attributes</a></li>
51 <li><a href="#gc">Garbage Collector Names</a></li>
52 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
53 <li><a href="#datalayout">Data Layout</a></li>
54 <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
55 <li><a href="#volatile">Volatile Memory Accesses</a></li>
56 <li><a href="#memmodel">Memory Model for Concurrent Operations</a></li>
57 <li><a href="#ordering">Atomic Memory Ordering Constraints</a></li>
60 <li><a href="#typesystem">Type System</a>
62 <li><a href="#t_classifications">Type Classifications</a></li>
63 <li><a href="#t_primitive">Primitive Types</a>
65 <li><a href="#t_integer">Integer Type</a></li>
66 <li><a href="#t_floating">Floating Point Types</a></li>
67 <li><a href="#t_x86mmx">X86mmx Type</a></li>
68 <li><a href="#t_void">Void Type</a></li>
69 <li><a href="#t_label">Label Type</a></li>
70 <li><a href="#t_metadata">Metadata Type</a></li>
73 <li><a href="#t_derived">Derived Types</a>
75 <li><a href="#t_aggregate">Aggregate Types</a>
77 <li><a href="#t_array">Array Type</a></li>
78 <li><a href="#t_struct">Structure Type</a></li>
79 <li><a href="#t_opaque">Opaque Structure Types</a></li>
80 <li><a href="#t_vector">Vector Type</a></li>
83 <li><a href="#t_function">Function Type</a></li>
84 <li><a href="#t_pointer">Pointer Type</a></li>
89 <li><a href="#constants">Constants</a>
91 <li><a href="#simpleconstants">Simple Constants</a></li>
92 <li><a href="#complexconstants">Complex Constants</a></li>
93 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
94 <li><a href="#undefvalues">Undefined Values</a></li>
95 <li><a href="#trapvalues">Trap Values</a></li>
96 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
97 <li><a href="#constantexprs">Constant Expressions</a></li>
100 <li><a href="#othervalues">Other Values</a>
102 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
103 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a>
105 <li><a href="#tbaa">'<tt>tbaa</tt>' Metadata</a></li>
110 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
112 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
113 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
114 Global Variable</a></li>
115 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
116 Global Variable</a></li>
117 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
118 Global Variable</a></li>
121 <li><a href="#instref">Instruction Reference</a>
123 <li><a href="#terminators">Terminator Instructions</a>
125 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
126 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
127 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
128 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
129 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
130 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
131 <li><a href="#i_resume">'<tt>resume</tt>' Instruction</a></li>
132 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
135 <li><a href="#binaryops">Binary Operations</a>
137 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
138 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
139 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
140 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
141 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
142 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
143 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
144 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
145 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
146 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
147 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
148 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
151 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
153 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
154 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
155 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
156 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
157 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
158 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
161 <li><a href="#vectorops">Vector Operations</a>
163 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
164 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
165 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
168 <li><a href="#aggregateops">Aggregate Operations</a>
170 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
171 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
174 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
176 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
177 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
178 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
179 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
180 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
181 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
182 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
185 <li><a href="#convertops">Conversion Operations</a>
187 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
188 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
189 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
190 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
191 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
192 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
193 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
194 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
195 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
196 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
197 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
198 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
201 <li><a href="#otherops">Other Operations</a>
203 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
204 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
205 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
206 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
207 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
208 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
209 <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
214 <li><a href="#intrinsics">Intrinsic Functions</a>
216 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
218 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
219 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
220 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
223 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
225 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
226 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
227 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
230 <li><a href="#int_codegen">Code Generator Intrinsics</a>
232 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
233 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
234 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
235 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
236 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
237 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
238 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
241 <li><a href="#int_libc">Standard C Library Intrinsics</a>
243 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
244 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
245 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
246 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
247 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
248 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
249 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
250 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
251 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
252 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
253 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
256 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
258 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
259 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
260 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
261 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
264 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
266 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
267 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
268 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
269 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
270 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
271 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
274 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
276 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
277 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
280 <li><a href="#int_debugger">Debugger intrinsics</a></li>
281 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
282 <li><a href="#int_trampoline">Trampoline Intrinsics</a>
284 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
285 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
288 <li><a href="#int_memorymarkers">Memory Use Markers</a>
290 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
291 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
292 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
293 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
296 <li><a href="#int_general">General intrinsics</a>
298 <li><a href="#int_var_annotation">
299 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
300 <li><a href="#int_annotation">
301 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
302 <li><a href="#int_trap">
303 '<tt>llvm.trap</tt>' Intrinsic</a></li>
304 <li><a href="#int_stackprotector">
305 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
306 <li><a href="#int_objectsize">
307 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
314 <div class="doc_author">
315 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
316 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
319 <!-- *********************************************************************** -->
320 <h2><a name="abstract">Abstract</a></h2>
321 <!-- *********************************************************************** -->
325 <p>This document is a reference manual for the LLVM assembly language. LLVM is
326 a Static Single Assignment (SSA) based representation that provides type
327 safety, low-level operations, flexibility, and the capability of representing
328 'all' high-level languages cleanly. It is the common code representation
329 used throughout all phases of the LLVM compilation strategy.</p>
333 <!-- *********************************************************************** -->
334 <h2><a name="introduction">Introduction</a></h2>
335 <!-- *********************************************************************** -->
339 <p>The LLVM code representation is designed to be used in three different forms:
340 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
341 for fast loading by a Just-In-Time compiler), and as a human readable
342 assembly language representation. This allows LLVM to provide a powerful
343 intermediate representation for efficient compiler transformations and
344 analysis, while providing a natural means to debug and visualize the
345 transformations. The three different forms of LLVM are all equivalent. This
346 document describes the human readable representation and notation.</p>
348 <p>The LLVM representation aims to be light-weight and low-level while being
349 expressive, typed, and extensible at the same time. It aims to be a
350 "universal IR" of sorts, by being at a low enough level that high-level ideas
351 may be cleanly mapped to it (similar to how microprocessors are "universal
352 IR's", allowing many source languages to be mapped to them). By providing
353 type information, LLVM can be used as the target of optimizations: for
354 example, through pointer analysis, it can be proven that a C automatic
355 variable is never accessed outside of the current function, allowing it to
356 be promoted to a simple SSA value instead of a memory location.</p>
358 <!-- _______________________________________________________________________ -->
360 <a name="wellformed">Well-Formedness</a>
365 <p>It is important to note that this document describes 'well formed' LLVM
366 assembly language. There is a difference between what the parser accepts and
367 what is considered 'well formed'. For example, the following instruction is
368 syntactically okay, but not well formed:</p>
370 <pre class="doc_code">
371 %x = <a href="#i_add">add</a> i32 1, %x
374 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
375 LLVM infrastructure provides a verification pass that may be used to verify
376 that an LLVM module is well formed. This pass is automatically run by the
377 parser after parsing input assembly and by the optimizer before it outputs
378 bitcode. The violations pointed out by the verifier pass indicate bugs in
379 transformation passes or input to the parser.</p>
385 <!-- Describe the typesetting conventions here. -->
387 <!-- *********************************************************************** -->
388 <h2><a name="identifiers">Identifiers</a></h2>
389 <!-- *********************************************************************** -->
393 <p>LLVM identifiers come in two basic types: global and local. Global
394 identifiers (functions, global variables) begin with the <tt>'@'</tt>
395 character. Local identifiers (register names, types) begin with
396 the <tt>'%'</tt> character. Additionally, there are three different formats
397 for identifiers, for different purposes:</p>
400 <li>Named values are represented as a string of characters with their prefix.
401 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
402 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
403 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
404 other characters in their names can be surrounded with quotes. Special
405 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
406 ASCII code for the character in hexadecimal. In this way, any character
407 can be used in a name value, even quotes themselves.</li>
409 <li>Unnamed values are represented as an unsigned numeric value with their
410 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
412 <li>Constants, which are described in a <a href="#constants">section about
413 constants</a>, below.</li>
416 <p>LLVM requires that values start with a prefix for two reasons: Compilers
417 don't need to worry about name clashes with reserved words, and the set of
418 reserved words may be expanded in the future without penalty. Additionally,
419 unnamed identifiers allow a compiler to quickly come up with a temporary
420 variable without having to avoid symbol table conflicts.</p>
422 <p>Reserved words in LLVM are very similar to reserved words in other
423 languages. There are keywords for different opcodes
424 ('<tt><a href="#i_add">add</a></tt>',
425 '<tt><a href="#i_bitcast">bitcast</a></tt>',
426 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
427 ('<tt><a href="#t_void">void</a></tt>',
428 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
429 reserved words cannot conflict with variable names, because none of them
430 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
432 <p>Here is an example of LLVM code to multiply the integer variable
433 '<tt>%X</tt>' by 8:</p>
437 <pre class="doc_code">
438 %result = <a href="#i_mul">mul</a> i32 %X, 8
441 <p>After strength reduction:</p>
443 <pre class="doc_code">
444 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
447 <p>And the hard way:</p>
449 <pre class="doc_code">
450 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
451 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
452 %result = <a href="#i_add">add</a> i32 %1, %1
455 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
456 lexical features of LLVM:</p>
459 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
462 <li>Unnamed temporaries are created when the result of a computation is not
463 assigned to a named value.</li>
465 <li>Unnamed temporaries are numbered sequentially</li>
468 <p>It also shows a convention that we follow in this document. When
469 demonstrating instructions, we will follow an instruction with a comment that
470 defines the type and name of value produced. Comments are shown in italic
475 <!-- *********************************************************************** -->
476 <h2><a name="highlevel">High Level Structure</a></h2>
477 <!-- *********************************************************************** -->
479 <!-- ======================================================================= -->
481 <a name="modulestructure">Module Structure</a>
486 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
487 of the input programs. Each module consists of functions, global variables,
488 and symbol table entries. Modules may be combined together with the LLVM
489 linker, which merges function (and global variable) definitions, resolves
490 forward declarations, and merges symbol table entries. Here is an example of
491 the "hello world" module:</p>
493 <pre class="doc_code">
494 <i>; Declare the string constant as a global constant.</i>
495 <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>
497 <i>; External declaration of the puts function</i>
498 <a href="#functionstructure">declare</a> i32 @puts(i8*) <i>; i32 (i8*)* </i>
500 <i>; Definition of main function</i>
501 define i32 @main() { <i>; i32()* </i>
502 <i>; Convert [13 x i8]* to i8 *...</i>
503 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8*</i>
505 <i>; Call puts function to write out the string to stdout.</i>
506 <a href="#i_call">call</a> i32 @puts(i8* %cast210) <i>; i32</i>
507 <a href="#i_ret">ret</a> i32 0
510 <i>; Named metadata</i>
511 !1 = metadata !{i32 41}
515 <p>This example is made up of a <a href="#globalvars">global variable</a> named
516 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
517 a <a href="#functionstructure">function definition</a> for
518 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
521 <p>In general, a module is made up of a list of global values, where both
522 functions and global variables are global values. Global values are
523 represented by a pointer to a memory location (in this case, a pointer to an
524 array of char, and a pointer to a function), and have one of the
525 following <a href="#linkage">linkage types</a>.</p>
529 <!-- ======================================================================= -->
531 <a name="linkage">Linkage Types</a>
536 <p>All Global Variables and Functions have one of the following types of
540 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
541 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
542 by objects in the current module. In particular, linking code into a
543 module with an private global value may cause the private to be renamed as
544 necessary to avoid collisions. Because the symbol is private to the
545 module, all references can be updated. This doesn't show up in any symbol
546 table in the object file.</dd>
548 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
549 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
550 assembler and evaluated by the linker. Unlike normal strong symbols, they
551 are removed by the linker from the final linked image (executable or
552 dynamic library).</dd>
554 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
555 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
556 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
557 linker. The symbols are removed by the linker from the final linked image
558 (executable or dynamic library).</dd>
560 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
561 <dd>Similar to "<tt>linker_private_weak</tt>", but it's known that the address
562 of the object is not taken. For instance, functions that had an inline
563 definition, but the compiler decided not to inline it. Note,
564 unlike <tt>linker_private</tt> and <tt>linker_private_weak</tt>,
565 <tt>linker_private_weak_def_auto</tt> may have only <tt>default</tt>
566 visibility. The symbols are removed by the linker from the final linked
567 image (executable or dynamic library).</dd>
569 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
570 <dd>Similar to private, but the value shows as a local symbol
571 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
572 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
574 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
575 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
576 into the object file corresponding to the LLVM module. They exist to
577 allow inlining and other optimizations to take place given knowledge of
578 the definition of the global, which is known to be somewhere outside the
579 module. Globals with <tt>available_externally</tt> linkage are allowed to
580 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
581 This linkage type is only allowed on definitions, not declarations.</dd>
583 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
584 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
585 the same name when linkage occurs. This can be used to implement
586 some forms of inline functions, templates, or other code which must be
587 generated in each translation unit that uses it, but where the body may
588 be overridden with a more definitive definition later. Unreferenced
589 <tt>linkonce</tt> globals are allowed to be discarded. Note that
590 <tt>linkonce</tt> linkage does not actually allow the optimizer to
591 inline the body of this function into callers because it doesn't know if
592 this definition of the function is the definitive definition within the
593 program or whether it will be overridden by a stronger definition.
594 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
597 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
598 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
599 <tt>linkonce</tt> linkage, except that unreferenced globals with
600 <tt>weak</tt> linkage may not be discarded. This is used for globals that
601 are declared "weak" in C source code.</dd>
603 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
604 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
605 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
607 Symbols with "<tt>common</tt>" linkage are merged in the same way as
608 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
609 <tt>common</tt> symbols may not have an explicit section,
610 must have a zero initializer, and may not be marked '<a
611 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
612 have common linkage.</dd>
615 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
616 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
617 pointer to array type. When two global variables with appending linkage
618 are linked together, the two global arrays are appended together. This is
619 the LLVM, typesafe, equivalent of having the system linker append together
620 "sections" with identical names when .o files are linked.</dd>
622 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
623 <dd>The semantics of this linkage follow the ELF object file model: the symbol
624 is weak until linked, if not linked, the symbol becomes null instead of
625 being an undefined reference.</dd>
627 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
628 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
629 <dd>Some languages allow differing globals to be merged, such as two functions
630 with different semantics. Other languages, such as <tt>C++</tt>, ensure
631 that only equivalent globals are ever merged (the "one definition rule"
632 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
633 and <tt>weak_odr</tt> linkage types to indicate that the global will only
634 be merged with equivalent globals. These linkage types are otherwise the
635 same as their non-<tt>odr</tt> versions.</dd>
637 <dt><tt><b><a name="linkage_external">external</a></b></tt>:</dt>
638 <dd>If none of the above identifiers are used, the global is externally
639 visible, meaning that it participates in linkage and can be used to
640 resolve external symbol references.</dd>
643 <p>The next two types of linkage are targeted for Microsoft Windows platform
644 only. They are designed to support importing (exporting) symbols from (to)
645 DLLs (Dynamic Link Libraries).</p>
648 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
649 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
650 or variable via a global pointer to a pointer that is set up by the DLL
651 exporting the symbol. On Microsoft Windows targets, the pointer name is
652 formed by combining <code>__imp_</code> and the function or variable
655 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
656 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
657 pointer to a pointer in a DLL, so that it can be referenced with the
658 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
659 name is formed by combining <code>__imp_</code> and the function or
663 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
664 another module defined a "<tt>.LC0</tt>" variable and was linked with this
665 one, one of the two would be renamed, preventing a collision. Since
666 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
667 declarations), they are accessible outside of the current module.</p>
669 <p>It is illegal for a function <i>declaration</i> to have any linkage type
670 other than <tt>external</tt>, <tt>dllimport</tt>
671 or <tt>extern_weak</tt>.</p>
673 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
674 or <tt>weak_odr</tt> linkages.</p>
678 <!-- ======================================================================= -->
680 <a name="callingconv">Calling Conventions</a>
685 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
686 and <a href="#i_invoke">invokes</a> can all have an optional calling
687 convention specified for the call. The calling convention of any pair of
688 dynamic caller/callee must match, or the behavior of the program is
689 undefined. The following calling conventions are supported by LLVM, and more
690 may be added in the future:</p>
693 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
694 <dd>This calling convention (the default if no other calling convention is
695 specified) matches the target C calling conventions. This calling
696 convention supports varargs function calls and tolerates some mismatch in
697 the declared prototype and implemented declaration of the function (as
700 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
701 <dd>This calling convention attempts to make calls as fast as possible
702 (e.g. by passing things in registers). This calling convention allows the
703 target to use whatever tricks it wants to produce fast code for the
704 target, without having to conform to an externally specified ABI
705 (Application Binary Interface).
706 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
707 when this or the GHC convention is used.</a> This calling convention
708 does not support varargs and requires the prototype of all callees to
709 exactly match the prototype of the function definition.</dd>
711 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
712 <dd>This calling convention attempts to make code in the caller as efficient
713 as possible under the assumption that the call is not commonly executed.
714 As such, these calls often preserve all registers so that the call does
715 not break any live ranges in the caller side. This calling convention
716 does not support varargs and requires the prototype of all callees to
717 exactly match the prototype of the function definition.</dd>
719 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
720 <dd>This calling convention has been implemented specifically for use by the
721 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
722 It passes everything in registers, going to extremes to achieve this by
723 disabling callee save registers. This calling convention should not be
724 used lightly but only for specific situations such as an alternative to
725 the <em>register pinning</em> performance technique often used when
726 implementing functional programming languages.At the moment only X86
727 supports this convention and it has the following limitations:
729 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
730 floating point types are supported.</li>
731 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
732 6 floating point parameters.</li>
734 This calling convention supports
735 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
736 requires both the caller and callee are using it.
739 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
740 <dd>Any calling convention may be specified by number, allowing
741 target-specific calling conventions to be used. Target specific calling
742 conventions start at 64.</dd>
745 <p>More calling conventions can be added/defined on an as-needed basis, to
746 support Pascal conventions or any other well-known target-independent
751 <!-- ======================================================================= -->
753 <a name="visibility">Visibility Styles</a>
758 <p>All Global Variables and Functions have one of the following visibility
762 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
763 <dd>On targets that use the ELF object file format, default visibility means
764 that the declaration is visible to other modules and, in shared libraries,
765 means that the declared entity may be overridden. On Darwin, default
766 visibility means that the declaration is visible to other modules. Default
767 visibility corresponds to "external linkage" in the language.</dd>
769 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
770 <dd>Two declarations of an object with hidden visibility refer to the same
771 object if they are in the same shared object. Usually, hidden visibility
772 indicates that the symbol will not be placed into the dynamic symbol
773 table, so no other module (executable or shared library) can reference it
776 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
777 <dd>On ELF, protected visibility indicates that the symbol will be placed in
778 the dynamic symbol table, but that references within the defining module
779 will bind to the local symbol. That is, the symbol cannot be overridden by
785 <!-- ======================================================================= -->
787 <a name="namedtypes">Named Types</a>
792 <p>LLVM IR allows you to specify name aliases for certain types. This can make
793 it easier to read the IR and make the IR more condensed (particularly when
794 recursive types are involved). An example of a name specification is:</p>
796 <pre class="doc_code">
797 %mytype = type { %mytype*, i32 }
800 <p>You may give a name to any <a href="#typesystem">type</a> except
801 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
802 is expected with the syntax "%mytype".</p>
804 <p>Note that type names are aliases for the structural type that they indicate,
805 and that you can therefore specify multiple names for the same type. This
806 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
807 uses structural typing, the name is not part of the type. When printing out
808 LLVM IR, the printer will pick <em>one name</em> to render all types of a
809 particular shape. This means that if you have code where two different
810 source types end up having the same LLVM type, that the dumper will sometimes
811 print the "wrong" or unexpected type. This is an important design point and
812 isn't going to change.</p>
816 <!-- ======================================================================= -->
818 <a name="globalvars">Global Variables</a>
823 <p>Global variables define regions of memory allocated at compilation time
824 instead of run-time. Global variables may optionally be initialized, may
825 have an explicit section to be placed in, and may have an optional explicit
826 alignment specified. A variable may be defined as "thread_local", which
827 means that it will not be shared by threads (each thread will have a
828 separated copy of the variable). A variable may be defined as a global
829 "constant," which indicates that the contents of the variable
830 will <b>never</b> be modified (enabling better optimization, allowing the
831 global data to be placed in the read-only section of an executable, etc).
832 Note that variables that need runtime initialization cannot be marked
833 "constant" as there is a store to the variable.</p>
835 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
836 constant, even if the final definition of the global is not. This capability
837 can be used to enable slightly better optimization of the program, but
838 requires the language definition to guarantee that optimizations based on the
839 'constantness' are valid for the translation units that do not include the
842 <p>As SSA values, global variables define pointer values that are in scope
843 (i.e. they dominate) all basic blocks in the program. Global variables
844 always define a pointer to their "content" type because they describe a
845 region of memory, and all memory objects in LLVM are accessed through
848 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
849 that the address is not significant, only the content. Constants marked
850 like this can be merged with other constants if they have the same
851 initializer. Note that a constant with significant address <em>can</em>
852 be merged with a <tt>unnamed_addr</tt> constant, the result being a
853 constant whose address is significant.</p>
855 <p>A global variable may be declared to reside in a target-specific numbered
856 address space. For targets that support them, address spaces may affect how
857 optimizations are performed and/or what target instructions are used to
858 access the variable. The default address space is zero. The address space
859 qualifier must precede any other attributes.</p>
861 <p>LLVM allows an explicit section to be specified for globals. If the target
862 supports it, it will emit globals to the section specified.</p>
864 <p>An explicit alignment may be specified for a global, which must be a power
865 of 2. If not present, or if the alignment is set to zero, the alignment of
866 the global is set by the target to whatever it feels convenient. If an
867 explicit alignment is specified, the global is forced to have exactly that
868 alignment. Targets and optimizers are not allowed to over-align the global
869 if the global has an assigned section. In this case, the extra alignment
870 could be observable: for example, code could assume that the globals are
871 densely packed in their section and try to iterate over them as an array,
872 alignment padding would break this iteration.</p>
874 <p>For example, the following defines a global in a numbered address space with
875 an initializer, section, and alignment:</p>
877 <pre class="doc_code">
878 @G = addrspace(5) constant float 1.0, section "foo", align 4
884 <!-- ======================================================================= -->
886 <a name="functionstructure">Functions</a>
891 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
892 optional <a href="#linkage">linkage type</a>, an optional
893 <a href="#visibility">visibility style</a>, an optional
894 <a href="#callingconv">calling convention</a>,
895 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
896 <a href="#paramattrs">parameter attribute</a> for the return type, a function
897 name, a (possibly empty) argument list (each with optional
898 <a href="#paramattrs">parameter attributes</a>), optional
899 <a href="#fnattrs">function attributes</a>, an optional section, an optional
900 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
901 curly brace, a list of basic blocks, and a closing curly brace.</p>
903 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
904 optional <a href="#linkage">linkage type</a>, an optional
905 <a href="#visibility">visibility style</a>, an optional
906 <a href="#callingconv">calling convention</a>,
907 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
908 <a href="#paramattrs">parameter attribute</a> for the return type, a function
909 name, a possibly empty list of arguments, an optional alignment, and an
910 optional <a href="#gc">garbage collector name</a>.</p>
912 <p>A function definition contains a list of basic blocks, forming the CFG
913 (Control Flow Graph) for the function. Each basic block may optionally start
914 with a label (giving the basic block a symbol table entry), contains a list
915 of instructions, and ends with a <a href="#terminators">terminator</a>
916 instruction (such as a branch or function return).</p>
918 <p>The first basic block in a function is special in two ways: it is immediately
919 executed on entrance to the function, and it is not allowed to have
920 predecessor basic blocks (i.e. there can not be any branches to the entry
921 block of a function). Because the block can have no predecessors, it also
922 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
924 <p>LLVM allows an explicit section to be specified for functions. If the target
925 supports it, it will emit functions to the section specified.</p>
927 <p>An explicit alignment may be specified for a function. If not present, or if
928 the alignment is set to zero, the alignment of the function is set by the
929 target to whatever it feels convenient. If an explicit alignment is
930 specified, the function is forced to have at least that much alignment. All
931 alignments must be a power of 2.</p>
933 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
934 be significant and two identical functions can be merged</p>.
937 <pre class="doc_code">
938 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
939 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
940 <ResultType> @<FunctionName> ([argument list])
941 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
942 [<a href="#gc">gc</a>] { ... }
947 <!-- ======================================================================= -->
949 <a name="aliasstructure">Aliases</a>
954 <p>Aliases act as "second name" for the aliasee value (which can be either
955 function, global variable, another alias or bitcast of global value). Aliases
956 may have an optional <a href="#linkage">linkage type</a>, and an
957 optional <a href="#visibility">visibility style</a>.</p>
960 <pre class="doc_code">
961 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
966 <!-- ======================================================================= -->
968 <a name="namedmetadatastructure">Named Metadata</a>
973 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
974 nodes</a> (but not metadata strings) are the only valid operands for
975 a named metadata.</p>
978 <pre class="doc_code">
979 ; Some unnamed metadata nodes, which are referenced by the named metadata.
980 !0 = metadata !{metadata !"zero"}
981 !1 = metadata !{metadata !"one"}
982 !2 = metadata !{metadata !"two"}
984 !name = !{!0, !1, !2}
989 <!-- ======================================================================= -->
991 <a name="paramattrs">Parameter Attributes</a>
996 <p>The return type and each parameter of a function type may have a set of
997 <i>parameter attributes</i> associated with them. Parameter attributes are
998 used to communicate additional information about the result or parameters of
999 a function. Parameter attributes are considered to be part of the function,
1000 not of the function type, so functions with different parameter attributes
1001 can have the same function type.</p>
1003 <p>Parameter attributes are simple keywords that follow the type specified. If
1004 multiple parameter attributes are needed, they are space separated. For
1007 <pre class="doc_code">
1008 declare i32 @printf(i8* noalias nocapture, ...)
1009 declare i32 @atoi(i8 zeroext)
1010 declare signext i8 @returns_signed_char()
1013 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1014 <tt>readonly</tt>) come immediately after the argument list.</p>
1016 <p>Currently, only the following parameter attributes are defined:</p>
1019 <dt><tt><b>zeroext</b></tt></dt>
1020 <dd>This indicates to the code generator that the parameter or return value
1021 should be zero-extended to the extent required by the target's ABI (which
1022 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1023 parameter) or the callee (for a return value).</dd>
1025 <dt><tt><b>signext</b></tt></dt>
1026 <dd>This indicates to the code generator that the parameter or return value
1027 should be sign-extended to the extent required by the target's ABI (which
1028 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1031 <dt><tt><b>inreg</b></tt></dt>
1032 <dd>This indicates that this parameter or return value should be treated in a
1033 special target-dependent fashion during while emitting code for a function
1034 call or return (usually, by putting it in a register as opposed to memory,
1035 though some targets use it to distinguish between two different kinds of
1036 registers). Use of this attribute is target-specific.</dd>
1038 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1039 <dd><p>This indicates that the pointer parameter should really be passed by
1040 value to the function. The attribute implies that a hidden copy of the
1042 is made between the caller and the callee, so the callee is unable to
1043 modify the value in the callee. This attribute is only valid on LLVM
1044 pointer arguments. It is generally used to pass structs and arrays by
1045 value, but is also valid on pointers to scalars. The copy is considered
1046 to belong to the caller not the callee (for example,
1047 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1048 <tt>byval</tt> parameters). This is not a valid attribute for return
1051 <p>The byval attribute also supports specifying an alignment with
1052 the align attribute. It indicates the alignment of the stack slot to
1053 form and the known alignment of the pointer specified to the call site. If
1054 the alignment is not specified, then the code generator makes a
1055 target-specific assumption.</p></dd>
1057 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1058 <dd>This indicates that the pointer parameter specifies the address of a
1059 structure that is the return value of the function in the source program.
1060 This pointer must be guaranteed by the caller to be valid: loads and
1061 stores to the structure may be assumed by the callee to not to trap. This
1062 may only be applied to the first parameter. This is not a valid attribute
1063 for return values. </dd>
1065 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1066 <dd>This indicates that pointer values
1067 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1068 value do not alias pointer values which are not <i>based</i> on it,
1069 ignoring certain "irrelevant" dependencies.
1070 For a call to the parent function, dependencies between memory
1071 references from before or after the call and from those during the call
1072 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1073 return value used in that call.
1074 The caller shares the responsibility with the callee for ensuring that
1075 these requirements are met.
1076 For further details, please see the discussion of the NoAlias response in
1077 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1079 Note that this definition of <tt>noalias</tt> is intentionally
1080 similar to the definition of <tt>restrict</tt> in C99 for function
1081 arguments, though it is slightly weaker.
1083 For function return values, C99's <tt>restrict</tt> is not meaningful,
1084 while LLVM's <tt>noalias</tt> is.
1087 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1088 <dd>This indicates that the callee does not make any copies of the pointer
1089 that outlive the callee itself. This is not a valid attribute for return
1092 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1093 <dd>This indicates that the pointer parameter can be excised using the
1094 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1095 attribute for return values.</dd>
1100 <!-- ======================================================================= -->
1102 <a name="gc">Garbage Collector Names</a>
1107 <p>Each function may specify a garbage collector name, which is simply a
1110 <pre class="doc_code">
1111 define void @f() gc "name" { ... }
1114 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1115 collector which will cause the compiler to alter its output in order to
1116 support the named garbage collection algorithm.</p>
1120 <!-- ======================================================================= -->
1122 <a name="fnattrs">Function Attributes</a>
1127 <p>Function attributes are set to communicate additional information about a
1128 function. Function attributes are considered to be part of the function, not
1129 of the function type, so functions with different parameter attributes can
1130 have the same function type.</p>
1132 <p>Function attributes are simple keywords that follow the type specified. If
1133 multiple attributes are needed, they are space separated. For example:</p>
1135 <pre class="doc_code">
1136 define void @f() noinline { ... }
1137 define void @f() alwaysinline { ... }
1138 define void @f() alwaysinline optsize { ... }
1139 define void @f() optsize { ... }
1143 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1144 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1145 the backend should forcibly align the stack pointer. Specify the
1146 desired alignment, which must be a power of two, in parentheses.
1148 <dt><tt><b>alwaysinline</b></tt></dt>
1149 <dd>This attribute indicates that the inliner should attempt to inline this
1150 function into callers whenever possible, ignoring any active inlining size
1151 threshold for this caller.</dd>
1153 <dt><tt><b>nonlazybind</b></tt></dt>
1154 <dd>This attribute suppresses lazy symbol binding for the function. This
1155 may make calls to the function faster, at the cost of extra program
1156 startup time if the function is not called during program startup.</dd>
1158 <dt><tt><b>inlinehint</b></tt></dt>
1159 <dd>This attribute indicates that the source code contained a hint that inlining
1160 this function is desirable (such as the "inline" keyword in C/C++). It
1161 is just a hint; it imposes no requirements on the inliner.</dd>
1163 <dt><tt><b>naked</b></tt></dt>
1164 <dd>This attribute disables prologue / epilogue emission for the function.
1165 This can have very system-specific consequences.</dd>
1167 <dt><tt><b>noimplicitfloat</b></tt></dt>
1168 <dd>This attributes disables implicit floating point instructions.</dd>
1170 <dt><tt><b>noinline</b></tt></dt>
1171 <dd>This attribute indicates that the inliner should never inline this
1172 function in any situation. This attribute may not be used together with
1173 the <tt>alwaysinline</tt> attribute.</dd>
1175 <dt><tt><b>noredzone</b></tt></dt>
1176 <dd>This attribute indicates that the code generator should not use a red
1177 zone, even if the target-specific ABI normally permits it.</dd>
1179 <dt><tt><b>noreturn</b></tt></dt>
1180 <dd>This function attribute indicates that the function never returns
1181 normally. This produces undefined behavior at runtime if the function
1182 ever does dynamically return.</dd>
1184 <dt><tt><b>nounwind</b></tt></dt>
1185 <dd>This function attribute indicates that the function never returns with an
1186 unwind or exceptional control flow. If the function does unwind, its
1187 runtime behavior is undefined.</dd>
1189 <dt><tt><b>optsize</b></tt></dt>
1190 <dd>This attribute suggests that optimization passes and code generator passes
1191 make choices that keep the code size of this function low, and otherwise
1192 do optimizations specifically to reduce code size.</dd>
1194 <dt><tt><b>readnone</b></tt></dt>
1195 <dd>This attribute indicates that the function computes its result (or decides
1196 to unwind an exception) based strictly on its arguments, without
1197 dereferencing any pointer arguments or otherwise accessing any mutable
1198 state (e.g. memory, control registers, etc) visible to caller functions.
1199 It does not write through any pointer arguments
1200 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1201 changes any state visible to callers. This means that it cannot unwind
1202 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1203 could use the <tt>unwind</tt> instruction.</dd>
1205 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1206 <dd>This attribute indicates that the function does not write through any
1207 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1208 arguments) or otherwise modify any state (e.g. memory, control registers,
1209 etc) visible to caller functions. It may dereference pointer arguments
1210 and read state that may be set in the caller. A readonly function always
1211 returns the same value (or unwinds an exception identically) when called
1212 with the same set of arguments and global state. It cannot unwind an
1213 exception by calling the <tt>C++</tt> exception throwing methods, but may
1214 use the <tt>unwind</tt> instruction.</dd>
1216 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1217 <dd>This attribute indicates that the function should emit a stack smashing
1218 protector. It is in the form of a "canary"—a random value placed on
1219 the stack before the local variables that's checked upon return from the
1220 function to see if it has been overwritten. A heuristic is used to
1221 determine if a function needs stack protectors or not.<br>
1223 If a function that has an <tt>ssp</tt> attribute is inlined into a
1224 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1225 function will have an <tt>ssp</tt> attribute.</dd>
1227 <dt><tt><b>sspreq</b></tt></dt>
1228 <dd>This attribute indicates that the function should <em>always</em> emit a
1229 stack smashing protector. This overrides
1230 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1232 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1233 function that doesn't have an <tt>sspreq</tt> attribute or which has
1234 an <tt>ssp</tt> attribute, then the resulting function will have
1235 an <tt>sspreq</tt> attribute.</dd>
1237 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1238 <dd>This attribute indicates that the ABI being targeted requires that
1239 an unwind table entry be produce for this function even if we can
1240 show that no exceptions passes by it. This is normally the case for
1241 the ELF x86-64 abi, but it can be disabled for some compilation
1244 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1245 <dd>This attribute indicates that this function can return
1246 twice. The C <code>setjmp</code> is an example of such a function.
1247 The compiler disables some optimizations (like tail calls) in the caller of
1248 these functions.</dd>
1253 <!-- ======================================================================= -->
1255 <a name="moduleasm">Module-Level Inline Assembly</a>
1260 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1261 the GCC "file scope inline asm" blocks. These blocks are internally
1262 concatenated by LLVM and treated as a single unit, but may be separated in
1263 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1265 <pre class="doc_code">
1266 module asm "inline asm code goes here"
1267 module asm "more can go here"
1270 <p>The strings can contain any character by escaping non-printable characters.
1271 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1274 <p>The inline asm code is simply printed to the machine code .s file when
1275 assembly code is generated.</p>
1279 <!-- ======================================================================= -->
1281 <a name="datalayout">Data Layout</a>
1286 <p>A module may specify a target specific data layout string that specifies how
1287 data is to be laid out in memory. The syntax for the data layout is
1290 <pre class="doc_code">
1291 target datalayout = "<i>layout specification</i>"
1294 <p>The <i>layout specification</i> consists of a list of specifications
1295 separated by the minus sign character ('-'). Each specification starts with
1296 a letter and may include other information after the letter to define some
1297 aspect of the data layout. The specifications accepted are as follows:</p>
1301 <dd>Specifies that the target lays out data in big-endian form. That is, the
1302 bits with the most significance have the lowest address location.</dd>
1305 <dd>Specifies that the target lays out data in little-endian form. That is,
1306 the bits with the least significance have the lowest address
1309 <dt><tt>S<i>size</i></tt></dt>
1310 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1311 of stack variables is limited to the natural stack alignment to avoid
1312 dynamic stack realignment. The stack alignment must be a multiple of
1313 8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1314 which does not prevent any alignment promotions.</dd>
1316 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1317 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1318 <i>preferred</i> alignments. All sizes are in bits. Specifying
1319 the <i>pref</i> alignment is optional. If omitted, the
1320 preceding <tt>:</tt> should be omitted too.</dd>
1322 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1323 <dd>This specifies the alignment for an integer type of a given bit
1324 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1326 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1327 <dd>This specifies the alignment for a vector type of a given bit
1330 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1331 <dd>This specifies the alignment for a floating point type of a given bit
1332 <i>size</i>. Only values of <i>size</i> that are supported by the target
1333 will work. 32 (float) and 64 (double) are supported on all targets;
1334 80 or 128 (different flavors of long double) are also supported on some
1337 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1338 <dd>This specifies the alignment for an aggregate type of a given bit
1341 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1342 <dd>This specifies the alignment for a stack object of a given bit
1345 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1346 <dd>This specifies a set of native integer widths for the target CPU
1347 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1348 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1349 this set are considered to support most general arithmetic
1350 operations efficiently.</dd>
1353 <p>When constructing the data layout for a given target, LLVM starts with a
1354 default set of specifications which are then (possibly) overridden by the
1355 specifications in the <tt>datalayout</tt> keyword. The default specifications
1356 are given in this list:</p>
1359 <li><tt>E</tt> - big endian</li>
1360 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1361 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1362 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1363 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1364 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1365 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1366 alignment of 64-bits</li>
1367 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1368 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1369 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1370 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1371 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1372 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1375 <p>When LLVM is determining the alignment for a given type, it uses the
1376 following rules:</p>
1379 <li>If the type sought is an exact match for one of the specifications, that
1380 specification is used.</li>
1382 <li>If no match is found, and the type sought is an integer type, then the
1383 smallest integer type that is larger than the bitwidth of the sought type
1384 is used. If none of the specifications are larger than the bitwidth then
1385 the the largest integer type is used. For example, given the default
1386 specifications above, the i7 type will use the alignment of i8 (next
1387 largest) while both i65 and i256 will use the alignment of i64 (largest
1390 <li>If no match is found, and the type sought is a vector type, then the
1391 largest vector type that is smaller than the sought vector type will be
1392 used as a fall back. This happens because <128 x double> can be
1393 implemented in terms of 64 <2 x double>, for example.</li>
1396 <p>The function of the data layout string may not be what you expect. Notably,
1397 this is not a specification from the frontend of what alignment the code
1398 generator should use.</p>
1400 <p>Instead, if specified, the target data layout is required to match what the
1401 ultimate <em>code generator</em> expects. This string is used by the
1402 mid-level optimizers to
1403 improve code, and this only works if it matches what the ultimate code
1404 generator uses. If you would like to generate IR that does not embed this
1405 target-specific detail into the IR, then you don't have to specify the
1406 string. This will disable some optimizations that require precise layout
1407 information, but this also prevents those optimizations from introducing
1408 target specificity into the IR.</p>
1414 <!-- ======================================================================= -->
1416 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1421 <p>Any memory access must be done through a pointer value associated
1422 with an address range of the memory access, otherwise the behavior
1423 is undefined. Pointer values are associated with address ranges
1424 according to the following rules:</p>
1427 <li>A pointer value is associated with the addresses associated with
1428 any value it is <i>based</i> on.
1429 <li>An address of a global variable is associated with the address
1430 range of the variable's storage.</li>
1431 <li>The result value of an allocation instruction is associated with
1432 the address range of the allocated storage.</li>
1433 <li>A null pointer in the default address-space is associated with
1435 <li>An integer constant other than zero or a pointer value returned
1436 from a function not defined within LLVM may be associated with address
1437 ranges allocated through mechanisms other than those provided by
1438 LLVM. Such ranges shall not overlap with any ranges of addresses
1439 allocated by mechanisms provided by LLVM.</li>
1442 <p>A pointer value is <i>based</i> on another pointer value according
1443 to the following rules:</p>
1446 <li>A pointer value formed from a
1447 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1448 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1449 <li>The result value of a
1450 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1451 of the <tt>bitcast</tt>.</li>
1452 <li>A pointer value formed by an
1453 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1454 pointer values that contribute (directly or indirectly) to the
1455 computation of the pointer's value.</li>
1456 <li>The "<i>based</i> on" relationship is transitive.</li>
1459 <p>Note that this definition of <i>"based"</i> is intentionally
1460 similar to the definition of <i>"based"</i> in C99, though it is
1461 slightly weaker.</p>
1463 <p>LLVM IR does not associate types with memory. The result type of a
1464 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1465 alignment of the memory from which to load, as well as the
1466 interpretation of the value. The first operand type of a
1467 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1468 and alignment of the store.</p>
1470 <p>Consequently, type-based alias analysis, aka TBAA, aka
1471 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1472 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1473 additional information which specialized optimization passes may use
1474 to implement type-based alias analysis.</p>
1478 <!-- ======================================================================= -->
1480 <a name="volatile">Volatile Memory Accesses</a>
1485 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1486 href="#i_store"><tt>store</tt></a>s, and <a
1487 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1488 The optimizers must not change the number of volatile operations or change their
1489 order of execution relative to other volatile operations. The optimizers
1490 <i>may</i> change the order of volatile operations relative to non-volatile
1491 operations. This is not Java's "volatile" and has no cross-thread
1492 synchronization behavior.</p>
1496 <!-- ======================================================================= -->
1498 <a name="memmodel">Memory Model for Concurrent Operations</a>
1503 <p>The LLVM IR does not define any way to start parallel threads of execution
1504 or to register signal handlers. Nonetheless, there are platform-specific
1505 ways to create them, and we define LLVM IR's behavior in their presence. This
1506 model is inspired by the C++0x memory model.</p>
1508 <p>For a more informal introduction to this model, see the
1509 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1511 <p>We define a <i>happens-before</i> partial order as the least partial order
1514 <li>Is a superset of single-thread program order, and</li>
1515 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1516 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1517 by platform-specific techniques, like pthread locks, thread
1518 creation, thread joining, etc., and by atomic instructions.
1519 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1523 <p>Note that program order does not introduce <i>happens-before</i> edges
1524 between a thread and signals executing inside that thread.</p>
1526 <p>Every (defined) read operation (load instructions, memcpy, atomic
1527 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1528 (defined) write operations (store instructions, atomic
1529 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1530 initialized globals are considered to have a write of the initializer which is
1531 atomic and happens before any other read or write of the memory in question.
1532 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1533 any write to the same byte, except:</p>
1536 <li>If <var>write<sub>1</sub></var> happens before
1537 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1538 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1539 does not see <var>write<sub>1</sub></var>.
1540 <li>If <var>R<sub>byte</sub></var> happens before
1541 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1542 see <var>write<sub>3</sub></var>.
1545 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1547 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1548 is supposed to give guarantees which can support
1549 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1550 addresses which do not behave like normal memory. It does not generally
1551 provide cross-thread synchronization.)
1552 <li>Otherwise, if there is no write to the same byte that happens before
1553 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1554 <tt>undef</tt> for that byte.
1555 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1556 <var>R<sub>byte</sub></var> returns the value written by that
1558 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1559 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1560 values written. See the <a href="#ordering">Atomic Memory Ordering
1561 Constraints</a> section for additional constraints on how the choice
1563 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1566 <p><var>R</var> returns the value composed of the series of bytes it read.
1567 This implies that some bytes within the value may be <tt>undef</tt>
1568 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1569 defines the semantics of the operation; it doesn't mean that targets will
1570 emit more than one instruction to read the series of bytes.</p>
1572 <p>Note that in cases where none of the atomic intrinsics are used, this model
1573 places only one restriction on IR transformations on top of what is required
1574 for single-threaded execution: introducing a store to a byte which might not
1575 otherwise be stored is not allowed in general. (Specifically, in the case
1576 where another thread might write to and read from an address, introducing a
1577 store can change a load that may see exactly one write into a load that may
1578 see multiple writes.)</p>
1580 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1581 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1582 none of the backends currently in the tree fall into this category; however,
1583 there might be targets which care. If there are, we want a paragraph
1586 Targets may specify that stores narrower than a certain width are not
1587 available; on such a target, for the purposes of this model, treat any
1588 non-atomic write with an alignment or width less than the minimum width
1589 as if it writes to the relevant surrounding bytes.
1594 <!-- ======================================================================= -->
1596 <a name="ordering">Atomic Memory Ordering Constraints</a>
1601 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1602 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1603 <a href="#i_fence"><code>fence</code></a>,
1604 <a href="#i_load"><code>atomic load</code></a>, and
1605 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1606 that determines which other atomic instructions on the same address they
1607 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1608 but are somewhat more colloquial. If these descriptions aren't precise enough,
1609 check those specs (see spec references in the
1610 <a href="Atomic.html#introduction">atomics guide</a>).
1611 <a href="#i_fence"><code>fence</code></a> instructions
1612 treat these orderings somewhat differently since they don't take an address.
1613 See that instruction's documentation for details.</p>
1615 <p>For a simpler introduction to the ordering constraints, see the
1616 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1619 <dt><code>unordered</code></dt>
1620 <dd>The set of values that can be read is governed by the happens-before
1621 partial order. A value cannot be read unless some operation wrote it.
1622 This is intended to provide a guarantee strong enough to model Java's
1623 non-volatile shared variables. This ordering cannot be specified for
1624 read-modify-write operations; it is not strong enough to make them atomic
1625 in any interesting way.</dd>
1626 <dt><code>monotonic</code></dt>
1627 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1628 total order for modifications by <code>monotonic</code> operations on each
1629 address. All modification orders must be compatible with the happens-before
1630 order. There is no guarantee that the modification orders can be combined to
1631 a global total order for the whole program (and this often will not be
1632 possible). The read in an atomic read-modify-write operation
1633 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1634 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1635 reads the value in the modification order immediately before the value it
1636 writes. If one atomic read happens before another atomic read of the same
1637 address, the later read must see the same value or a later value in the
1638 address's modification order. This disallows reordering of
1639 <code>monotonic</code> (or stronger) operations on the same address. If an
1640 address is written <code>monotonic</code>ally by one thread, and other threads
1641 <code>monotonic</code>ally read that address repeatedly, the other threads must
1642 eventually see the write. This corresponds to the C++0x/C1x
1643 <code>memory_order_relaxed</code>.</dd>
1644 <dt><code>acquire</code></dt>
1645 <dd>In addition to the guarantees of <code>monotonic</code>,
1646 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1647 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1648 <dt><code>release</code></dt>
1649 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1650 writes a value which is subsequently read by an <code>acquire</code> operation,
1651 it <i>synchronizes-with</i> that operation. (This isn't a complete
1652 description; see the C++0x definition of a release sequence.) This corresponds
1653 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1654 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1655 <code>acquire</code> and <code>release</code> operation on its address.
1656 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1657 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1658 <dd>In addition to the guarantees of <code>acq_rel</code>
1659 (<code>acquire</code> for an operation which only reads, <code>release</code>
1660 for an operation which only writes), there is a global total order on all
1661 sequentially-consistent operations on all addresses, which is consistent with
1662 the <i>happens-before</i> partial order and with the modification orders of
1663 all the affected addresses. Each sequentially-consistent read sees the last
1664 preceding write to the same address in this global order. This corresponds
1665 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1668 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1669 it only <i>synchronizes with</i> or participates in modification and seq_cst
1670 total orderings with other operations running in the same thread (for example,
1671 in signal handlers).</p>
1677 <!-- *********************************************************************** -->
1678 <h2><a name="typesystem">Type System</a></h2>
1679 <!-- *********************************************************************** -->
1683 <p>The LLVM type system is one of the most important features of the
1684 intermediate representation. Being typed enables a number of optimizations
1685 to be performed on the intermediate representation directly, without having
1686 to do extra analyses on the side before the transformation. A strong type
1687 system makes it easier to read the generated code and enables novel analyses
1688 and transformations that are not feasible to perform on normal three address
1689 code representations.</p>
1691 <!-- ======================================================================= -->
1693 <a name="t_classifications">Type Classifications</a>
1698 <p>The types fall into a few useful classifications:</p>
1700 <table border="1" cellspacing="0" cellpadding="4">
1702 <tr><th>Classification</th><th>Types</th></tr>
1704 <td><a href="#t_integer">integer</a></td>
1705 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1708 <td><a href="#t_floating">floating point</a></td>
1709 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1712 <td><a name="t_firstclass">first class</a></td>
1713 <td><a href="#t_integer">integer</a>,
1714 <a href="#t_floating">floating point</a>,
1715 <a href="#t_pointer">pointer</a>,
1716 <a href="#t_vector">vector</a>,
1717 <a href="#t_struct">structure</a>,
1718 <a href="#t_array">array</a>,
1719 <a href="#t_label">label</a>,
1720 <a href="#t_metadata">metadata</a>.
1724 <td><a href="#t_primitive">primitive</a></td>
1725 <td><a href="#t_label">label</a>,
1726 <a href="#t_void">void</a>,
1727 <a href="#t_integer">integer</a>,
1728 <a href="#t_floating">floating point</a>,
1729 <a href="#t_x86mmx">x86mmx</a>,
1730 <a href="#t_metadata">metadata</a>.</td>
1733 <td><a href="#t_derived">derived</a></td>
1734 <td><a href="#t_array">array</a>,
1735 <a href="#t_function">function</a>,
1736 <a href="#t_pointer">pointer</a>,
1737 <a href="#t_struct">structure</a>,
1738 <a href="#t_vector">vector</a>,
1739 <a href="#t_opaque">opaque</a>.
1745 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1746 important. Values of these types are the only ones which can be produced by
1751 <!-- ======================================================================= -->
1753 <a name="t_primitive">Primitive Types</a>
1758 <p>The primitive types are the fundamental building blocks of the LLVM
1761 <!-- _______________________________________________________________________ -->
1763 <a name="t_integer">Integer Type</a>
1769 <p>The integer type is a very simple type that simply specifies an arbitrary
1770 bit width for the integer type desired. Any bit width from 1 bit to
1771 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1778 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1782 <table class="layout">
1784 <td class="left"><tt>i1</tt></td>
1785 <td class="left">a single-bit integer.</td>
1788 <td class="left"><tt>i32</tt></td>
1789 <td class="left">a 32-bit integer.</td>
1792 <td class="left"><tt>i1942652</tt></td>
1793 <td class="left">a really big integer of over 1 million bits.</td>
1799 <!-- _______________________________________________________________________ -->
1801 <a name="t_floating">Floating Point Types</a>
1808 <tr><th>Type</th><th>Description</th></tr>
1809 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1810 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1811 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1812 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1813 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1819 <!-- _______________________________________________________________________ -->
1821 <a name="t_x86mmx">X86mmx Type</a>
1827 <p>The x86mmx type represents a value held in an MMX register on an x86 machine. The operations allowed on it are quite limited: parameters and return values, load and store, and bitcast. User-specified MMX instructions are represented as intrinsic or asm calls with arguments and/or results of this type. There are no arrays, vectors or constants of this type.</p>
1836 <!-- _______________________________________________________________________ -->
1838 <a name="t_void">Void Type</a>
1844 <p>The void type does not represent any value and has no size.</p>
1853 <!-- _______________________________________________________________________ -->
1855 <a name="t_label">Label Type</a>
1861 <p>The label type represents code labels.</p>
1870 <!-- _______________________________________________________________________ -->
1872 <a name="t_metadata">Metadata Type</a>
1878 <p>The metadata type represents embedded metadata. No derived types may be
1879 created from metadata except for <a href="#t_function">function</a>
1891 <!-- ======================================================================= -->
1893 <a name="t_derived">Derived Types</a>
1898 <p>The real power in LLVM comes from the derived types in the system. This is
1899 what allows a programmer to represent arrays, functions, pointers, and other
1900 useful types. Each of these types contain one or more element types which
1901 may be a primitive type, or another derived type. For example, it is
1902 possible to have a two dimensional array, using an array as the element type
1903 of another array.</p>
1908 <!-- _______________________________________________________________________ -->
1910 <a name="t_aggregate">Aggregate Types</a>
1915 <p>Aggregate Types are a subset of derived types that can contain multiple
1916 member types. <a href="#t_array">Arrays</a>,
1917 <a href="#t_struct">structs</a>, and <a href="#t_vector">vectors</a> are
1918 aggregate types.</p>
1922 <!-- _______________________________________________________________________ -->
1924 <a name="t_array">Array Type</a>
1930 <p>The array type is a very simple derived type that arranges elements
1931 sequentially in memory. The array type requires a size (number of elements)
1932 and an underlying data type.</p>
1936 [<# elements> x <elementtype>]
1939 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1940 be any type with a size.</p>
1943 <table class="layout">
1945 <td class="left"><tt>[40 x i32]</tt></td>
1946 <td class="left">Array of 40 32-bit integer values.</td>
1949 <td class="left"><tt>[41 x i32]</tt></td>
1950 <td class="left">Array of 41 32-bit integer values.</td>
1953 <td class="left"><tt>[4 x i8]</tt></td>
1954 <td class="left">Array of 4 8-bit integer values.</td>
1957 <p>Here are some examples of multidimensional arrays:</p>
1958 <table class="layout">
1960 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1961 <td class="left">3x4 array of 32-bit integer values.</td>
1964 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1965 <td class="left">12x10 array of single precision floating point values.</td>
1968 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1969 <td class="left">2x3x4 array of 16-bit integer values.</td>
1973 <p>There is no restriction on indexing beyond the end of the array implied by
1974 a static type (though there are restrictions on indexing beyond the bounds
1975 of an allocated object in some cases). This means that single-dimension
1976 'variable sized array' addressing can be implemented in LLVM with a zero
1977 length array type. An implementation of 'pascal style arrays' in LLVM could
1978 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1982 <!-- _______________________________________________________________________ -->
1984 <a name="t_function">Function Type</a>
1990 <p>The function type can be thought of as a function signature. It consists of
1991 a return type and a list of formal parameter types. The return type of a
1992 function type is a first class type or a void type.</p>
1996 <returntype> (<parameter list>)
1999 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
2000 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
2001 which indicates that the function takes a variable number of arguments.
2002 Variable argument functions can access their arguments with
2003 the <a href="#int_varargs">variable argument handling intrinsic</a>
2004 functions. '<tt><returntype></tt>' is any type except
2005 <a href="#t_label">label</a>.</p>
2008 <table class="layout">
2010 <td class="left"><tt>i32 (i32)</tt></td>
2011 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2013 </tr><tr class="layout">
2014 <td class="left"><tt>float (i16, i32 *) *
2016 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2017 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2018 returning <tt>float</tt>.
2020 </tr><tr class="layout">
2021 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2022 <td class="left">A vararg function that takes at least one
2023 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2024 which returns an integer. This is the signature for <tt>printf</tt> in
2027 </tr><tr class="layout">
2028 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2029 <td class="left">A function taking an <tt>i32</tt>, returning a
2030 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2037 <!-- _______________________________________________________________________ -->
2039 <a name="t_struct">Structure Type</a>
2045 <p>The structure type is used to represent a collection of data members together
2046 in memory. The elements of a structure may be any type that has a size.</p>
2048 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2049 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2050 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2051 Structures in registers are accessed using the
2052 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2053 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2055 <p>Structures may optionally be "packed" structures, which indicate that the
2056 alignment of the struct is one byte, and that there is no padding between
2057 the elements. In non-packed structs, padding between field types is inserted
2058 as defined by the TargetData string in the module, which is required to match
2059 what the underlying code generator expects.</p>
2061 <p>Structures can either be "literal" or "identified". A literal structure is
2062 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2063 types are always defined at the top level with a name. Literal types are
2064 uniqued by their contents and can never be recursive or opaque since there is
2065 no way to write one. Identified types can be recursive, can be opaqued, and are
2071 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2072 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2076 <table class="layout">
2078 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2079 <td class="left">A triple of three <tt>i32</tt> values</td>
2082 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2083 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2084 second element is a <a href="#t_pointer">pointer</a> to a
2085 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2086 an <tt>i32</tt>.</td>
2089 <td class="left"><tt><{ i8, i32 }></tt></td>
2090 <td class="left">A packed struct known to be 5 bytes in size.</td>
2096 <!-- _______________________________________________________________________ -->
2098 <a name="t_opaque">Opaque Structure Types</a>
2104 <p>Opaque structure types are used to represent named structure types that do
2105 not have a body specified. This corresponds (for example) to the C notion of
2106 a forward declared structure.</p>
2115 <table class="layout">
2117 <td class="left"><tt>opaque</tt></td>
2118 <td class="left">An opaque type.</td>
2126 <!-- _______________________________________________________________________ -->
2128 <a name="t_pointer">Pointer Type</a>
2134 <p>The pointer type is used to specify memory locations.
2135 Pointers are commonly used to reference objects in memory.</p>
2137 <p>Pointer types may have an optional address space attribute defining the
2138 numbered address space where the pointed-to object resides. The default
2139 address space is number zero. The semantics of non-zero address
2140 spaces are target-specific.</p>
2142 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2143 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2151 <table class="layout">
2153 <td class="left"><tt>[4 x i32]*</tt></td>
2154 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2155 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2158 <td class="left"><tt>i32 (i32*) *</tt></td>
2159 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2160 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2164 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2165 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2166 that resides in address space #5.</td>
2172 <!-- _______________________________________________________________________ -->
2174 <a name="t_vector">Vector Type</a>
2180 <p>A vector type is a simple derived type that represents a vector of elements.
2181 Vector types are used when multiple primitive data are operated in parallel
2182 using a single instruction (SIMD). A vector type requires a size (number of
2183 elements) and an underlying primitive data type. Vector types are considered
2184 <a href="#t_firstclass">first class</a>.</p>
2188 < <# elements> x <elementtype> >
2191 <p>The number of elements is a constant integer value larger than 0; elementtype
2192 may be any integer or floating point type. Vectors of size zero are not
2193 allowed, and pointers are not allowed as the element type.</p>
2196 <table class="layout">
2198 <td class="left"><tt><4 x i32></tt></td>
2199 <td class="left">Vector of 4 32-bit integer values.</td>
2202 <td class="left"><tt><8 x float></tt></td>
2203 <td class="left">Vector of 8 32-bit floating-point values.</td>
2206 <td class="left"><tt><2 x i64></tt></td>
2207 <td class="left">Vector of 2 64-bit integer values.</td>
2215 <!-- *********************************************************************** -->
2216 <h2><a name="constants">Constants</a></h2>
2217 <!-- *********************************************************************** -->
2221 <p>LLVM has several different basic types of constants. This section describes
2222 them all and their syntax.</p>
2224 <!-- ======================================================================= -->
2226 <a name="simpleconstants">Simple Constants</a>
2232 <dt><b>Boolean constants</b></dt>
2233 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2234 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2236 <dt><b>Integer constants</b></dt>
2237 <dd>Standard integers (such as '4') are constants of
2238 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2239 with integer types.</dd>
2241 <dt><b>Floating point constants</b></dt>
2242 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2243 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2244 notation (see below). The assembler requires the exact decimal value of a
2245 floating-point constant. For example, the assembler accepts 1.25 but
2246 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2247 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2249 <dt><b>Null pointer constants</b></dt>
2250 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2251 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2254 <p>The one non-intuitive notation for constants is the hexadecimal form of
2255 floating point constants. For example, the form '<tt>double
2256 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2257 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2258 constants are required (and the only time that they are generated by the
2259 disassembler) is when a floating point constant must be emitted but it cannot
2260 be represented as a decimal floating point number in a reasonable number of
2261 digits. For example, NaN's, infinities, and other special values are
2262 represented in their IEEE hexadecimal format so that assembly and disassembly
2263 do not cause any bits to change in the constants.</p>
2265 <p>When using the hexadecimal form, constants of types float and double are
2266 represented using the 16-digit form shown above (which matches the IEEE754
2267 representation for double); float values must, however, be exactly
2268 representable as IEE754 single precision. Hexadecimal format is always used
2269 for long double, and there are three forms of long double. The 80-bit format
2270 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2271 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2272 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2273 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2274 currently supported target uses this format. Long doubles will only work if
2275 they match the long double format on your target. All hexadecimal formats
2276 are big-endian (sign bit at the left).</p>
2278 <p>There are no constants of type x86mmx.</p>
2281 <!-- ======================================================================= -->
2283 <a name="aggregateconstants"></a> <!-- old anchor -->
2284 <a name="complexconstants">Complex Constants</a>
2289 <p>Complex constants are a (potentially recursive) combination of simple
2290 constants and smaller complex constants.</p>
2293 <dt><b>Structure constants</b></dt>
2294 <dd>Structure constants are represented with notation similar to structure
2295 type definitions (a comma separated list of elements, surrounded by braces
2296 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2297 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2298 Structure constants must have <a href="#t_struct">structure type</a>, and
2299 the number and types of elements must match those specified by the
2302 <dt><b>Array constants</b></dt>
2303 <dd>Array constants are represented with notation similar to array type
2304 definitions (a comma separated list of elements, surrounded by square
2305 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2306 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2307 the number and types of elements must match those specified by the
2310 <dt><b>Vector constants</b></dt>
2311 <dd>Vector constants are represented with notation similar to vector type
2312 definitions (a comma separated list of elements, surrounded by
2313 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2314 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2315 have <a href="#t_vector">vector type</a>, and the number and types of
2316 elements must match those specified by the type.</dd>
2318 <dt><b>Zero initialization</b></dt>
2319 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2320 value to zero of <em>any</em> type, including scalar and
2321 <a href="#t_aggregate">aggregate</a> types.
2322 This is often used to avoid having to print large zero initializers
2323 (e.g. for large arrays) and is always exactly equivalent to using explicit
2324 zero initializers.</dd>
2326 <dt><b>Metadata node</b></dt>
2327 <dd>A metadata node is a structure-like constant with
2328 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2329 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2330 be interpreted as part of the instruction stream, metadata is a place to
2331 attach additional information such as debug info.</dd>
2336 <!-- ======================================================================= -->
2338 <a name="globalconstants">Global Variable and Function Addresses</a>
2343 <p>The addresses of <a href="#globalvars">global variables</a>
2344 and <a href="#functionstructure">functions</a> are always implicitly valid
2345 (link-time) constants. These constants are explicitly referenced when
2346 the <a href="#identifiers">identifier for the global</a> is used and always
2347 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2348 legal LLVM file:</p>
2350 <pre class="doc_code">
2353 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2358 <!-- ======================================================================= -->
2360 <a name="undefvalues">Undefined Values</a>
2365 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2366 indicates that the user of the value may receive an unspecified bit-pattern.
2367 Undefined values may be of any type (other than '<tt>label</tt>'
2368 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2370 <p>Undefined values are useful because they indicate to the compiler that the
2371 program is well defined no matter what value is used. This gives the
2372 compiler more freedom to optimize. Here are some examples of (potentially
2373 surprising) transformations that are valid (in pseudo IR):</p>
2376 <pre class="doc_code">
2386 <p>This is safe because all of the output bits are affected by the undef bits.
2387 Any output bit can have a zero or one depending on the input bits.</p>
2389 <pre class="doc_code">
2400 <p>These logical operations have bits that are not always affected by the input.
2401 For example, if <tt>%X</tt> has a zero bit, then the output of the
2402 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2403 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2404 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2405 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2406 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2407 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2408 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2410 <pre class="doc_code">
2411 %A = select undef, %X, %Y
2412 %B = select undef, 42, %Y
2413 %C = select %X, %Y, undef
2424 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2425 branch) conditions can go <em>either way</em>, but they have to come from one
2426 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2427 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2428 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2429 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2430 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2433 <pre class="doc_code">
2434 %A = xor undef, undef
2452 <p>This example points out that two '<tt>undef</tt>' operands are not
2453 necessarily the same. This can be surprising to people (and also matches C
2454 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2455 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2456 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2457 its value over its "live range". This is true because the variable doesn't
2458 actually <em>have a live range</em>. Instead, the value is logically read
2459 from arbitrary registers that happen to be around when needed, so the value
2460 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2461 need to have the same semantics or the core LLVM "replace all uses with"
2462 concept would not hold.</p>
2464 <pre class="doc_code">
2472 <p>These examples show the crucial difference between an <em>undefined
2473 value</em> and <em>undefined behavior</em>. An undefined value (like
2474 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2475 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2476 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2477 defined on SNaN's. However, in the second example, we can make a more
2478 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2479 arbitrary value, we are allowed to assume that it could be zero. Since a
2480 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2481 the operation does not execute at all. This allows us to delete the divide and
2482 all code after it. Because the undefined operation "can't happen", the
2483 optimizer can assume that it occurs in dead code.</p>
2485 <pre class="doc_code">
2486 a: store undef -> %X
2487 b: store %X -> undef
2493 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2494 undefined value can be assumed to not have any effect; we can assume that the
2495 value is overwritten with bits that happen to match what was already there.
2496 However, a store <em>to</em> an undefined location could clobber arbitrary
2497 memory, therefore, it has undefined behavior.</p>
2501 <!-- ======================================================================= -->
2503 <a name="trapvalues">Trap Values</a>
2508 <p>Trap values are similar to <a href="#undefvalues">undef values</a>, however
2509 instead of representing an unspecified bit pattern, they represent the
2510 fact that an instruction or constant expression which cannot evoke side
2511 effects has nevertheless detected a condition which results in undefined
2514 <p>There is currently no way of representing a trap value in the IR; they
2515 only exist when produced by operations such as
2516 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2518 <p>Trap value behavior is defined in terms of value <i>dependence</i>:</p>
2521 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2522 their operands.</li>
2524 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2525 to their dynamic predecessor basic block.</li>
2527 <li>Function arguments depend on the corresponding actual argument values in
2528 the dynamic callers of their functions.</li>
2530 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2531 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2532 control back to them.</li>
2534 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2535 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_unwind"><tt>unwind</tt></a>,
2536 or exception-throwing call instructions that dynamically transfer control
2539 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2540 referenced memory addresses, following the order in the IR
2541 (including loads and stores implied by intrinsics such as
2542 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2544 <!-- TODO: In the case of multiple threads, this only applies if the store
2545 "happens-before" the load or store. -->
2547 <!-- TODO: floating-point exception state -->
2549 <li>An instruction with externally visible side effects depends on the most
2550 recent preceding instruction with externally visible side effects, following
2551 the order in the IR. (This includes
2552 <a href="#volatile">volatile operations</a>.)</li>
2554 <li>An instruction <i>control-depends</i> on a
2555 <a href="#terminators">terminator instruction</a>
2556 if the terminator instruction has multiple successors and the instruction
2557 is always executed when control transfers to one of the successors, and
2558 may not be executed when control is transferred to another.</li>
2560 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2561 instruction if the set of instructions it otherwise depends on would be
2562 different if the terminator had transferred control to a different
2565 <li>Dependence is transitive.</li>
2569 <p>Whenever a trap value is generated, all values which depend on it evaluate
2570 to trap. If they have side effects, they evoke their side effects as if each
2571 operand with a trap value were undef. If they have externally-visible side
2572 effects, the behavior is undefined.</p>
2574 <p>Here are some examples:</p>
2576 <pre class="doc_code">
2578 %trap = sub nuw i32 0, 1 ; Results in a trap value.
2579 %still_trap = and i32 %trap, 0 ; Whereas (and i32 undef, 0) would return 0.
2580 %trap_yet_again = getelementptr i32* @h, i32 %still_trap
2581 store i32 0, i32* %trap_yet_again ; undefined behavior
2583 store i32 %trap, i32* @g ; Trap value conceptually stored to memory.
2584 %trap2 = load i32* @g ; Returns a trap value, not just undef.
2586 volatile store i32 %trap, i32* @g ; External observation; undefined behavior.
2588 %narrowaddr = bitcast i32* @g to i16*
2589 %wideaddr = bitcast i32* @g to i64*
2590 %trap3 = load i16* %narrowaddr ; Returns a trap value.
2591 %trap4 = load i64* %wideaddr ; Returns a trap value.
2593 %cmp = icmp slt i32 %trap, 0 ; Returns a trap value.
2594 br i1 %cmp, label %true, label %end ; Branch to either destination.
2597 volatile store i32 0, i32* @g ; This is control-dependent on %cmp, so
2598 ; it has undefined behavior.
2602 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2603 ; Both edges into this PHI are
2604 ; control-dependent on %cmp, so this
2605 ; always results in a trap value.
2607 volatile store i32 0, i32* @g ; This would depend on the store in %true
2608 ; if %cmp is true, or the store in %entry
2609 ; otherwise, so this is undefined behavior.
2611 br i1 %cmp, label %second_true, label %second_end
2612 ; The same branch again, but this time the
2613 ; true block doesn't have side effects.
2620 volatile store i32 0, i32* @g ; This time, the instruction always depends
2621 ; on the store in %end. Also, it is
2622 ; control-equivalent to %end, so this is
2623 ; well-defined (again, ignoring earlier
2624 ; undefined behavior in this example).
2629 <!-- ======================================================================= -->
2631 <a name="blockaddress">Addresses of Basic Blocks</a>
2636 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2638 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2639 basic block in the specified function, and always has an i8* type. Taking
2640 the address of the entry block is illegal.</p>
2642 <p>This value only has defined behavior when used as an operand to the
2643 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2644 comparisons against null. Pointer equality tests between labels addresses
2645 results in undefined behavior — though, again, comparison against null
2646 is ok, and no label is equal to the null pointer. This may be passed around
2647 as an opaque pointer sized value as long as the bits are not inspected. This
2648 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2649 long as the original value is reconstituted before the <tt>indirectbr</tt>
2652 <p>Finally, some targets may provide defined semantics when using the value as
2653 the operand to an inline assembly, but that is target specific.</p>
2658 <!-- ======================================================================= -->
2660 <a name="constantexprs">Constant Expressions</a>
2665 <p>Constant expressions are used to allow expressions involving other constants
2666 to be used as constants. Constant expressions may be of
2667 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2668 operation that does not have side effects (e.g. load and call are not
2669 supported). The following is the syntax for constant expressions:</p>
2672 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2673 <dd>Truncate a constant to another type. The bit size of CST must be larger
2674 than the bit size of TYPE. Both types must be integers.</dd>
2676 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2677 <dd>Zero extend a constant to another type. The bit size of CST must be
2678 smaller than the bit size of TYPE. Both types must be integers.</dd>
2680 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2681 <dd>Sign extend a constant to another type. The bit size of CST must be
2682 smaller than the bit size of TYPE. Both types must be integers.</dd>
2684 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2685 <dd>Truncate a floating point constant to another floating point type. The
2686 size of CST must be larger than the size of TYPE. Both types must be
2687 floating point.</dd>
2689 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2690 <dd>Floating point extend a constant to another type. The size of CST must be
2691 smaller or equal to the size of TYPE. Both types must be floating
2694 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2695 <dd>Convert a floating point constant to the corresponding unsigned integer
2696 constant. TYPE must be a scalar or vector integer type. CST must be of
2697 scalar or vector floating point type. Both CST and TYPE must be scalars,
2698 or vectors of the same number of elements. If the value won't fit in the
2699 integer type, the results are undefined.</dd>
2701 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2702 <dd>Convert a floating point constant to the corresponding signed integer
2703 constant. TYPE must be a scalar or vector integer type. CST must be of
2704 scalar or vector floating point type. Both CST and TYPE must be scalars,
2705 or vectors of the same number of elements. If the value won't fit in the
2706 integer type, the results are undefined.</dd>
2708 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2709 <dd>Convert an unsigned integer constant to the corresponding floating point
2710 constant. TYPE must be a scalar or vector floating point type. CST must be
2711 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2712 vectors of the same number of elements. If the value won't fit in the
2713 floating point type, the results are undefined.</dd>
2715 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2716 <dd>Convert a signed integer constant to the corresponding floating point
2717 constant. TYPE must be a scalar or vector floating point type. CST must be
2718 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2719 vectors of the same number of elements. If the value won't fit in the
2720 floating point type, the results are undefined.</dd>
2722 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2723 <dd>Convert a pointer typed constant to the corresponding integer constant
2724 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2725 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2726 make it fit in <tt>TYPE</tt>.</dd>
2728 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2729 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2730 type. CST must be of integer type. The CST value is zero extended,
2731 truncated, or unchanged to make it fit in a pointer size. This one is
2732 <i>really</i> dangerous!</dd>
2734 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2735 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2736 are the same as those for the <a href="#i_bitcast">bitcast
2737 instruction</a>.</dd>
2739 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2740 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2741 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2742 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2743 instruction, the index list may have zero or more indexes, which are
2744 required to make sense for the type of "CSTPTR".</dd>
2746 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2747 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2749 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2750 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2752 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2753 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2755 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2756 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2759 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2760 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2763 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2764 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2767 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2768 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2769 constants. The index list is interpreted in a similar manner as indices in
2770 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2771 index value must be specified.</dd>
2773 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2774 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2775 constants. The index list is interpreted in a similar manner as indices in
2776 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2777 index value must be specified.</dd>
2779 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2780 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2781 be any of the <a href="#binaryops">binary</a>
2782 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2783 on operands are the same as those for the corresponding instruction
2784 (e.g. no bitwise operations on floating point values are allowed).</dd>
2791 <!-- *********************************************************************** -->
2792 <h2><a name="othervalues">Other Values</a></h2>
2793 <!-- *********************************************************************** -->
2795 <!-- ======================================================================= -->
2797 <a name="inlineasm">Inline Assembler Expressions</a>
2802 <p>LLVM supports inline assembler expressions (as opposed
2803 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2804 a special value. This value represents the inline assembler as a string
2805 (containing the instructions to emit), a list of operand constraints (stored
2806 as a string), a flag that indicates whether or not the inline asm
2807 expression has side effects, and a flag indicating whether the function
2808 containing the asm needs to align its stack conservatively. An example
2809 inline assembler expression is:</p>
2811 <pre class="doc_code">
2812 i32 (i32) asm "bswap $0", "=r,r"
2815 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2816 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2819 <pre class="doc_code">
2820 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2823 <p>Inline asms with side effects not visible in the constraint list must be
2824 marked as having side effects. This is done through the use of the
2825 '<tt>sideeffect</tt>' keyword, like so:</p>
2827 <pre class="doc_code">
2828 call void asm sideeffect "eieio", ""()
2831 <p>In some cases inline asms will contain code that will not work unless the
2832 stack is aligned in some way, such as calls or SSE instructions on x86,
2833 yet will not contain code that does that alignment within the asm.
2834 The compiler should make conservative assumptions about what the asm might
2835 contain and should generate its usual stack alignment code in the prologue
2836 if the '<tt>alignstack</tt>' keyword is present:</p>
2838 <pre class="doc_code">
2839 call void asm alignstack "eieio", ""()
2842 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2845 <p>TODO: The format of the asm and constraints string still need to be
2846 documented here. Constraints on what can be done (e.g. duplication, moving,
2847 etc need to be documented). This is probably best done by reference to
2848 another document that covers inline asm from a holistic perspective.</p>
2851 <a name="inlineasm_md">Inline Asm Metadata</a>
2856 <p>The call instructions that wrap inline asm nodes may have a "!srcloc" MDNode
2857 attached to it that contains a list of constant integers. If present, the
2858 code generator will use the integer as the location cookie value when report
2859 errors through the LLVMContext error reporting mechanisms. This allows a
2860 front-end to correlate backend errors that occur with inline asm back to the
2861 source code that produced it. For example:</p>
2863 <pre class="doc_code">
2864 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2866 !42 = !{ i32 1234567 }
2869 <p>It is up to the front-end to make sense of the magic numbers it places in the
2870 IR. If the MDNode contains multiple constants, the code generator will use
2871 the one that corresponds to the line of the asm that the error occurs on.</p>
2877 <!-- ======================================================================= -->
2879 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2884 <p>LLVM IR allows metadata to be attached to instructions in the program that
2885 can convey extra information about the code to the optimizers and code
2886 generator. One example application of metadata is source-level debug
2887 information. There are two metadata primitives: strings and nodes. All
2888 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2889 preceding exclamation point ('<tt>!</tt>').</p>
2891 <p>A metadata string is a string surrounded by double quotes. It can contain
2892 any character by escaping non-printable characters with "\xx" where "xx" is
2893 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2895 <p>Metadata nodes are represented with notation similar to structure constants
2896 (a comma separated list of elements, surrounded by braces and preceded by an
2897 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2898 10}</tt>". Metadata nodes can have any values as their operand.</p>
2900 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2901 metadata nodes, which can be looked up in the module symbol table. For
2902 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2904 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2905 function is using two metadata arguments.</p>
2907 <div class="doc_code">
2909 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2913 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2914 attached with <tt>add</tt> instruction using <tt>!dbg</tt> identifier.</p>
2916 <div class="doc_code">
2918 %indvar.next = add i64 %indvar, 1, !dbg !21
2922 <p>More information about specific metadata nodes recognized by the optimizers
2923 and code generator is found below.</p>
2926 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
2931 <p>In LLVM IR, memory does not have types, so LLVM's own type system is not
2932 suitable for doing TBAA. Instead, metadata is added to the IR to describe
2933 a type system of a higher level language. This can be used to implement
2934 typical C/C++ TBAA, but it can also be used to implement custom alias
2935 analysis behavior for other languages.</p>
2937 <p>The current metadata format is very simple. TBAA metadata nodes have up to
2938 three fields, e.g.:</p>
2940 <div class="doc_code">
2942 !0 = metadata !{ metadata !"an example type tree" }
2943 !1 = metadata !{ metadata !"int", metadata !0 }
2944 !2 = metadata !{ metadata !"float", metadata !0 }
2945 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2949 <p>The first field is an identity field. It can be any value, usually
2950 a metadata string, which uniquely identifies the type. The most important
2951 name in the tree is the name of the root node. Two trees with
2952 different root node names are entirely disjoint, even if they
2953 have leaves with common names.</p>
2955 <p>The second field identifies the type's parent node in the tree, or
2956 is null or omitted for a root node. A type is considered to alias
2957 all of its descendants and all of its ancestors in the tree. Also,
2958 a type is considered to alias all types in other trees, so that
2959 bitcode produced from multiple front-ends is handled conservatively.</p>
2961 <p>If the third field is present, it's an integer which if equal to 1
2962 indicates that the type is "constant" (meaning
2963 <tt>pointsToConstantMemory</tt> should return true; see
2964 <a href="AliasAnalysis.html#OtherItfs">other useful
2965 <tt>AliasAnalysis</tt> methods</a>).</p>
2973 <!-- *********************************************************************** -->
2975 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2977 <!-- *********************************************************************** -->
2979 <p>LLVM has a number of "magic" global variables that contain data that affect
2980 code generation or other IR semantics. These are documented here. All globals
2981 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2982 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2985 <!-- ======================================================================= -->
2987 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2992 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2993 href="#linkage_appending">appending linkage</a>. This array contains a list of
2994 pointers to global variables and functions which may optionally have a pointer
2995 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
3001 @llvm.used = appending global [2 x i8*] [
3003 i8* bitcast (i32* @Y to i8*)
3004 ], section "llvm.metadata"
3007 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
3008 compiler, assembler, and linker are required to treat the symbol as if there is
3009 a reference to the global that it cannot see. For example, if a variable has
3010 internal linkage and no references other than that from the <tt>@llvm.used</tt>
3011 list, it cannot be deleted. This is commonly used to represent references from
3012 inline asms and other things the compiler cannot "see", and corresponds to
3013 "attribute((used))" in GNU C.</p>
3015 <p>On some targets, the code generator must emit a directive to the assembler or
3016 object file to prevent the assembler and linker from molesting the symbol.</p>
3020 <!-- ======================================================================= -->
3022 <a name="intg_compiler_used">
3023 The '<tt>llvm.compiler.used</tt>' Global Variable
3029 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
3030 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
3031 touching the symbol. On targets that support it, this allows an intelligent
3032 linker to optimize references to the symbol without being impeded as it would be
3033 by <tt>@llvm.used</tt>.</p>
3035 <p>This is a rare construct that should only be used in rare circumstances, and
3036 should not be exposed to source languages.</p>
3040 <!-- ======================================================================= -->
3042 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
3047 %0 = type { i32, void ()* }
3048 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3050 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor functions and associated priorities. The functions referenced by this array will be called in ascending order of priority (i.e. lowest first) when the module is loaded. The order of functions with the same priority is not defined.
3055 <!-- ======================================================================= -->
3057 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3062 %0 = type { i32, void ()* }
3063 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3066 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions and associated priorities. The functions referenced by this array will be called in descending order of priority (i.e. highest first) when the module is loaded. The order of functions with the same priority is not defined.
3073 <!-- *********************************************************************** -->
3074 <h2><a name="instref">Instruction Reference</a></h2>
3075 <!-- *********************************************************************** -->
3079 <p>The LLVM instruction set consists of several different classifications of
3080 instructions: <a href="#terminators">terminator
3081 instructions</a>, <a href="#binaryops">binary instructions</a>,
3082 <a href="#bitwiseops">bitwise binary instructions</a>,
3083 <a href="#memoryops">memory instructions</a>, and
3084 <a href="#otherops">other instructions</a>.</p>
3086 <!-- ======================================================================= -->
3088 <a name="terminators">Terminator Instructions</a>
3093 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3094 in a program ends with a "Terminator" instruction, which indicates which
3095 block should be executed after the current block is finished. These
3096 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3097 control flow, not values (the one exception being the
3098 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3100 <p>The terminator instructions are:
3101 '<a href="#i_ret"><tt>ret</tt></a>',
3102 '<a href="#i_br"><tt>br</tt></a>',
3103 '<a href="#i_switch"><tt>switch</tt></a>',
3104 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3105 '<a href="#i_invoke"><tt>invoke</tt></a>',
3106 '<a href="#i_unwind"><tt>unwind</tt></a>',
3107 '<a href="#i_resume"><tt>resume</tt></a>', and
3108 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3110 <!-- _______________________________________________________________________ -->
3112 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3119 ret <type> <value> <i>; Return a value from a non-void function</i>
3120 ret void <i>; Return from void function</i>
3124 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3125 a value) from a function back to the caller.</p>
3127 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3128 value and then causes control flow, and one that just causes control flow to
3132 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3133 return value. The type of the return value must be a
3134 '<a href="#t_firstclass">first class</a>' type.</p>
3136 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3137 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3138 value or a return value with a type that does not match its type, or if it
3139 has a void return type and contains a '<tt>ret</tt>' instruction with a
3143 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3144 the calling function's context. If the caller is a
3145 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3146 instruction after the call. If the caller was an
3147 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3148 the beginning of the "normal" destination block. If the instruction returns
3149 a value, that value shall set the call or invoke instruction's return
3154 ret i32 5 <i>; Return an integer value of 5</i>
3155 ret void <i>; Return from a void function</i>
3156 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3160 <!-- _______________________________________________________________________ -->
3162 <a name="i_br">'<tt>br</tt>' Instruction</a>
3169 br i1 <cond>, label <iftrue>, label <iffalse>
3170 br label <dest> <i>; Unconditional branch</i>
3174 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3175 different basic block in the current function. There are two forms of this
3176 instruction, corresponding to a conditional branch and an unconditional
3180 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3181 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3182 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3186 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3187 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3188 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3189 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3194 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3195 br i1 %cond, label %IfEqual, label %IfUnequal
3197 <a href="#i_ret">ret</a> i32 1
3199 <a href="#i_ret">ret</a> i32 0
3204 <!-- _______________________________________________________________________ -->
3206 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3213 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3217 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3218 several different places. It is a generalization of the '<tt>br</tt>'
3219 instruction, allowing a branch to occur to one of many possible
3223 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3224 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3225 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3226 The table is not allowed to contain duplicate constant entries.</p>
3229 <p>The <tt>switch</tt> instruction specifies a table of values and
3230 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3231 is searched for the given value. If the value is found, control flow is
3232 transferred to the corresponding destination; otherwise, control flow is
3233 transferred to the default destination.</p>
3235 <h5>Implementation:</h5>
3236 <p>Depending on properties of the target machine and the particular
3237 <tt>switch</tt> instruction, this instruction may be code generated in
3238 different ways. For example, it could be generated as a series of chained
3239 conditional branches or with a lookup table.</p>
3243 <i>; Emulate a conditional br instruction</i>
3244 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3245 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3247 <i>; Emulate an unconditional br instruction</i>
3248 switch i32 0, label %dest [ ]
3250 <i>; Implement a jump table:</i>
3251 switch i32 %val, label %otherwise [ i32 0, label %onzero
3253 i32 2, label %ontwo ]
3259 <!-- _______________________________________________________________________ -->
3261 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3268 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3273 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3274 within the current function, whose address is specified by
3275 "<tt>address</tt>". Address must be derived from a <a
3276 href="#blockaddress">blockaddress</a> constant.</p>
3280 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3281 rest of the arguments indicate the full set of possible destinations that the
3282 address may point to. Blocks are allowed to occur multiple times in the
3283 destination list, though this isn't particularly useful.</p>
3285 <p>This destination list is required so that dataflow analysis has an accurate
3286 understanding of the CFG.</p>
3290 <p>Control transfers to the block specified in the address argument. All
3291 possible destination blocks must be listed in the label list, otherwise this
3292 instruction has undefined behavior. This implies that jumps to labels
3293 defined in other functions have undefined behavior as well.</p>
3295 <h5>Implementation:</h5>
3297 <p>This is typically implemented with a jump through a register.</p>
3301 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3307 <!-- _______________________________________________________________________ -->
3309 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3316 <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>]
3317 to label <normal label> unwind label <exception label>
3321 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3322 function, with the possibility of control flow transfer to either the
3323 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3324 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3325 control flow will return to the "normal" label. If the callee (or any
3326 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
3327 instruction, control is interrupted and continued at the dynamically nearest
3328 "exception" label.</p>
3330 <p>The '<tt>exception</tt>' label is a
3331 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3332 exception. As such, '<tt>exception</tt>' label is required to have the
3333 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3334 the information about about the behavior of the program after unwinding
3335 happens, as its first non-PHI instruction. The restrictions on the
3336 "<tt>landingpad</tt>" instruction's tightly couples it to the
3337 "<tt>invoke</tt>" instruction, so that the important information contained
3338 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3342 <p>This instruction requires several arguments:</p>
3345 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3346 convention</a> the call should use. If none is specified, the call
3347 defaults to using C calling conventions.</li>
3349 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3350 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3351 '<tt>inreg</tt>' attributes are valid here.</li>
3353 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3354 function value being invoked. In most cases, this is a direct function
3355 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3356 off an arbitrary pointer to function value.</li>
3358 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3359 function to be invoked. </li>
3361 <li>'<tt>function args</tt>': argument list whose types match the function
3362 signature argument types and parameter attributes. All arguments must be
3363 of <a href="#t_firstclass">first class</a> type. If the function
3364 signature indicates the function accepts a variable number of arguments,
3365 the extra arguments can be specified.</li>
3367 <li>'<tt>normal label</tt>': the label reached when the called function
3368 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3370 <li>'<tt>exception label</tt>': the label reached when a callee returns with
3371 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
3373 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3374 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3375 '<tt>readnone</tt>' attributes are valid here.</li>
3379 <p>This instruction is designed to operate as a standard
3380 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3381 primary difference is that it establishes an association with a label, which
3382 is used by the runtime library to unwind the stack.</p>
3384 <p>This instruction is used in languages with destructors to ensure that proper
3385 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3386 exception. Additionally, this is important for implementation of
3387 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3389 <p>For the purposes of the SSA form, the definition of the value returned by the
3390 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3391 block to the "normal" label. If the callee unwinds then no return value is
3394 <p>Note that the code generator does not yet completely support unwind, and
3395 that the invoke/unwind semantics are likely to change in future versions.</p>
3399 %retval = invoke i32 @Test(i32 15) to label %Continue
3400 unwind label %TestCleanup <i>; {i32}:retval set</i>
3401 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3402 unwind label %TestCleanup <i>; {i32}:retval set</i>
3407 <!-- _______________________________________________________________________ -->
3410 <a name="i_unwind">'<tt>unwind</tt>' Instruction</a>
3421 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
3422 at the first callee in the dynamic call stack which used
3423 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
3424 This is primarily used to implement exception handling.</p>
3427 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
3428 immediately halt. The dynamic call stack is then searched for the
3429 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
3430 Once found, execution continues at the "exceptional" destination block
3431 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
3432 instruction in the dynamic call chain, undefined behavior results.</p>
3434 <p>Note that the code generator does not yet completely support unwind, and
3435 that the invoke/unwind semantics are likely to change in future versions.</p>
3439 <!-- _______________________________________________________________________ -->
3442 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3449 resume <type> <value>
3453 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3457 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3458 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3462 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3463 (in-flight) exception whose unwinding was interrupted with
3464 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3468 resume { i8*, i32 } %exn
3473 <!-- _______________________________________________________________________ -->
3476 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3487 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3488 instruction is used to inform the optimizer that a particular portion of the
3489 code is not reachable. This can be used to indicate that the code after a
3490 no-return function cannot be reached, and other facts.</p>
3493 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3499 <!-- ======================================================================= -->
3501 <a name="binaryops">Binary Operations</a>
3506 <p>Binary operators are used to do most of the computation in a program. They
3507 require two operands of the same type, execute an operation on them, and
3508 produce a single value. The operands might represent multiple data, as is
3509 the case with the <a href="#t_vector">vector</a> data type. The result value
3510 has the same type as its operands.</p>
3512 <p>There are several different binary operators:</p>
3514 <!-- _______________________________________________________________________ -->
3516 <a name="i_add">'<tt>add</tt>' Instruction</a>
3523 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3524 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3525 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3526 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3530 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3533 <p>The two arguments to the '<tt>add</tt>' instruction must
3534 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3535 integer values. Both arguments must have identical types.</p>
3538 <p>The value produced is the integer sum of the two operands.</p>
3540 <p>If the sum has unsigned overflow, the result returned is the mathematical
3541 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3543 <p>Because LLVM integers use a two's complement representation, this instruction
3544 is appropriate for both signed and unsigned integers.</p>
3546 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3547 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3548 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3549 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3550 respectively, occurs.</p>
3554 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3559 <!-- _______________________________________________________________________ -->
3561 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3568 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3572 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3575 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3576 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3577 floating point values. Both arguments must have identical types.</p>
3580 <p>The value produced is the floating point sum of the two operands.</p>
3584 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3589 <!-- _______________________________________________________________________ -->
3591 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3598 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3599 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3600 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3601 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3605 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3608 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3609 '<tt>neg</tt>' instruction present in most other intermediate
3610 representations.</p>
3613 <p>The two arguments to the '<tt>sub</tt>' instruction must
3614 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3615 integer values. Both arguments must have identical types.</p>
3618 <p>The value produced is the integer difference of the two operands.</p>
3620 <p>If the difference has unsigned overflow, the result returned is the
3621 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3624 <p>Because LLVM integers use a two's complement representation, this instruction
3625 is appropriate for both signed and unsigned integers.</p>
3627 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3628 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3629 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3630 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3631 respectively, occurs.</p>
3635 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3636 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3641 <!-- _______________________________________________________________________ -->
3643 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3650 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3654 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3657 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3658 '<tt>fneg</tt>' instruction present in most other intermediate
3659 representations.</p>
3662 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3663 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3664 floating point values. Both arguments must have identical types.</p>
3667 <p>The value produced is the floating point difference of the two operands.</p>
3671 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3672 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3677 <!-- _______________________________________________________________________ -->
3679 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3686 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3687 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3688 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3689 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3693 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3696 <p>The two arguments to the '<tt>mul</tt>' instruction must
3697 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3698 integer values. Both arguments must have identical types.</p>
3701 <p>The value produced is the integer product of the two operands.</p>
3703 <p>If the result of the multiplication has unsigned overflow, the result
3704 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3705 width of the result.</p>
3707 <p>Because LLVM integers use a two's complement representation, and the result
3708 is the same width as the operands, this instruction returns the correct
3709 result for both signed and unsigned integers. If a full product
3710 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3711 be sign-extended or zero-extended as appropriate to the width of the full
3714 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3715 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3716 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3717 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3718 respectively, occurs.</p>
3722 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3727 <!-- _______________________________________________________________________ -->
3729 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3736 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3740 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3743 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3744 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3745 floating point values. Both arguments must have identical types.</p>
3748 <p>The value produced is the floating point product of the two operands.</p>
3752 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3757 <!-- _______________________________________________________________________ -->
3759 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
3766 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3767 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3771 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3774 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3775 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3776 values. Both arguments must have identical types.</p>
3779 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3781 <p>Note that unsigned integer division and signed integer division are distinct
3782 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3784 <p>Division by zero leads to undefined behavior.</p>
3786 <p>If the <tt>exact</tt> keyword is present, the result value of the
3787 <tt>udiv</tt> is a <a href="#trapvalues">trap value</a> if %op1 is not a
3788 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
3793 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3798 <!-- _______________________________________________________________________ -->
3800 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
3807 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3808 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3812 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3815 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3816 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3817 values. Both arguments must have identical types.</p>
3820 <p>The value produced is the signed integer quotient of the two operands rounded
3823 <p>Note that signed integer division and unsigned integer division are distinct
3824 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3826 <p>Division by zero leads to undefined behavior. Overflow also leads to
3827 undefined behavior; this is a rare case, but can occur, for example, by doing
3828 a 32-bit division of -2147483648 by -1.</p>
3830 <p>If the <tt>exact</tt> keyword is present, the result value of the
3831 <tt>sdiv</tt> is a <a href="#trapvalues">trap value</a> if the result would
3836 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3841 <!-- _______________________________________________________________________ -->
3843 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
3850 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3854 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3857 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3858 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3859 floating point values. Both arguments must have identical types.</p>
3862 <p>The value produced is the floating point quotient of the two operands.</p>
3866 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3871 <!-- _______________________________________________________________________ -->
3873 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3880 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3884 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3885 division of its two arguments.</p>
3888 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3889 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3890 values. Both arguments must have identical types.</p>
3893 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3894 This instruction always performs an unsigned division to get the
3897 <p>Note that unsigned integer remainder and signed integer remainder are
3898 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3900 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3904 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3909 <!-- _______________________________________________________________________ -->
3911 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3918 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3922 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3923 division of its two operands. This instruction can also take
3924 <a href="#t_vector">vector</a> versions of the values in which case the
3925 elements must be integers.</p>
3928 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3929 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3930 values. Both arguments must have identical types.</p>
3933 <p>This instruction returns the <i>remainder</i> of a division (where the result
3934 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
3935 <i>modulo</i> operator (where the result is either zero or has the same sign
3936 as the divisor, <tt>op2</tt>) of a value.
3937 For more information about the difference,
3938 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3939 Math Forum</a>. For a table of how this is implemented in various languages,
3940 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3941 Wikipedia: modulo operation</a>.</p>
3943 <p>Note that signed integer remainder and unsigned integer remainder are
3944 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3946 <p>Taking the remainder of a division by zero leads to undefined behavior.
3947 Overflow also leads to undefined behavior; this is a rare case, but can
3948 occur, for example, by taking the remainder of a 32-bit division of
3949 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3950 lets srem be implemented using instructions that return both the result of
3951 the division and the remainder.)</p>
3955 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3960 <!-- _______________________________________________________________________ -->
3962 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
3969 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3973 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3974 its two operands.</p>
3977 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3978 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3979 floating point values. Both arguments must have identical types.</p>
3982 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3983 has the same sign as the dividend.</p>
3987 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3994 <!-- ======================================================================= -->
3996 <a name="bitwiseops">Bitwise Binary Operations</a>
4001 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
4002 program. They are generally very efficient instructions and can commonly be
4003 strength reduced from other instructions. They require two operands of the
4004 same type, execute an operation on them, and produce a single value. The
4005 resulting value is the same type as its operands.</p>
4007 <!-- _______________________________________________________________________ -->
4009 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
4016 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4017 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4018 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4019 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4023 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
4024 a specified number of bits.</p>
4027 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
4028 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4029 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4032 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
4033 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
4034 is (statically or dynamically) negative or equal to or larger than the number
4035 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4036 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4037 shift amount in <tt>op2</tt>.</p>
4039 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
4040 <a href="#trapvalues">trap value</a> if it shifts out any non-zero bits. If
4041 the <tt>nsw</tt> keyword is present, then the shift produces a
4042 <a href="#trapvalues">trap value</a> if it shifts out any bits that disagree
4043 with the resultant sign bit. As such, NUW/NSW have the same semantics as
4044 they would if the shift were expressed as a mul instruction with the same
4045 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
4049 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
4050 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
4051 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
4052 <result> = shl i32 1, 32 <i>; undefined</i>
4053 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4058 <!-- _______________________________________________________________________ -->
4060 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4067 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4068 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4072 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4073 operand shifted to the right a specified number of bits with zero fill.</p>
4076 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4077 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4078 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4081 <p>This instruction always performs a logical shift right operation. The most
4082 significant bits of the result will be filled with zero bits after the shift.
4083 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4084 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4085 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4086 shift amount in <tt>op2</tt>.</p>
4088 <p>If the <tt>exact</tt> keyword is present, the result value of the
4089 <tt>lshr</tt> is a <a href="#trapvalues">trap value</a> if any of the bits
4090 shifted out are non-zero.</p>
4095 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4096 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4097 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4098 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4099 <result> = lshr i32 1, 32 <i>; undefined</i>
4100 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4105 <!-- _______________________________________________________________________ -->
4107 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4114 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4115 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4119 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4120 operand shifted to the right a specified number of bits with sign
4124 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4125 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4126 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4129 <p>This instruction always performs an arithmetic shift right operation, The
4130 most significant bits of the result will be filled with the sign bit
4131 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4132 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4133 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4134 the corresponding shift amount in <tt>op2</tt>.</p>
4136 <p>If the <tt>exact</tt> keyword is present, the result value of the
4137 <tt>ashr</tt> is a <a href="#trapvalues">trap value</a> if any of the bits
4138 shifted out are non-zero.</p>
4142 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4143 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4144 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4145 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4146 <result> = ashr i32 1, 32 <i>; undefined</i>
4147 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4152 <!-- _______________________________________________________________________ -->
4154 <a name="i_and">'<tt>and</tt>' Instruction</a>
4161 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4165 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4169 <p>The two arguments to the '<tt>and</tt>' instruction must be
4170 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4171 values. Both arguments must have identical types.</p>
4174 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4176 <table border="1" cellspacing="0" cellpadding="4">
4208 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4209 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4210 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4213 <!-- _______________________________________________________________________ -->
4215 <a name="i_or">'<tt>or</tt>' Instruction</a>
4222 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4226 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4230 <p>The two arguments to the '<tt>or</tt>' instruction must be
4231 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4232 values. Both arguments must have identical types.</p>
4235 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4237 <table border="1" cellspacing="0" cellpadding="4">
4269 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4270 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4271 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4276 <!-- _______________________________________________________________________ -->
4278 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4285 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4289 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4290 its two operands. The <tt>xor</tt> is used to implement the "one's
4291 complement" operation, which is the "~" operator in C.</p>
4294 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4295 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4296 values. Both arguments must have identical types.</p>
4299 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4301 <table border="1" cellspacing="0" cellpadding="4">
4333 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4334 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4335 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4336 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4343 <!-- ======================================================================= -->
4345 <a name="vectorops">Vector Operations</a>
4350 <p>LLVM supports several instructions to represent vector operations in a
4351 target-independent manner. These instructions cover the element-access and
4352 vector-specific operations needed to process vectors effectively. While LLVM
4353 does directly support these vector operations, many sophisticated algorithms
4354 will want to use target-specific intrinsics to take full advantage of a
4355 specific target.</p>
4357 <!-- _______________________________________________________________________ -->
4359 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4366 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4370 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4371 from a vector at a specified index.</p>
4375 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4376 of <a href="#t_vector">vector</a> type. The second operand is an index
4377 indicating the position from which to extract the element. The index may be
4381 <p>The result is a scalar of the same type as the element type of
4382 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4383 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4384 results are undefined.</p>
4388 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4393 <!-- _______________________________________________________________________ -->
4395 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4402 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4406 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4407 vector at a specified index.</p>
4410 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4411 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4412 whose type must equal the element type of the first operand. The third
4413 operand is an index indicating the position at which to insert the value.
4414 The index may be a variable.</p>
4417 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4418 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4419 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4420 results are undefined.</p>
4424 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4429 <!-- _______________________________________________________________________ -->
4431 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4438 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4442 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4443 from two input vectors, returning a vector with the same element type as the
4444 input and length that is the same as the shuffle mask.</p>
4447 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4448 with types that match each other. The third argument is a shuffle mask whose
4449 element type is always 'i32'. The result of the instruction is a vector
4450 whose length is the same as the shuffle mask and whose element type is the
4451 same as the element type of the first two operands.</p>
4453 <p>The shuffle mask operand is required to be a constant vector with either
4454 constant integer or undef values.</p>
4457 <p>The elements of the two input vectors are numbered from left to right across
4458 both of the vectors. The shuffle mask operand specifies, for each element of
4459 the result vector, which element of the two input vectors the result element
4460 gets. The element selector may be undef (meaning "don't care") and the
4461 second operand may be undef if performing a shuffle from only one vector.</p>
4465 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4466 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4467 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4468 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4469 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4470 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4471 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4472 <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>
4479 <!-- ======================================================================= -->
4481 <a name="aggregateops">Aggregate Operations</a>
4486 <p>LLVM supports several instructions for working with
4487 <a href="#t_aggregate">aggregate</a> values.</p>
4489 <!-- _______________________________________________________________________ -->
4491 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4498 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4502 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4503 from an <a href="#t_aggregate">aggregate</a> value.</p>
4506 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4507 of <a href="#t_struct">struct</a> or
4508 <a href="#t_array">array</a> type. The operands are constant indices to
4509 specify which value to extract in a similar manner as indices in a
4510 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4511 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4513 <li>Since the value being indexed is not a pointer, the first index is
4514 omitted and assumed to be zero.</li>
4515 <li>At least one index must be specified.</li>
4516 <li>Not only struct indices but also array indices must be in
4521 <p>The result is the value at the position in the aggregate specified by the
4526 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4531 <!-- _______________________________________________________________________ -->
4533 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4540 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4544 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4545 in an <a href="#t_aggregate">aggregate</a> value.</p>
4548 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4549 of <a href="#t_struct">struct</a> or
4550 <a href="#t_array">array</a> type. The second operand is a first-class
4551 value to insert. The following operands are constant indices indicating
4552 the position at which to insert the value in a similar manner as indices in a
4553 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4554 value to insert must have the same type as the value identified by the
4558 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4559 that of <tt>val</tt> except that the value at the position specified by the
4560 indices is that of <tt>elt</tt>.</p>
4564 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4565 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4566 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4573 <!-- ======================================================================= -->
4575 <a name="memoryops">Memory Access and Addressing Operations</a>
4580 <p>A key design point of an SSA-based representation is how it represents
4581 memory. In LLVM, no memory locations are in SSA form, which makes things
4582 very simple. This section describes how to read, write, and allocate
4585 <!-- _______________________________________________________________________ -->
4587 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4594 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4598 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4599 currently executing function, to be automatically released when this function
4600 returns to its caller. The object is always allocated in the generic address
4601 space (address space zero).</p>
4604 <p>The '<tt>alloca</tt>' instruction
4605 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4606 runtime stack, returning a pointer of the appropriate type to the program.
4607 If "NumElements" is specified, it is the number of elements allocated,
4608 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4609 specified, the value result of the allocation is guaranteed to be aligned to
4610 at least that boundary. If not specified, or if zero, the target can choose
4611 to align the allocation on any convenient boundary compatible with the
4614 <p>'<tt>type</tt>' may be any sized type.</p>
4617 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4618 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4619 memory is automatically released when the function returns. The
4620 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4621 variables that must have an address available. When the function returns
4622 (either with the <tt><a href="#i_ret">ret</a></tt>
4623 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
4624 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4628 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4629 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4630 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4631 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4636 <!-- _______________________________________________________________________ -->
4638 <a name="i_load">'<tt>load</tt>' Instruction</a>
4645 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4646 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4647 !<index> = !{ i32 1 }
4651 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4654 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4655 from which to load. The pointer must point to
4656 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4657 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4658 number or order of execution of this <tt>load</tt> with other <a
4659 href="#volatile">volatile operations</a>.</p>
4661 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
4662 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4663 argument. The <code>release</code> and <code>acq_rel</code> orderings are
4664 not valid on <code>load</code> instructions. Atomic loads produce <a
4665 href="#memorymodel">defined</a> results when they may see multiple atomic
4666 stores. The type of the pointee must be an integer type whose bit width
4667 is a power of two greater than or equal to eight and less than or equal
4668 to a target-specific size limit. <code>align</code> must be explicitly
4669 specified on atomic loads, and the load has undefined behavior if the
4670 alignment is not set to a value which is at least the size in bytes of
4671 the pointee. <code>!nontemporal</code> does not have any defined semantics
4672 for atomic loads.</p>
4674 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4675 operation (that is, the alignment of the memory address). A value of 0 or an
4676 omitted <tt>align</tt> argument means that the operation has the preferential
4677 alignment for the target. It is the responsibility of the code emitter to
4678 ensure that the alignment information is correct. Overestimating the
4679 alignment results in undefined behavior. Underestimating the alignment may
4680 produce less efficient code. An alignment of 1 is always safe.</p>
4682 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4683 metatadata name <index> corresponding to a metadata node with
4684 one <tt>i32</tt> entry of value 1. The existence of
4685 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4686 and code generator that this load is not expected to be reused in the cache.
4687 The code generator may select special instructions to save cache bandwidth,
4688 such as the <tt>MOVNT</tt> instruction on x86.</p>
4691 <p>The location of memory pointed to is loaded. If the value being loaded is of
4692 scalar type then the number of bytes read does not exceed the minimum number
4693 of bytes needed to hold all bits of the type. For example, loading an
4694 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4695 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4696 is undefined if the value was not originally written using a store of the
4701 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4702 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4703 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4708 <!-- _______________________________________________________________________ -->
4710 <a name="i_store">'<tt>store</tt>' Instruction</a>
4717 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4718 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
4722 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4725 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4726 and an address at which to store it. The type of the
4727 '<tt><pointer></tt>' operand must be a pointer to
4728 the <a href="#t_firstclass">first class</a> type of the
4729 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
4730 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
4731 order of execution of this <tt>store</tt> with other <a
4732 href="#volatile">volatile operations</a>.</p>
4734 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
4735 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4736 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
4737 valid on <code>store</code> instructions. Atomic loads produce <a
4738 href="#memorymodel">defined</a> results when they may see multiple atomic
4739 stores. The type of the pointee must be an integer type whose bit width
4740 is a power of two greater than or equal to eight and less than or equal
4741 to a target-specific size limit. <code>align</code> must be explicitly
4742 specified on atomic stores, and the store has undefined behavior if the
4743 alignment is not set to a value which is at least the size in bytes of
4744 the pointee. <code>!nontemporal</code> does not have any defined semantics
4745 for atomic stores.</p>
4747 <p>The optional constant "align" argument specifies the alignment of the
4748 operation (that is, the alignment of the memory address). A value of 0 or an
4749 omitted "align" argument means that the operation has the preferential
4750 alignment for the target. It is the responsibility of the code emitter to
4751 ensure that the alignment information is correct. Overestimating the
4752 alignment results in an undefined behavior. Underestimating the alignment may
4753 produce less efficient code. An alignment of 1 is always safe.</p>
4755 <p>The optional !nontemporal metadata must reference a single metatadata
4756 name <index> corresponding to a metadata node with one i32 entry of
4757 value 1. The existence of the !nontemporal metatadata on the
4758 instruction tells the optimizer and code generator that this load is
4759 not expected to be reused in the cache. The code generator may
4760 select special instructions to save cache bandwidth, such as the
4761 MOVNT instruction on x86.</p>
4765 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4766 location specified by the '<tt><pointer></tt>' operand. If
4767 '<tt><value></tt>' is of scalar type then the number of bytes written
4768 does not exceed the minimum number of bytes needed to hold all bits of the
4769 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4770 writing a value of a type like <tt>i20</tt> with a size that is not an
4771 integral number of bytes, it is unspecified what happens to the extra bits
4772 that do not belong to the type, but they will typically be overwritten.</p>
4776 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4777 store i32 3, i32* %ptr <i>; yields {void}</i>
4778 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4783 <!-- _______________________________________________________________________ -->
4785 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
4792 fence [singlethread] <ordering> <i>; yields {void}</i>
4796 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
4797 between operations.</p>
4799 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
4800 href="#ordering">ordering</a> argument which defines what
4801 <i>synchronizes-with</i> edges they add. They can only be given
4802 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
4803 <code>seq_cst</code> orderings.</p>
4806 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
4807 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
4808 <code>acquire</code> ordering semantics if and only if there exist atomic
4809 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
4810 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
4811 <var>X</var> modifies <var>M</var> (either directly or through some side effect
4812 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
4813 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
4814 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
4815 than an explicit <code>fence</code>, one (but not both) of the atomic operations
4816 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
4817 <code>acquire</code> (resp.) ordering constraint and still
4818 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
4819 <i>happens-before</i> edge.</p>
4821 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
4822 having both <code>acquire</code> and <code>release</code> semantics specified
4823 above, participates in the global program order of other <code>seq_cst</code>
4824 operations and/or fences.</p>
4826 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
4827 specifies that the fence only synchronizes with other fences in the same
4828 thread. (This is useful for interacting with signal handlers.)</p>
4832 fence acquire <i>; yields {void}</i>
4833 fence singlethread seq_cst <i>; yields {void}</i>
4838 <!-- _______________________________________________________________________ -->
4840 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
4847 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
4851 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
4852 It loads a value in memory and compares it to a given value. If they are
4853 equal, it stores a new value into the memory.</p>
4856 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
4857 address to operate on, a value to compare to the value currently be at that
4858 address, and a new value to place at that address if the compared values are
4859 equal. The type of '<var><cmp></var>' must be an integer type whose
4860 bit width is a power of two greater than or equal to eight and less than
4861 or equal to a target-specific size limit. '<var><cmp></var>' and
4862 '<var><new></var>' must have the same type, and the type of
4863 '<var><pointer></var>' must be a pointer to that type. If the
4864 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
4865 optimizer is not allowed to modify the number or order of execution
4866 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
4869 <!-- FIXME: Extend allowed types. -->
4871 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
4872 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
4874 <p>The optional "<code>singlethread</code>" argument declares that the
4875 <code>cmpxchg</code> is only atomic with respect to code (usually signal
4876 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
4877 cmpxchg is atomic with respect to all other code in the system.</p>
4879 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
4880 the size in memory of the operand.
4883 <p>The contents of memory at the location specified by the
4884 '<tt><pointer></tt>' operand is read and compared to
4885 '<tt><cmp></tt>'; if the read value is the equal,
4886 '<tt><new></tt>' is written. The original value at the location
4889 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
4890 purpose of identifying <a href="#release_sequence">release sequences</a>. A
4891 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
4892 parameter determined by dropping any <code>release</code> part of the
4893 <code>cmpxchg</code>'s ordering.</p>
4896 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
4897 optimization work on ARM.)
4899 FIXME: Is a weaker ordering constraint on failure helpful in practice?
4905 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
4906 <a href="#i_br">br</a> label %loop
4909 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
4910 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
4911 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
4912 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
4913 <a href="#i_br">br</a> i1 %success, label %done, label %loop
4921 <!-- _______________________________________________________________________ -->
4923 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
4930 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
4934 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
4937 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
4938 operation to apply, an address whose value to modify, an argument to the
4939 operation. The operation must be one of the following keywords:</p>
4954 <p>The type of '<var><value></var>' must be an integer type whose
4955 bit width is a power of two greater than or equal to eight and less than
4956 or equal to a target-specific size limit. The type of the
4957 '<code><pointer></code>' operand must be a pointer to that type.
4958 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
4959 optimizer is not allowed to modify the number or order of execution of this
4960 <code>atomicrmw</code> with other <a href="#volatile">volatile
4963 <!-- FIXME: Extend allowed types. -->
4966 <p>The contents of memory at the location specified by the
4967 '<tt><pointer></tt>' operand are atomically read, modified, and written
4968 back. The original value at the location is returned. The modification is
4969 specified by the <var>operation</var> argument:</p>
4972 <li>xchg: <code>*ptr = val</code></li>
4973 <li>add: <code>*ptr = *ptr + val</code></li>
4974 <li>sub: <code>*ptr = *ptr - val</code></li>
4975 <li>and: <code>*ptr = *ptr & val</code></li>
4976 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
4977 <li>or: <code>*ptr = *ptr | val</code></li>
4978 <li>xor: <code>*ptr = *ptr ^ val</code></li>
4979 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
4980 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
4981 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
4982 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
4987 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
4992 <!-- _______________________________________________________________________ -->
4994 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
5001 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5002 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5006 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
5007 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
5008 It performs address calculation only and does not access memory.</p>
5011 <p>The first argument is always a pointer, and forms the basis of the
5012 calculation. The remaining arguments are indices that indicate which of the
5013 elements of the aggregate object are indexed. The interpretation of each
5014 index is dependent on the type being indexed into. The first index always
5015 indexes the pointer value given as the first argument, the second index
5016 indexes a value of the type pointed to (not necessarily the value directly
5017 pointed to, since the first index can be non-zero), etc. The first type
5018 indexed into must be a pointer value, subsequent types can be arrays,
5019 vectors, and structs. Note that subsequent types being indexed into
5020 can never be pointers, since that would require loading the pointer before
5021 continuing calculation.</p>
5023 <p>The type of each index argument depends on the type it is indexing into.
5024 When indexing into a (optionally packed) structure, only <tt>i32</tt>
5025 integer <b>constants</b> are allowed. When indexing into an array, pointer
5026 or vector, integers of any width are allowed, and they are not required to be
5027 constant. These integers are treated as signed values where relevant.</p>
5029 <p>For example, let's consider a C code fragment and how it gets compiled to
5032 <pre class="doc_code">
5044 int *foo(struct ST *s) {
5045 return &s[1].Z.B[5][13];
5049 <p>The LLVM code generated by the GCC frontend is:</p>
5051 <pre class="doc_code">
5052 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
5053 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
5055 define i32* @foo(%ST* %s) {
5057 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
5063 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
5064 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
5065 }</tt>' type, a structure. The second index indexes into the third element
5066 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
5067 i8 }</tt>' type, another structure. The third index indexes into the second
5068 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
5069 array. The two dimensions of the array are subscripted into, yielding an
5070 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
5071 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
5073 <p>Note that it is perfectly legal to index partially through a structure,
5074 returning a pointer to an inner element. Because of this, the LLVM code for
5075 the given testcase is equivalent to:</p>
5078 define i32* @foo(%ST* %s) {
5079 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
5080 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
5081 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5082 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5083 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5088 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5089 <tt>getelementptr</tt> is a <a href="#trapvalues">trap value</a> if the
5090 base pointer is not an <i>in bounds</i> address of an allocated object,
5091 or if any of the addresses that would be formed by successive addition of
5092 the offsets implied by the indices to the base address with infinitely
5093 precise signed arithmetic are not an <i>in bounds</i> address of that
5094 allocated object. The <i>in bounds</i> addresses for an allocated object
5095 are all the addresses that point into the object, plus the address one
5096 byte past the end.</p>
5098 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5099 the base address with silently-wrapping two's complement arithmetic. If the
5100 offsets have a different width from the pointer, they are sign-extended or
5101 truncated to the width of the pointer. The result value of the
5102 <tt>getelementptr</tt> may be outside the object pointed to by the base
5103 pointer. The result value may not necessarily be used to access memory
5104 though, even if it happens to point into allocated storage. See the
5105 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5108 <p>The getelementptr instruction is often confusing. For some more insight into
5109 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5113 <i>; yields [12 x i8]*:aptr</i>
5114 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5115 <i>; yields i8*:vptr</i>
5116 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5117 <i>; yields i8*:eptr</i>
5118 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5119 <i>; yields i32*:iptr</i>
5120 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5127 <!-- ======================================================================= -->
5129 <a name="convertops">Conversion Operations</a>
5134 <p>The instructions in this category are the conversion instructions (casting)
5135 which all take a single operand and a type. They perform various bit
5136 conversions on the operand.</p>
5138 <!-- _______________________________________________________________________ -->
5140 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5147 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5151 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5152 type <tt>ty2</tt>.</p>
5155 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5156 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5157 of the same number of integers.
5158 The bit size of the <tt>value</tt> must be larger than
5159 the bit size of the destination type, <tt>ty2</tt>.
5160 Equal sized types are not allowed.</p>
5163 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5164 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5165 source size must be larger than the destination size, <tt>trunc</tt> cannot
5166 be a <i>no-op cast</i>. It will always truncate bits.</p>
5170 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5171 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5172 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5173 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5178 <!-- _______________________________________________________________________ -->
5180 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5187 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5191 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5196 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5197 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5198 of the same number of integers.
5199 The bit size of the <tt>value</tt> must be smaller than
5200 the bit size of the destination type,
5204 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5205 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5207 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5211 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5212 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5213 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5218 <!-- _______________________________________________________________________ -->
5220 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5227 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5231 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5234 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5235 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5236 of the same number of integers.
5237 The bit size of the <tt>value</tt> must be smaller than
5238 the bit size of the destination type,
5242 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5243 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5244 of the type <tt>ty2</tt>.</p>
5246 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5250 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5251 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5252 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5257 <!-- _______________________________________________________________________ -->
5259 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5266 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5270 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5274 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5275 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5276 to cast it to. The size of <tt>value</tt> must be larger than the size of
5277 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5278 <i>no-op cast</i>.</p>
5281 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5282 <a href="#t_floating">floating point</a> type to a smaller
5283 <a href="#t_floating">floating point</a> type. If the value cannot fit
5284 within the destination type, <tt>ty2</tt>, then the results are
5289 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5290 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5295 <!-- _______________________________________________________________________ -->
5297 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5304 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5308 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5309 floating point value.</p>
5312 <p>The '<tt>fpext</tt>' instruction takes a
5313 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5314 a <a href="#t_floating">floating point</a> type to cast it to. The source
5315 type must be smaller than the destination type.</p>
5318 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5319 <a href="#t_floating">floating point</a> type to a larger
5320 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5321 used to make a <i>no-op cast</i> because it always changes bits. Use
5322 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5326 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5327 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5332 <!-- _______________________________________________________________________ -->
5334 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5341 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5345 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5346 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5349 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5350 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5351 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5352 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5353 vector integer type with the same number of elements as <tt>ty</tt></p>
5356 <p>The '<tt>fptoui</tt>' instruction converts its
5357 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5358 towards zero) unsigned integer value. If the value cannot fit
5359 in <tt>ty2</tt>, the results are undefined.</p>
5363 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5364 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5365 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5370 <!-- _______________________________________________________________________ -->
5372 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5379 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5383 <p>The '<tt>fptosi</tt>' instruction converts
5384 <a href="#t_floating">floating point</a> <tt>value</tt> to
5385 type <tt>ty2</tt>.</p>
5388 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5389 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5390 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5391 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5392 vector integer type with the same number of elements as <tt>ty</tt></p>
5395 <p>The '<tt>fptosi</tt>' instruction converts its
5396 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5397 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5398 the results are undefined.</p>
5402 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5403 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5404 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5409 <!-- _______________________________________________________________________ -->
5411 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5418 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5422 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5423 integer and converts that value to the <tt>ty2</tt> type.</p>
5426 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5427 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5428 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5429 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5430 floating point type with the same number of elements as <tt>ty</tt></p>
5433 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5434 integer quantity and converts it to the corresponding floating point
5435 value. If the value cannot fit in the floating point value, the results are
5440 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5441 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5446 <!-- _______________________________________________________________________ -->
5448 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5455 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5459 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5460 and converts that value to the <tt>ty2</tt> type.</p>
5463 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5464 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5465 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5466 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5467 floating point type with the same number of elements as <tt>ty</tt></p>
5470 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5471 quantity and converts it to the corresponding floating point value. If the
5472 value cannot fit in the floating point value, the results are undefined.</p>
5476 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5477 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5482 <!-- _______________________________________________________________________ -->
5484 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5491 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5495 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
5496 the integer type <tt>ty2</tt>.</p>
5499 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5500 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
5501 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
5504 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5505 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5506 truncating or zero extending that value to the size of the integer type. If
5507 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5508 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5509 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5514 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
5515 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
5520 <!-- _______________________________________________________________________ -->
5522 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5529 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5533 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5534 pointer type, <tt>ty2</tt>.</p>
5537 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5538 value to cast, and a type to cast it to, which must be a
5539 <a href="#t_pointer">pointer</a> type.</p>
5542 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5543 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5544 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5545 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5546 than the size of a pointer then a zero extension is done. If they are the
5547 same size, nothing is done (<i>no-op cast</i>).</p>
5551 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5552 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5553 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5558 <!-- _______________________________________________________________________ -->
5560 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5567 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5571 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5572 <tt>ty2</tt> without changing any bits.</p>
5575 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5576 non-aggregate first class value, and a type to cast it to, which must also be
5577 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5578 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5579 identical. If the source type is a pointer, the destination type must also be
5580 a pointer. This instruction supports bitwise conversion of vectors to
5581 integers and to vectors of other types (as long as they have the same
5585 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5586 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5587 this conversion. The conversion is done as if the <tt>value</tt> had been
5588 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
5589 be converted to other pointer types with this instruction. To convert
5590 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5591 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5595 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
5596 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
5597 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
5604 <!-- ======================================================================= -->
5606 <a name="otherops">Other Operations</a>
5611 <p>The instructions in this category are the "miscellaneous" instructions, which
5612 defy better classification.</p>
5614 <!-- _______________________________________________________________________ -->
5616 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5623 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5627 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5628 boolean values based on comparison of its two integer, integer vector, or
5629 pointer operands.</p>
5632 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
5633 the condition code indicating the kind of comparison to perform. It is not a
5634 value, just a keyword. The possible condition code are:</p>
5637 <li><tt>eq</tt>: equal</li>
5638 <li><tt>ne</tt>: not equal </li>
5639 <li><tt>ugt</tt>: unsigned greater than</li>
5640 <li><tt>uge</tt>: unsigned greater or equal</li>
5641 <li><tt>ult</tt>: unsigned less than</li>
5642 <li><tt>ule</tt>: unsigned less or equal</li>
5643 <li><tt>sgt</tt>: signed greater than</li>
5644 <li><tt>sge</tt>: signed greater or equal</li>
5645 <li><tt>slt</tt>: signed less than</li>
5646 <li><tt>sle</tt>: signed less or equal</li>
5649 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
5650 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
5651 typed. They must also be identical types.</p>
5654 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
5655 condition code given as <tt>cond</tt>. The comparison performed always yields
5656 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
5657 result, as follows:</p>
5660 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
5661 <tt>false</tt> otherwise. No sign interpretation is necessary or
5664 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
5665 <tt>false</tt> otherwise. No sign interpretation is necessary or
5668 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
5669 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5671 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
5672 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5673 to <tt>op2</tt>.</li>
5675 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
5676 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5678 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
5679 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5681 <li><tt>sgt</tt>: interprets the operands as signed values and yields
5682 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5684 <li><tt>sge</tt>: interprets the operands as signed values and yields
5685 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5686 to <tt>op2</tt>.</li>
5688 <li><tt>slt</tt>: interprets the operands as signed values and yields
5689 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5691 <li><tt>sle</tt>: interprets the operands as signed values and yields
5692 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5695 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
5696 values are compared as if they were integers.</p>
5698 <p>If the operands are integer vectors, then they are compared element by
5699 element. The result is an <tt>i1</tt> vector with the same number of elements
5700 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
5704 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
5705 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
5706 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
5707 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
5708 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
5709 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
5712 <p>Note that the code generator does not yet support vector types with
5713 the <tt>icmp</tt> instruction.</p>
5717 <!-- _______________________________________________________________________ -->
5719 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
5726 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5730 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
5731 values based on comparison of its operands.</p>
5733 <p>If the operands are floating point scalars, then the result type is a boolean
5734 (<a href="#t_integer"><tt>i1</tt></a>).</p>
5736 <p>If the operands are floating point vectors, then the result type is a vector
5737 of boolean with the same number of elements as the operands being
5741 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
5742 the condition code indicating the kind of comparison to perform. It is not a
5743 value, just a keyword. The possible condition code are:</p>
5746 <li><tt>false</tt>: no comparison, always returns false</li>
5747 <li><tt>oeq</tt>: ordered and equal</li>
5748 <li><tt>ogt</tt>: ordered and greater than </li>
5749 <li><tt>oge</tt>: ordered and greater than or equal</li>
5750 <li><tt>olt</tt>: ordered and less than </li>
5751 <li><tt>ole</tt>: ordered and less than or equal</li>
5752 <li><tt>one</tt>: ordered and not equal</li>
5753 <li><tt>ord</tt>: ordered (no nans)</li>
5754 <li><tt>ueq</tt>: unordered or equal</li>
5755 <li><tt>ugt</tt>: unordered or greater than </li>
5756 <li><tt>uge</tt>: unordered or greater than or equal</li>
5757 <li><tt>ult</tt>: unordered or less than </li>
5758 <li><tt>ule</tt>: unordered or less than or equal</li>
5759 <li><tt>une</tt>: unordered or not equal</li>
5760 <li><tt>uno</tt>: unordered (either nans)</li>
5761 <li><tt>true</tt>: no comparison, always returns true</li>
5764 <p><i>Ordered</i> means that neither operand is a QNAN while
5765 <i>unordered</i> means that either operand may be a QNAN.</p>
5767 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
5768 a <a href="#t_floating">floating point</a> type or
5769 a <a href="#t_vector">vector</a> of floating point type. They must have
5770 identical types.</p>
5773 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
5774 according to the condition code given as <tt>cond</tt>. If the operands are
5775 vectors, then the vectors are compared element by element. Each comparison
5776 performed always yields an <a href="#t_integer">i1</a> result, as
5780 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
5782 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5783 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5785 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5786 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5788 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5789 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5791 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5792 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5794 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5795 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5797 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5798 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5800 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
5802 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
5803 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5805 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
5806 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5808 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
5809 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5811 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
5812 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5814 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
5815 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5817 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
5818 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5820 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
5822 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
5827 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
5828 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
5829 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
5830 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
5833 <p>Note that the code generator does not yet support vector types with
5834 the <tt>fcmp</tt> instruction.</p>
5838 <!-- _______________________________________________________________________ -->
5840 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
5847 <result> = phi <ty> [ <val0>, <label0>], ...
5851 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
5852 SSA graph representing the function.</p>
5855 <p>The type of the incoming values is specified with the first type field. After
5856 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
5857 one pair for each predecessor basic block of the current block. Only values
5858 of <a href="#t_firstclass">first class</a> type may be used as the value
5859 arguments to the PHI node. Only labels may be used as the label
5862 <p>There must be no non-phi instructions between the start of a basic block and
5863 the PHI instructions: i.e. PHI instructions must be first in a basic
5866 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5867 occur on the edge from the corresponding predecessor block to the current
5868 block (but after any definition of an '<tt>invoke</tt>' instruction's return
5869 value on the same edge).</p>
5872 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
5873 specified by the pair corresponding to the predecessor basic block that
5874 executed just prior to the current block.</p>
5878 Loop: ; Infinite loop that counts from 0 on up...
5879 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5880 %nextindvar = add i32 %indvar, 1
5886 <!-- _______________________________________________________________________ -->
5888 <a name="i_select">'<tt>select</tt>' Instruction</a>
5895 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5897 <i>selty</i> is either i1 or {<N x i1>}
5901 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
5902 condition, without branching.</p>
5906 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
5907 values indicating the condition, and two values of the
5908 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
5909 vectors and the condition is a scalar, then entire vectors are selected, not
5910 individual elements.</p>
5913 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
5914 first value argument; otherwise, it returns the second value argument.</p>
5916 <p>If the condition is a vector of i1, then the value arguments must be vectors
5917 of the same size, and the selection is done element by element.</p>
5921 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
5924 <p>Note that the code generator does not yet support conditions
5925 with vector type.</p>
5929 <!-- _______________________________________________________________________ -->
5931 <a name="i_call">'<tt>call</tt>' Instruction</a>
5938 <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>]
5942 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
5945 <p>This instruction requires several arguments:</p>
5948 <li>The optional "tail" marker indicates that the callee function does not
5949 access any allocas or varargs in the caller. Note that calls may be
5950 marked "tail" even if they do not occur before
5951 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
5952 present, the function call is eligible for tail call optimization,
5953 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
5954 optimized into a jump</a>. The code generator may optimize calls marked
5955 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
5956 sibling call optimization</a> when the caller and callee have
5957 matching signatures, or 2) forced tail call optimization when the
5958 following extra requirements are met:
5960 <li>Caller and callee both have the calling
5961 convention <tt>fastcc</tt>.</li>
5962 <li>The call is in tail position (ret immediately follows call and ret
5963 uses value of call or is void).</li>
5964 <li>Option <tt>-tailcallopt</tt> is enabled,
5965 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
5966 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
5967 constraints are met.</a></li>
5971 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
5972 convention</a> the call should use. If none is specified, the call
5973 defaults to using C calling conventions. The calling convention of the
5974 call must match the calling convention of the target function, or else the
5975 behavior is undefined.</li>
5977 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
5978 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
5979 '<tt>inreg</tt>' attributes are valid here.</li>
5981 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
5982 type of the return value. Functions that return no value are marked
5983 <tt><a href="#t_void">void</a></tt>.</li>
5985 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5986 being invoked. The argument types must match the types implied by this
5987 signature. This type can be omitted if the function is not varargs and if
5988 the function type does not return a pointer to a function.</li>
5990 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5991 be invoked. In most cases, this is a direct function invocation, but
5992 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5993 to function value.</li>
5995 <li>'<tt>function args</tt>': argument list whose types match the function
5996 signature argument types and parameter attributes. All arguments must be
5997 of <a href="#t_firstclass">first class</a> type. If the function
5998 signature indicates the function accepts a variable number of arguments,
5999 the extra arguments can be specified.</li>
6001 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
6002 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
6003 '<tt>readnone</tt>' attributes are valid here.</li>
6007 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
6008 a specified function, with its incoming arguments bound to the specified
6009 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
6010 function, control flow continues with the instruction after the function
6011 call, and the return value of the function is bound to the result
6016 %retval = call i32 @test(i32 %argc)
6017 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
6018 %X = tail call i32 @foo() <i>; yields i32</i>
6019 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
6020 call void %foo(i8 97 signext)
6022 %struct.A = type { i32, i8 }
6023 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
6024 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
6025 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
6026 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
6027 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
6030 <p>llvm treats calls to some functions with names and arguments that match the
6031 standard C99 library as being the C99 library functions, and may perform
6032 optimizations or generate code for them under that assumption. This is
6033 something we'd like to change in the future to provide better support for
6034 freestanding environments and non-C-based languages.</p>
6038 <!-- _______________________________________________________________________ -->
6040 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
6047 <resultval> = va_arg <va_list*> <arglist>, <argty>
6051 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6052 the "variable argument" area of a function call. It is used to implement the
6053 <tt>va_arg</tt> macro in C.</p>
6056 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6057 argument. It returns a value of the specified argument type and increments
6058 the <tt>va_list</tt> to point to the next argument. The actual type
6059 of <tt>va_list</tt> is target specific.</p>
6062 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6063 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6064 to the next argument. For more information, see the variable argument
6065 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6067 <p>It is legal for this instruction to be called in a function which does not
6068 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6071 <p><tt>va_arg</tt> is an LLVM instruction instead of
6072 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6076 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6078 <p>Note that the code generator does not yet fully support va_arg on many
6079 targets. Also, it does not currently support va_arg with aggregate types on
6084 <!-- _______________________________________________________________________ -->
6086 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6093 <resultval> = landingpad <somety> personality <type> <pers_fn> <clause>+
6094 <resultval> = landingpad <somety> personality <type> <pers_fn> cleanup <clause>*
6096 <clause> := catch <type> <value>
6097 <clause> := filter <array constant type> <array constant>
6101 <p>The '<tt>landingpad</tt>' instruction is used by
6102 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6103 system</a> to specify that a basic block is a landing pad — one where
6104 the exception lands, and corresponds to the code found in the
6105 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6106 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6107 re-entry to the function. The <tt>resultval</tt> has the
6108 type <tt>somety</tt>.</p>
6111 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6112 function associated with the unwinding mechanism. The optional
6113 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6115 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6116 or <tt>filter</tt> — and contains the global variable representing the
6117 "type" that may be caught or filtered respectively. Unlike the
6118 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6119 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6120 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6121 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6124 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6125 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6126 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6127 calling conventions, how the personality function results are represented in
6128 LLVM IR is target specific.</p>
6130 <p>The clauses are applied in order from top to bottom. If two
6131 <tt>landingpad</tt> instructions are merged together through inlining, the
6132 clauses from the calling function are appended to the list of clauses.</p>
6134 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6137 <li>A landing pad block is a basic block which is the unwind destination of an
6138 '<tt>invoke</tt>' instruction.</li>
6139 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6140 first non-PHI instruction.</li>
6141 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6143 <li>A basic block that is not a landing pad block may not include a
6144 '<tt>landingpad</tt>' instruction.</li>
6145 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6146 personality function.</li>
6151 ;; A landing pad which can catch an integer.
6152 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6154 ;; A landing pad that is a cleanup.
6155 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6157 ;; A landing pad which can catch an integer and can only throw a double.
6158 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6160 filter [1 x i8**] [@_ZTId]
6169 <!-- *********************************************************************** -->
6170 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6171 <!-- *********************************************************************** -->
6175 <p>LLVM supports the notion of an "intrinsic function". These functions have
6176 well known names and semantics and are required to follow certain
6177 restrictions. Overall, these intrinsics represent an extension mechanism for
6178 the LLVM language that does not require changing all of the transformations
6179 in LLVM when adding to the language (or the bitcode reader/writer, the
6180 parser, etc...).</p>
6182 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6183 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6184 begin with this prefix. Intrinsic functions must always be external
6185 functions: you cannot define the body of intrinsic functions. Intrinsic
6186 functions may only be used in call or invoke instructions: it is illegal to
6187 take the address of an intrinsic function. Additionally, because intrinsic
6188 functions are part of the LLVM language, it is required if any are added that
6189 they be documented here.</p>
6191 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6192 family of functions that perform the same operation but on different data
6193 types. Because LLVM can represent over 8 million different integer types,
6194 overloading is used commonly to allow an intrinsic function to operate on any
6195 integer type. One or more of the argument types or the result type can be
6196 overloaded to accept any integer type. Argument types may also be defined as
6197 exactly matching a previous argument's type or the result type. This allows
6198 an intrinsic function which accepts multiple arguments, but needs all of them
6199 to be of the same type, to only be overloaded with respect to a single
6200 argument or the result.</p>
6202 <p>Overloaded intrinsics will have the names of its overloaded argument types
6203 encoded into its function name, each preceded by a period. Only those types
6204 which are overloaded result in a name suffix. Arguments whose type is matched
6205 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6206 can take an integer of any width and returns an integer of exactly the same
6207 integer width. This leads to a family of functions such as
6208 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6209 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6210 suffix is required. Because the argument's type is matched against the return
6211 type, it does not require its own name suffix.</p>
6213 <p>To learn how to add an intrinsic function, please see the
6214 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6216 <!-- ======================================================================= -->
6218 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6223 <p>Variable argument support is defined in LLVM with
6224 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6225 intrinsic functions. These functions are related to the similarly named
6226 macros defined in the <tt><stdarg.h></tt> header file.</p>
6228 <p>All of these functions operate on arguments that use a target-specific value
6229 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6230 not define what this type is, so all transformations should be prepared to
6231 handle these functions regardless of the type used.</p>
6233 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6234 instruction and the variable argument handling intrinsic functions are
6237 <pre class="doc_code">
6238 define i32 @test(i32 %X, ...) {
6239 ; Initialize variable argument processing
6241 %ap2 = bitcast i8** %ap to i8*
6242 call void @llvm.va_start(i8* %ap2)
6244 ; Read a single integer argument
6245 %tmp = va_arg i8** %ap, i32
6247 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6249 %aq2 = bitcast i8** %aq to i8*
6250 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6251 call void @llvm.va_end(i8* %aq2)
6253 ; Stop processing of arguments.
6254 call void @llvm.va_end(i8* %ap2)
6258 declare void @llvm.va_start(i8*)
6259 declare void @llvm.va_copy(i8*, i8*)
6260 declare void @llvm.va_end(i8*)
6263 <!-- _______________________________________________________________________ -->
6265 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6273 declare void %llvm.va_start(i8* <arglist>)
6277 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6278 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6281 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6284 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6285 macro available in C. In a target-dependent way, it initializes
6286 the <tt>va_list</tt> element to which the argument points, so that the next
6287 call to <tt>va_arg</tt> will produce the first variable argument passed to
6288 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6289 need to know the last argument of the function as the compiler can figure
6294 <!-- _______________________________________________________________________ -->
6296 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6303 declare void @llvm.va_end(i8* <arglist>)
6307 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6308 which has been initialized previously
6309 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6310 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6313 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6316 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6317 macro available in C. In a target-dependent way, it destroys
6318 the <tt>va_list</tt> element to which the argument points. Calls
6319 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6320 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6321 with calls to <tt>llvm.va_end</tt>.</p>
6325 <!-- _______________________________________________________________________ -->
6327 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6334 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6338 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6339 from the source argument list to the destination argument list.</p>
6342 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6343 The second argument is a pointer to a <tt>va_list</tt> element to copy
6347 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6348 macro available in C. In a target-dependent way, it copies the
6349 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6350 element. This intrinsic is necessary because
6351 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6352 arbitrarily complex and require, for example, memory allocation.</p>
6360 <!-- ======================================================================= -->
6362 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6367 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6368 Collection</a> (GC) requires the implementation and generation of these
6369 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6370 roots on the stack</a>, as well as garbage collector implementations that
6371 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6372 barriers. Front-ends for type-safe garbage collected languages should generate
6373 these intrinsics to make use of the LLVM garbage collectors. For more details,
6374 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6377 <p>The garbage collection intrinsics only operate on objects in the generic
6378 address space (address space zero).</p>
6380 <!-- _______________________________________________________________________ -->
6382 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6389 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6393 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6394 the code generator, and allows some metadata to be associated with it.</p>
6397 <p>The first argument specifies the address of a stack object that contains the
6398 root pointer. The second pointer (which must be either a constant or a
6399 global value address) contains the meta-data to be associated with the
6403 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6404 location. At compile-time, the code generator generates information to allow
6405 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6406 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6411 <!-- _______________________________________________________________________ -->
6413 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6420 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6424 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6425 locations, allowing garbage collector implementations that require read
6429 <p>The second argument is the address to read from, which should be an address
6430 allocated from the garbage collector. The first object is a pointer to the
6431 start of the referenced object, if needed by the language runtime (otherwise
6435 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6436 instruction, but may be replaced with substantially more complex code by the
6437 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6438 may only be used in a function which <a href="#gc">specifies a GC
6443 <!-- _______________________________________________________________________ -->
6445 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6452 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6456 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6457 locations, allowing garbage collector implementations that require write
6458 barriers (such as generational or reference counting collectors).</p>
6461 <p>The first argument is the reference to store, the second is the start of the
6462 object to store it to, and the third is the address of the field of Obj to
6463 store to. If the runtime does not require a pointer to the object, Obj may
6467 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6468 instruction, but may be replaced with substantially more complex code by the
6469 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6470 may only be used in a function which <a href="#gc">specifies a GC
6477 <!-- ======================================================================= -->
6479 <a name="int_codegen">Code Generator Intrinsics</a>
6484 <p>These intrinsics are provided by LLVM to expose special features that may
6485 only be implemented with code generator support.</p>
6487 <!-- _______________________________________________________________________ -->
6489 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6496 declare i8 *@llvm.returnaddress(i32 <level>)
6500 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6501 target-specific value indicating the return address of the current function
6502 or one of its callers.</p>
6505 <p>The argument to this intrinsic indicates which function to return the address
6506 for. Zero indicates the calling function, one indicates its caller, etc.
6507 The argument is <b>required</b> to be a constant integer value.</p>
6510 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6511 indicating the return address of the specified call frame, or zero if it
6512 cannot be identified. The value returned by this intrinsic is likely to be
6513 incorrect or 0 for arguments other than zero, so it should only be used for
6514 debugging purposes.</p>
6516 <p>Note that calling this intrinsic does not prevent function inlining or other
6517 aggressive transformations, so the value returned may not be that of the
6518 obvious source-language caller.</p>
6522 <!-- _______________________________________________________________________ -->
6524 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6531 declare i8* @llvm.frameaddress(i32 <level>)
6535 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6536 target-specific frame pointer value for the specified stack frame.</p>
6539 <p>The argument to this intrinsic indicates which function to return the frame
6540 pointer for. Zero indicates the calling function, one indicates its caller,
6541 etc. The argument is <b>required</b> to be a constant integer value.</p>
6544 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6545 indicating the frame address of the specified call frame, or zero if it
6546 cannot be identified. The value returned by this intrinsic is likely to be
6547 incorrect or 0 for arguments other than zero, so it should only be used for
6548 debugging purposes.</p>
6550 <p>Note that calling this intrinsic does not prevent function inlining or other
6551 aggressive transformations, so the value returned may not be that of the
6552 obvious source-language caller.</p>
6556 <!-- _______________________________________________________________________ -->
6558 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6565 declare i8* @llvm.stacksave()
6569 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6570 of the function stack, for use
6571 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6572 useful for implementing language features like scoped automatic variable
6573 sized arrays in C99.</p>
6576 <p>This intrinsic returns a opaque pointer value that can be passed
6577 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6578 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6579 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6580 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6581 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6582 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6586 <!-- _______________________________________________________________________ -->
6588 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6595 declare void @llvm.stackrestore(i8* %ptr)
6599 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6600 the function stack to the state it was in when the
6601 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6602 executed. This is useful for implementing language features like scoped
6603 automatic variable sized arrays in C99.</p>
6606 <p>See the description
6607 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6611 <!-- _______________________________________________________________________ -->
6613 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6620 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6624 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6625 insert a prefetch instruction if supported; otherwise, it is a noop.
6626 Prefetches have no effect on the behavior of the program but can change its
6627 performance characteristics.</p>
6630 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6631 specifier determining if the fetch should be for a read (0) or write (1),
6632 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
6633 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6634 specifies whether the prefetch is performed on the data (1) or instruction (0)
6635 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
6636 must be constant integers.</p>
6639 <p>This intrinsic does not modify the behavior of the program. In particular,
6640 prefetches cannot trap and do not produce a value. On targets that support
6641 this intrinsic, the prefetch can provide hints to the processor cache for
6642 better performance.</p>
6646 <!-- _______________________________________________________________________ -->
6648 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
6655 declare void @llvm.pcmarker(i32 <id>)
6659 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
6660 Counter (PC) in a region of code to simulators and other tools. The method
6661 is target specific, but it is expected that the marker will use exported
6662 symbols to transmit the PC of the marker. The marker makes no guarantees
6663 that it will remain with any specific instruction after optimizations. It is
6664 possible that the presence of a marker will inhibit optimizations. The
6665 intended use is to be inserted after optimizations to allow correlations of
6666 simulation runs.</p>
6669 <p><tt>id</tt> is a numerical id identifying the marker.</p>
6672 <p>This intrinsic does not modify the behavior of the program. Backends that do
6673 not support this intrinsic may ignore it.</p>
6677 <!-- _______________________________________________________________________ -->
6679 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
6686 declare i64 @llvm.readcyclecounter()
6690 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
6691 counter register (or similar low latency, high accuracy clocks) on those
6692 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6693 should map to RPCC. As the backing counters overflow quickly (on the order
6694 of 9 seconds on alpha), this should only be used for small timings.</p>
6697 <p>When directly supported, reading the cycle counter should not modify any
6698 memory. Implementations are allowed to either return a application specific
6699 value or a system wide value. On backends without support, this is lowered
6700 to a constant 0.</p>
6706 <!-- ======================================================================= -->
6708 <a name="int_libc">Standard C Library Intrinsics</a>
6713 <p>LLVM provides intrinsics for a few important standard C library functions.
6714 These intrinsics allow source-language front-ends to pass information about
6715 the alignment of the pointer arguments to the code generator, providing
6716 opportunity for more efficient code generation.</p>
6718 <!-- _______________________________________________________________________ -->
6720 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
6726 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
6727 integer bit width and for different address spaces. Not all targets support
6728 all bit widths however.</p>
6731 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6732 i32 <len>, i32 <align>, i1 <isvolatile>)
6733 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6734 i64 <len>, i32 <align>, i1 <isvolatile>)
6738 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6739 source location to the destination location.</p>
6741 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
6742 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6743 and the pointers can be in specified address spaces.</p>
6747 <p>The first argument is a pointer to the destination, the second is a pointer
6748 to the source. The third argument is an integer argument specifying the
6749 number of bytes to copy, the fourth argument is the alignment of the
6750 source and destination locations, and the fifth is a boolean indicating a
6751 volatile access.</p>
6753 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6754 then the caller guarantees that both the source and destination pointers are
6755 aligned to that boundary.</p>
6757 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6758 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
6759 The detailed access behavior is not very cleanly specified and it is unwise
6760 to depend on it.</p>
6764 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6765 source location to the destination location, which are not allowed to
6766 overlap. It copies "len" bytes of memory over. If the argument is known to
6767 be aligned to some boundary, this can be specified as the fourth argument,
6768 otherwise it should be set to 0 or 1.</p>
6772 <!-- _______________________________________________________________________ -->
6774 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
6780 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
6781 width and for different address space. Not all targets support all bit
6785 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6786 i32 <len>, i32 <align>, i1 <isvolatile>)
6787 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6788 i64 <len>, i32 <align>, i1 <isvolatile>)
6792 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
6793 source location to the destination location. It is similar to the
6794 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
6797 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
6798 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6799 and the pointers can be in specified address spaces.</p>
6803 <p>The first argument is a pointer to the destination, the second is a pointer
6804 to the source. The third argument is an integer argument specifying the
6805 number of bytes to copy, the fourth argument is the alignment of the
6806 source and destination locations, and the fifth is a boolean indicating a
6807 volatile access.</p>
6809 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6810 then the caller guarantees that the source and destination pointers are
6811 aligned to that boundary.</p>
6813 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6814 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
6815 The detailed access behavior is not very cleanly specified and it is unwise
6816 to depend on it.</p>
6820 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
6821 source location to the destination location, which may overlap. It copies
6822 "len" bytes of memory over. If the argument is known to be aligned to some
6823 boundary, this can be specified as the fourth argument, otherwise it should
6824 be set to 0 or 1.</p>
6828 <!-- _______________________________________________________________________ -->
6830 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
6836 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6837 width and for different address spaces. However, not all targets support all
6841 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6842 i32 <len>, i32 <align>, i1 <isvolatile>)
6843 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6844 i64 <len>, i32 <align>, i1 <isvolatile>)
6848 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
6849 particular byte value.</p>
6851 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
6852 intrinsic does not return a value and takes extra alignment/volatile
6853 arguments. Also, the destination can be in an arbitrary address space.</p>
6856 <p>The first argument is a pointer to the destination to fill, the second is the
6857 byte value with which to fill it, the third argument is an integer argument
6858 specifying the number of bytes to fill, and the fourth argument is the known
6859 alignment of the destination location.</p>
6861 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6862 then the caller guarantees that the destination pointer is aligned to that
6865 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6866 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
6867 The detailed access behavior is not very cleanly specified and it is unwise
6868 to depend on it.</p>
6871 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
6872 at the destination location. If the argument is known to be aligned to some
6873 boundary, this can be specified as the fourth argument, otherwise it should
6874 be set to 0 or 1.</p>
6878 <!-- _______________________________________________________________________ -->
6880 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
6886 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
6887 floating point or vector of floating point type. Not all targets support all
6891 declare float @llvm.sqrt.f32(float %Val)
6892 declare double @llvm.sqrt.f64(double %Val)
6893 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6894 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6895 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6899 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
6900 returning the same value as the libm '<tt>sqrt</tt>' functions would.
6901 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
6902 behavior for negative numbers other than -0.0 (which allows for better
6903 optimization, because there is no need to worry about errno being
6904 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
6907 <p>The argument and return value are floating point numbers of the same
6911 <p>This function returns the sqrt of the specified operand if it is a
6912 nonnegative floating point number.</p>
6916 <!-- _______________________________________________________________________ -->
6918 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
6924 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
6925 floating point or vector of floating point type. Not all targets support all
6929 declare float @llvm.powi.f32(float %Val, i32 %power)
6930 declare double @llvm.powi.f64(double %Val, i32 %power)
6931 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6932 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6933 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6937 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
6938 specified (positive or negative) power. The order of evaluation of
6939 multiplications is not defined. When a vector of floating point type is
6940 used, the second argument remains a scalar integer value.</p>
6943 <p>The second argument is an integer power, and the first is a value to raise to
6947 <p>This function returns the first value raised to the second power with an
6948 unspecified sequence of rounding operations.</p>
6952 <!-- _______________________________________________________________________ -->
6954 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
6960 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
6961 floating point or vector of floating point type. Not all targets support all
6965 declare float @llvm.sin.f32(float %Val)
6966 declare double @llvm.sin.f64(double %Val)
6967 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6968 declare fp128 @llvm.sin.f128(fp128 %Val)
6969 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6973 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
6976 <p>The argument and return value are floating point numbers of the same
6980 <p>This function returns the sine of the specified operand, returning the same
6981 values as the libm <tt>sin</tt> functions would, and handles error conditions
6982 in the same way.</p>
6986 <!-- _______________________________________________________________________ -->
6988 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
6994 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
6995 floating point or vector of floating point type. Not all targets support all
6999 declare float @llvm.cos.f32(float %Val)
7000 declare double @llvm.cos.f64(double %Val)
7001 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7002 declare fp128 @llvm.cos.f128(fp128 %Val)
7003 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7007 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
7010 <p>The argument and return value are floating point numbers of the same
7014 <p>This function returns the cosine of the specified operand, returning the same
7015 values as the libm <tt>cos</tt> functions would, and handles error conditions
7016 in the same way.</p>
7020 <!-- _______________________________________________________________________ -->
7022 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
7028 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
7029 floating point or vector of floating point type. Not all targets support all
7033 declare float @llvm.pow.f32(float %Val, float %Power)
7034 declare double @llvm.pow.f64(double %Val, double %Power)
7035 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7036 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7037 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7041 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
7042 specified (positive or negative) power.</p>
7045 <p>The second argument is a floating point power, and the first is a value to
7046 raise to that power.</p>
7049 <p>This function returns the first value raised to the second power, returning
7050 the same values as the libm <tt>pow</tt> functions would, and handles error
7051 conditions in the same way.</p>
7057 <!-- _______________________________________________________________________ -->
7059 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7065 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7066 floating point or vector of floating point type. Not all targets support all
7070 declare float @llvm.exp.f32(float %Val)
7071 declare double @llvm.exp.f64(double %Val)
7072 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7073 declare fp128 @llvm.exp.f128(fp128 %Val)
7074 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7078 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7081 <p>The argument and return value are floating point numbers of the same
7085 <p>This function returns the same values as the libm <tt>exp</tt> functions
7086 would, and handles error conditions in the same way.</p>
7090 <!-- _______________________________________________________________________ -->
7092 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7098 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7099 floating point or vector of floating point type. Not all targets support all
7103 declare float @llvm.log.f32(float %Val)
7104 declare double @llvm.log.f64(double %Val)
7105 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7106 declare fp128 @llvm.log.f128(fp128 %Val)
7107 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7111 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7114 <p>The argument and return value are floating point numbers of the same
7118 <p>This function returns the same values as the libm <tt>log</tt> functions
7119 would, and handles error conditions in the same way.</p>
7122 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7128 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7129 floating point or vector of floating point type. Not all targets support all
7133 declare float @llvm.fma.f32(float %a, float %b, float %c)
7134 declare double @llvm.fma.f64(double %a, double %b, double %c)
7135 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7136 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7137 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7141 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7145 <p>The argument and return value are floating point numbers of the same
7149 <p>This function returns the same values as the libm <tt>fma</tt> functions
7154 <!-- ======================================================================= -->
7156 <a name="int_manip">Bit Manipulation Intrinsics</a>
7161 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7162 These allow efficient code generation for some algorithms.</p>
7164 <!-- _______________________________________________________________________ -->
7166 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7172 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7173 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7176 declare i16 @llvm.bswap.i16(i16 <id>)
7177 declare i32 @llvm.bswap.i32(i32 <id>)
7178 declare i64 @llvm.bswap.i64(i64 <id>)
7182 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7183 values with an even number of bytes (positive multiple of 16 bits). These
7184 are useful for performing operations on data that is not in the target's
7185 native byte order.</p>
7188 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7189 and low byte of the input i16 swapped. Similarly,
7190 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7191 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7192 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7193 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7194 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7195 more, respectively).</p>
7199 <!-- _______________________________________________________________________ -->
7201 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7207 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7208 width, or on any vector with integer elements. Not all targets support all
7209 bit widths or vector types, however.</p>
7212 declare i8 @llvm.ctpop.i8(i8 <src>)
7213 declare i16 @llvm.ctpop.i16(i16 <src>)
7214 declare i32 @llvm.ctpop.i32(i32 <src>)
7215 declare i64 @llvm.ctpop.i64(i64 <src>)
7216 declare i256 @llvm.ctpop.i256(i256 <src>)
7217 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7221 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7225 <p>The only argument is the value to be counted. The argument may be of any
7226 integer type, or a vector with integer elements.
7227 The return type must match the argument type.</p>
7230 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7231 element of a vector.</p>
7235 <!-- _______________________________________________________________________ -->
7237 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7243 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7244 integer bit width, or any vector whose elements are integers. Not all
7245 targets support all bit widths or vector types, however.</p>
7248 declare i8 @llvm.ctlz.i8 (i8 <src>)
7249 declare i16 @llvm.ctlz.i16(i16 <src>)
7250 declare i32 @llvm.ctlz.i32(i32 <src>)
7251 declare i64 @llvm.ctlz.i64(i64 <src>)
7252 declare i256 @llvm.ctlz.i256(i256 <src>)
7253 declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src;gt)
7257 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7258 leading zeros in a variable.</p>
7261 <p>The only argument is the value to be counted. The argument may be of any
7262 integer type, or any vector type with integer element type.
7263 The return type must match the argument type.</p>
7266 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7267 zeros in a variable, or within each element of the vector if the operation
7268 is of vector type. If the src == 0 then the result is the size in bits of
7269 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7273 <!-- _______________________________________________________________________ -->
7275 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7281 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7282 integer bit width, or any vector of integer elements. Not all targets
7283 support all bit widths or vector types, however.</p>
7286 declare i8 @llvm.cttz.i8 (i8 <src>)
7287 declare i16 @llvm.cttz.i16(i16 <src>)
7288 declare i32 @llvm.cttz.i32(i32 <src>)
7289 declare i64 @llvm.cttz.i64(i64 <src>)
7290 declare i256 @llvm.cttz.i256(i256 <src>)
7291 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>)
7295 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7299 <p>The only argument is the value to be counted. The argument may be of any
7300 integer type, or a vectory with integer element type.. The return type
7301 must match the argument type.</p>
7304 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7305 zeros in a variable, or within each element of a vector.
7306 If the src == 0 then the result is the size in bits of
7307 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7313 <!-- ======================================================================= -->
7315 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7320 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7322 <!-- _______________________________________________________________________ -->
7324 <a name="int_sadd_overflow">
7325 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7332 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7333 on any integer bit width.</p>
7336 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7337 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7338 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7342 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7343 a signed addition of the two arguments, and indicate whether an overflow
7344 occurred during the signed summation.</p>
7347 <p>The arguments (%a and %b) and the first element of the result structure may
7348 be of integer types of any bit width, but they must have the same bit
7349 width. The second element of the result structure must be of
7350 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7351 undergo signed addition.</p>
7354 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7355 a signed addition of the two variables. They return a structure — the
7356 first element of which is the signed summation, and the second element of
7357 which is a bit specifying if the signed summation resulted in an
7362 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7363 %sum = extractvalue {i32, i1} %res, 0
7364 %obit = extractvalue {i32, i1} %res, 1
7365 br i1 %obit, label %overflow, label %normal
7370 <!-- _______________________________________________________________________ -->
7372 <a name="int_uadd_overflow">
7373 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7380 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7381 on any integer bit width.</p>
7384 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7385 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7386 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7390 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7391 an unsigned addition of the two arguments, and indicate whether a carry
7392 occurred during the unsigned summation.</p>
7395 <p>The arguments (%a and %b) and the first element of the result structure may
7396 be of integer types of any bit width, but they must have the same bit
7397 width. The second element of the result structure must be of
7398 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7399 undergo unsigned addition.</p>
7402 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7403 an unsigned addition of the two arguments. They return a structure —
7404 the first element of which is the sum, and the second element of which is a
7405 bit specifying if the unsigned summation resulted in a carry.</p>
7409 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7410 %sum = extractvalue {i32, i1} %res, 0
7411 %obit = extractvalue {i32, i1} %res, 1
7412 br i1 %obit, label %carry, label %normal
7417 <!-- _______________________________________________________________________ -->
7419 <a name="int_ssub_overflow">
7420 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7427 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7428 on any integer bit width.</p>
7431 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7432 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7433 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7437 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7438 a signed subtraction of the two arguments, and indicate whether an overflow
7439 occurred during the signed subtraction.</p>
7442 <p>The arguments (%a and %b) and the first element of the result structure may
7443 be of integer types of any bit width, but they must have the same bit
7444 width. The second element of the result structure must be of
7445 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7446 undergo signed subtraction.</p>
7449 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7450 a signed subtraction of the two arguments. They return a structure —
7451 the first element of which is the subtraction, and the second element of
7452 which is a bit specifying if the signed subtraction resulted in an
7457 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7458 %sum = extractvalue {i32, i1} %res, 0
7459 %obit = extractvalue {i32, i1} %res, 1
7460 br i1 %obit, label %overflow, label %normal
7465 <!-- _______________________________________________________________________ -->
7467 <a name="int_usub_overflow">
7468 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7475 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7476 on any integer bit width.</p>
7479 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7480 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7481 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7485 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7486 an unsigned subtraction of the two arguments, and indicate whether an
7487 overflow occurred during the unsigned subtraction.</p>
7490 <p>The arguments (%a and %b) and the first element of the result structure may
7491 be of integer types of any bit width, but they must have the same bit
7492 width. The second element of the result structure must be of
7493 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7494 undergo unsigned subtraction.</p>
7497 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7498 an unsigned subtraction of the two arguments. They return a structure —
7499 the first element of which is the subtraction, and the second element of
7500 which is a bit specifying if the unsigned subtraction resulted in an
7505 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7506 %sum = extractvalue {i32, i1} %res, 0
7507 %obit = extractvalue {i32, i1} %res, 1
7508 br i1 %obit, label %overflow, label %normal
7513 <!-- _______________________________________________________________________ -->
7515 <a name="int_smul_overflow">
7516 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7523 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7524 on any integer bit width.</p>
7527 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7528 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7529 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7534 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7535 a signed multiplication of the two arguments, and indicate whether an
7536 overflow occurred during the signed multiplication.</p>
7539 <p>The arguments (%a and %b) and the first element of the result structure may
7540 be of integer types of any bit width, but they must have the same bit
7541 width. The second element of the result structure must be of
7542 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7543 undergo signed multiplication.</p>
7546 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7547 a signed multiplication of the two arguments. They return a structure —
7548 the first element of which is the multiplication, and the second element of
7549 which is a bit specifying if the signed multiplication resulted in an
7554 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7555 %sum = extractvalue {i32, i1} %res, 0
7556 %obit = extractvalue {i32, i1} %res, 1
7557 br i1 %obit, label %overflow, label %normal
7562 <!-- _______________________________________________________________________ -->
7564 <a name="int_umul_overflow">
7565 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
7572 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
7573 on any integer bit width.</p>
7576 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7577 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7578 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7582 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7583 a unsigned multiplication of the two arguments, and indicate whether an
7584 overflow occurred during the unsigned multiplication.</p>
7587 <p>The arguments (%a and %b) and the first element of the result structure may
7588 be of integer types of any bit width, but they must have the same bit
7589 width. The second element of the result structure must be of
7590 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7591 undergo unsigned multiplication.</p>
7594 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7595 an unsigned multiplication of the two arguments. They return a structure
7596 — the first element of which is the multiplication, and the second
7597 element of which is a bit specifying if the unsigned multiplication resulted
7602 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7603 %sum = extractvalue {i32, i1} %res, 0
7604 %obit = extractvalue {i32, i1} %res, 1
7605 br i1 %obit, label %overflow, label %normal
7612 <!-- ======================================================================= -->
7614 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
7619 <p>Half precision floating point is a storage-only format. This means that it is
7620 a dense encoding (in memory) but does not support computation in the
7623 <p>This means that code must first load the half-precision floating point
7624 value as an i16, then convert it to float with <a
7625 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
7626 Computation can then be performed on the float value (including extending to
7627 double etc). To store the value back to memory, it is first converted to
7628 float if needed, then converted to i16 with
7629 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
7630 storing as an i16 value.</p>
7632 <!-- _______________________________________________________________________ -->
7634 <a name="int_convert_to_fp16">
7635 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
7643 declare i16 @llvm.convert.to.fp16(f32 %a)
7647 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7648 a conversion from single precision floating point format to half precision
7649 floating point format.</p>
7652 <p>The intrinsic function contains single argument - the value to be
7656 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7657 a conversion from single precision floating point format to half precision
7658 floating point format. The return value is an <tt>i16</tt> which
7659 contains the converted number.</p>
7663 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7664 store i16 %res, i16* @x, align 2
7669 <!-- _______________________________________________________________________ -->
7671 <a name="int_convert_from_fp16">
7672 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
7680 declare f32 @llvm.convert.from.fp16(i16 %a)
7684 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
7685 a conversion from half precision floating point format to single precision
7686 floating point format.</p>
7689 <p>The intrinsic function contains single argument - the value to be
7693 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
7694 conversion from half single precision floating point format to single
7695 precision floating point format. The input half-float value is represented by
7696 an <tt>i16</tt> value.</p>
7700 %a = load i16* @x, align 2
7701 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7708 <!-- ======================================================================= -->
7710 <a name="int_debugger">Debugger Intrinsics</a>
7715 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
7716 prefix), are described in
7717 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
7718 Level Debugging</a> document.</p>
7722 <!-- ======================================================================= -->
7724 <a name="int_eh">Exception Handling Intrinsics</a>
7729 <p>The LLVM exception handling intrinsics (which all start with
7730 <tt>llvm.eh.</tt> prefix), are described in
7731 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
7732 Handling</a> document.</p>
7736 <!-- ======================================================================= -->
7738 <a name="int_trampoline">Trampoline Intrinsics</a>
7743 <p>These intrinsics make it possible to excise one parameter, marked with
7744 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
7745 The result is a callable
7746 function pointer lacking the nest parameter - the caller does not need to
7747 provide a value for it. Instead, the value to use is stored in advance in a
7748 "trampoline", a block of memory usually allocated on the stack, which also
7749 contains code to splice the nest value into the argument list. This is used
7750 to implement the GCC nested function address extension.</p>
7752 <p>For example, if the function is
7753 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
7754 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
7757 <pre class="doc_code">
7758 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
7759 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
7760 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
7761 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
7762 %fp = bitcast i8* %p to i32 (i32, i32)*
7765 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
7766 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
7768 <!-- _______________________________________________________________________ -->
7771 '<tt>llvm.init.trampoline</tt>' Intrinsic
7779 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
7783 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
7784 turning it into a trampoline.</p>
7787 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
7788 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
7789 sufficiently aligned block of memory; this memory is written to by the
7790 intrinsic. Note that the size and the alignment are target-specific - LLVM
7791 currently provides no portable way of determining them, so a front-end that
7792 generates this intrinsic needs to have some target-specific knowledge.
7793 The <tt>func</tt> argument must hold a function bitcast to
7794 an <tt>i8*</tt>.</p>
7797 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
7798 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
7799 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
7800 which can be <a href="#int_trampoline">bitcast (to a new function) and
7801 called</a>. The new function's signature is the same as that of
7802 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
7803 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
7804 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
7805 with the same argument list, but with <tt>nval</tt> used for the missing
7806 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
7807 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
7808 to the returned function pointer is undefined.</p>
7811 <!-- _______________________________________________________________________ -->
7814 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
7822 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
7826 <p>This performs any required machine-specific adjustment to the address of a
7827 trampoline (passed as <tt>tramp</tt>).</p>
7830 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
7831 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
7835 <p>On some architectures the address of the code to be executed needs to be
7836 different to the address where the trampoline is actually stored. This
7837 intrinsic returns the executable address corresponding to <tt>tramp</tt>
7838 after performing the required machine specific adjustments.
7839 The pointer returned can then be <a href="#int_trampoline"> bitcast and
7847 <!-- ======================================================================= -->
7849 <a name="int_memorymarkers">Memory Use Markers</a>
7854 <p>This class of intrinsics exists to information about the lifetime of memory
7855 objects and ranges where variables are immutable.</p>
7857 <!-- _______________________________________________________________________ -->
7859 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
7866 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
7870 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
7871 object's lifetime.</p>
7874 <p>The first argument is a constant integer representing the size of the
7875 object, or -1 if it is variable sized. The second argument is a pointer to
7879 <p>This intrinsic indicates that before this point in the code, the value of the
7880 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7881 never be used and has an undefined value. A load from the pointer that
7882 precedes this intrinsic can be replaced with
7883 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
7887 <!-- _______________________________________________________________________ -->
7889 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
7896 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
7900 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
7901 object's lifetime.</p>
7904 <p>The first argument is a constant integer representing the size of the
7905 object, or -1 if it is variable sized. The second argument is a pointer to
7909 <p>This intrinsic indicates that after this point in the code, the value of the
7910 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7911 never be used and has an undefined value. Any stores into the memory object
7912 following this intrinsic may be removed as dead.
7916 <!-- _______________________________________________________________________ -->
7918 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
7925 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
7929 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
7930 a memory object will not change.</p>
7933 <p>The first argument is a constant integer representing the size of the
7934 object, or -1 if it is variable sized. The second argument is a pointer to
7938 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
7939 the return value, the referenced memory location is constant and
7944 <!-- _______________________________________________________________________ -->
7946 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
7953 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7957 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
7958 a memory object are mutable.</p>
7961 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
7962 The second argument is a constant integer representing the size of the
7963 object, or -1 if it is variable sized and the third argument is a pointer
7967 <p>This intrinsic indicates that the memory is mutable again.</p>
7973 <!-- ======================================================================= -->
7975 <a name="int_general">General Intrinsics</a>
7980 <p>This class of intrinsics is designed to be generic and has no specific
7983 <!-- _______________________________________________________________________ -->
7985 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
7992 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
7996 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
7999 <p>The first argument is a pointer to a value, the second is a pointer to a
8000 global string, the third is a pointer to a global string which is the source
8001 file name, and the last argument is the line number.</p>
8004 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8005 This can be useful for special purpose optimizations that want to look for
8006 these annotations. These have no other defined use; they are ignored by code
8007 generation and optimization.</p>
8011 <!-- _______________________________________________________________________ -->
8013 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8019 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8020 any integer bit width.</p>
8023 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8024 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8025 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8026 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8027 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8031 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8034 <p>The first argument is an integer value (result of some expression), the
8035 second is a pointer to a global string, the third is a pointer to a global
8036 string which is the source file name, and the last argument is the line
8037 number. It returns the value of the first argument.</p>
8040 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8041 arbitrary strings. This can be useful for special purpose optimizations that
8042 want to look for these annotations. These have no other defined use; they
8043 are ignored by code generation and optimization.</p>
8047 <!-- _______________________________________________________________________ -->
8049 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8056 declare void @llvm.trap()
8060 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8066 <p>This intrinsics is lowered to the target dependent trap instruction. If the
8067 target does not have a trap instruction, this intrinsic will be lowered to
8068 the call of the <tt>abort()</tt> function.</p>
8072 <!-- _______________________________________________________________________ -->
8074 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8081 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8085 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8086 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8087 ensure that it is placed on the stack before local variables.</p>
8090 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8091 arguments. The first argument is the value loaded from the stack
8092 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8093 that has enough space to hold the value of the guard.</p>
8096 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8097 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8098 stack. This is to ensure that if a local variable on the stack is
8099 overwritten, it will destroy the value of the guard. When the function exits,
8100 the guard on the stack is checked against the original guard. If they are
8101 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8106 <!-- _______________________________________________________________________ -->
8108 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8115 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <type>)
8116 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <type>)
8120 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8121 the optimizers to determine at compile time whether a) an operation (like
8122 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8123 runtime check for overflow isn't necessary. An object in this context means
8124 an allocation of a specific class, structure, array, or other object.</p>
8127 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8128 argument is a pointer to or into the <tt>object</tt>. The second argument
8129 is a boolean 0 or 1. This argument determines whether you want the
8130 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
8131 1, variables are not allowed.</p>
8134 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
8135 representing the size of the object concerned, or <tt>i32/i64 -1 or 0</tt>,
8136 depending on the <tt>type</tt> argument, if the size cannot be determined at
8145 <!-- *********************************************************************** -->
8148 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8149 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8150 <a href="http://validator.w3.org/check/referer"><img
8151 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8153 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8154 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8155 Last modified: $Date$