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></li>
106 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
108 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
109 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
110 Global Variable</a></li>
111 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
112 Global Variable</a></li>
113 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
114 Global Variable</a></li>
117 <li><a href="#instref">Instruction Reference</a>
119 <li><a href="#terminators">Terminator Instructions</a>
121 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
122 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
123 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
124 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
125 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
126 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
127 <li><a href="#i_resume">'<tt>resume</tt>' Instruction</a></li>
128 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
131 <li><a href="#binaryops">Binary Operations</a>
133 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
134 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
135 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
136 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
137 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
138 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
139 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
140 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
141 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
142 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
143 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
144 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
147 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
149 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
150 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
151 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
152 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
153 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
154 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
157 <li><a href="#vectorops">Vector Operations</a>
159 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
160 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
161 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
164 <li><a href="#aggregateops">Aggregate Operations</a>
166 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
167 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
170 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
172 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
173 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
174 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
175 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
176 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
177 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
178 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
181 <li><a href="#convertops">Conversion Operations</a>
183 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
184 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
185 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
186 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
187 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
188 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
189 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
190 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
191 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
192 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
193 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
194 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
197 <li><a href="#otherops">Other Operations</a>
199 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
200 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
201 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
202 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
203 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
204 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
205 <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
210 <li><a href="#intrinsics">Intrinsic Functions</a>
212 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
214 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
215 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
216 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
219 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
221 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
222 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
223 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
226 <li><a href="#int_codegen">Code Generator Intrinsics</a>
228 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
229 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
230 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
231 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
232 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
233 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
234 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
237 <li><a href="#int_libc">Standard C Library Intrinsics</a>
239 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
240 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
241 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
242 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
243 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
244 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
245 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
246 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
247 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
248 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
249 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
252 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
254 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
255 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
256 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
257 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
260 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
262 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
263 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
264 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
265 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
266 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
267 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
270 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
272 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
273 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
276 <li><a href="#int_debugger">Debugger intrinsics</a></li>
277 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
278 <li><a href="#int_trampoline">Trampoline Intrinsics</a>
280 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
281 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
284 <li><a href="#int_memorymarkers">Memory Use Markers</a>
286 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
287 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
288 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
289 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
292 <li><a href="#int_general">General intrinsics</a>
294 <li><a href="#int_var_annotation">
295 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
296 <li><a href="#int_annotation">
297 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
298 <li><a href="#int_trap">
299 '<tt>llvm.trap</tt>' Intrinsic</a></li>
300 <li><a href="#int_stackprotector">
301 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
302 <li><a href="#int_objectsize">
303 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
310 <div class="doc_author">
311 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
312 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
315 <!-- *********************************************************************** -->
316 <h2><a name="abstract">Abstract</a></h2>
317 <!-- *********************************************************************** -->
321 <p>This document is a reference manual for the LLVM assembly language. LLVM is
322 a Static Single Assignment (SSA) based representation that provides type
323 safety, low-level operations, flexibility, and the capability of representing
324 'all' high-level languages cleanly. It is the common code representation
325 used throughout all phases of the LLVM compilation strategy.</p>
329 <!-- *********************************************************************** -->
330 <h2><a name="introduction">Introduction</a></h2>
331 <!-- *********************************************************************** -->
335 <p>The LLVM code representation is designed to be used in three different forms:
336 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
337 for fast loading by a Just-In-Time compiler), and as a human readable
338 assembly language representation. This allows LLVM to provide a powerful
339 intermediate representation for efficient compiler transformations and
340 analysis, while providing a natural means to debug and visualize the
341 transformations. The three different forms of LLVM are all equivalent. This
342 document describes the human readable representation and notation.</p>
344 <p>The LLVM representation aims to be light-weight and low-level while being
345 expressive, typed, and extensible at the same time. It aims to be a
346 "universal IR" of sorts, by being at a low enough level that high-level ideas
347 may be cleanly mapped to it (similar to how microprocessors are "universal
348 IR's", allowing many source languages to be mapped to them). By providing
349 type information, LLVM can be used as the target of optimizations: for
350 example, through pointer analysis, it can be proven that a C automatic
351 variable is never accessed outside of the current function, allowing it to
352 be promoted to a simple SSA value instead of a memory location.</p>
354 <!-- _______________________________________________________________________ -->
356 <a name="wellformed">Well-Formedness</a>
361 <p>It is important to note that this document describes 'well formed' LLVM
362 assembly language. There is a difference between what the parser accepts and
363 what is considered 'well formed'. For example, the following instruction is
364 syntactically okay, but not well formed:</p>
366 <pre class="doc_code">
367 %x = <a href="#i_add">add</a> i32 1, %x
370 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
371 LLVM infrastructure provides a verification pass that may be used to verify
372 that an LLVM module is well formed. This pass is automatically run by the
373 parser after parsing input assembly and by the optimizer before it outputs
374 bitcode. The violations pointed out by the verifier pass indicate bugs in
375 transformation passes or input to the parser.</p>
381 <!-- Describe the typesetting conventions here. -->
383 <!-- *********************************************************************** -->
384 <h2><a name="identifiers">Identifiers</a></h2>
385 <!-- *********************************************************************** -->
389 <p>LLVM identifiers come in two basic types: global and local. Global
390 identifiers (functions, global variables) begin with the <tt>'@'</tt>
391 character. Local identifiers (register names, types) begin with
392 the <tt>'%'</tt> character. Additionally, there are three different formats
393 for identifiers, for different purposes:</p>
396 <li>Named values are represented as a string of characters with their prefix.
397 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
398 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
399 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
400 other characters in their names can be surrounded with quotes. Special
401 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
402 ASCII code for the character in hexadecimal. In this way, any character
403 can be used in a name value, even quotes themselves.</li>
405 <li>Unnamed values are represented as an unsigned numeric value with their
406 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
408 <li>Constants, which are described in a <a href="#constants">section about
409 constants</a>, below.</li>
412 <p>LLVM requires that values start with a prefix for two reasons: Compilers
413 don't need to worry about name clashes with reserved words, and the set of
414 reserved words may be expanded in the future without penalty. Additionally,
415 unnamed identifiers allow a compiler to quickly come up with a temporary
416 variable without having to avoid symbol table conflicts.</p>
418 <p>Reserved words in LLVM are very similar to reserved words in other
419 languages. There are keywords for different opcodes
420 ('<tt><a href="#i_add">add</a></tt>',
421 '<tt><a href="#i_bitcast">bitcast</a></tt>',
422 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
423 ('<tt><a href="#t_void">void</a></tt>',
424 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
425 reserved words cannot conflict with variable names, because none of them
426 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
428 <p>Here is an example of LLVM code to multiply the integer variable
429 '<tt>%X</tt>' by 8:</p>
433 <pre class="doc_code">
434 %result = <a href="#i_mul">mul</a> i32 %X, 8
437 <p>After strength reduction:</p>
439 <pre class="doc_code">
440 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
443 <p>And the hard way:</p>
445 <pre class="doc_code">
446 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
447 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
448 %result = <a href="#i_add">add</a> i32 %1, %1
451 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
452 lexical features of LLVM:</p>
455 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
458 <li>Unnamed temporaries are created when the result of a computation is not
459 assigned to a named value.</li>
461 <li>Unnamed temporaries are numbered sequentially</li>
464 <p>It also shows a convention that we follow in this document. When
465 demonstrating instructions, we will follow an instruction with a comment that
466 defines the type and name of value produced. Comments are shown in italic
471 <!-- *********************************************************************** -->
472 <h2><a name="highlevel">High Level Structure</a></h2>
473 <!-- *********************************************************************** -->
475 <!-- ======================================================================= -->
477 <a name="modulestructure">Module Structure</a>
482 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
483 of the input programs. Each module consists of functions, global variables,
484 and symbol table entries. Modules may be combined together with the LLVM
485 linker, which merges function (and global variable) definitions, resolves
486 forward declarations, and merges symbol table entries. Here is an example of
487 the "hello world" module:</p>
489 <pre class="doc_code">
490 <i>; Declare the string constant as a global constant.</i>
491 <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>
493 <i>; External declaration of the puts function</i>
494 <a href="#functionstructure">declare</a> i32 @puts(i8*) <i>; i32 (i8*)* </i>
496 <i>; Definition of main function</i>
497 define i32 @main() { <i>; i32()* </i>
498 <i>; Convert [13 x i8]* to i8 *...</i>
499 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8*</i>
501 <i>; Call puts function to write out the string to stdout.</i>
502 <a href="#i_call">call</a> i32 @puts(i8* %cast210) <i>; i32</i>
503 <a href="#i_ret">ret</a> i32 0
506 <i>; Named metadata</i>
507 !1 = metadata !{i32 41}
511 <p>This example is made up of a <a href="#globalvars">global variable</a> named
512 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
513 a <a href="#functionstructure">function definition</a> for
514 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
517 <p>In general, a module is made up of a list of global values, where both
518 functions and global variables are global values. Global values are
519 represented by a pointer to a memory location (in this case, a pointer to an
520 array of char, and a pointer to a function), and have one of the
521 following <a href="#linkage">linkage types</a>.</p>
525 <!-- ======================================================================= -->
527 <a name="linkage">Linkage Types</a>
532 <p>All Global Variables and Functions have one of the following types of
536 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
537 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
538 by objects in the current module. In particular, linking code into a
539 module with an private global value may cause the private to be renamed as
540 necessary to avoid collisions. Because the symbol is private to the
541 module, all references can be updated. This doesn't show up in any symbol
542 table in the object file.</dd>
544 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
545 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
546 assembler and evaluated by the linker. Unlike normal strong symbols, they
547 are removed by the linker from the final linked image (executable or
548 dynamic library).</dd>
550 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
551 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
552 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
553 linker. The symbols are removed by the linker from the final linked image
554 (executable or dynamic library).</dd>
556 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
557 <dd>Similar to "<tt>linker_private_weak</tt>", but it's known that the address
558 of the object is not taken. For instance, functions that had an inline
559 definition, but the compiler decided not to inline it. Note,
560 unlike <tt>linker_private</tt> and <tt>linker_private_weak</tt>,
561 <tt>linker_private_weak_def_auto</tt> may have only <tt>default</tt>
562 visibility. The symbols are removed by the linker from the final linked
563 image (executable or dynamic library).</dd>
565 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
566 <dd>Similar to private, but the value shows as a local symbol
567 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
568 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
570 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
571 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
572 into the object file corresponding to the LLVM module. They exist to
573 allow inlining and other optimizations to take place given knowledge of
574 the definition of the global, which is known to be somewhere outside the
575 module. Globals with <tt>available_externally</tt> linkage are allowed to
576 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
577 This linkage type is only allowed on definitions, not declarations.</dd>
579 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
580 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
581 the same name when linkage occurs. This can be used to implement
582 some forms of inline functions, templates, or other code which must be
583 generated in each translation unit that uses it, but where the body may
584 be overridden with a more definitive definition later. Unreferenced
585 <tt>linkonce</tt> globals are allowed to be discarded. Note that
586 <tt>linkonce</tt> linkage does not actually allow the optimizer to
587 inline the body of this function into callers because it doesn't know if
588 this definition of the function is the definitive definition within the
589 program or whether it will be overridden by a stronger definition.
590 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
593 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
594 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
595 <tt>linkonce</tt> linkage, except that unreferenced globals with
596 <tt>weak</tt> linkage may not be discarded. This is used for globals that
597 are declared "weak" in C source code.</dd>
599 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
600 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
601 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
603 Symbols with "<tt>common</tt>" linkage are merged in the same way as
604 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
605 <tt>common</tt> symbols may not have an explicit section,
606 must have a zero initializer, and may not be marked '<a
607 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
608 have common linkage.</dd>
611 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
612 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
613 pointer to array type. When two global variables with appending linkage
614 are linked together, the two global arrays are appended together. This is
615 the LLVM, typesafe, equivalent of having the system linker append together
616 "sections" with identical names when .o files are linked.</dd>
618 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
619 <dd>The semantics of this linkage follow the ELF object file model: the symbol
620 is weak until linked, if not linked, the symbol becomes null instead of
621 being an undefined reference.</dd>
623 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
624 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
625 <dd>Some languages allow differing globals to be merged, such as two functions
626 with different semantics. Other languages, such as <tt>C++</tt>, ensure
627 that only equivalent globals are ever merged (the "one definition rule"
628 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
629 and <tt>weak_odr</tt> linkage types to indicate that the global will only
630 be merged with equivalent globals. These linkage types are otherwise the
631 same as their non-<tt>odr</tt> versions.</dd>
633 <dt><tt><b><a name="linkage_external">external</a></b></tt>:</dt>
634 <dd>If none of the above identifiers are used, the global is externally
635 visible, meaning that it participates in linkage and can be used to
636 resolve external symbol references.</dd>
639 <p>The next two types of linkage are targeted for Microsoft Windows platform
640 only. They are designed to support importing (exporting) symbols from (to)
641 DLLs (Dynamic Link Libraries).</p>
644 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
645 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
646 or variable via a global pointer to a pointer that is set up by the DLL
647 exporting the symbol. On Microsoft Windows targets, the pointer name is
648 formed by combining <code>__imp_</code> and the function or variable
651 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
652 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
653 pointer to a pointer in a DLL, so that it can be referenced with the
654 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
655 name is formed by combining <code>__imp_</code> and the function or
659 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
660 another module defined a "<tt>.LC0</tt>" variable and was linked with this
661 one, one of the two would be renamed, preventing a collision. Since
662 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
663 declarations), they are accessible outside of the current module.</p>
665 <p>It is illegal for a function <i>declaration</i> to have any linkage type
666 other than <tt>external</tt>, <tt>dllimport</tt>
667 or <tt>extern_weak</tt>.</p>
669 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
670 or <tt>weak_odr</tt> linkages.</p>
674 <!-- ======================================================================= -->
676 <a name="callingconv">Calling Conventions</a>
681 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
682 and <a href="#i_invoke">invokes</a> can all have an optional calling
683 convention specified for the call. The calling convention of any pair of
684 dynamic caller/callee must match, or the behavior of the program is
685 undefined. The following calling conventions are supported by LLVM, and more
686 may be added in the future:</p>
689 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
690 <dd>This calling convention (the default if no other calling convention is
691 specified) matches the target C calling conventions. This calling
692 convention supports varargs function calls and tolerates some mismatch in
693 the declared prototype and implemented declaration of the function (as
696 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
697 <dd>This calling convention attempts to make calls as fast as possible
698 (e.g. by passing things in registers). This calling convention allows the
699 target to use whatever tricks it wants to produce fast code for the
700 target, without having to conform to an externally specified ABI
701 (Application Binary Interface).
702 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
703 when this or the GHC convention is used.</a> This calling convention
704 does not support varargs and requires the prototype of all callees to
705 exactly match the prototype of the function definition.</dd>
707 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
708 <dd>This calling convention attempts to make code in the caller as efficient
709 as possible under the assumption that the call is not commonly executed.
710 As such, these calls often preserve all registers so that the call does
711 not break any live ranges in the caller side. This calling convention
712 does not support varargs and requires the prototype of all callees to
713 exactly match the prototype of the function definition.</dd>
715 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
716 <dd>This calling convention has been implemented specifically for use by the
717 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
718 It passes everything in registers, going to extremes to achieve this by
719 disabling callee save registers. This calling convention should not be
720 used lightly but only for specific situations such as an alternative to
721 the <em>register pinning</em> performance technique often used when
722 implementing functional programming languages.At the moment only X86
723 supports this convention and it has the following limitations:
725 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
726 floating point types are supported.</li>
727 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
728 6 floating point parameters.</li>
730 This calling convention supports
731 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
732 requires both the caller and callee are using it.
735 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
736 <dd>Any calling convention may be specified by number, allowing
737 target-specific calling conventions to be used. Target specific calling
738 conventions start at 64.</dd>
741 <p>More calling conventions can be added/defined on an as-needed basis, to
742 support Pascal conventions or any other well-known target-independent
747 <!-- ======================================================================= -->
749 <a name="visibility">Visibility Styles</a>
754 <p>All Global Variables and Functions have one of the following visibility
758 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
759 <dd>On targets that use the ELF object file format, default visibility means
760 that the declaration is visible to other modules and, in shared libraries,
761 means that the declared entity may be overridden. On Darwin, default
762 visibility means that the declaration is visible to other modules. Default
763 visibility corresponds to "external linkage" in the language.</dd>
765 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
766 <dd>Two declarations of an object with hidden visibility refer to the same
767 object if they are in the same shared object. Usually, hidden visibility
768 indicates that the symbol will not be placed into the dynamic symbol
769 table, so no other module (executable or shared library) can reference it
772 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
773 <dd>On ELF, protected visibility indicates that the symbol will be placed in
774 the dynamic symbol table, but that references within the defining module
775 will bind to the local symbol. That is, the symbol cannot be overridden by
781 <!-- ======================================================================= -->
783 <a name="namedtypes">Named Types</a>
788 <p>LLVM IR allows you to specify name aliases for certain types. This can make
789 it easier to read the IR and make the IR more condensed (particularly when
790 recursive types are involved). An example of a name specification is:</p>
792 <pre class="doc_code">
793 %mytype = type { %mytype*, i32 }
796 <p>You may give a name to any <a href="#typesystem">type</a> except
797 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
798 is expected with the syntax "%mytype".</p>
800 <p>Note that type names are aliases for the structural type that they indicate,
801 and that you can therefore specify multiple names for the same type. This
802 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
803 uses structural typing, the name is not part of the type. When printing out
804 LLVM IR, the printer will pick <em>one name</em> to render all types of a
805 particular shape. This means that if you have code where two different
806 source types end up having the same LLVM type, that the dumper will sometimes
807 print the "wrong" or unexpected type. This is an important design point and
808 isn't going to change.</p>
812 <!-- ======================================================================= -->
814 <a name="globalvars">Global Variables</a>
819 <p>Global variables define regions of memory allocated at compilation time
820 instead of run-time. Global variables may optionally be initialized, may
821 have an explicit section to be placed in, and may have an optional explicit
822 alignment specified. A variable may be defined as "thread_local", which
823 means that it will not be shared by threads (each thread will have a
824 separated copy of the variable). A variable may be defined as a global
825 "constant," which indicates that the contents of the variable
826 will <b>never</b> be modified (enabling better optimization, allowing the
827 global data to be placed in the read-only section of an executable, etc).
828 Note that variables that need runtime initialization cannot be marked
829 "constant" as there is a store to the variable.</p>
831 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
832 constant, even if the final definition of the global is not. This capability
833 can be used to enable slightly better optimization of the program, but
834 requires the language definition to guarantee that optimizations based on the
835 'constantness' are valid for the translation units that do not include the
838 <p>As SSA values, global variables define pointer values that are in scope
839 (i.e. they dominate) all basic blocks in the program. Global variables
840 always define a pointer to their "content" type because they describe a
841 region of memory, and all memory objects in LLVM are accessed through
844 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
845 that the address is not significant, only the content. Constants marked
846 like this can be merged with other constants if they have the same
847 initializer. Note that a constant with significant address <em>can</em>
848 be merged with a <tt>unnamed_addr</tt> constant, the result being a
849 constant whose address is significant.</p>
851 <p>A global variable may be declared to reside in a target-specific numbered
852 address space. For targets that support them, address spaces may affect how
853 optimizations are performed and/or what target instructions are used to
854 access the variable. The default address space is zero. The address space
855 qualifier must precede any other attributes.</p>
857 <p>LLVM allows an explicit section to be specified for globals. If the target
858 supports it, it will emit globals to the section specified.</p>
860 <p>An explicit alignment may be specified for a global, which must be a power
861 of 2. If not present, or if the alignment is set to zero, the alignment of
862 the global is set by the target to whatever it feels convenient. If an
863 explicit alignment is specified, the global is forced to have exactly that
864 alignment. Targets and optimizers are not allowed to over-align the global
865 if the global has an assigned section. In this case, the extra alignment
866 could be observable: for example, code could assume that the globals are
867 densely packed in their section and try to iterate over them as an array,
868 alignment padding would break this iteration.</p>
870 <p>For example, the following defines a global in a numbered address space with
871 an initializer, section, and alignment:</p>
873 <pre class="doc_code">
874 @G = addrspace(5) constant float 1.0, section "foo", align 4
880 <!-- ======================================================================= -->
882 <a name="functionstructure">Functions</a>
887 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
888 optional <a href="#linkage">linkage type</a>, an optional
889 <a href="#visibility">visibility style</a>, an optional
890 <a href="#callingconv">calling convention</a>,
891 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
892 <a href="#paramattrs">parameter attribute</a> for the return type, a function
893 name, a (possibly empty) argument list (each with optional
894 <a href="#paramattrs">parameter attributes</a>), optional
895 <a href="#fnattrs">function attributes</a>, an optional section, an optional
896 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
897 curly brace, a list of basic blocks, and a closing curly brace.</p>
899 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
900 optional <a href="#linkage">linkage type</a>, an optional
901 <a href="#visibility">visibility style</a>, an optional
902 <a href="#callingconv">calling convention</a>,
903 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
904 <a href="#paramattrs">parameter attribute</a> for the return type, a function
905 name, a possibly empty list of arguments, an optional alignment, and an
906 optional <a href="#gc">garbage collector name</a>.</p>
908 <p>A function definition contains a list of basic blocks, forming the CFG
909 (Control Flow Graph) for the function. Each basic block may optionally start
910 with a label (giving the basic block a symbol table entry), contains a list
911 of instructions, and ends with a <a href="#terminators">terminator</a>
912 instruction (such as a branch or function return).</p>
914 <p>The first basic block in a function is special in two ways: it is immediately
915 executed on entrance to the function, and it is not allowed to have
916 predecessor basic blocks (i.e. there can not be any branches to the entry
917 block of a function). Because the block can have no predecessors, it also
918 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
920 <p>LLVM allows an explicit section to be specified for functions. If the target
921 supports it, it will emit functions to the section specified.</p>
923 <p>An explicit alignment may be specified for a function. If not present, or if
924 the alignment is set to zero, the alignment of the function is set by the
925 target to whatever it feels convenient. If an explicit alignment is
926 specified, the function is forced to have at least that much alignment. All
927 alignments must be a power of 2.</p>
929 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
930 be significant and two identical functions can be merged</p>.
933 <pre class="doc_code">
934 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
935 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
936 <ResultType> @<FunctionName> ([argument list])
937 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
938 [<a href="#gc">gc</a>] { ... }
943 <!-- ======================================================================= -->
945 <a name="aliasstructure">Aliases</a>
950 <p>Aliases act as "second name" for the aliasee value (which can be either
951 function, global variable, another alias or bitcast of global value). Aliases
952 may have an optional <a href="#linkage">linkage type</a>, and an
953 optional <a href="#visibility">visibility style</a>.</p>
956 <pre class="doc_code">
957 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
962 <!-- ======================================================================= -->
964 <a name="namedmetadatastructure">Named Metadata</a>
969 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
970 nodes</a> (but not metadata strings) are the only valid operands for
971 a named metadata.</p>
974 <pre class="doc_code">
975 ; Some unnamed metadata nodes, which are referenced by the named metadata.
976 !0 = metadata !{metadata !"zero"}
977 !1 = metadata !{metadata !"one"}
978 !2 = metadata !{metadata !"two"}
980 !name = !{!0, !1, !2}
985 <!-- ======================================================================= -->
987 <a name="paramattrs">Parameter Attributes</a>
992 <p>The return type and each parameter of a function type may have a set of
993 <i>parameter attributes</i> associated with them. Parameter attributes are
994 used to communicate additional information about the result or parameters of
995 a function. Parameter attributes are considered to be part of the function,
996 not of the function type, so functions with different parameter attributes
997 can have the same function type.</p>
999 <p>Parameter attributes are simple keywords that follow the type specified. If
1000 multiple parameter attributes are needed, they are space separated. For
1003 <pre class="doc_code">
1004 declare i32 @printf(i8* noalias nocapture, ...)
1005 declare i32 @atoi(i8 zeroext)
1006 declare signext i8 @returns_signed_char()
1009 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1010 <tt>readonly</tt>) come immediately after the argument list.</p>
1012 <p>Currently, only the following parameter attributes are defined:</p>
1015 <dt><tt><b>zeroext</b></tt></dt>
1016 <dd>This indicates to the code generator that the parameter or return value
1017 should be zero-extended to the extent required by the target's ABI (which
1018 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1019 parameter) or the callee (for a return value).</dd>
1021 <dt><tt><b>signext</b></tt></dt>
1022 <dd>This indicates to the code generator that the parameter or return value
1023 should be sign-extended to the extent required by the target's ABI (which
1024 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1027 <dt><tt><b>inreg</b></tt></dt>
1028 <dd>This indicates that this parameter or return value should be treated in a
1029 special target-dependent fashion during while emitting code for a function
1030 call or return (usually, by putting it in a register as opposed to memory,
1031 though some targets use it to distinguish between two different kinds of
1032 registers). Use of this attribute is target-specific.</dd>
1034 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1035 <dd><p>This indicates that the pointer parameter should really be passed by
1036 value to the function. The attribute implies that a hidden copy of the
1038 is made between the caller and the callee, so the callee is unable to
1039 modify the value in the callee. This attribute is only valid on LLVM
1040 pointer arguments. It is generally used to pass structs and arrays by
1041 value, but is also valid on pointers to scalars. The copy is considered
1042 to belong to the caller not the callee (for example,
1043 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1044 <tt>byval</tt> parameters). This is not a valid attribute for return
1047 <p>The byval attribute also supports specifying an alignment with
1048 the align attribute. It indicates the alignment of the stack slot to
1049 form and the known alignment of the pointer specified to the call site. If
1050 the alignment is not specified, then the code generator makes a
1051 target-specific assumption.</p></dd>
1053 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1054 <dd>This indicates that the pointer parameter specifies the address of a
1055 structure that is the return value of the function in the source program.
1056 This pointer must be guaranteed by the caller to be valid: loads and
1057 stores to the structure may be assumed by the callee to not to trap. This
1058 may only be applied to the first parameter. This is not a valid attribute
1059 for return values. </dd>
1061 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1062 <dd>This indicates that pointer values
1063 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1064 value do not alias pointer values which are not <i>based</i> on it,
1065 ignoring certain "irrelevant" dependencies.
1066 For a call to the parent function, dependencies between memory
1067 references from before or after the call and from those during the call
1068 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1069 return value used in that call.
1070 The caller shares the responsibility with the callee for ensuring that
1071 these requirements are met.
1072 For further details, please see the discussion of the NoAlias response in
1073 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1075 Note that this definition of <tt>noalias</tt> is intentionally
1076 similar to the definition of <tt>restrict</tt> in C99 for function
1077 arguments, though it is slightly weaker.
1079 For function return values, C99's <tt>restrict</tt> is not meaningful,
1080 while LLVM's <tt>noalias</tt> is.
1083 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1084 <dd>This indicates that the callee does not make any copies of the pointer
1085 that outlive the callee itself. This is not a valid attribute for return
1088 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1089 <dd>This indicates that the pointer parameter can be excised using the
1090 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1091 attribute for return values.</dd>
1096 <!-- ======================================================================= -->
1098 <a name="gc">Garbage Collector Names</a>
1103 <p>Each function may specify a garbage collector name, which is simply a
1106 <pre class="doc_code">
1107 define void @f() gc "name" { ... }
1110 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1111 collector which will cause the compiler to alter its output in order to
1112 support the named garbage collection algorithm.</p>
1116 <!-- ======================================================================= -->
1118 <a name="fnattrs">Function Attributes</a>
1123 <p>Function attributes are set to communicate additional information about a
1124 function. Function attributes are considered to be part of the function, not
1125 of the function type, so functions with different parameter attributes can
1126 have the same function type.</p>
1128 <p>Function attributes are simple keywords that follow the type specified. If
1129 multiple attributes are needed, they are space separated. For example:</p>
1131 <pre class="doc_code">
1132 define void @f() noinline { ... }
1133 define void @f() alwaysinline { ... }
1134 define void @f() alwaysinline optsize { ... }
1135 define void @f() optsize { ... }
1139 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1140 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1141 the backend should forcibly align the stack pointer. Specify the
1142 desired alignment, which must be a power of two, in parentheses.
1144 <dt><tt><b>alwaysinline</b></tt></dt>
1145 <dd>This attribute indicates that the inliner should attempt to inline this
1146 function into callers whenever possible, ignoring any active inlining size
1147 threshold for this caller.</dd>
1149 <dt><tt><b>nonlazybind</b></tt></dt>
1150 <dd>This attribute suppresses lazy symbol binding for the function. This
1151 may make calls to the function faster, at the cost of extra program
1152 startup time if the function is not called during program startup.</dd>
1154 <dt><tt><b>inlinehint</b></tt></dt>
1155 <dd>This attribute indicates that the source code contained a hint that inlining
1156 this function is desirable (such as the "inline" keyword in C/C++). It
1157 is just a hint; it imposes no requirements on the inliner.</dd>
1159 <dt><tt><b>naked</b></tt></dt>
1160 <dd>This attribute disables prologue / epilogue emission for the function.
1161 This can have very system-specific consequences.</dd>
1163 <dt><tt><b>noimplicitfloat</b></tt></dt>
1164 <dd>This attributes disables implicit floating point instructions.</dd>
1166 <dt><tt><b>noinline</b></tt></dt>
1167 <dd>This attribute indicates that the inliner should never inline this
1168 function in any situation. This attribute may not be used together with
1169 the <tt>alwaysinline</tt> attribute.</dd>
1171 <dt><tt><b>noredzone</b></tt></dt>
1172 <dd>This attribute indicates that the code generator should not use a red
1173 zone, even if the target-specific ABI normally permits it.</dd>
1175 <dt><tt><b>noreturn</b></tt></dt>
1176 <dd>This function attribute indicates that the function never returns
1177 normally. This produces undefined behavior at runtime if the function
1178 ever does dynamically return.</dd>
1180 <dt><tt><b>nounwind</b></tt></dt>
1181 <dd>This function attribute indicates that the function never returns with an
1182 unwind or exceptional control flow. If the function does unwind, its
1183 runtime behavior is undefined.</dd>
1185 <dt><tt><b>optsize</b></tt></dt>
1186 <dd>This attribute suggests that optimization passes and code generator passes
1187 make choices that keep the code size of this function low, and otherwise
1188 do optimizations specifically to reduce code size.</dd>
1190 <dt><tt><b>readnone</b></tt></dt>
1191 <dd>This attribute indicates that the function computes its result (or decides
1192 to unwind an exception) based strictly on its arguments, without
1193 dereferencing any pointer arguments or otherwise accessing any mutable
1194 state (e.g. memory, control registers, etc) visible to caller functions.
1195 It does not write through any pointer arguments
1196 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1197 changes any state visible to callers. This means that it cannot unwind
1198 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1199 could use the <tt>unwind</tt> instruction.</dd>
1201 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1202 <dd>This attribute indicates that the function does not write through any
1203 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1204 arguments) or otherwise modify any state (e.g. memory, control registers,
1205 etc) visible to caller functions. It may dereference pointer arguments
1206 and read state that may be set in the caller. A readonly function always
1207 returns the same value (or unwinds an exception identically) when called
1208 with the same set of arguments and global state. It cannot unwind an
1209 exception by calling the <tt>C++</tt> exception throwing methods, but may
1210 use the <tt>unwind</tt> instruction.</dd>
1212 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1213 <dd>This attribute indicates that the function should emit a stack smashing
1214 protector. It is in the form of a "canary"—a random value placed on
1215 the stack before the local variables that's checked upon return from the
1216 function to see if it has been overwritten. A heuristic is used to
1217 determine if a function needs stack protectors or not.<br>
1219 If a function that has an <tt>ssp</tt> attribute is inlined into a
1220 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1221 function will have an <tt>ssp</tt> attribute.</dd>
1223 <dt><tt><b>sspreq</b></tt></dt>
1224 <dd>This attribute indicates that the function should <em>always</em> emit a
1225 stack smashing protector. This overrides
1226 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1228 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1229 function that doesn't have an <tt>sspreq</tt> attribute or which has
1230 an <tt>ssp</tt> attribute, then the resulting function will have
1231 an <tt>sspreq</tt> attribute.</dd>
1233 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1234 <dd>This attribute indicates that the ABI being targeted requires that
1235 an unwind table entry be produce for this function even if we can
1236 show that no exceptions passes by it. This is normally the case for
1237 the ELF x86-64 abi, but it can be disabled for some compilation
1240 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1241 <dd>This attribute indicates that this function can return
1242 twice. The C <code>setjmp</code> is an example of such a function.
1243 The compiler disables some optimizations (like tail calls) in the caller of
1244 these functions.</dd>
1249 <!-- ======================================================================= -->
1251 <a name="moduleasm">Module-Level Inline Assembly</a>
1256 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1257 the GCC "file scope inline asm" blocks. These blocks are internally
1258 concatenated by LLVM and treated as a single unit, but may be separated in
1259 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1261 <pre class="doc_code">
1262 module asm "inline asm code goes here"
1263 module asm "more can go here"
1266 <p>The strings can contain any character by escaping non-printable characters.
1267 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1270 <p>The inline asm code is simply printed to the machine code .s file when
1271 assembly code is generated.</p>
1275 <!-- ======================================================================= -->
1277 <a name="datalayout">Data Layout</a>
1282 <p>A module may specify a target specific data layout string that specifies how
1283 data is to be laid out in memory. The syntax for the data layout is
1286 <pre class="doc_code">
1287 target datalayout = "<i>layout specification</i>"
1290 <p>The <i>layout specification</i> consists of a list of specifications
1291 separated by the minus sign character ('-'). Each specification starts with
1292 a letter and may include other information after the letter to define some
1293 aspect of the data layout. The specifications accepted are as follows:</p>
1297 <dd>Specifies that the target lays out data in big-endian form. That is, the
1298 bits with the most significance have the lowest address location.</dd>
1301 <dd>Specifies that the target lays out data in little-endian form. That is,
1302 the bits with the least significance have the lowest address
1305 <dt><tt>S<i>size</i></tt></dt>
1306 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1307 of stack variables is limited to the natural stack alignment to avoid
1308 dynamic stack realignment. The stack alignment must be a multiple of
1309 8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1310 which does not prevent any alignment promotions.</dd>
1312 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1313 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1314 <i>preferred</i> alignments. All sizes are in bits. Specifying
1315 the <i>pref</i> alignment is optional. If omitted, the
1316 preceding <tt>:</tt> should be omitted too.</dd>
1318 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1319 <dd>This specifies the alignment for an integer type of a given bit
1320 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1322 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1323 <dd>This specifies the alignment for a vector type of a given bit
1326 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1327 <dd>This specifies the alignment for a floating point type of a given bit
1328 <i>size</i>. Only values of <i>size</i> that are supported by the target
1329 will work. 32 (float) and 64 (double) are supported on all targets;
1330 80 or 128 (different flavors of long double) are also supported on some
1333 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1334 <dd>This specifies the alignment for an aggregate type of a given bit
1337 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1338 <dd>This specifies the alignment for a stack object of a given bit
1341 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1342 <dd>This specifies a set of native integer widths for the target CPU
1343 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1344 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1345 this set are considered to support most general arithmetic
1346 operations efficiently.</dd>
1349 <p>When constructing the data layout for a given target, LLVM starts with a
1350 default set of specifications which are then (possibly) overridden by the
1351 specifications in the <tt>datalayout</tt> keyword. The default specifications
1352 are given in this list:</p>
1355 <li><tt>E</tt> - big endian</li>
1356 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1357 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1358 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1359 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1360 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1361 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1362 alignment of 64-bits</li>
1363 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1364 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1365 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1366 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1367 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1368 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1371 <p>When LLVM is determining the alignment for a given type, it uses the
1372 following rules:</p>
1375 <li>If the type sought is an exact match for one of the specifications, that
1376 specification is used.</li>
1378 <li>If no match is found, and the type sought is an integer type, then the
1379 smallest integer type that is larger than the bitwidth of the sought type
1380 is used. If none of the specifications are larger than the bitwidth then
1381 the the largest integer type is used. For example, given the default
1382 specifications above, the i7 type will use the alignment of i8 (next
1383 largest) while both i65 and i256 will use the alignment of i64 (largest
1386 <li>If no match is found, and the type sought is a vector type, then the
1387 largest vector type that is smaller than the sought vector type will be
1388 used as a fall back. This happens because <128 x double> can be
1389 implemented in terms of 64 <2 x double>, for example.</li>
1392 <p>The function of the data layout string may not be what you expect. Notably,
1393 this is not a specification from the frontend of what alignment the code
1394 generator should use.</p>
1396 <p>Instead, if specified, the target data layout is required to match what the
1397 ultimate <em>code generator</em> expects. This string is used by the
1398 mid-level optimizers to
1399 improve code, and this only works if it matches what the ultimate code
1400 generator uses. If you would like to generate IR that does not embed this
1401 target-specific detail into the IR, then you don't have to specify the
1402 string. This will disable some optimizations that require precise layout
1403 information, but this also prevents those optimizations from introducing
1404 target specificity into the IR.</p>
1410 <!-- ======================================================================= -->
1412 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1417 <p>Any memory access must be done through a pointer value associated
1418 with an address range of the memory access, otherwise the behavior
1419 is undefined. Pointer values are associated with address ranges
1420 according to the following rules:</p>
1423 <li>A pointer value is associated with the addresses associated with
1424 any value it is <i>based</i> on.
1425 <li>An address of a global variable is associated with the address
1426 range of the variable's storage.</li>
1427 <li>The result value of an allocation instruction is associated with
1428 the address range of the allocated storage.</li>
1429 <li>A null pointer in the default address-space is associated with
1431 <li>An integer constant other than zero or a pointer value returned
1432 from a function not defined within LLVM may be associated with address
1433 ranges allocated through mechanisms other than those provided by
1434 LLVM. Such ranges shall not overlap with any ranges of addresses
1435 allocated by mechanisms provided by LLVM.</li>
1438 <p>A pointer value is <i>based</i> on another pointer value according
1439 to the following rules:</p>
1442 <li>A pointer value formed from a
1443 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1444 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1445 <li>The result value of a
1446 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1447 of the <tt>bitcast</tt>.</li>
1448 <li>A pointer value formed by an
1449 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1450 pointer values that contribute (directly or indirectly) to the
1451 computation of the pointer's value.</li>
1452 <li>The "<i>based</i> on" relationship is transitive.</li>
1455 <p>Note that this definition of <i>"based"</i> is intentionally
1456 similar to the definition of <i>"based"</i> in C99, though it is
1457 slightly weaker.</p>
1459 <p>LLVM IR does not associate types with memory. The result type of a
1460 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1461 alignment of the memory from which to load, as well as the
1462 interpretation of the value. The first operand type of a
1463 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1464 and alignment of the store.</p>
1466 <p>Consequently, type-based alias analysis, aka TBAA, aka
1467 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1468 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1469 additional information which specialized optimization passes may use
1470 to implement type-based alias analysis.</p>
1474 <!-- ======================================================================= -->
1476 <a name="volatile">Volatile Memory Accesses</a>
1481 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1482 href="#i_store"><tt>store</tt></a>s, and <a
1483 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1484 The optimizers must not change the number of volatile operations or change their
1485 order of execution relative to other volatile operations. The optimizers
1486 <i>may</i> change the order of volatile operations relative to non-volatile
1487 operations. This is not Java's "volatile" and has no cross-thread
1488 synchronization behavior.</p>
1492 <!-- ======================================================================= -->
1494 <a name="memmodel">Memory Model for Concurrent Operations</a>
1499 <p>The LLVM IR does not define any way to start parallel threads of execution
1500 or to register signal handlers. Nonetheless, there are platform-specific
1501 ways to create them, and we define LLVM IR's behavior in their presence. This
1502 model is inspired by the C++0x memory model.</p>
1504 <p>For a more informal introduction to this model, see the
1505 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1507 <p>We define a <i>happens-before</i> partial order as the least partial order
1510 <li>Is a superset of single-thread program order, and</li>
1511 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1512 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1513 by platform-specific techniques, like pthread locks, thread
1514 creation, thread joining, etc., and by atomic instructions.
1515 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1519 <p>Note that program order does not introduce <i>happens-before</i> edges
1520 between a thread and signals executing inside that thread.</p>
1522 <p>Every (defined) read operation (load instructions, memcpy, atomic
1523 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1524 (defined) write operations (store instructions, atomic
1525 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1526 initialized globals are considered to have a write of the initializer which is
1527 atomic and happens before any other read or write of the memory in question.
1528 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1529 any write to the same byte, except:</p>
1532 <li>If <var>write<sub>1</sub></var> happens before
1533 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1534 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1535 does not see <var>write<sub>1</sub></var>.
1536 <li>If <var>R<sub>byte</sub></var> happens before
1537 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1538 see <var>write<sub>3</sub></var>.
1541 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1543 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1544 is supposed to give guarantees which can support
1545 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1546 addresses which do not behave like normal memory. It does not generally
1547 provide cross-thread synchronization.)
1548 <li>Otherwise, if there is no write to the same byte that happens before
1549 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1550 <tt>undef</tt> for that byte.
1551 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1552 <var>R<sub>byte</sub></var> returns the value written by that
1554 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1555 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1556 values written. See the <a href="#ordering">Atomic Memory Ordering
1557 Constraints</a> section for additional constraints on how the choice
1559 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1562 <p><var>R</var> returns the value composed of the series of bytes it read.
1563 This implies that some bytes within the value may be <tt>undef</tt>
1564 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1565 defines the semantics of the operation; it doesn't mean that targets will
1566 emit more than one instruction to read the series of bytes.</p>
1568 <p>Note that in cases where none of the atomic intrinsics are used, this model
1569 places only one restriction on IR transformations on top of what is required
1570 for single-threaded execution: introducing a store to a byte which might not
1571 otherwise be stored is not allowed in general. (Specifically, in the case
1572 where another thread might write to and read from an address, introducing a
1573 store can change a load that may see exactly one write into a load that may
1574 see multiple writes.)</p>
1576 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1577 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1578 none of the backends currently in the tree fall into this category; however,
1579 there might be targets which care. If there are, we want a paragraph
1582 Targets may specify that stores narrower than a certain width are not
1583 available; on such a target, for the purposes of this model, treat any
1584 non-atomic write with an alignment or width less than the minimum width
1585 as if it writes to the relevant surrounding bytes.
1590 <!-- ======================================================================= -->
1592 <a name="ordering">Atomic Memory Ordering Constraints</a>
1597 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1598 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1599 <a href="#i_fence"><code>fence</code></a>,
1600 <a href="#i_load"><code>atomic load</code></a>, and
1601 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1602 that determines which other atomic instructions on the same address they
1603 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1604 but are somewhat more colloquial. If these descriptions aren't precise enough,
1605 check those specs (see spec references in the
1606 <a href="Atomic.html#introduction">atomics guide</a>).
1607 <a href="#i_fence"><code>fence</code></a> instructions
1608 treat these orderings somewhat differently since they don't take an address.
1609 See that instruction's documentation for details.</p>
1611 <p>For a simpler introduction to the ordering constraints, see the
1612 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1615 <dt><code>unordered</code></dt>
1616 <dd>The set of values that can be read is governed by the happens-before
1617 partial order. A value cannot be read unless some operation wrote it.
1618 This is intended to provide a guarantee strong enough to model Java's
1619 non-volatile shared variables. This ordering cannot be specified for
1620 read-modify-write operations; it is not strong enough to make them atomic
1621 in any interesting way.</dd>
1622 <dt><code>monotonic</code></dt>
1623 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1624 total order for modifications by <code>monotonic</code> operations on each
1625 address. All modification orders must be compatible with the happens-before
1626 order. There is no guarantee that the modification orders can be combined to
1627 a global total order for the whole program (and this often will not be
1628 possible). The read in an atomic read-modify-write operation
1629 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1630 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1631 reads the value in the modification order immediately before the value it
1632 writes. If one atomic read happens before another atomic read of the same
1633 address, the later read must see the same value or a later value in the
1634 address's modification order. This disallows reordering of
1635 <code>monotonic</code> (or stronger) operations on the same address. If an
1636 address is written <code>monotonic</code>ally by one thread, and other threads
1637 <code>monotonic</code>ally read that address repeatedly, the other threads must
1638 eventually see the write. This corresponds to the C++0x/C1x
1639 <code>memory_order_relaxed</code>.</dd>
1640 <dt><code>acquire</code></dt>
1641 <dd>In addition to the guarantees of <code>monotonic</code>,
1642 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1643 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1644 <dt><code>release</code></dt>
1645 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1646 writes a value which is subsequently read by an <code>acquire</code> operation,
1647 it <i>synchronizes-with</i> that operation. (This isn't a complete
1648 description; see the C++0x definition of a release sequence.) This corresponds
1649 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1650 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1651 <code>acquire</code> and <code>release</code> operation on its address.
1652 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1653 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1654 <dd>In addition to the guarantees of <code>acq_rel</code>
1655 (<code>acquire</code> for an operation which only reads, <code>release</code>
1656 for an operation which only writes), there is a global total order on all
1657 sequentially-consistent operations on all addresses, which is consistent with
1658 the <i>happens-before</i> partial order and with the modification orders of
1659 all the affected addresses. Each sequentially-consistent read sees the last
1660 preceding write to the same address in this global order. This corresponds
1661 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1664 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1665 it only <i>synchronizes with</i> or participates in modification and seq_cst
1666 total orderings with other operations running in the same thread (for example,
1667 in signal handlers).</p>
1673 <!-- *********************************************************************** -->
1674 <h2><a name="typesystem">Type System</a></h2>
1675 <!-- *********************************************************************** -->
1679 <p>The LLVM type system is one of the most important features of the
1680 intermediate representation. Being typed enables a number of optimizations
1681 to be performed on the intermediate representation directly, without having
1682 to do extra analyses on the side before the transformation. A strong type
1683 system makes it easier to read the generated code and enables novel analyses
1684 and transformations that are not feasible to perform on normal three address
1685 code representations.</p>
1687 <!-- ======================================================================= -->
1689 <a name="t_classifications">Type Classifications</a>
1694 <p>The types fall into a few useful classifications:</p>
1696 <table border="1" cellspacing="0" cellpadding="4">
1698 <tr><th>Classification</th><th>Types</th></tr>
1700 <td><a href="#t_integer">integer</a></td>
1701 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1704 <td><a href="#t_floating">floating point</a></td>
1705 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1708 <td><a name="t_firstclass">first class</a></td>
1709 <td><a href="#t_integer">integer</a>,
1710 <a href="#t_floating">floating point</a>,
1711 <a href="#t_pointer">pointer</a>,
1712 <a href="#t_vector">vector</a>,
1713 <a href="#t_struct">structure</a>,
1714 <a href="#t_array">array</a>,
1715 <a href="#t_label">label</a>,
1716 <a href="#t_metadata">metadata</a>.
1720 <td><a href="#t_primitive">primitive</a></td>
1721 <td><a href="#t_label">label</a>,
1722 <a href="#t_void">void</a>,
1723 <a href="#t_integer">integer</a>,
1724 <a href="#t_floating">floating point</a>,
1725 <a href="#t_x86mmx">x86mmx</a>,
1726 <a href="#t_metadata">metadata</a>.</td>
1729 <td><a href="#t_derived">derived</a></td>
1730 <td><a href="#t_array">array</a>,
1731 <a href="#t_function">function</a>,
1732 <a href="#t_pointer">pointer</a>,
1733 <a href="#t_struct">structure</a>,
1734 <a href="#t_vector">vector</a>,
1735 <a href="#t_opaque">opaque</a>.
1741 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1742 important. Values of these types are the only ones which can be produced by
1747 <!-- ======================================================================= -->
1749 <a name="t_primitive">Primitive Types</a>
1754 <p>The primitive types are the fundamental building blocks of the LLVM
1757 <!-- _______________________________________________________________________ -->
1759 <a name="t_integer">Integer Type</a>
1765 <p>The integer type is a very simple type that simply specifies an arbitrary
1766 bit width for the integer type desired. Any bit width from 1 bit to
1767 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1774 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1778 <table class="layout">
1780 <td class="left"><tt>i1</tt></td>
1781 <td class="left">a single-bit integer.</td>
1784 <td class="left"><tt>i32</tt></td>
1785 <td class="left">a 32-bit integer.</td>
1788 <td class="left"><tt>i1942652</tt></td>
1789 <td class="left">a really big integer of over 1 million bits.</td>
1795 <!-- _______________________________________________________________________ -->
1797 <a name="t_floating">Floating Point Types</a>
1804 <tr><th>Type</th><th>Description</th></tr>
1805 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1806 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1807 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1808 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1809 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1815 <!-- _______________________________________________________________________ -->
1817 <a name="t_x86mmx">X86mmx Type</a>
1823 <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>
1832 <!-- _______________________________________________________________________ -->
1834 <a name="t_void">Void Type</a>
1840 <p>The void type does not represent any value and has no size.</p>
1849 <!-- _______________________________________________________________________ -->
1851 <a name="t_label">Label Type</a>
1857 <p>The label type represents code labels.</p>
1866 <!-- _______________________________________________________________________ -->
1868 <a name="t_metadata">Metadata Type</a>
1874 <p>The metadata type represents embedded metadata. No derived types may be
1875 created from metadata except for <a href="#t_function">function</a>
1887 <!-- ======================================================================= -->
1889 <a name="t_derived">Derived Types</a>
1894 <p>The real power in LLVM comes from the derived types in the system. This is
1895 what allows a programmer to represent arrays, functions, pointers, and other
1896 useful types. Each of these types contain one or more element types which
1897 may be a primitive type, or another derived type. For example, it is
1898 possible to have a two dimensional array, using an array as the element type
1899 of another array.</p>
1904 <!-- _______________________________________________________________________ -->
1906 <a name="t_aggregate">Aggregate Types</a>
1911 <p>Aggregate Types are a subset of derived types that can contain multiple
1912 member types. <a href="#t_array">Arrays</a>,
1913 <a href="#t_struct">structs</a>, and <a href="#t_vector">vectors</a> are
1914 aggregate types.</p>
1918 <!-- _______________________________________________________________________ -->
1920 <a name="t_array">Array Type</a>
1926 <p>The array type is a very simple derived type that arranges elements
1927 sequentially in memory. The array type requires a size (number of elements)
1928 and an underlying data type.</p>
1932 [<# elements> x <elementtype>]
1935 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1936 be any type with a size.</p>
1939 <table class="layout">
1941 <td class="left"><tt>[40 x i32]</tt></td>
1942 <td class="left">Array of 40 32-bit integer values.</td>
1945 <td class="left"><tt>[41 x i32]</tt></td>
1946 <td class="left">Array of 41 32-bit integer values.</td>
1949 <td class="left"><tt>[4 x i8]</tt></td>
1950 <td class="left">Array of 4 8-bit integer values.</td>
1953 <p>Here are some examples of multidimensional arrays:</p>
1954 <table class="layout">
1956 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1957 <td class="left">3x4 array of 32-bit integer values.</td>
1960 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1961 <td class="left">12x10 array of single precision floating point values.</td>
1964 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1965 <td class="left">2x3x4 array of 16-bit integer values.</td>
1969 <p>There is no restriction on indexing beyond the end of the array implied by
1970 a static type (though there are restrictions on indexing beyond the bounds
1971 of an allocated object in some cases). This means that single-dimension
1972 'variable sized array' addressing can be implemented in LLVM with a zero
1973 length array type. An implementation of 'pascal style arrays' in LLVM could
1974 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1978 <!-- _______________________________________________________________________ -->
1980 <a name="t_function">Function Type</a>
1986 <p>The function type can be thought of as a function signature. It consists of
1987 a return type and a list of formal parameter types. The return type of a
1988 function type is a first class type or a void type.</p>
1992 <returntype> (<parameter list>)
1995 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1996 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1997 which indicates that the function takes a variable number of arguments.
1998 Variable argument functions can access their arguments with
1999 the <a href="#int_varargs">variable argument handling intrinsic</a>
2000 functions. '<tt><returntype></tt>' is any type except
2001 <a href="#t_label">label</a>.</p>
2004 <table class="layout">
2006 <td class="left"><tt>i32 (i32)</tt></td>
2007 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2009 </tr><tr class="layout">
2010 <td class="left"><tt>float (i16, i32 *) *
2012 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2013 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2014 returning <tt>float</tt>.
2016 </tr><tr class="layout">
2017 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2018 <td class="left">A vararg function that takes at least one
2019 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2020 which returns an integer. This is the signature for <tt>printf</tt> in
2023 </tr><tr class="layout">
2024 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2025 <td class="left">A function taking an <tt>i32</tt>, returning a
2026 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2033 <!-- _______________________________________________________________________ -->
2035 <a name="t_struct">Structure Type</a>
2041 <p>The structure type is used to represent a collection of data members together
2042 in memory. The elements of a structure may be any type that has a size.</p>
2044 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2045 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2046 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2047 Structures in registers are accessed using the
2048 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2049 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2051 <p>Structures may optionally be "packed" structures, which indicate that the
2052 alignment of the struct is one byte, and that there is no padding between
2053 the elements. In non-packed structs, padding between field types is inserted
2054 as defined by the TargetData string in the module, which is required to match
2055 what the underlying code generator expects.</p>
2057 <p>Structures can either be "literal" or "identified". A literal structure is
2058 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2059 types are always defined at the top level with a name. Literal types are
2060 uniqued by their contents and can never be recursive or opaque since there is
2061 no way to write one. Identified types can be recursive, can be opaqued, and are
2067 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2068 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2072 <table class="layout">
2074 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2075 <td class="left">A triple of three <tt>i32</tt> values</td>
2078 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2079 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2080 second element is a <a href="#t_pointer">pointer</a> to a
2081 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2082 an <tt>i32</tt>.</td>
2085 <td class="left"><tt><{ i8, i32 }></tt></td>
2086 <td class="left">A packed struct known to be 5 bytes in size.</td>
2092 <!-- _______________________________________________________________________ -->
2094 <a name="t_opaque">Opaque Structure Types</a>
2100 <p>Opaque structure types are used to represent named structure types that do
2101 not have a body specified. This corresponds (for example) to the C notion of
2102 a forward declared structure.</p>
2111 <table class="layout">
2113 <td class="left"><tt>opaque</tt></td>
2114 <td class="left">An opaque type.</td>
2122 <!-- _______________________________________________________________________ -->
2124 <a name="t_pointer">Pointer Type</a>
2130 <p>The pointer type is used to specify memory locations.
2131 Pointers are commonly used to reference objects in memory.</p>
2133 <p>Pointer types may have an optional address space attribute defining the
2134 numbered address space where the pointed-to object resides. The default
2135 address space is number zero. The semantics of non-zero address
2136 spaces are target-specific.</p>
2138 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2139 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2147 <table class="layout">
2149 <td class="left"><tt>[4 x i32]*</tt></td>
2150 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2151 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2154 <td class="left"><tt>i32 (i32*) *</tt></td>
2155 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2156 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2160 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2161 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2162 that resides in address space #5.</td>
2168 <!-- _______________________________________________________________________ -->
2170 <a name="t_vector">Vector Type</a>
2176 <p>A vector type is a simple derived type that represents a vector of elements.
2177 Vector types are used when multiple primitive data are operated in parallel
2178 using a single instruction (SIMD). A vector type requires a size (number of
2179 elements) and an underlying primitive data type. Vector types are considered
2180 <a href="#t_firstclass">first class</a>.</p>
2184 < <# elements> x <elementtype> >
2187 <p>The number of elements is a constant integer value larger than 0; elementtype
2188 may be any integer or floating point type. Vectors of size zero are not
2189 allowed, and pointers are not allowed as the element type.</p>
2192 <table class="layout">
2194 <td class="left"><tt><4 x i32></tt></td>
2195 <td class="left">Vector of 4 32-bit integer values.</td>
2198 <td class="left"><tt><8 x float></tt></td>
2199 <td class="left">Vector of 8 32-bit floating-point values.</td>
2202 <td class="left"><tt><2 x i64></tt></td>
2203 <td class="left">Vector of 2 64-bit integer values.</td>
2211 <!-- *********************************************************************** -->
2212 <h2><a name="constants">Constants</a></h2>
2213 <!-- *********************************************************************** -->
2217 <p>LLVM has several different basic types of constants. This section describes
2218 them all and their syntax.</p>
2220 <!-- ======================================================================= -->
2222 <a name="simpleconstants">Simple Constants</a>
2228 <dt><b>Boolean constants</b></dt>
2229 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2230 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2232 <dt><b>Integer constants</b></dt>
2233 <dd>Standard integers (such as '4') are constants of
2234 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2235 with integer types.</dd>
2237 <dt><b>Floating point constants</b></dt>
2238 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2239 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2240 notation (see below). The assembler requires the exact decimal value of a
2241 floating-point constant. For example, the assembler accepts 1.25 but
2242 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2243 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2245 <dt><b>Null pointer constants</b></dt>
2246 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2247 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2250 <p>The one non-intuitive notation for constants is the hexadecimal form of
2251 floating point constants. For example, the form '<tt>double
2252 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2253 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2254 constants are required (and the only time that they are generated by the
2255 disassembler) is when a floating point constant must be emitted but it cannot
2256 be represented as a decimal floating point number in a reasonable number of
2257 digits. For example, NaN's, infinities, and other special values are
2258 represented in their IEEE hexadecimal format so that assembly and disassembly
2259 do not cause any bits to change in the constants.</p>
2261 <p>When using the hexadecimal form, constants of types float and double are
2262 represented using the 16-digit form shown above (which matches the IEEE754
2263 representation for double); float values must, however, be exactly
2264 representable as IEE754 single precision. Hexadecimal format is always used
2265 for long double, and there are three forms of long double. The 80-bit format
2266 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2267 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2268 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2269 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2270 currently supported target uses this format. Long doubles will only work if
2271 they match the long double format on your target. All hexadecimal formats
2272 are big-endian (sign bit at the left).</p>
2274 <p>There are no constants of type x86mmx.</p>
2277 <!-- ======================================================================= -->
2279 <a name="aggregateconstants"></a> <!-- old anchor -->
2280 <a name="complexconstants">Complex Constants</a>
2285 <p>Complex constants are a (potentially recursive) combination of simple
2286 constants and smaller complex constants.</p>
2289 <dt><b>Structure constants</b></dt>
2290 <dd>Structure constants are represented with notation similar to structure
2291 type definitions (a comma separated list of elements, surrounded by braces
2292 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2293 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2294 Structure constants must have <a href="#t_struct">structure type</a>, and
2295 the number and types of elements must match those specified by the
2298 <dt><b>Array constants</b></dt>
2299 <dd>Array constants are represented with notation similar to array type
2300 definitions (a comma separated list of elements, surrounded by square
2301 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2302 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2303 the number and types of elements must match those specified by the
2306 <dt><b>Vector constants</b></dt>
2307 <dd>Vector constants are represented with notation similar to vector type
2308 definitions (a comma separated list of elements, surrounded by
2309 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2310 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2311 have <a href="#t_vector">vector type</a>, and the number and types of
2312 elements must match those specified by the type.</dd>
2314 <dt><b>Zero initialization</b></dt>
2315 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2316 value to zero of <em>any</em> type, including scalar and
2317 <a href="#t_aggregate">aggregate</a> types.
2318 This is often used to avoid having to print large zero initializers
2319 (e.g. for large arrays) and is always exactly equivalent to using explicit
2320 zero initializers.</dd>
2322 <dt><b>Metadata node</b></dt>
2323 <dd>A metadata node is a structure-like constant with
2324 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2325 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2326 be interpreted as part of the instruction stream, metadata is a place to
2327 attach additional information such as debug info.</dd>
2332 <!-- ======================================================================= -->
2334 <a name="globalconstants">Global Variable and Function Addresses</a>
2339 <p>The addresses of <a href="#globalvars">global variables</a>
2340 and <a href="#functionstructure">functions</a> are always implicitly valid
2341 (link-time) constants. These constants are explicitly referenced when
2342 the <a href="#identifiers">identifier for the global</a> is used and always
2343 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2344 legal LLVM file:</p>
2346 <pre class="doc_code">
2349 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2354 <!-- ======================================================================= -->
2356 <a name="undefvalues">Undefined Values</a>
2361 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2362 indicates that the user of the value may receive an unspecified bit-pattern.
2363 Undefined values may be of any type (other than '<tt>label</tt>'
2364 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2366 <p>Undefined values are useful because they indicate to the compiler that the
2367 program is well defined no matter what value is used. This gives the
2368 compiler more freedom to optimize. Here are some examples of (potentially
2369 surprising) transformations that are valid (in pseudo IR):</p>
2372 <pre class="doc_code">
2382 <p>This is safe because all of the output bits are affected by the undef bits.
2383 Any output bit can have a zero or one depending on the input bits.</p>
2385 <pre class="doc_code">
2396 <p>These logical operations have bits that are not always affected by the input.
2397 For example, if <tt>%X</tt> has a zero bit, then the output of the
2398 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2399 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2400 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2401 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2402 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2403 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2404 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2406 <pre class="doc_code">
2407 %A = select undef, %X, %Y
2408 %B = select undef, 42, %Y
2409 %C = select %X, %Y, undef
2420 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2421 branch) conditions can go <em>either way</em>, but they have to come from one
2422 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2423 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2424 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2425 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2426 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2429 <pre class="doc_code">
2430 %A = xor undef, undef
2448 <p>This example points out that two '<tt>undef</tt>' operands are not
2449 necessarily the same. This can be surprising to people (and also matches C
2450 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2451 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2452 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2453 its value over its "live range". This is true because the variable doesn't
2454 actually <em>have a live range</em>. Instead, the value is logically read
2455 from arbitrary registers that happen to be around when needed, so the value
2456 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2457 need to have the same semantics or the core LLVM "replace all uses with"
2458 concept would not hold.</p>
2460 <pre class="doc_code">
2468 <p>These examples show the crucial difference between an <em>undefined
2469 value</em> and <em>undefined behavior</em>. An undefined value (like
2470 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2471 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2472 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2473 defined on SNaN's. However, in the second example, we can make a more
2474 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2475 arbitrary value, we are allowed to assume that it could be zero. Since a
2476 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2477 the operation does not execute at all. This allows us to delete the divide and
2478 all code after it. Because the undefined operation "can't happen", the
2479 optimizer can assume that it occurs in dead code.</p>
2481 <pre class="doc_code">
2482 a: store undef -> %X
2483 b: store %X -> undef
2489 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2490 undefined value can be assumed to not have any effect; we can assume that the
2491 value is overwritten with bits that happen to match what was already there.
2492 However, a store <em>to</em> an undefined location could clobber arbitrary
2493 memory, therefore, it has undefined behavior.</p>
2497 <!-- ======================================================================= -->
2499 <a name="trapvalues">Trap Values</a>
2504 <p>Trap values are similar to <a href="#undefvalues">undef values</a>, however
2505 instead of representing an unspecified bit pattern, they represent the
2506 fact that an instruction or constant expression which cannot evoke side
2507 effects has nevertheless detected a condition which results in undefined
2510 <p>There is currently no way of representing a trap value in the IR; they
2511 only exist when produced by operations such as
2512 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2514 <p>Trap value behavior is defined in terms of value <i>dependence</i>:</p>
2517 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2518 their operands.</li>
2520 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2521 to their dynamic predecessor basic block.</li>
2523 <li>Function arguments depend on the corresponding actual argument values in
2524 the dynamic callers of their functions.</li>
2526 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2527 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2528 control back to them.</li>
2530 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2531 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_unwind"><tt>unwind</tt></a>,
2532 or exception-throwing call instructions that dynamically transfer control
2535 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2536 referenced memory addresses, following the order in the IR
2537 (including loads and stores implied by intrinsics such as
2538 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2540 <!-- TODO: In the case of multiple threads, this only applies if the store
2541 "happens-before" the load or store. -->
2543 <!-- TODO: floating-point exception state -->
2545 <li>An instruction with externally visible side effects depends on the most
2546 recent preceding instruction with externally visible side effects, following
2547 the order in the IR. (This includes
2548 <a href="#volatile">volatile operations</a>.)</li>
2550 <li>An instruction <i>control-depends</i> on a
2551 <a href="#terminators">terminator instruction</a>
2552 if the terminator instruction has multiple successors and the instruction
2553 is always executed when control transfers to one of the successors, and
2554 may not be executed when control is transferred to another.</li>
2556 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2557 instruction if the set of instructions it otherwise depends on would be
2558 different if the terminator had transferred control to a different
2561 <li>Dependence is transitive.</li>
2565 <p>Whenever a trap value is generated, all values which depend on it evaluate
2566 to trap. If they have side effects, they evoke their side effects as if each
2567 operand with a trap value were undef. If they have externally-visible side
2568 effects, the behavior is undefined.</p>
2570 <p>Here are some examples:</p>
2572 <pre class="doc_code">
2574 %trap = sub nuw i32 0, 1 ; Results in a trap value.
2575 %still_trap = and i32 %trap, 0 ; Whereas (and i32 undef, 0) would return 0.
2576 %trap_yet_again = getelementptr i32* @h, i32 %still_trap
2577 store i32 0, i32* %trap_yet_again ; undefined behavior
2579 store i32 %trap, i32* @g ; Trap value conceptually stored to memory.
2580 %trap2 = load i32* @g ; Returns a trap value, not just undef.
2582 volatile store i32 %trap, i32* @g ; External observation; undefined behavior.
2584 %narrowaddr = bitcast i32* @g to i16*
2585 %wideaddr = bitcast i32* @g to i64*
2586 %trap3 = load i16* %narrowaddr ; Returns a trap value.
2587 %trap4 = load i64* %wideaddr ; Returns a trap value.
2589 %cmp = icmp slt i32 %trap, 0 ; Returns a trap value.
2590 br i1 %cmp, label %true, label %end ; Branch to either destination.
2593 volatile store i32 0, i32* @g ; This is control-dependent on %cmp, so
2594 ; it has undefined behavior.
2598 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2599 ; Both edges into this PHI are
2600 ; control-dependent on %cmp, so this
2601 ; always results in a trap value.
2603 volatile store i32 0, i32* @g ; This would depend on the store in %true
2604 ; if %cmp is true, or the store in %entry
2605 ; otherwise, so this is undefined behavior.
2607 br i1 %cmp, label %second_true, label %second_end
2608 ; The same branch again, but this time the
2609 ; true block doesn't have side effects.
2616 volatile store i32 0, i32* @g ; This time, the instruction always depends
2617 ; on the store in %end. Also, it is
2618 ; control-equivalent to %end, so this is
2619 ; well-defined (again, ignoring earlier
2620 ; undefined behavior in this example).
2625 <!-- ======================================================================= -->
2627 <a name="blockaddress">Addresses of Basic Blocks</a>
2632 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2634 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2635 basic block in the specified function, and always has an i8* type. Taking
2636 the address of the entry block is illegal.</p>
2638 <p>This value only has defined behavior when used as an operand to the
2639 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2640 comparisons against null. Pointer equality tests between labels addresses
2641 results in undefined behavior — though, again, comparison against null
2642 is ok, and no label is equal to the null pointer. This may be passed around
2643 as an opaque pointer sized value as long as the bits are not inspected. This
2644 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2645 long as the original value is reconstituted before the <tt>indirectbr</tt>
2648 <p>Finally, some targets may provide defined semantics when using the value as
2649 the operand to an inline assembly, but that is target specific.</p>
2654 <!-- ======================================================================= -->
2656 <a name="constantexprs">Constant Expressions</a>
2661 <p>Constant expressions are used to allow expressions involving other constants
2662 to be used as constants. Constant expressions may be of
2663 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2664 operation that does not have side effects (e.g. load and call are not
2665 supported). The following is the syntax for constant expressions:</p>
2668 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2669 <dd>Truncate a constant to another type. The bit size of CST must be larger
2670 than the bit size of TYPE. Both types must be integers.</dd>
2672 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2673 <dd>Zero extend a constant to another type. The bit size of CST must be
2674 smaller than the bit size of TYPE. Both types must be integers.</dd>
2676 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2677 <dd>Sign 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>fptrunc (CST to TYPE)</tt></b></dt>
2681 <dd>Truncate a floating point constant to another floating point type. The
2682 size of CST must be larger than the size of TYPE. Both types must be
2683 floating point.</dd>
2685 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2686 <dd>Floating point extend a constant to another type. The size of CST must be
2687 smaller or equal to the size of TYPE. Both types must be floating
2690 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2691 <dd>Convert a floating point constant to the corresponding unsigned integer
2692 constant. TYPE must be a scalar or vector integer type. CST must be of
2693 scalar or vector floating point type. Both CST and TYPE must be scalars,
2694 or vectors of the same number of elements. If the value won't fit in the
2695 integer type, the results are undefined.</dd>
2697 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2698 <dd>Convert a floating point constant to the corresponding signed integer
2699 constant. TYPE must be a scalar or vector integer type. CST must be of
2700 scalar or vector floating point type. Both CST and TYPE must be scalars,
2701 or vectors of the same number of elements. If the value won't fit in the
2702 integer type, the results are undefined.</dd>
2704 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2705 <dd>Convert an unsigned integer constant to the corresponding floating point
2706 constant. TYPE must be a scalar or vector floating point type. CST must be
2707 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2708 vectors of the same number of elements. If the value won't fit in the
2709 floating point type, the results are undefined.</dd>
2711 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2712 <dd>Convert a signed integer constant to the corresponding floating point
2713 constant. TYPE must be a scalar or vector floating point type. CST must be
2714 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2715 vectors of the same number of elements. If the value won't fit in the
2716 floating point type, the results are undefined.</dd>
2718 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2719 <dd>Convert a pointer typed constant to the corresponding integer constant
2720 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2721 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2722 make it fit in <tt>TYPE</tt>.</dd>
2724 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2725 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2726 type. CST must be of integer type. The CST value is zero extended,
2727 truncated, or unchanged to make it fit in a pointer size. This one is
2728 <i>really</i> dangerous!</dd>
2730 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2731 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2732 are the same as those for the <a href="#i_bitcast">bitcast
2733 instruction</a>.</dd>
2735 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2736 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2737 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2738 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2739 instruction, the index list may have zero or more indexes, which are
2740 required to make sense for the type of "CSTPTR".</dd>
2742 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2743 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2745 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2746 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2748 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2749 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2751 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2752 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2755 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2756 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2759 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2760 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2763 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2764 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2765 constants. The index list is interpreted in a similar manner as indices in
2766 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2767 index value must be specified.</dd>
2769 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2770 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2771 constants. The index list is interpreted in a similar manner as indices in
2772 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2773 index value must be specified.</dd>
2775 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2776 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2777 be any of the <a href="#binaryops">binary</a>
2778 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2779 on operands are the same as those for the corresponding instruction
2780 (e.g. no bitwise operations on floating point values are allowed).</dd>
2787 <!-- *********************************************************************** -->
2788 <h2><a name="othervalues">Other Values</a></h2>
2789 <!-- *********************************************************************** -->
2791 <!-- ======================================================================= -->
2793 <a name="inlineasm">Inline Assembler Expressions</a>
2798 <p>LLVM supports inline assembler expressions (as opposed
2799 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2800 a special value. This value represents the inline assembler as a string
2801 (containing the instructions to emit), a list of operand constraints (stored
2802 as a string), a flag that indicates whether or not the inline asm
2803 expression has side effects, and a flag indicating whether the function
2804 containing the asm needs to align its stack conservatively. An example
2805 inline assembler expression is:</p>
2807 <pre class="doc_code">
2808 i32 (i32) asm "bswap $0", "=r,r"
2811 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2812 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2815 <pre class="doc_code">
2816 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2819 <p>Inline asms with side effects not visible in the constraint list must be
2820 marked as having side effects. This is done through the use of the
2821 '<tt>sideeffect</tt>' keyword, like so:</p>
2823 <pre class="doc_code">
2824 call void asm sideeffect "eieio", ""()
2827 <p>In some cases inline asms will contain code that will not work unless the
2828 stack is aligned in some way, such as calls or SSE instructions on x86,
2829 yet will not contain code that does that alignment within the asm.
2830 The compiler should make conservative assumptions about what the asm might
2831 contain and should generate its usual stack alignment code in the prologue
2832 if the '<tt>alignstack</tt>' keyword is present:</p>
2834 <pre class="doc_code">
2835 call void asm alignstack "eieio", ""()
2838 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2841 <p>TODO: The format of the asm and constraints string still need to be
2842 documented here. Constraints on what can be done (e.g. duplication, moving,
2843 etc need to be documented). This is probably best done by reference to
2844 another document that covers inline asm from a holistic perspective.</p>
2847 <a name="inlineasm_md">Inline Asm Metadata</a>
2852 <p>The call instructions that wrap inline asm nodes may have a "!srcloc" MDNode
2853 attached to it that contains a list of constant integers. If present, the
2854 code generator will use the integer as the location cookie value when report
2855 errors through the LLVMContext error reporting mechanisms. This allows a
2856 front-end to correlate backend errors that occur with inline asm back to the
2857 source code that produced it. For example:</p>
2859 <pre class="doc_code">
2860 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2862 !42 = !{ i32 1234567 }
2865 <p>It is up to the front-end to make sense of the magic numbers it places in the
2866 IR. If the MDNode contains multiple constants, the code generator will use
2867 the one that corresponds to the line of the asm that the error occurs on.</p>
2873 <!-- ======================================================================= -->
2875 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2880 <p>LLVM IR allows metadata to be attached to instructions in the program that
2881 can convey extra information about the code to the optimizers and code
2882 generator. One example application of metadata is source-level debug
2883 information. There are two metadata primitives: strings and nodes. All
2884 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2885 preceding exclamation point ('<tt>!</tt>').</p>
2887 <p>A metadata string is a string surrounded by double quotes. It can contain
2888 any character by escaping non-printable characters with "\xx" where "xx" is
2889 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2891 <p>Metadata nodes are represented with notation similar to structure constants
2892 (a comma separated list of elements, surrounded by braces and preceded by an
2893 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2894 10}</tt>". Metadata nodes can have any values as their operand.</p>
2896 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2897 metadata nodes, which can be looked up in the module symbol table. For
2898 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2900 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2901 function is using two metadata arguments.</p>
2903 <div class="doc_code">
2905 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2909 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2910 attached with <tt>add</tt> instruction using <tt>!dbg</tt> identifier.</p>
2912 <div class="doc_code">
2914 %indvar.next = add i64 %indvar, 1, !dbg !21
2922 <!-- *********************************************************************** -->
2924 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2926 <!-- *********************************************************************** -->
2928 <p>LLVM has a number of "magic" global variables that contain data that affect
2929 code generation or other IR semantics. These are documented here. All globals
2930 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2931 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2934 <!-- ======================================================================= -->
2936 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2941 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2942 href="#linkage_appending">appending linkage</a>. This array contains a list of
2943 pointers to global variables and functions which may optionally have a pointer
2944 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2950 @llvm.used = appending global [2 x i8*] [
2952 i8* bitcast (i32* @Y to i8*)
2953 ], section "llvm.metadata"
2956 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2957 compiler, assembler, and linker are required to treat the symbol as if there is
2958 a reference to the global that it cannot see. For example, if a variable has
2959 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2960 list, it cannot be deleted. This is commonly used to represent references from
2961 inline asms and other things the compiler cannot "see", and corresponds to
2962 "attribute((used))" in GNU C.</p>
2964 <p>On some targets, the code generator must emit a directive to the assembler or
2965 object file to prevent the assembler and linker from molesting the symbol.</p>
2969 <!-- ======================================================================= -->
2971 <a name="intg_compiler_used">
2972 The '<tt>llvm.compiler.used</tt>' Global Variable
2978 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2979 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2980 touching the symbol. On targets that support it, this allows an intelligent
2981 linker to optimize references to the symbol without being impeded as it would be
2982 by <tt>@llvm.used</tt>.</p>
2984 <p>This is a rare construct that should only be used in rare circumstances, and
2985 should not be exposed to source languages.</p>
2989 <!-- ======================================================================= -->
2991 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2996 %0 = type { i32, void ()* }
2997 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2999 <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.
3004 <!-- ======================================================================= -->
3006 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3011 %0 = type { i32, void ()* }
3012 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3015 <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.
3022 <!-- *********************************************************************** -->
3023 <h2><a name="instref">Instruction Reference</a></h2>
3024 <!-- *********************************************************************** -->
3028 <p>The LLVM instruction set consists of several different classifications of
3029 instructions: <a href="#terminators">terminator
3030 instructions</a>, <a href="#binaryops">binary instructions</a>,
3031 <a href="#bitwiseops">bitwise binary instructions</a>,
3032 <a href="#memoryops">memory instructions</a>, and
3033 <a href="#otherops">other instructions</a>.</p>
3035 <!-- ======================================================================= -->
3037 <a name="terminators">Terminator Instructions</a>
3042 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3043 in a program ends with a "Terminator" instruction, which indicates which
3044 block should be executed after the current block is finished. These
3045 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3046 control flow, not values (the one exception being the
3047 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3049 <p>The terminator instructions are:
3050 '<a href="#i_ret"><tt>ret</tt></a>',
3051 '<a href="#i_br"><tt>br</tt></a>',
3052 '<a href="#i_switch"><tt>switch</tt></a>',
3053 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3054 '<a href="#i_invoke"><tt>invoke</tt></a>',
3055 '<a href="#i_unwind"><tt>unwind</tt></a>',
3056 '<a href="#i_resume"><tt>resume</tt></a>', and
3057 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3059 <!-- _______________________________________________________________________ -->
3061 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3068 ret <type> <value> <i>; Return a value from a non-void function</i>
3069 ret void <i>; Return from void function</i>
3073 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3074 a value) from a function back to the caller.</p>
3076 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3077 value and then causes control flow, and one that just causes control flow to
3081 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3082 return value. The type of the return value must be a
3083 '<a href="#t_firstclass">first class</a>' type.</p>
3085 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3086 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3087 value or a return value with a type that does not match its type, or if it
3088 has a void return type and contains a '<tt>ret</tt>' instruction with a
3092 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3093 the calling function's context. If the caller is a
3094 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3095 instruction after the call. If the caller was an
3096 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3097 the beginning of the "normal" destination block. If the instruction returns
3098 a value, that value shall set the call or invoke instruction's return
3103 ret i32 5 <i>; Return an integer value of 5</i>
3104 ret void <i>; Return from a void function</i>
3105 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3109 <!-- _______________________________________________________________________ -->
3111 <a name="i_br">'<tt>br</tt>' Instruction</a>
3118 br i1 <cond>, label <iftrue>, label <iffalse>
3119 br label <dest> <i>; Unconditional branch</i>
3123 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3124 different basic block in the current function. There are two forms of this
3125 instruction, corresponding to a conditional branch and an unconditional
3129 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3130 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3131 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3135 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3136 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3137 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3138 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3143 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3144 br i1 %cond, label %IfEqual, label %IfUnequal
3146 <a href="#i_ret">ret</a> i32 1
3148 <a href="#i_ret">ret</a> i32 0
3153 <!-- _______________________________________________________________________ -->
3155 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3162 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3166 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3167 several different places. It is a generalization of the '<tt>br</tt>'
3168 instruction, allowing a branch to occur to one of many possible
3172 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3173 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3174 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3175 The table is not allowed to contain duplicate constant entries.</p>
3178 <p>The <tt>switch</tt> instruction specifies a table of values and
3179 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3180 is searched for the given value. If the value is found, control flow is
3181 transferred to the corresponding destination; otherwise, control flow is
3182 transferred to the default destination.</p>
3184 <h5>Implementation:</h5>
3185 <p>Depending on properties of the target machine and the particular
3186 <tt>switch</tt> instruction, this instruction may be code generated in
3187 different ways. For example, it could be generated as a series of chained
3188 conditional branches or with a lookup table.</p>
3192 <i>; Emulate a conditional br instruction</i>
3193 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3194 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3196 <i>; Emulate an unconditional br instruction</i>
3197 switch i32 0, label %dest [ ]
3199 <i>; Implement a jump table:</i>
3200 switch i32 %val, label %otherwise [ i32 0, label %onzero
3202 i32 2, label %ontwo ]
3208 <!-- _______________________________________________________________________ -->
3210 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3217 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3222 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3223 within the current function, whose address is specified by
3224 "<tt>address</tt>". Address must be derived from a <a
3225 href="#blockaddress">blockaddress</a> constant.</p>
3229 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3230 rest of the arguments indicate the full set of possible destinations that the
3231 address may point to. Blocks are allowed to occur multiple times in the
3232 destination list, though this isn't particularly useful.</p>
3234 <p>This destination list is required so that dataflow analysis has an accurate
3235 understanding of the CFG.</p>
3239 <p>Control transfers to the block specified in the address argument. All
3240 possible destination blocks must be listed in the label list, otherwise this
3241 instruction has undefined behavior. This implies that jumps to labels
3242 defined in other functions have undefined behavior as well.</p>
3244 <h5>Implementation:</h5>
3246 <p>This is typically implemented with a jump through a register.</p>
3250 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3256 <!-- _______________________________________________________________________ -->
3258 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3265 <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>]
3266 to label <normal label> unwind label <exception label>
3270 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3271 function, with the possibility of control flow transfer to either the
3272 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3273 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3274 control flow will return to the "normal" label. If the callee (or any
3275 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
3276 instruction, control is interrupted and continued at the dynamically nearest
3277 "exception" label.</p>
3279 <p>The '<tt>exception</tt>' label is a
3280 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3281 exception. As such, '<tt>exception</tt>' label is required to have the
3282 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3283 the information about about the behavior of the program after unwinding
3284 happens, as its first non-PHI instruction. The restrictions on the
3285 "<tt>landingpad</tt>" instruction's tightly couples it to the
3286 "<tt>invoke</tt>" instruction, so that the important information contained
3287 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3291 <p>This instruction requires several arguments:</p>
3294 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3295 convention</a> the call should use. If none is specified, the call
3296 defaults to using C calling conventions.</li>
3298 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3299 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3300 '<tt>inreg</tt>' attributes are valid here.</li>
3302 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3303 function value being invoked. In most cases, this is a direct function
3304 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3305 off an arbitrary pointer to function value.</li>
3307 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3308 function to be invoked. </li>
3310 <li>'<tt>function args</tt>': argument list whose types match the function
3311 signature argument types and parameter attributes. All arguments must be
3312 of <a href="#t_firstclass">first class</a> type. If the function
3313 signature indicates the function accepts a variable number of arguments,
3314 the extra arguments can be specified.</li>
3316 <li>'<tt>normal label</tt>': the label reached when the called function
3317 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3319 <li>'<tt>exception label</tt>': the label reached when a callee returns with
3320 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
3322 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3323 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3324 '<tt>readnone</tt>' attributes are valid here.</li>
3328 <p>This instruction is designed to operate as a standard
3329 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3330 primary difference is that it establishes an association with a label, which
3331 is used by the runtime library to unwind the stack.</p>
3333 <p>This instruction is used in languages with destructors to ensure that proper
3334 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3335 exception. Additionally, this is important for implementation of
3336 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3338 <p>For the purposes of the SSA form, the definition of the value returned by the
3339 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3340 block to the "normal" label. If the callee unwinds then no return value is
3343 <p>Note that the code generator does not yet completely support unwind, and
3344 that the invoke/unwind semantics are likely to change in future versions.</p>
3348 %retval = invoke i32 @Test(i32 15) to label %Continue
3349 unwind label %TestCleanup <i>; {i32}:retval set</i>
3350 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3351 unwind label %TestCleanup <i>; {i32}:retval set</i>
3356 <!-- _______________________________________________________________________ -->
3359 <a name="i_unwind">'<tt>unwind</tt>' Instruction</a>
3370 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
3371 at the first callee in the dynamic call stack which used
3372 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
3373 This is primarily used to implement exception handling.</p>
3376 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
3377 immediately halt. The dynamic call stack is then searched for the
3378 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
3379 Once found, execution continues at the "exceptional" destination block
3380 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
3381 instruction in the dynamic call chain, undefined behavior results.</p>
3383 <p>Note that the code generator does not yet completely support unwind, and
3384 that the invoke/unwind semantics are likely to change in future versions.</p>
3388 <!-- _______________________________________________________________________ -->
3391 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3398 resume <type> <value>
3402 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3406 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3407 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3411 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3412 (in-flight) exception whose unwinding was interrupted with
3413 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3417 resume { i8*, i32 } %exn
3422 <!-- _______________________________________________________________________ -->
3425 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3436 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3437 instruction is used to inform the optimizer that a particular portion of the
3438 code is not reachable. This can be used to indicate that the code after a
3439 no-return function cannot be reached, and other facts.</p>
3442 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3448 <!-- ======================================================================= -->
3450 <a name="binaryops">Binary Operations</a>
3455 <p>Binary operators are used to do most of the computation in a program. They
3456 require two operands of the same type, execute an operation on them, and
3457 produce a single value. The operands might represent multiple data, as is
3458 the case with the <a href="#t_vector">vector</a> data type. The result value
3459 has the same type as its operands.</p>
3461 <p>There are several different binary operators:</p>
3463 <!-- _______________________________________________________________________ -->
3465 <a name="i_add">'<tt>add</tt>' Instruction</a>
3472 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3473 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3474 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3475 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3479 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3482 <p>The two arguments to the '<tt>add</tt>' instruction must
3483 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3484 integer values. Both arguments must have identical types.</p>
3487 <p>The value produced is the integer sum of the two operands.</p>
3489 <p>If the sum has unsigned overflow, the result returned is the mathematical
3490 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3492 <p>Because LLVM integers use a two's complement representation, this instruction
3493 is appropriate for both signed and unsigned integers.</p>
3495 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3496 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3497 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3498 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3499 respectively, occurs.</p>
3503 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3508 <!-- _______________________________________________________________________ -->
3510 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3517 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3521 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3524 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3525 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3526 floating point values. Both arguments must have identical types.</p>
3529 <p>The value produced is the floating point sum of the two operands.</p>
3533 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3538 <!-- _______________________________________________________________________ -->
3540 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3547 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3548 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3549 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3550 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3554 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3557 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3558 '<tt>neg</tt>' instruction present in most other intermediate
3559 representations.</p>
3562 <p>The two arguments to the '<tt>sub</tt>' instruction must
3563 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3564 integer values. Both arguments must have identical types.</p>
3567 <p>The value produced is the integer difference of the two operands.</p>
3569 <p>If the difference has unsigned overflow, the result returned is the
3570 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3573 <p>Because LLVM integers use a two's complement representation, this instruction
3574 is appropriate for both signed and unsigned integers.</p>
3576 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3577 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3578 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3579 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3580 respectively, occurs.</p>
3584 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3585 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3590 <!-- _______________________________________________________________________ -->
3592 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3599 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3603 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3606 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3607 '<tt>fneg</tt>' instruction present in most other intermediate
3608 representations.</p>
3611 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3612 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3613 floating point values. Both arguments must have identical types.</p>
3616 <p>The value produced is the floating point difference of the two operands.</p>
3620 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3621 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3626 <!-- _______________________________________________________________________ -->
3628 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3635 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3636 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3637 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3638 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3642 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3645 <p>The two arguments to the '<tt>mul</tt>' instruction must
3646 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3647 integer values. Both arguments must have identical types.</p>
3650 <p>The value produced is the integer product of the two operands.</p>
3652 <p>If the result of the multiplication has unsigned overflow, the result
3653 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3654 width of the result.</p>
3656 <p>Because LLVM integers use a two's complement representation, and the result
3657 is the same width as the operands, this instruction returns the correct
3658 result for both signed and unsigned integers. If a full product
3659 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3660 be sign-extended or zero-extended as appropriate to the width of the full
3663 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3664 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3665 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3666 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3667 respectively, occurs.</p>
3671 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3676 <!-- _______________________________________________________________________ -->
3678 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3685 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3689 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3692 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3693 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3694 floating point values. Both arguments must have identical types.</p>
3697 <p>The value produced is the floating point product of the two operands.</p>
3701 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3706 <!-- _______________________________________________________________________ -->
3708 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
3715 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3716 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3720 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3723 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3724 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3725 values. Both arguments must have identical types.</p>
3728 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3730 <p>Note that unsigned integer division and signed integer division are distinct
3731 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3733 <p>Division by zero leads to undefined behavior.</p>
3735 <p>If the <tt>exact</tt> keyword is present, the result value of the
3736 <tt>udiv</tt> is a <a href="#trapvalues">trap value</a> if %op1 is not a
3737 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
3742 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3747 <!-- _______________________________________________________________________ -->
3749 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
3756 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3757 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3761 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3764 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3765 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3766 values. Both arguments must have identical types.</p>
3769 <p>The value produced is the signed integer quotient of the two operands rounded
3772 <p>Note that signed integer division and unsigned integer division are distinct
3773 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3775 <p>Division by zero leads to undefined behavior. Overflow also leads to
3776 undefined behavior; this is a rare case, but can occur, for example, by doing
3777 a 32-bit division of -2147483648 by -1.</p>
3779 <p>If the <tt>exact</tt> keyword is present, the result value of the
3780 <tt>sdiv</tt> is a <a href="#trapvalues">trap value</a> if the result would
3785 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3790 <!-- _______________________________________________________________________ -->
3792 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
3799 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3803 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3806 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3807 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3808 floating point values. Both arguments must have identical types.</p>
3811 <p>The value produced is the floating point quotient of the two operands.</p>
3815 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3820 <!-- _______________________________________________________________________ -->
3822 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3829 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3833 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3834 division of its two arguments.</p>
3837 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3838 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3839 values. Both arguments must have identical types.</p>
3842 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3843 This instruction always performs an unsigned division to get the
3846 <p>Note that unsigned integer remainder and signed integer remainder are
3847 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3849 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3853 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3858 <!-- _______________________________________________________________________ -->
3860 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3867 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3871 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3872 division of its two operands. This instruction can also take
3873 <a href="#t_vector">vector</a> versions of the values in which case the
3874 elements must be integers.</p>
3877 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3878 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3879 values. Both arguments must have identical types.</p>
3882 <p>This instruction returns the <i>remainder</i> of a division (where the result
3883 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
3884 <i>modulo</i> operator (where the result is either zero or has the same sign
3885 as the divisor, <tt>op2</tt>) of a value.
3886 For more information about the difference,
3887 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3888 Math Forum</a>. For a table of how this is implemented in various languages,
3889 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3890 Wikipedia: modulo operation</a>.</p>
3892 <p>Note that signed integer remainder and unsigned integer remainder are
3893 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3895 <p>Taking the remainder of a division by zero leads to undefined behavior.
3896 Overflow also leads to undefined behavior; this is a rare case, but can
3897 occur, for example, by taking the remainder of a 32-bit division of
3898 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3899 lets srem be implemented using instructions that return both the result of
3900 the division and the remainder.)</p>
3904 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3909 <!-- _______________________________________________________________________ -->
3911 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
3918 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3922 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3923 its two operands.</p>
3926 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3927 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3928 floating point values. Both arguments must have identical types.</p>
3931 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3932 has the same sign as the dividend.</p>
3936 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3943 <!-- ======================================================================= -->
3945 <a name="bitwiseops">Bitwise Binary Operations</a>
3950 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3951 program. They are generally very efficient instructions and can commonly be
3952 strength reduced from other instructions. They require two operands of the
3953 same type, execute an operation on them, and produce a single value. The
3954 resulting value is the same type as its operands.</p>
3956 <!-- _______________________________________________________________________ -->
3958 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
3965 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3966 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3967 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3968 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3972 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3973 a specified number of bits.</p>
3976 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3977 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3978 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3981 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3982 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3983 is (statically or dynamically) negative or equal to or larger than the number
3984 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3985 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3986 shift amount in <tt>op2</tt>.</p>
3988 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
3989 <a href="#trapvalues">trap value</a> if it shifts out any non-zero bits. If
3990 the <tt>nsw</tt> keyword is present, then the shift produces a
3991 <a href="#trapvalues">trap value</a> if it shifts out any bits that disagree
3992 with the resultant sign bit. As such, NUW/NSW have the same semantics as
3993 they would if the shift were expressed as a mul instruction with the same
3994 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
3998 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3999 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
4000 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
4001 <result> = shl i32 1, 32 <i>; undefined</i>
4002 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4007 <!-- _______________________________________________________________________ -->
4009 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4016 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4017 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4021 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4022 operand shifted to the right a specified number of bits with zero fill.</p>
4025 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4026 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4027 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4030 <p>This instruction always performs a logical shift right operation. The most
4031 significant bits of the result will be filled with zero bits after the shift.
4032 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4033 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4034 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4035 shift amount in <tt>op2</tt>.</p>
4037 <p>If the <tt>exact</tt> keyword is present, the result value of the
4038 <tt>lshr</tt> is a <a href="#trapvalues">trap value</a> if any of the bits
4039 shifted out are non-zero.</p>
4044 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4045 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4046 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4047 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4048 <result> = lshr i32 1, 32 <i>; undefined</i>
4049 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4054 <!-- _______________________________________________________________________ -->
4056 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4063 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4064 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4068 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4069 operand shifted to the right a specified number of bits with sign
4073 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4074 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4075 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4078 <p>This instruction always performs an arithmetic shift right operation, The
4079 most significant bits of the result will be filled with the sign bit
4080 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4081 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4082 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4083 the corresponding shift amount in <tt>op2</tt>.</p>
4085 <p>If the <tt>exact</tt> keyword is present, the result value of the
4086 <tt>ashr</tt> is a <a href="#trapvalues">trap value</a> if any of the bits
4087 shifted out are non-zero.</p>
4091 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4092 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4093 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4094 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4095 <result> = ashr i32 1, 32 <i>; undefined</i>
4096 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4101 <!-- _______________________________________________________________________ -->
4103 <a name="i_and">'<tt>and</tt>' Instruction</a>
4110 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4114 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4118 <p>The two arguments to the '<tt>and</tt>' instruction must be
4119 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4120 values. Both arguments must have identical types.</p>
4123 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4125 <table border="1" cellspacing="0" cellpadding="4">
4157 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4158 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4159 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4162 <!-- _______________________________________________________________________ -->
4164 <a name="i_or">'<tt>or</tt>' Instruction</a>
4171 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4175 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4179 <p>The two arguments to the '<tt>or</tt>' instruction must be
4180 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4181 values. Both arguments must have identical types.</p>
4184 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4186 <table border="1" cellspacing="0" cellpadding="4">
4218 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4219 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4220 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4225 <!-- _______________________________________________________________________ -->
4227 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4234 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4238 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4239 its two operands. The <tt>xor</tt> is used to implement the "one's
4240 complement" operation, which is the "~" operator in C.</p>
4243 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4244 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4245 values. Both arguments must have identical types.</p>
4248 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4250 <table border="1" cellspacing="0" cellpadding="4">
4282 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4283 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4284 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4285 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4292 <!-- ======================================================================= -->
4294 <a name="vectorops">Vector Operations</a>
4299 <p>LLVM supports several instructions to represent vector operations in a
4300 target-independent manner. These instructions cover the element-access and
4301 vector-specific operations needed to process vectors effectively. While LLVM
4302 does directly support these vector operations, many sophisticated algorithms
4303 will want to use target-specific intrinsics to take full advantage of a
4304 specific target.</p>
4306 <!-- _______________________________________________________________________ -->
4308 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4315 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4319 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4320 from a vector at a specified index.</p>
4324 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4325 of <a href="#t_vector">vector</a> type. The second operand is an index
4326 indicating the position from which to extract the element. The index may be
4330 <p>The result is a scalar of the same type as the element type of
4331 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4332 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4333 results are undefined.</p>
4337 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4342 <!-- _______________________________________________________________________ -->
4344 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4351 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4355 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4356 vector at a specified index.</p>
4359 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4360 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4361 whose type must equal the element type of the first operand. The third
4362 operand is an index indicating the position at which to insert the value.
4363 The index may be a variable.</p>
4366 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4367 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4368 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4369 results are undefined.</p>
4373 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4378 <!-- _______________________________________________________________________ -->
4380 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4387 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4391 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4392 from two input vectors, returning a vector with the same element type as the
4393 input and length that is the same as the shuffle mask.</p>
4396 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4397 with types that match each other. The third argument is a shuffle mask whose
4398 element type is always 'i32'. The result of the instruction is a vector
4399 whose length is the same as the shuffle mask and whose element type is the
4400 same as the element type of the first two operands.</p>
4402 <p>The shuffle mask operand is required to be a constant vector with either
4403 constant integer or undef values.</p>
4406 <p>The elements of the two input vectors are numbered from left to right across
4407 both of the vectors. The shuffle mask operand specifies, for each element of
4408 the result vector, which element of the two input vectors the result element
4409 gets. The element selector may be undef (meaning "don't care") and the
4410 second operand may be undef if performing a shuffle from only one vector.</p>
4414 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4415 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4416 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4417 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4418 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4419 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4420 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4421 <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>
4428 <!-- ======================================================================= -->
4430 <a name="aggregateops">Aggregate Operations</a>
4435 <p>LLVM supports several instructions for working with
4436 <a href="#t_aggregate">aggregate</a> values.</p>
4438 <!-- _______________________________________________________________________ -->
4440 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4447 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4451 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4452 from an <a href="#t_aggregate">aggregate</a> value.</p>
4455 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4456 of <a href="#t_struct">struct</a> or
4457 <a href="#t_array">array</a> type. The operands are constant indices to
4458 specify which value to extract in a similar manner as indices in a
4459 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4460 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4462 <li>Since the value being indexed is not a pointer, the first index is
4463 omitted and assumed to be zero.</li>
4464 <li>At least one index must be specified.</li>
4465 <li>Not only struct indices but also array indices must be in
4470 <p>The result is the value at the position in the aggregate specified by the
4475 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4480 <!-- _______________________________________________________________________ -->
4482 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4489 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4493 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4494 in an <a href="#t_aggregate">aggregate</a> value.</p>
4497 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4498 of <a href="#t_struct">struct</a> or
4499 <a href="#t_array">array</a> type. The second operand is a first-class
4500 value to insert. The following operands are constant indices indicating
4501 the position at which to insert the value in a similar manner as indices in a
4502 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4503 value to insert must have the same type as the value identified by the
4507 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4508 that of <tt>val</tt> except that the value at the position specified by the
4509 indices is that of <tt>elt</tt>.</p>
4513 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4514 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4515 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4522 <!-- ======================================================================= -->
4524 <a name="memoryops">Memory Access and Addressing Operations</a>
4529 <p>A key design point of an SSA-based representation is how it represents
4530 memory. In LLVM, no memory locations are in SSA form, which makes things
4531 very simple. This section describes how to read, write, and allocate
4534 <!-- _______________________________________________________________________ -->
4536 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4543 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4547 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4548 currently executing function, to be automatically released when this function
4549 returns to its caller. The object is always allocated in the generic address
4550 space (address space zero).</p>
4553 <p>The '<tt>alloca</tt>' instruction
4554 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4555 runtime stack, returning a pointer of the appropriate type to the program.
4556 If "NumElements" is specified, it is the number of elements allocated,
4557 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4558 specified, the value result of the allocation is guaranteed to be aligned to
4559 at least that boundary. If not specified, or if zero, the target can choose
4560 to align the allocation on any convenient boundary compatible with the
4563 <p>'<tt>type</tt>' may be any sized type.</p>
4566 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4567 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4568 memory is automatically released when the function returns. The
4569 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4570 variables that must have an address available. When the function returns
4571 (either with the <tt><a href="#i_ret">ret</a></tt>
4572 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
4573 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4577 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4578 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4579 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4580 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4585 <!-- _______________________________________________________________________ -->
4587 <a name="i_load">'<tt>load</tt>' Instruction</a>
4594 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4595 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4596 !<index> = !{ i32 1 }
4600 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4603 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4604 from which to load. The pointer must point to
4605 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4606 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4607 number or order of execution of this <tt>load</tt> with other <a
4608 href="#volatile">volatile operations</a>.</p>
4610 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
4611 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4612 argument. The <code>release</code> and <code>acq_rel</code> orderings are
4613 not valid on <code>load</code> instructions. Atomic loads produce <a
4614 href="#memorymodel">defined</a> results when they may see multiple atomic
4615 stores. The type of the pointee must be an integer type whose bit width
4616 is a power of two greater than or equal to eight and less than or equal
4617 to a target-specific size limit. <code>align</code> must be explicitly
4618 specified on atomic loads, and the load has undefined behavior if the
4619 alignment is not set to a value which is at least the size in bytes of
4620 the pointee. <code>!nontemporal</code> does not have any defined semantics
4621 for atomic loads.</p>
4623 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4624 operation (that is, the alignment of the memory address). A value of 0 or an
4625 omitted <tt>align</tt> argument means that the operation has the preferential
4626 alignment for the target. It is the responsibility of the code emitter to
4627 ensure that the alignment information is correct. Overestimating the
4628 alignment results in undefined behavior. Underestimating the alignment may
4629 produce less efficient code. An alignment of 1 is always safe.</p>
4631 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4632 metatadata name <index> corresponding to a metadata node with
4633 one <tt>i32</tt> entry of value 1. The existence of
4634 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4635 and code generator that this load is not expected to be reused in the cache.
4636 The code generator may select special instructions to save cache bandwidth,
4637 such as the <tt>MOVNT</tt> instruction on x86.</p>
4640 <p>The location of memory pointed to is loaded. If the value being loaded is of
4641 scalar type then the number of bytes read does not exceed the minimum number
4642 of bytes needed to hold all bits of the type. For example, loading an
4643 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4644 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4645 is undefined if the value was not originally written using a store of the
4650 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4651 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4652 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4657 <!-- _______________________________________________________________________ -->
4659 <a name="i_store">'<tt>store</tt>' Instruction</a>
4666 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4667 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
4671 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4674 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4675 and an address at which to store it. The type of the
4676 '<tt><pointer></tt>' operand must be a pointer to
4677 the <a href="#t_firstclass">first class</a> type of the
4678 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
4679 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
4680 order of execution of this <tt>store</tt> with other <a
4681 href="#volatile">volatile operations</a>.</p>
4683 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
4684 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4685 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
4686 valid on <code>store</code> instructions. Atomic loads produce <a
4687 href="#memorymodel">defined</a> results when they may see multiple atomic
4688 stores. The type of the pointee must be an integer type whose bit width
4689 is a power of two greater than or equal to eight and less than or equal
4690 to a target-specific size limit. <code>align</code> must be explicitly
4691 specified on atomic stores, and the store has undefined behavior if the
4692 alignment is not set to a value which is at least the size in bytes of
4693 the pointee. <code>!nontemporal</code> does not have any defined semantics
4694 for atomic stores.</p>
4696 <p>The optional constant "align" argument specifies the alignment of the
4697 operation (that is, the alignment of the memory address). A value of 0 or an
4698 omitted "align" argument means that the operation has the preferential
4699 alignment for the target. It is the responsibility of the code emitter to
4700 ensure that the alignment information is correct. Overestimating the
4701 alignment results in an undefined behavior. Underestimating the alignment may
4702 produce less efficient code. An alignment of 1 is always safe.</p>
4704 <p>The optional !nontemporal metadata must reference a single metatadata
4705 name <index> corresponding to a metadata node with one i32 entry of
4706 value 1. The existence of the !nontemporal metatadata on the
4707 instruction tells the optimizer and code generator that this load is
4708 not expected to be reused in the cache. The code generator may
4709 select special instructions to save cache bandwidth, such as the
4710 MOVNT instruction on x86.</p>
4714 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4715 location specified by the '<tt><pointer></tt>' operand. If
4716 '<tt><value></tt>' is of scalar type then the number of bytes written
4717 does not exceed the minimum number of bytes needed to hold all bits of the
4718 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4719 writing a value of a type like <tt>i20</tt> with a size that is not an
4720 integral number of bytes, it is unspecified what happens to the extra bits
4721 that do not belong to the type, but they will typically be overwritten.</p>
4725 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4726 store i32 3, i32* %ptr <i>; yields {void}</i>
4727 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4732 <!-- _______________________________________________________________________ -->
4734 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
4741 fence [singlethread] <ordering> <i>; yields {void}</i>
4745 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
4746 between operations.</p>
4748 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
4749 href="#ordering">ordering</a> argument which defines what
4750 <i>synchronizes-with</i> edges they add. They can only be given
4751 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
4752 <code>seq_cst</code> orderings.</p>
4755 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
4756 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
4757 <code>acquire</code> ordering semantics if and only if there exist atomic
4758 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
4759 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
4760 <var>X</var> modifies <var>M</var> (either directly or through some side effect
4761 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
4762 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
4763 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
4764 than an explicit <code>fence</code>, one (but not both) of the atomic operations
4765 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
4766 <code>acquire</code> (resp.) ordering constraint and still
4767 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
4768 <i>happens-before</i> edge.</p>
4770 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
4771 having both <code>acquire</code> and <code>release</code> semantics specified
4772 above, participates in the global program order of other <code>seq_cst</code>
4773 operations and/or fences.</p>
4775 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
4776 specifies that the fence only synchronizes with other fences in the same
4777 thread. (This is useful for interacting with signal handlers.)</p>
4781 fence acquire <i>; yields {void}</i>
4782 fence singlethread seq_cst <i>; yields {void}</i>
4787 <!-- _______________________________________________________________________ -->
4789 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
4796 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
4800 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
4801 It loads a value in memory and compares it to a given value. If they are
4802 equal, it stores a new value into the memory.</p>
4805 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
4806 address to operate on, a value to compare to the value currently be at that
4807 address, and a new value to place at that address if the compared values are
4808 equal. The type of '<var><cmp></var>' must be an integer type whose
4809 bit width is a power of two greater than or equal to eight and less than
4810 or equal to a target-specific size limit. '<var><cmp></var>' and
4811 '<var><new></var>' must have the same type, and the type of
4812 '<var><pointer></var>' must be a pointer to that type. If the
4813 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
4814 optimizer is not allowed to modify the number or order of execution
4815 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
4818 <!-- FIXME: Extend allowed types. -->
4820 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
4821 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
4823 <p>The optional "<code>singlethread</code>" argument declares that the
4824 <code>cmpxchg</code> is only atomic with respect to code (usually signal
4825 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
4826 cmpxchg is atomic with respect to all other code in the system.</p>
4828 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
4829 the size in memory of the operand.
4832 <p>The contents of memory at the location specified by the
4833 '<tt><pointer></tt>' operand is read and compared to
4834 '<tt><cmp></tt>'; if the read value is the equal,
4835 '<tt><new></tt>' is written. The original value at the location
4838 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
4839 purpose of identifying <a href="#release_sequence">release sequences</a>. A
4840 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
4841 parameter determined by dropping any <code>release</code> part of the
4842 <code>cmpxchg</code>'s ordering.</p>
4845 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
4846 optimization work on ARM.)
4848 FIXME: Is a weaker ordering constraint on failure helpful in practice?
4854 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
4855 <a href="#i_br">br</a> label %loop
4858 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
4859 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
4860 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
4861 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
4862 <a href="#i_br">br</a> i1 %success, label %done, label %loop
4870 <!-- _______________________________________________________________________ -->
4872 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
4879 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
4883 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
4886 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
4887 operation to apply, an address whose value to modify, an argument to the
4888 operation. The operation must be one of the following keywords:</p>
4903 <p>The type of '<var><value></var>' must be an integer type whose
4904 bit width is a power of two greater than or equal to eight and less than
4905 or equal to a target-specific size limit. The type of the
4906 '<code><pointer></code>' operand must be a pointer to that type.
4907 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
4908 optimizer is not allowed to modify the number or order of execution of this
4909 <code>atomicrmw</code> with other <a href="#volatile">volatile
4912 <!-- FIXME: Extend allowed types. -->
4915 <p>The contents of memory at the location specified by the
4916 '<tt><pointer></tt>' operand are atomically read, modified, and written
4917 back. The original value at the location is returned. The modification is
4918 specified by the <var>operation</var> argument:</p>
4921 <li>xchg: <code>*ptr = val</code></li>
4922 <li>add: <code>*ptr = *ptr + val</code></li>
4923 <li>sub: <code>*ptr = *ptr - val</code></li>
4924 <li>and: <code>*ptr = *ptr & val</code></li>
4925 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
4926 <li>or: <code>*ptr = *ptr | val</code></li>
4927 <li>xor: <code>*ptr = *ptr ^ val</code></li>
4928 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
4929 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
4930 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
4931 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
4936 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
4941 <!-- _______________________________________________________________________ -->
4943 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
4950 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4951 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4955 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
4956 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
4957 It performs address calculation only and does not access memory.</p>
4960 <p>The first argument is always a pointer, and forms the basis of the
4961 calculation. The remaining arguments are indices that indicate which of the
4962 elements of the aggregate object are indexed. The interpretation of each
4963 index is dependent on the type being indexed into. The first index always
4964 indexes the pointer value given as the first argument, the second index
4965 indexes a value of the type pointed to (not necessarily the value directly
4966 pointed to, since the first index can be non-zero), etc. The first type
4967 indexed into must be a pointer value, subsequent types can be arrays,
4968 vectors, and structs. Note that subsequent types being indexed into
4969 can never be pointers, since that would require loading the pointer before
4970 continuing calculation.</p>
4972 <p>The type of each index argument depends on the type it is indexing into.
4973 When indexing into a (optionally packed) structure, only <tt>i32</tt>
4974 integer <b>constants</b> are allowed. When indexing into an array, pointer
4975 or vector, integers of any width are allowed, and they are not required to be
4976 constant. These integers are treated as signed values where relevant.</p>
4978 <p>For example, let's consider a C code fragment and how it gets compiled to
4981 <pre class="doc_code">
4993 int *foo(struct ST *s) {
4994 return &s[1].Z.B[5][13];
4998 <p>The LLVM code generated by the GCC frontend is:</p>
5000 <pre class="doc_code">
5001 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
5002 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
5004 define i32* @foo(%ST* %s) {
5006 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
5012 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
5013 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
5014 }</tt>' type, a structure. The second index indexes into the third element
5015 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
5016 i8 }</tt>' type, another structure. The third index indexes into the second
5017 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
5018 array. The two dimensions of the array are subscripted into, yielding an
5019 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
5020 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
5022 <p>Note that it is perfectly legal to index partially through a structure,
5023 returning a pointer to an inner element. Because of this, the LLVM code for
5024 the given testcase is equivalent to:</p>
5027 define i32* @foo(%ST* %s) {
5028 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
5029 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
5030 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5031 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5032 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5037 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5038 <tt>getelementptr</tt> is a <a href="#trapvalues">trap value</a> if the
5039 base pointer is not an <i>in bounds</i> address of an allocated object,
5040 or if any of the addresses that would be formed by successive addition of
5041 the offsets implied by the indices to the base address with infinitely
5042 precise signed arithmetic are not an <i>in bounds</i> address of that
5043 allocated object. The <i>in bounds</i> addresses for an allocated object
5044 are all the addresses that point into the object, plus the address one
5045 byte past the end.</p>
5047 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5048 the base address with silently-wrapping two's complement arithmetic. If the
5049 offsets have a different width from the pointer, they are sign-extended or
5050 truncated to the width of the pointer. The result value of the
5051 <tt>getelementptr</tt> may be outside the object pointed to by the base
5052 pointer. The result value may not necessarily be used to access memory
5053 though, even if it happens to point into allocated storage. See the
5054 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5057 <p>The getelementptr instruction is often confusing. For some more insight into
5058 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5062 <i>; yields [12 x i8]*:aptr</i>
5063 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5064 <i>; yields i8*:vptr</i>
5065 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5066 <i>; yields i8*:eptr</i>
5067 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5068 <i>; yields i32*:iptr</i>
5069 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5076 <!-- ======================================================================= -->
5078 <a name="convertops">Conversion Operations</a>
5083 <p>The instructions in this category are the conversion instructions (casting)
5084 which all take a single operand and a type. They perform various bit
5085 conversions on the operand.</p>
5087 <!-- _______________________________________________________________________ -->
5089 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5096 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5100 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5101 type <tt>ty2</tt>.</p>
5104 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5105 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5106 of the same number of integers.
5107 The bit size of the <tt>value</tt> must be larger than
5108 the bit size of the destination type, <tt>ty2</tt>.
5109 Equal sized types are not allowed.</p>
5112 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5113 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5114 source size must be larger than the destination size, <tt>trunc</tt> cannot
5115 be a <i>no-op cast</i>. It will always truncate bits.</p>
5119 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5120 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5121 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5122 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5127 <!-- _______________________________________________________________________ -->
5129 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5136 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5140 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5145 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5146 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5147 of the same number of integers.
5148 The bit size of the <tt>value</tt> must be smaller than
5149 the bit size of the destination type,
5153 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5154 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5156 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5160 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5161 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5162 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5167 <!-- _______________________________________________________________________ -->
5169 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5176 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5180 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5183 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5184 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5185 of the same number of integers.
5186 The bit size of the <tt>value</tt> must be smaller than
5187 the bit size of the destination type,
5191 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5192 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5193 of the type <tt>ty2</tt>.</p>
5195 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5199 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5200 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5201 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5206 <!-- _______________________________________________________________________ -->
5208 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5215 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5219 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5223 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5224 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5225 to cast it to. The size of <tt>value</tt> must be larger than the size of
5226 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5227 <i>no-op cast</i>.</p>
5230 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5231 <a href="#t_floating">floating point</a> type to a smaller
5232 <a href="#t_floating">floating point</a> type. If the value cannot fit
5233 within the destination type, <tt>ty2</tt>, then the results are
5238 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5239 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5244 <!-- _______________________________________________________________________ -->
5246 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5253 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5257 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5258 floating point value.</p>
5261 <p>The '<tt>fpext</tt>' instruction takes a
5262 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5263 a <a href="#t_floating">floating point</a> type to cast it to. The source
5264 type must be smaller than the destination type.</p>
5267 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5268 <a href="#t_floating">floating point</a> type to a larger
5269 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5270 used to make a <i>no-op cast</i> because it always changes bits. Use
5271 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5275 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5276 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5281 <!-- _______________________________________________________________________ -->
5283 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5290 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5294 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5295 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5298 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5299 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5300 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5301 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5302 vector integer type with the same number of elements as <tt>ty</tt></p>
5305 <p>The '<tt>fptoui</tt>' instruction converts its
5306 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5307 towards zero) unsigned integer value. If the value cannot fit
5308 in <tt>ty2</tt>, the results are undefined.</p>
5312 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5313 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5314 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5319 <!-- _______________________________________________________________________ -->
5321 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5328 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5332 <p>The '<tt>fptosi</tt>' instruction converts
5333 <a href="#t_floating">floating point</a> <tt>value</tt> to
5334 type <tt>ty2</tt>.</p>
5337 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5338 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5339 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5340 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5341 vector integer type with the same number of elements as <tt>ty</tt></p>
5344 <p>The '<tt>fptosi</tt>' instruction converts its
5345 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5346 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5347 the results are undefined.</p>
5351 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5352 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5353 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5358 <!-- _______________________________________________________________________ -->
5360 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5367 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5371 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5372 integer and converts that value to the <tt>ty2</tt> type.</p>
5375 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5376 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5377 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5378 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5379 floating point type with the same number of elements as <tt>ty</tt></p>
5382 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5383 integer quantity and converts it to the corresponding floating point
5384 value. If the value cannot fit in the floating point value, the results are
5389 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5390 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5395 <!-- _______________________________________________________________________ -->
5397 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5404 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5408 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5409 and converts that value to the <tt>ty2</tt> type.</p>
5412 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5413 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5414 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5415 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5416 floating point type with the same number of elements as <tt>ty</tt></p>
5419 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5420 quantity and converts it to the corresponding floating point value. If the
5421 value cannot fit in the floating point value, the results are undefined.</p>
5425 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5426 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5431 <!-- _______________________________________________________________________ -->
5433 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5440 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5444 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
5445 the integer type <tt>ty2</tt>.</p>
5448 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5449 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
5450 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
5453 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5454 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5455 truncating or zero extending that value to the size of the integer type. If
5456 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5457 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5458 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5463 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
5464 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
5469 <!-- _______________________________________________________________________ -->
5471 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5478 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5482 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5483 pointer type, <tt>ty2</tt>.</p>
5486 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5487 value to cast, and a type to cast it to, which must be a
5488 <a href="#t_pointer">pointer</a> type.</p>
5491 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5492 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5493 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5494 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5495 than the size of a pointer then a zero extension is done. If they are the
5496 same size, nothing is done (<i>no-op cast</i>).</p>
5500 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5501 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5502 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5507 <!-- _______________________________________________________________________ -->
5509 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5516 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5520 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5521 <tt>ty2</tt> without changing any bits.</p>
5524 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5525 non-aggregate first class value, and a type to cast it to, which must also be
5526 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5527 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5528 identical. If the source type is a pointer, the destination type must also be
5529 a pointer. This instruction supports bitwise conversion of vectors to
5530 integers and to vectors of other types (as long as they have the same
5534 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5535 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5536 this conversion. The conversion is done as if the <tt>value</tt> had been
5537 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
5538 be converted to other pointer types with this instruction. To convert
5539 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5540 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5544 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
5545 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
5546 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
5553 <!-- ======================================================================= -->
5555 <a name="otherops">Other Operations</a>
5560 <p>The instructions in this category are the "miscellaneous" instructions, which
5561 defy better classification.</p>
5563 <!-- _______________________________________________________________________ -->
5565 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5572 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5576 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5577 boolean values based on comparison of its two integer, integer vector, or
5578 pointer operands.</p>
5581 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
5582 the condition code indicating the kind of comparison to perform. It is not a
5583 value, just a keyword. The possible condition code are:</p>
5586 <li><tt>eq</tt>: equal</li>
5587 <li><tt>ne</tt>: not equal </li>
5588 <li><tt>ugt</tt>: unsigned greater than</li>
5589 <li><tt>uge</tt>: unsigned greater or equal</li>
5590 <li><tt>ult</tt>: unsigned less than</li>
5591 <li><tt>ule</tt>: unsigned less or equal</li>
5592 <li><tt>sgt</tt>: signed greater than</li>
5593 <li><tt>sge</tt>: signed greater or equal</li>
5594 <li><tt>slt</tt>: signed less than</li>
5595 <li><tt>sle</tt>: signed less or equal</li>
5598 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
5599 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
5600 typed. They must also be identical types.</p>
5603 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
5604 condition code given as <tt>cond</tt>. The comparison performed always yields
5605 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
5606 result, as follows:</p>
5609 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
5610 <tt>false</tt> otherwise. No sign interpretation is necessary or
5613 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
5614 <tt>false</tt> otherwise. No sign interpretation is necessary or
5617 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
5618 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5620 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
5621 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5622 to <tt>op2</tt>.</li>
5624 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
5625 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5627 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
5628 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5630 <li><tt>sgt</tt>: interprets the operands as signed values and yields
5631 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5633 <li><tt>sge</tt>: interprets the operands as signed values and yields
5634 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5635 to <tt>op2</tt>.</li>
5637 <li><tt>slt</tt>: interprets the operands as signed values and yields
5638 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5640 <li><tt>sle</tt>: interprets the operands as signed values and yields
5641 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5644 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
5645 values are compared as if they were integers.</p>
5647 <p>If the operands are integer vectors, then they are compared element by
5648 element. The result is an <tt>i1</tt> vector with the same number of elements
5649 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
5653 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
5654 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
5655 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
5656 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
5657 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
5658 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
5661 <p>Note that the code generator does not yet support vector types with
5662 the <tt>icmp</tt> instruction.</p>
5666 <!-- _______________________________________________________________________ -->
5668 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
5675 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5679 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
5680 values based on comparison of its operands.</p>
5682 <p>If the operands are floating point scalars, then the result type is a boolean
5683 (<a href="#t_integer"><tt>i1</tt></a>).</p>
5685 <p>If the operands are floating point vectors, then the result type is a vector
5686 of boolean with the same number of elements as the operands being
5690 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
5691 the condition code indicating the kind of comparison to perform. It is not a
5692 value, just a keyword. The possible condition code are:</p>
5695 <li><tt>false</tt>: no comparison, always returns false</li>
5696 <li><tt>oeq</tt>: ordered and equal</li>
5697 <li><tt>ogt</tt>: ordered and greater than </li>
5698 <li><tt>oge</tt>: ordered and greater than or equal</li>
5699 <li><tt>olt</tt>: ordered and less than </li>
5700 <li><tt>ole</tt>: ordered and less than or equal</li>
5701 <li><tt>one</tt>: ordered and not equal</li>
5702 <li><tt>ord</tt>: ordered (no nans)</li>
5703 <li><tt>ueq</tt>: unordered or equal</li>
5704 <li><tt>ugt</tt>: unordered or greater than </li>
5705 <li><tt>uge</tt>: unordered or greater than or equal</li>
5706 <li><tt>ult</tt>: unordered or less than </li>
5707 <li><tt>ule</tt>: unordered or less than or equal</li>
5708 <li><tt>une</tt>: unordered or not equal</li>
5709 <li><tt>uno</tt>: unordered (either nans)</li>
5710 <li><tt>true</tt>: no comparison, always returns true</li>
5713 <p><i>Ordered</i> means that neither operand is a QNAN while
5714 <i>unordered</i> means that either operand may be a QNAN.</p>
5716 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
5717 a <a href="#t_floating">floating point</a> type or
5718 a <a href="#t_vector">vector</a> of floating point type. They must have
5719 identical types.</p>
5722 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
5723 according to the condition code given as <tt>cond</tt>. If the operands are
5724 vectors, then the vectors are compared element by element. Each comparison
5725 performed always yields an <a href="#t_integer">i1</a> result, as
5729 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
5731 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5732 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5734 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5735 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5737 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5738 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5740 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5741 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5743 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5744 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5746 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5747 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5749 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
5751 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
5752 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5754 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
5755 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5757 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
5758 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5760 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
5761 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5763 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
5764 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5766 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
5767 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5769 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
5771 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
5776 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
5777 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
5778 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
5779 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
5782 <p>Note that the code generator does not yet support vector types with
5783 the <tt>fcmp</tt> instruction.</p>
5787 <!-- _______________________________________________________________________ -->
5789 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
5796 <result> = phi <ty> [ <val0>, <label0>], ...
5800 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
5801 SSA graph representing the function.</p>
5804 <p>The type of the incoming values is specified with the first type field. After
5805 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
5806 one pair for each predecessor basic block of the current block. Only values
5807 of <a href="#t_firstclass">first class</a> type may be used as the value
5808 arguments to the PHI node. Only labels may be used as the label
5811 <p>There must be no non-phi instructions between the start of a basic block and
5812 the PHI instructions: i.e. PHI instructions must be first in a basic
5815 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5816 occur on the edge from the corresponding predecessor block to the current
5817 block (but after any definition of an '<tt>invoke</tt>' instruction's return
5818 value on the same edge).</p>
5821 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
5822 specified by the pair corresponding to the predecessor basic block that
5823 executed just prior to the current block.</p>
5827 Loop: ; Infinite loop that counts from 0 on up...
5828 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5829 %nextindvar = add i32 %indvar, 1
5835 <!-- _______________________________________________________________________ -->
5837 <a name="i_select">'<tt>select</tt>' Instruction</a>
5844 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5846 <i>selty</i> is either i1 or {<N x i1>}
5850 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
5851 condition, without branching.</p>
5855 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
5856 values indicating the condition, and two values of the
5857 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
5858 vectors and the condition is a scalar, then entire vectors are selected, not
5859 individual elements.</p>
5862 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
5863 first value argument; otherwise, it returns the second value argument.</p>
5865 <p>If the condition is a vector of i1, then the value arguments must be vectors
5866 of the same size, and the selection is done element by element.</p>
5870 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
5873 <p>Note that the code generator does not yet support conditions
5874 with vector type.</p>
5878 <!-- _______________________________________________________________________ -->
5880 <a name="i_call">'<tt>call</tt>' Instruction</a>
5887 <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>]
5891 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
5894 <p>This instruction requires several arguments:</p>
5897 <li>The optional "tail" marker indicates that the callee function does not
5898 access any allocas or varargs in the caller. Note that calls may be
5899 marked "tail" even if they do not occur before
5900 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
5901 present, the function call is eligible for tail call optimization,
5902 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
5903 optimized into a jump</a>. The code generator may optimize calls marked
5904 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
5905 sibling call optimization</a> when the caller and callee have
5906 matching signatures, or 2) forced tail call optimization when the
5907 following extra requirements are met:
5909 <li>Caller and callee both have the calling
5910 convention <tt>fastcc</tt>.</li>
5911 <li>The call is in tail position (ret immediately follows call and ret
5912 uses value of call or is void).</li>
5913 <li>Option <tt>-tailcallopt</tt> is enabled,
5914 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
5915 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
5916 constraints are met.</a></li>
5920 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
5921 convention</a> the call should use. If none is specified, the call
5922 defaults to using C calling conventions. The calling convention of the
5923 call must match the calling convention of the target function, or else the
5924 behavior is undefined.</li>
5926 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
5927 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
5928 '<tt>inreg</tt>' attributes are valid here.</li>
5930 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
5931 type of the return value. Functions that return no value are marked
5932 <tt><a href="#t_void">void</a></tt>.</li>
5934 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5935 being invoked. The argument types must match the types implied by this
5936 signature. This type can be omitted if the function is not varargs and if
5937 the function type does not return a pointer to a function.</li>
5939 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5940 be invoked. In most cases, this is a direct function invocation, but
5941 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5942 to function value.</li>
5944 <li>'<tt>function args</tt>': argument list whose types match the function
5945 signature argument types and parameter attributes. All arguments must be
5946 of <a href="#t_firstclass">first class</a> type. If the function
5947 signature indicates the function accepts a variable number of arguments,
5948 the extra arguments can be specified.</li>
5950 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
5951 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
5952 '<tt>readnone</tt>' attributes are valid here.</li>
5956 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
5957 a specified function, with its incoming arguments bound to the specified
5958 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
5959 function, control flow continues with the instruction after the function
5960 call, and the return value of the function is bound to the result
5965 %retval = call i32 @test(i32 %argc)
5966 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
5967 %X = tail call i32 @foo() <i>; yields i32</i>
5968 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
5969 call void %foo(i8 97 signext)
5971 %struct.A = type { i32, i8 }
5972 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
5973 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
5974 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
5975 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
5976 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
5979 <p>llvm treats calls to some functions with names and arguments that match the
5980 standard C99 library as being the C99 library functions, and may perform
5981 optimizations or generate code for them under that assumption. This is
5982 something we'd like to change in the future to provide better support for
5983 freestanding environments and non-C-based languages.</p>
5987 <!-- _______________________________________________________________________ -->
5989 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
5996 <resultval> = va_arg <va_list*> <arglist>, <argty>
6000 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6001 the "variable argument" area of a function call. It is used to implement the
6002 <tt>va_arg</tt> macro in C.</p>
6005 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6006 argument. It returns a value of the specified argument type and increments
6007 the <tt>va_list</tt> to point to the next argument. The actual type
6008 of <tt>va_list</tt> is target specific.</p>
6011 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6012 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6013 to the next argument. For more information, see the variable argument
6014 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6016 <p>It is legal for this instruction to be called in a function which does not
6017 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6020 <p><tt>va_arg</tt> is an LLVM instruction instead of
6021 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6025 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6027 <p>Note that the code generator does not yet fully support va_arg on many
6028 targets. Also, it does not currently support va_arg with aggregate types on
6033 <!-- _______________________________________________________________________ -->
6035 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6042 <resultval> = landingpad <somety> personality <type> <pers_fn> <clause>+
6043 <resultval> = landingpad <somety> personality <type> <pers_fn> cleanup <clause>*
6045 <clause> := catch <type> <value>
6046 <clause> := filter <array constant type> <array constant>
6050 <p>The '<tt>landingpad</tt>' instruction is used by
6051 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6052 system</a> to specify that a basic block is a landing pad — one where
6053 the exception lands, and corresponds to the code found in the
6054 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6055 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6056 re-entry to the function. The <tt>resultval</tt> has the
6057 type <tt>somety</tt>.</p>
6060 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6061 function associated with the unwinding mechanism. The optional
6062 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6064 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6065 or <tt>filter</tt> — and contains the global variable representing the
6066 "type" that may be caught or filtered respectively. Unlike the
6067 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6068 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6069 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6070 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6073 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6074 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6075 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6076 calling conventions, how the personality function results are represented in
6077 LLVM IR is target specific.</p>
6079 <p>The clauses are applied in order from top to bottom. If two
6080 <tt>landingpad</tt> instructions are merged together through inlining, the
6081 clauses from the calling function are appended to the list of clauses.</p>
6083 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6086 <li>A landing pad block is a basic block which is the unwind destination of an
6087 '<tt>invoke</tt>' instruction.</li>
6088 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6089 first non-PHI instruction.</li>
6090 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6092 <li>A basic block that is not a landing pad block may not include a
6093 '<tt>landingpad</tt>' instruction.</li>
6094 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6095 personality function.</li>
6100 ;; A landing pad which can catch an integer.
6101 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6103 ;; A landing pad that is a cleanup.
6104 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6106 ;; A landing pad which can catch an integer and can only throw a double.
6107 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6109 filter [1 x i8**] [@_ZTId]
6118 <!-- *********************************************************************** -->
6119 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6120 <!-- *********************************************************************** -->
6124 <p>LLVM supports the notion of an "intrinsic function". These functions have
6125 well known names and semantics and are required to follow certain
6126 restrictions. Overall, these intrinsics represent an extension mechanism for
6127 the LLVM language that does not require changing all of the transformations
6128 in LLVM when adding to the language (or the bitcode reader/writer, the
6129 parser, etc...).</p>
6131 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6132 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6133 begin with this prefix. Intrinsic functions must always be external
6134 functions: you cannot define the body of intrinsic functions. Intrinsic
6135 functions may only be used in call or invoke instructions: it is illegal to
6136 take the address of an intrinsic function. Additionally, because intrinsic
6137 functions are part of the LLVM language, it is required if any are added that
6138 they be documented here.</p>
6140 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6141 family of functions that perform the same operation but on different data
6142 types. Because LLVM can represent over 8 million different integer types,
6143 overloading is used commonly to allow an intrinsic function to operate on any
6144 integer type. One or more of the argument types or the result type can be
6145 overloaded to accept any integer type. Argument types may also be defined as
6146 exactly matching a previous argument's type or the result type. This allows
6147 an intrinsic function which accepts multiple arguments, but needs all of them
6148 to be of the same type, to only be overloaded with respect to a single
6149 argument or the result.</p>
6151 <p>Overloaded intrinsics will have the names of its overloaded argument types
6152 encoded into its function name, each preceded by a period. Only those types
6153 which are overloaded result in a name suffix. Arguments whose type is matched
6154 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6155 can take an integer of any width and returns an integer of exactly the same
6156 integer width. This leads to a family of functions such as
6157 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6158 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6159 suffix is required. Because the argument's type is matched against the return
6160 type, it does not require its own name suffix.</p>
6162 <p>To learn how to add an intrinsic function, please see the
6163 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6165 <!-- ======================================================================= -->
6167 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6172 <p>Variable argument support is defined in LLVM with
6173 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6174 intrinsic functions. These functions are related to the similarly named
6175 macros defined in the <tt><stdarg.h></tt> header file.</p>
6177 <p>All of these functions operate on arguments that use a target-specific value
6178 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6179 not define what this type is, so all transformations should be prepared to
6180 handle these functions regardless of the type used.</p>
6182 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6183 instruction and the variable argument handling intrinsic functions are
6186 <pre class="doc_code">
6187 define i32 @test(i32 %X, ...) {
6188 ; Initialize variable argument processing
6190 %ap2 = bitcast i8** %ap to i8*
6191 call void @llvm.va_start(i8* %ap2)
6193 ; Read a single integer argument
6194 %tmp = va_arg i8** %ap, i32
6196 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6198 %aq2 = bitcast i8** %aq to i8*
6199 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6200 call void @llvm.va_end(i8* %aq2)
6202 ; Stop processing of arguments.
6203 call void @llvm.va_end(i8* %ap2)
6207 declare void @llvm.va_start(i8*)
6208 declare void @llvm.va_copy(i8*, i8*)
6209 declare void @llvm.va_end(i8*)
6212 <!-- _______________________________________________________________________ -->
6214 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6222 declare void %llvm.va_start(i8* <arglist>)
6226 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6227 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6230 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6233 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6234 macro available in C. In a target-dependent way, it initializes
6235 the <tt>va_list</tt> element to which the argument points, so that the next
6236 call to <tt>va_arg</tt> will produce the first variable argument passed to
6237 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6238 need to know the last argument of the function as the compiler can figure
6243 <!-- _______________________________________________________________________ -->
6245 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6252 declare void @llvm.va_end(i8* <arglist>)
6256 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6257 which has been initialized previously
6258 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6259 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6262 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6265 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6266 macro available in C. In a target-dependent way, it destroys
6267 the <tt>va_list</tt> element to which the argument points. Calls
6268 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6269 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6270 with calls to <tt>llvm.va_end</tt>.</p>
6274 <!-- _______________________________________________________________________ -->
6276 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6283 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6287 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6288 from the source argument list to the destination argument list.</p>
6291 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6292 The second argument is a pointer to a <tt>va_list</tt> element to copy
6296 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6297 macro available in C. In a target-dependent way, it copies the
6298 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6299 element. This intrinsic is necessary because
6300 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6301 arbitrarily complex and require, for example, memory allocation.</p>
6309 <!-- ======================================================================= -->
6311 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6316 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6317 Collection</a> (GC) requires the implementation and generation of these
6318 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6319 roots on the stack</a>, as well as garbage collector implementations that
6320 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6321 barriers. Front-ends for type-safe garbage collected languages should generate
6322 these intrinsics to make use of the LLVM garbage collectors. For more details,
6323 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6326 <p>The garbage collection intrinsics only operate on objects in the generic
6327 address space (address space zero).</p>
6329 <!-- _______________________________________________________________________ -->
6331 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6338 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6342 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6343 the code generator, and allows some metadata to be associated with it.</p>
6346 <p>The first argument specifies the address of a stack object that contains the
6347 root pointer. The second pointer (which must be either a constant or a
6348 global value address) contains the meta-data to be associated with the
6352 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6353 location. At compile-time, the code generator generates information to allow
6354 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6355 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6360 <!-- _______________________________________________________________________ -->
6362 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6369 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6373 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6374 locations, allowing garbage collector implementations that require read
6378 <p>The second argument is the address to read from, which should be an address
6379 allocated from the garbage collector. The first object is a pointer to the
6380 start of the referenced object, if needed by the language runtime (otherwise
6384 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6385 instruction, but may be replaced with substantially more complex code by the
6386 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6387 may only be used in a function which <a href="#gc">specifies a GC
6392 <!-- _______________________________________________________________________ -->
6394 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6401 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6405 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6406 locations, allowing garbage collector implementations that require write
6407 barriers (such as generational or reference counting collectors).</p>
6410 <p>The first argument is the reference to store, the second is the start of the
6411 object to store it to, and the third is the address of the field of Obj to
6412 store to. If the runtime does not require a pointer to the object, Obj may
6416 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6417 instruction, but may be replaced with substantially more complex code by the
6418 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6419 may only be used in a function which <a href="#gc">specifies a GC
6426 <!-- ======================================================================= -->
6428 <a name="int_codegen">Code Generator Intrinsics</a>
6433 <p>These intrinsics are provided by LLVM to expose special features that may
6434 only be implemented with code generator support.</p>
6436 <!-- _______________________________________________________________________ -->
6438 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6445 declare i8 *@llvm.returnaddress(i32 <level>)
6449 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6450 target-specific value indicating the return address of the current function
6451 or one of its callers.</p>
6454 <p>The argument to this intrinsic indicates which function to return the address
6455 for. Zero indicates the calling function, one indicates its caller, etc.
6456 The argument is <b>required</b> to be a constant integer value.</p>
6459 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6460 indicating the return address of the specified call frame, or zero if it
6461 cannot be identified. The value returned by this intrinsic is likely to be
6462 incorrect or 0 for arguments other than zero, so it should only be used for
6463 debugging purposes.</p>
6465 <p>Note that calling this intrinsic does not prevent function inlining or other
6466 aggressive transformations, so the value returned may not be that of the
6467 obvious source-language caller.</p>
6471 <!-- _______________________________________________________________________ -->
6473 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6480 declare i8* @llvm.frameaddress(i32 <level>)
6484 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6485 target-specific frame pointer value for the specified stack frame.</p>
6488 <p>The argument to this intrinsic indicates which function to return the frame
6489 pointer for. Zero indicates the calling function, one indicates its caller,
6490 etc. The argument is <b>required</b> to be a constant integer value.</p>
6493 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6494 indicating the frame address of the specified call frame, or zero if it
6495 cannot be identified. The value returned by this intrinsic is likely to be
6496 incorrect or 0 for arguments other than zero, so it should only be used for
6497 debugging purposes.</p>
6499 <p>Note that calling this intrinsic does not prevent function inlining or other
6500 aggressive transformations, so the value returned may not be that of the
6501 obvious source-language caller.</p>
6505 <!-- _______________________________________________________________________ -->
6507 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6514 declare i8* @llvm.stacksave()
6518 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6519 of the function stack, for use
6520 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6521 useful for implementing language features like scoped automatic variable
6522 sized arrays in C99.</p>
6525 <p>This intrinsic returns a opaque pointer value that can be passed
6526 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6527 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6528 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6529 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6530 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6531 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6535 <!-- _______________________________________________________________________ -->
6537 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6544 declare void @llvm.stackrestore(i8* %ptr)
6548 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6549 the function stack to the state it was in when the
6550 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6551 executed. This is useful for implementing language features like scoped
6552 automatic variable sized arrays in C99.</p>
6555 <p>See the description
6556 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6560 <!-- _______________________________________________________________________ -->
6562 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6569 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6573 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6574 insert a prefetch instruction if supported; otherwise, it is a noop.
6575 Prefetches have no effect on the behavior of the program but can change its
6576 performance characteristics.</p>
6579 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6580 specifier determining if the fetch should be for a read (0) or write (1),
6581 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
6582 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6583 specifies whether the prefetch is performed on the data (1) or instruction (0)
6584 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
6585 must be constant integers.</p>
6588 <p>This intrinsic does not modify the behavior of the program. In particular,
6589 prefetches cannot trap and do not produce a value. On targets that support
6590 this intrinsic, the prefetch can provide hints to the processor cache for
6591 better performance.</p>
6595 <!-- _______________________________________________________________________ -->
6597 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
6604 declare void @llvm.pcmarker(i32 <id>)
6608 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
6609 Counter (PC) in a region of code to simulators and other tools. The method
6610 is target specific, but it is expected that the marker will use exported
6611 symbols to transmit the PC of the marker. The marker makes no guarantees
6612 that it will remain with any specific instruction after optimizations. It is
6613 possible that the presence of a marker will inhibit optimizations. The
6614 intended use is to be inserted after optimizations to allow correlations of
6615 simulation runs.</p>
6618 <p><tt>id</tt> is a numerical id identifying the marker.</p>
6621 <p>This intrinsic does not modify the behavior of the program. Backends that do
6622 not support this intrinsic may ignore it.</p>
6626 <!-- _______________________________________________________________________ -->
6628 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
6635 declare i64 @llvm.readcyclecounter()
6639 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
6640 counter register (or similar low latency, high accuracy clocks) on those
6641 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6642 should map to RPCC. As the backing counters overflow quickly (on the order
6643 of 9 seconds on alpha), this should only be used for small timings.</p>
6646 <p>When directly supported, reading the cycle counter should not modify any
6647 memory. Implementations are allowed to either return a application specific
6648 value or a system wide value. On backends without support, this is lowered
6649 to a constant 0.</p>
6655 <!-- ======================================================================= -->
6657 <a name="int_libc">Standard C Library Intrinsics</a>
6662 <p>LLVM provides intrinsics for a few important standard C library functions.
6663 These intrinsics allow source-language front-ends to pass information about
6664 the alignment of the pointer arguments to the code generator, providing
6665 opportunity for more efficient code generation.</p>
6667 <!-- _______________________________________________________________________ -->
6669 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
6675 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
6676 integer bit width and for different address spaces. Not all targets support
6677 all bit widths however.</p>
6680 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6681 i32 <len>, i32 <align>, i1 <isvolatile>)
6682 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6683 i64 <len>, i32 <align>, i1 <isvolatile>)
6687 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6688 source location to the destination location.</p>
6690 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
6691 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6692 and the pointers can be in specified address spaces.</p>
6696 <p>The first argument is a pointer to the destination, the second is a pointer
6697 to the source. The third argument is an integer argument specifying the
6698 number of bytes to copy, the fourth argument is the alignment of the
6699 source and destination locations, and the fifth is a boolean indicating a
6700 volatile access.</p>
6702 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6703 then the caller guarantees that both the source and destination pointers are
6704 aligned to that boundary.</p>
6706 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6707 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
6708 The detailed access behavior is not very cleanly specified and it is unwise
6709 to depend on it.</p>
6713 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6714 source location to the destination location, which are not allowed to
6715 overlap. It copies "len" bytes of memory over. If the argument is known to
6716 be aligned to some boundary, this can be specified as the fourth argument,
6717 otherwise it should be set to 0 or 1.</p>
6721 <!-- _______________________________________________________________________ -->
6723 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
6729 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
6730 width and for different address space. Not all targets support all bit
6734 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6735 i32 <len>, i32 <align>, i1 <isvolatile>)
6736 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6737 i64 <len>, i32 <align>, i1 <isvolatile>)
6741 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
6742 source location to the destination location. It is similar to the
6743 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
6746 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
6747 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6748 and the pointers can be in specified address spaces.</p>
6752 <p>The first argument is a pointer to the destination, the second is a pointer
6753 to the source. The third argument is an integer argument specifying the
6754 number of bytes to copy, the fourth argument is the alignment of the
6755 source and destination locations, and the fifth is a boolean indicating a
6756 volatile access.</p>
6758 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6759 then the caller guarantees that the source and destination pointers are
6760 aligned to that boundary.</p>
6762 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6763 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
6764 The detailed access behavior is not very cleanly specified and it is unwise
6765 to depend on it.</p>
6769 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
6770 source location to the destination location, which may overlap. It copies
6771 "len" bytes of memory over. If the argument is known to be aligned to some
6772 boundary, this can be specified as the fourth argument, otherwise it should
6773 be set to 0 or 1.</p>
6777 <!-- _______________________________________________________________________ -->
6779 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
6785 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6786 width and for different address spaces. However, not all targets support all
6790 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6791 i32 <len>, i32 <align>, i1 <isvolatile>)
6792 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6793 i64 <len>, i32 <align>, i1 <isvolatile>)
6797 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
6798 particular byte value.</p>
6800 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
6801 intrinsic does not return a value and takes extra alignment/volatile
6802 arguments. Also, the destination can be in an arbitrary address space.</p>
6805 <p>The first argument is a pointer to the destination to fill, the second is the
6806 byte value with which to fill it, the third argument is an integer argument
6807 specifying the number of bytes to fill, and the fourth argument is the known
6808 alignment of the destination location.</p>
6810 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6811 then the caller guarantees that the destination pointer is aligned to that
6814 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6815 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
6816 The detailed access behavior is not very cleanly specified and it is unwise
6817 to depend on it.</p>
6820 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
6821 at the destination location. If the argument is known to be aligned to some
6822 boundary, this can be specified as the fourth argument, otherwise it should
6823 be set to 0 or 1.</p>
6827 <!-- _______________________________________________________________________ -->
6829 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
6835 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
6836 floating point or vector of floating point type. Not all targets support all
6840 declare float @llvm.sqrt.f32(float %Val)
6841 declare double @llvm.sqrt.f64(double %Val)
6842 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6843 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6844 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6848 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
6849 returning the same value as the libm '<tt>sqrt</tt>' functions would.
6850 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
6851 behavior for negative numbers other than -0.0 (which allows for better
6852 optimization, because there is no need to worry about errno being
6853 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
6856 <p>The argument and return value are floating point numbers of the same
6860 <p>This function returns the sqrt of the specified operand if it is a
6861 nonnegative floating point number.</p>
6865 <!-- _______________________________________________________________________ -->
6867 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
6873 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
6874 floating point or vector of floating point type. Not all targets support all
6878 declare float @llvm.powi.f32(float %Val, i32 %power)
6879 declare double @llvm.powi.f64(double %Val, i32 %power)
6880 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6881 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6882 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6886 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
6887 specified (positive or negative) power. The order of evaluation of
6888 multiplications is not defined. When a vector of floating point type is
6889 used, the second argument remains a scalar integer value.</p>
6892 <p>The second argument is an integer power, and the first is a value to raise to
6896 <p>This function returns the first value raised to the second power with an
6897 unspecified sequence of rounding operations.</p>
6901 <!-- _______________________________________________________________________ -->
6903 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
6909 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
6910 floating point or vector of floating point type. Not all targets support all
6914 declare float @llvm.sin.f32(float %Val)
6915 declare double @llvm.sin.f64(double %Val)
6916 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6917 declare fp128 @llvm.sin.f128(fp128 %Val)
6918 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6922 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
6925 <p>The argument and return value are floating point numbers of the same
6929 <p>This function returns the sine of the specified operand, returning the same
6930 values as the libm <tt>sin</tt> functions would, and handles error conditions
6931 in the same way.</p>
6935 <!-- _______________________________________________________________________ -->
6937 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
6943 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
6944 floating point or vector of floating point type. Not all targets support all
6948 declare float @llvm.cos.f32(float %Val)
6949 declare double @llvm.cos.f64(double %Val)
6950 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6951 declare fp128 @llvm.cos.f128(fp128 %Val)
6952 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6956 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
6959 <p>The argument and return value are floating point numbers of the same
6963 <p>This function returns the cosine of the specified operand, returning the same
6964 values as the libm <tt>cos</tt> functions would, and handles error conditions
6965 in the same way.</p>
6969 <!-- _______________________________________________________________________ -->
6971 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
6977 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
6978 floating point or vector of floating point type. Not all targets support all
6982 declare float @llvm.pow.f32(float %Val, float %Power)
6983 declare double @llvm.pow.f64(double %Val, double %Power)
6984 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6985 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6986 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6990 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
6991 specified (positive or negative) power.</p>
6994 <p>The second argument is a floating point power, and the first is a value to
6995 raise to that power.</p>
6998 <p>This function returns the first value raised to the second power, returning
6999 the same values as the libm <tt>pow</tt> functions would, and handles error
7000 conditions in the same way.</p>
7006 <!-- _______________________________________________________________________ -->
7008 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7014 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7015 floating point or vector of floating point type. Not all targets support all
7019 declare float @llvm.exp.f32(float %Val)
7020 declare double @llvm.exp.f64(double %Val)
7021 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7022 declare fp128 @llvm.exp.f128(fp128 %Val)
7023 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7027 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7030 <p>The argument and return value are floating point numbers of the same
7034 <p>This function returns the same values as the libm <tt>exp</tt> functions
7035 would, and handles error conditions in the same way.</p>
7039 <!-- _______________________________________________________________________ -->
7041 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7047 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7048 floating point or vector of floating point type. Not all targets support all
7052 declare float @llvm.log.f32(float %Val)
7053 declare double @llvm.log.f64(double %Val)
7054 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7055 declare fp128 @llvm.log.f128(fp128 %Val)
7056 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7060 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7063 <p>The argument and return value are floating point numbers of the same
7067 <p>This function returns the same values as the libm <tt>log</tt> functions
7068 would, and handles error conditions in the same way.</p>
7071 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7077 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7078 floating point or vector of floating point type. Not all targets support all
7082 declare float @llvm.fma.f32(float %a, float %b, float %c)
7083 declare double @llvm.fma.f64(double %a, double %b, double %c)
7084 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7085 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7086 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7090 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7094 <p>The argument and return value are floating point numbers of the same
7098 <p>This function returns the same values as the libm <tt>fma</tt> functions
7103 <!-- ======================================================================= -->
7105 <a name="int_manip">Bit Manipulation Intrinsics</a>
7110 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7111 These allow efficient code generation for some algorithms.</p>
7113 <!-- _______________________________________________________________________ -->
7115 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7121 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7122 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7125 declare i16 @llvm.bswap.i16(i16 <id>)
7126 declare i32 @llvm.bswap.i32(i32 <id>)
7127 declare i64 @llvm.bswap.i64(i64 <id>)
7131 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7132 values with an even number of bytes (positive multiple of 16 bits). These
7133 are useful for performing operations on data that is not in the target's
7134 native byte order.</p>
7137 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7138 and low byte of the input i16 swapped. Similarly,
7139 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7140 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7141 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7142 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7143 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7144 more, respectively).</p>
7148 <!-- _______________________________________________________________________ -->
7150 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7156 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7157 width, or on any vector with integer elements. Not all targets support all
7158 bit widths or vector types, however.</p>
7161 declare i8 @llvm.ctpop.i8(i8 <src>)
7162 declare i16 @llvm.ctpop.i16(i16 <src>)
7163 declare i32 @llvm.ctpop.i32(i32 <src>)
7164 declare i64 @llvm.ctpop.i64(i64 <src>)
7165 declare i256 @llvm.ctpop.i256(i256 <src>)
7166 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7170 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7174 <p>The only argument is the value to be counted. The argument may be of any
7175 integer type, or a vector with integer elements.
7176 The return type must match the argument type.</p>
7179 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7180 element of a vector.</p>
7184 <!-- _______________________________________________________________________ -->
7186 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7192 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7193 integer bit width, or any vector whose elements are integers. Not all
7194 targets support all bit widths or vector types, however.</p>
7197 declare i8 @llvm.ctlz.i8 (i8 <src>)
7198 declare i16 @llvm.ctlz.i16(i16 <src>)
7199 declare i32 @llvm.ctlz.i32(i32 <src>)
7200 declare i64 @llvm.ctlz.i64(i64 <src>)
7201 declare i256 @llvm.ctlz.i256(i256 <src>)
7202 declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src;gt)
7206 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7207 leading zeros in a variable.</p>
7210 <p>The only argument is the value to be counted. The argument may be of any
7211 integer type, or any vector type with integer element type.
7212 The return type must match the argument type.</p>
7215 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7216 zeros in a variable, or within each element of the vector if the operation
7217 is of vector type. If the src == 0 then the result is the size in bits of
7218 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7222 <!-- _______________________________________________________________________ -->
7224 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7230 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7231 integer bit width, or any vector of integer elements. Not all targets
7232 support all bit widths or vector types, however.</p>
7235 declare i8 @llvm.cttz.i8 (i8 <src>)
7236 declare i16 @llvm.cttz.i16(i16 <src>)
7237 declare i32 @llvm.cttz.i32(i32 <src>)
7238 declare i64 @llvm.cttz.i64(i64 <src>)
7239 declare i256 @llvm.cttz.i256(i256 <src>)
7240 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>)
7244 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7248 <p>The only argument is the value to be counted. The argument may be of any
7249 integer type, or a vectory with integer element type.. The return type
7250 must match the argument type.</p>
7253 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7254 zeros in a variable, or within each element of a vector.
7255 If the src == 0 then the result is the size in bits of
7256 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7262 <!-- ======================================================================= -->
7264 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7269 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7271 <!-- _______________________________________________________________________ -->
7273 <a name="int_sadd_overflow">
7274 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7281 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7282 on any integer bit width.</p>
7285 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7286 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7287 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7291 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7292 a signed addition of the two arguments, and indicate whether an overflow
7293 occurred during the signed summation.</p>
7296 <p>The arguments (%a and %b) and the first element of the result structure may
7297 be of integer types of any bit width, but they must have the same bit
7298 width. The second element of the result structure must be of
7299 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7300 undergo signed addition.</p>
7303 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7304 a signed addition of the two variables. They return a structure — the
7305 first element of which is the signed summation, and the second element of
7306 which is a bit specifying if the signed summation resulted in an
7311 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7312 %sum = extractvalue {i32, i1} %res, 0
7313 %obit = extractvalue {i32, i1} %res, 1
7314 br i1 %obit, label %overflow, label %normal
7319 <!-- _______________________________________________________________________ -->
7321 <a name="int_uadd_overflow">
7322 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7329 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7330 on any integer bit width.</p>
7333 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7334 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7335 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7339 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7340 an unsigned addition of the two arguments, and indicate whether a carry
7341 occurred during the unsigned summation.</p>
7344 <p>The arguments (%a and %b) and the first element of the result structure may
7345 be of integer types of any bit width, but they must have the same bit
7346 width. The second element of the result structure must be of
7347 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7348 undergo unsigned addition.</p>
7351 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7352 an unsigned addition of the two arguments. They return a structure —
7353 the first element of which is the sum, and the second element of which is a
7354 bit specifying if the unsigned summation resulted in a carry.</p>
7358 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7359 %sum = extractvalue {i32, i1} %res, 0
7360 %obit = extractvalue {i32, i1} %res, 1
7361 br i1 %obit, label %carry, label %normal
7366 <!-- _______________________________________________________________________ -->
7368 <a name="int_ssub_overflow">
7369 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7376 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7377 on any integer bit width.</p>
7380 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7381 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7382 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7386 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7387 a signed subtraction of the two arguments, and indicate whether an overflow
7388 occurred during the signed subtraction.</p>
7391 <p>The arguments (%a and %b) and the first element of the result structure may
7392 be of integer types of any bit width, but they must have the same bit
7393 width. The second element of the result structure must be of
7394 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7395 undergo signed subtraction.</p>
7398 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7399 a signed subtraction of the two arguments. They return a structure —
7400 the first element of which is the subtraction, and the second element of
7401 which is a bit specifying if the signed subtraction resulted in an
7406 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7407 %sum = extractvalue {i32, i1} %res, 0
7408 %obit = extractvalue {i32, i1} %res, 1
7409 br i1 %obit, label %overflow, label %normal
7414 <!-- _______________________________________________________________________ -->
7416 <a name="int_usub_overflow">
7417 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7424 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7425 on any integer bit width.</p>
7428 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7429 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7430 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7434 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7435 an unsigned subtraction of the two arguments, and indicate whether an
7436 overflow occurred during the unsigned subtraction.</p>
7439 <p>The arguments (%a and %b) and the first element of the result structure may
7440 be of integer types of any bit width, but they must have the same bit
7441 width. The second element of the result structure must be of
7442 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7443 undergo unsigned subtraction.</p>
7446 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7447 an unsigned subtraction of the two arguments. They return a structure —
7448 the first element of which is the subtraction, and the second element of
7449 which is a bit specifying if the unsigned subtraction resulted in an
7454 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7455 %sum = extractvalue {i32, i1} %res, 0
7456 %obit = extractvalue {i32, i1} %res, 1
7457 br i1 %obit, label %overflow, label %normal
7462 <!-- _______________________________________________________________________ -->
7464 <a name="int_smul_overflow">
7465 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7472 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7473 on any integer bit width.</p>
7476 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7477 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7478 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7483 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7484 a signed multiplication of the two arguments, and indicate whether an
7485 overflow occurred during the signed multiplication.</p>
7488 <p>The arguments (%a and %b) and the first element of the result structure may
7489 be of integer types of any bit width, but they must have the same bit
7490 width. The second element of the result structure must be of
7491 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7492 undergo signed multiplication.</p>
7495 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7496 a signed multiplication of the two arguments. They return a structure —
7497 the first element of which is the multiplication, and the second element of
7498 which is a bit specifying if the signed multiplication resulted in an
7503 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7504 %sum = extractvalue {i32, i1} %res, 0
7505 %obit = extractvalue {i32, i1} %res, 1
7506 br i1 %obit, label %overflow, label %normal
7511 <!-- _______________________________________________________________________ -->
7513 <a name="int_umul_overflow">
7514 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
7521 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
7522 on any integer bit width.</p>
7525 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7526 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7527 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7531 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7532 a unsigned multiplication of the two arguments, and indicate whether an
7533 overflow occurred during the unsigned multiplication.</p>
7536 <p>The arguments (%a and %b) and the first element of the result structure may
7537 be of integer types of any bit width, but they must have the same bit
7538 width. The second element of the result structure must be of
7539 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7540 undergo unsigned multiplication.</p>
7543 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7544 an unsigned multiplication of the two arguments. They return a structure
7545 — the first element of which is the multiplication, and the second
7546 element of which is a bit specifying if the unsigned multiplication resulted
7551 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7552 %sum = extractvalue {i32, i1} %res, 0
7553 %obit = extractvalue {i32, i1} %res, 1
7554 br i1 %obit, label %overflow, label %normal
7561 <!-- ======================================================================= -->
7563 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
7568 <p>Half precision floating point is a storage-only format. This means that it is
7569 a dense encoding (in memory) but does not support computation in the
7572 <p>This means that code must first load the half-precision floating point
7573 value as an i16, then convert it to float with <a
7574 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
7575 Computation can then be performed on the float value (including extending to
7576 double etc). To store the value back to memory, it is first converted to
7577 float if needed, then converted to i16 with
7578 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
7579 storing as an i16 value.</p>
7581 <!-- _______________________________________________________________________ -->
7583 <a name="int_convert_to_fp16">
7584 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
7592 declare i16 @llvm.convert.to.fp16(f32 %a)
7596 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7597 a conversion from single precision floating point format to half precision
7598 floating point format.</p>
7601 <p>The intrinsic function contains single argument - the value to be
7605 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7606 a conversion from single precision floating point format to half precision
7607 floating point format. The return value is an <tt>i16</tt> which
7608 contains the converted number.</p>
7612 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7613 store i16 %res, i16* @x, align 2
7618 <!-- _______________________________________________________________________ -->
7620 <a name="int_convert_from_fp16">
7621 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
7629 declare f32 @llvm.convert.from.fp16(i16 %a)
7633 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
7634 a conversion from half precision floating point format to single precision
7635 floating point format.</p>
7638 <p>The intrinsic function contains single argument - the value to be
7642 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
7643 conversion from half single precision floating point format to single
7644 precision floating point format. The input half-float value is represented by
7645 an <tt>i16</tt> value.</p>
7649 %a = load i16* @x, align 2
7650 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7657 <!-- ======================================================================= -->
7659 <a name="int_debugger">Debugger Intrinsics</a>
7664 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
7665 prefix), are described in
7666 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
7667 Level Debugging</a> document.</p>
7671 <!-- ======================================================================= -->
7673 <a name="int_eh">Exception Handling Intrinsics</a>
7678 <p>The LLVM exception handling intrinsics (which all start with
7679 <tt>llvm.eh.</tt> prefix), are described in
7680 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
7681 Handling</a> document.</p>
7685 <!-- ======================================================================= -->
7687 <a name="int_trampoline">Trampoline Intrinsics</a>
7692 <p>These intrinsics make it possible to excise one parameter, marked with
7693 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
7694 The result is a callable
7695 function pointer lacking the nest parameter - the caller does not need to
7696 provide a value for it. Instead, the value to use is stored in advance in a
7697 "trampoline", a block of memory usually allocated on the stack, which also
7698 contains code to splice the nest value into the argument list. This is used
7699 to implement the GCC nested function address extension.</p>
7701 <p>For example, if the function is
7702 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
7703 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
7706 <pre class="doc_code">
7707 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
7708 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
7709 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
7710 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
7711 %fp = bitcast i8* %p to i32 (i32, i32)*
7714 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
7715 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
7717 <!-- _______________________________________________________________________ -->
7720 '<tt>llvm.init.trampoline</tt>' Intrinsic
7728 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
7732 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
7733 turning it into a trampoline.</p>
7736 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
7737 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
7738 sufficiently aligned block of memory; this memory is written to by the
7739 intrinsic. Note that the size and the alignment are target-specific - LLVM
7740 currently provides no portable way of determining them, so a front-end that
7741 generates this intrinsic needs to have some target-specific knowledge.
7742 The <tt>func</tt> argument must hold a function bitcast to
7743 an <tt>i8*</tt>.</p>
7746 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
7747 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
7748 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
7749 which can be <a href="#int_trampoline">bitcast (to a new function) and
7750 called</a>. The new function's signature is the same as that of
7751 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
7752 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
7753 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
7754 with the same argument list, but with <tt>nval</tt> used for the missing
7755 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
7756 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
7757 to the returned function pointer is undefined.</p>
7760 <!-- _______________________________________________________________________ -->
7763 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
7771 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
7775 <p>This performs any required machine-specific adjustment to the address of a
7776 trampoline (passed as <tt>tramp</tt>).</p>
7779 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
7780 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
7784 <p>On some architectures the address of the code to be executed needs to be
7785 different to the address where the trampoline is actually stored. This
7786 intrinsic returns the executable address corresponding to <tt>tramp</tt>
7787 after performing the required machine specific adjustments.
7788 The pointer returned can then be <a href="#int_trampoline"> bitcast and
7796 <!-- ======================================================================= -->
7798 <a name="int_memorymarkers">Memory Use Markers</a>
7803 <p>This class of intrinsics exists to information about the lifetime of memory
7804 objects and ranges where variables are immutable.</p>
7806 <!-- _______________________________________________________________________ -->
7808 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
7815 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
7819 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
7820 object's lifetime.</p>
7823 <p>The first argument is a constant integer representing the size of the
7824 object, or -1 if it is variable sized. The second argument is a pointer to
7828 <p>This intrinsic indicates that before this point in the code, the value of the
7829 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7830 never be used and has an undefined value. A load from the pointer that
7831 precedes this intrinsic can be replaced with
7832 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
7836 <!-- _______________________________________________________________________ -->
7838 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
7845 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
7849 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
7850 object's lifetime.</p>
7853 <p>The first argument is a constant integer representing the size of the
7854 object, or -1 if it is variable sized. The second argument is a pointer to
7858 <p>This intrinsic indicates that after this point in the code, the value of the
7859 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7860 never be used and has an undefined value. Any stores into the memory object
7861 following this intrinsic may be removed as dead.
7865 <!-- _______________________________________________________________________ -->
7867 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
7874 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
7878 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
7879 a memory object will not change.</p>
7882 <p>The first argument is a constant integer representing the size of the
7883 object, or -1 if it is variable sized. The second argument is a pointer to
7887 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
7888 the return value, the referenced memory location is constant and
7893 <!-- _______________________________________________________________________ -->
7895 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
7902 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7906 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
7907 a memory object are mutable.</p>
7910 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
7911 The second argument is a constant integer representing the size of the
7912 object, or -1 if it is variable sized and the third argument is a pointer
7916 <p>This intrinsic indicates that the memory is mutable again.</p>
7922 <!-- ======================================================================= -->
7924 <a name="int_general">General Intrinsics</a>
7929 <p>This class of intrinsics is designed to be generic and has no specific
7932 <!-- _______________________________________________________________________ -->
7934 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
7941 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
7945 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
7948 <p>The first argument is a pointer to a value, the second is a pointer to a
7949 global string, the third is a pointer to a global string which is the source
7950 file name, and the last argument is the line number.</p>
7953 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7954 This can be useful for special purpose optimizations that want to look for
7955 these annotations. These have no other defined use; they are ignored by code
7956 generation and optimization.</p>
7960 <!-- _______________________________________________________________________ -->
7962 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
7968 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
7969 any integer bit width.</p>
7972 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
7973 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
7974 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
7975 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
7976 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
7980 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
7983 <p>The first argument is an integer value (result of some expression), the
7984 second is a pointer to a global string, the third is a pointer to a global
7985 string which is the source file name, and the last argument is the line
7986 number. It returns the value of the first argument.</p>
7989 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7990 arbitrary strings. This can be useful for special purpose optimizations that
7991 want to look for these annotations. These have no other defined use; they
7992 are ignored by code generation and optimization.</p>
7996 <!-- _______________________________________________________________________ -->
7998 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8005 declare void @llvm.trap()
8009 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8015 <p>This intrinsics is lowered to the target dependent trap instruction. If the
8016 target does not have a trap instruction, this intrinsic will be lowered to
8017 the call of the <tt>abort()</tt> function.</p>
8021 <!-- _______________________________________________________________________ -->
8023 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8030 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8034 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8035 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8036 ensure that it is placed on the stack before local variables.</p>
8039 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8040 arguments. The first argument is the value loaded from the stack
8041 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8042 that has enough space to hold the value of the guard.</p>
8045 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8046 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8047 stack. This is to ensure that if a local variable on the stack is
8048 overwritten, it will destroy the value of the guard. When the function exits,
8049 the guard on the stack is checked against the original guard. If they are
8050 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8055 <!-- _______________________________________________________________________ -->
8057 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8064 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <type>)
8065 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <type>)
8069 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8070 the optimizers to determine at compile time whether a) an operation (like
8071 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8072 runtime check for overflow isn't necessary. An object in this context means
8073 an allocation of a specific class, structure, array, or other object.</p>
8076 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8077 argument is a pointer to or into the <tt>object</tt>. The second argument
8078 is a boolean 0 or 1. This argument determines whether you want the
8079 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
8080 1, variables are not allowed.</p>
8083 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
8084 representing the size of the object concerned, or <tt>i32/i64 -1 or 0</tt>,
8085 depending on the <tt>type</tt> argument, if the size cannot be determined at
8094 <!-- *********************************************************************** -->
8097 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8098 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8099 <a href="http://validator.w3.org/check/referer"><img
8100 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8102 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8103 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8104 Last modified: $Date$