1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <div class="doc_title"> LLVM Language Reference Manual </div>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a>
25 <li><a href="#linkage_private">'<tt>private</tt>' Linkage</a></li>
26 <li><a href="#linkage_linker_private">'<tt>linker_private</tt>' Linkage</a></li>
27 <li><a href="#linkage_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>externally visible</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>
58 <li><a href="#typesystem">Type System</a>
60 <li><a href="#t_classifications">Type Classifications</a></li>
61 <li><a href="#t_primitive">Primitive Types</a>
63 <li><a href="#t_integer">Integer Type</a></li>
64 <li><a href="#t_floating">Floating Point Types</a></li>
65 <li><a href="#t_void">Void Type</a></li>
66 <li><a href="#t_label">Label Type</a></li>
67 <li><a href="#t_metadata">Metadata Type</a></li>
70 <li><a href="#t_derived">Derived Types</a>
72 <li><a href="#t_aggregate">Aggregate Types</a>
74 <li><a href="#t_array">Array Type</a></li>
75 <li><a href="#t_struct">Structure Type</a></li>
76 <li><a href="#t_pstruct">Packed Structure Type</a></li>
77 <li><a href="#t_vector">Vector Type</a></li>
80 <li><a href="#t_function">Function Type</a></li>
81 <li><a href="#t_pointer">Pointer Type</a></li>
82 <li><a href="#t_opaque">Opaque Type</a></li>
85 <li><a href="#t_uprefs">Type Up-references</a></li>
88 <li><a href="#constants">Constants</a>
90 <li><a href="#simpleconstants">Simple Constants</a></li>
91 <li><a href="#complexconstants">Complex Constants</a></li>
92 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
93 <li><a href="#undefvalues">Undefined Values</a></li>
94 <li><a href="#trapvalues">Trap Values</a></li>
95 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
96 <li><a href="#constantexprs">Constant Expressions</a></li>
99 <li><a href="#othervalues">Other Values</a>
101 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
102 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a></li>
105 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
107 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
108 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
109 Global Variable</a></li>
110 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
111 Global Variable</a></li>
112 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
113 Global Variable</a></li>
116 <li><a href="#instref">Instruction Reference</a>
118 <li><a href="#terminators">Terminator Instructions</a>
120 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
121 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
122 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
123 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
124 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
125 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
126 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
129 <li><a href="#binaryops">Binary Operations</a>
131 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
132 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
133 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
134 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
135 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
136 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
137 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
138 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
139 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
140 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
141 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
142 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
145 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
147 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
148 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
149 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
150 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
151 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
152 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
155 <li><a href="#vectorops">Vector Operations</a>
157 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
158 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
159 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
162 <li><a href="#aggregateops">Aggregate Operations</a>
164 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
165 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
168 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
170 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
171 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
172 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
173 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
176 <li><a href="#convertops">Conversion Operations</a>
178 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
179 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
180 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
181 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
182 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
183 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
184 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
185 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
186 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
187 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
188 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
189 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
192 <li><a href="#otherops">Other Operations</a>
194 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
195 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
196 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
197 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
198 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
199 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
204 <li><a href="#intrinsics">Intrinsic Functions</a>
206 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
208 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
209 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
210 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
213 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
215 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
216 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
217 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
220 <li><a href="#int_codegen">Code Generator Intrinsics</a>
222 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
223 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
224 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
225 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
226 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
227 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
228 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
231 <li><a href="#int_libc">Standard C Library Intrinsics</a>
233 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
234 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
235 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
236 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
237 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
238 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
239 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
240 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
243 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
245 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
246 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
247 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
248 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
251 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
253 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
254 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
255 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
256 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
257 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
258 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
261 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
263 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
264 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
267 <li><a href="#int_debugger">Debugger intrinsics</a></li>
268 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
269 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
271 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
274 <li><a href="#int_atomics">Atomic intrinsics</a>
276 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
277 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
278 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
279 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
280 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
281 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
282 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
283 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
284 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
285 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
286 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
287 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
288 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
291 <li><a href="#int_memorymarkers">Memory Use Markers</a>
293 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
294 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
295 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
296 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
299 <li><a href="#int_general">General intrinsics</a>
301 <li><a href="#int_var_annotation">
302 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
303 <li><a href="#int_annotation">
304 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
305 <li><a href="#int_trap">
306 '<tt>llvm.trap</tt>' Intrinsic</a></li>
307 <li><a href="#int_stackprotector">
308 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
309 <li><a href="#int_objectsize">
310 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
317 <div class="doc_author">
318 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
319 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
322 <!-- *********************************************************************** -->
323 <div class="doc_section"> <a name="abstract">Abstract </a></div>
324 <!-- *********************************************************************** -->
326 <div class="doc_text">
328 <p>This document is a reference manual for the LLVM assembly language. LLVM is
329 a Static Single Assignment (SSA) based representation that provides type
330 safety, low-level operations, flexibility, and the capability of representing
331 'all' high-level languages cleanly. It is the common code representation
332 used throughout all phases of the LLVM compilation strategy.</p>
336 <!-- *********************************************************************** -->
337 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
338 <!-- *********************************************************************** -->
340 <div class="doc_text">
342 <p>The LLVM code representation is designed to be used in three different forms:
343 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
344 for fast loading by a Just-In-Time compiler), and as a human readable
345 assembly language representation. This allows LLVM to provide a powerful
346 intermediate representation for efficient compiler transformations and
347 analysis, while providing a natural means to debug and visualize the
348 transformations. The three different forms of LLVM are all equivalent. This
349 document describes the human readable representation and notation.</p>
351 <p>The LLVM representation aims to be light-weight and low-level while being
352 expressive, typed, and extensible at the same time. It aims to be a
353 "universal IR" of sorts, by being at a low enough level that high-level ideas
354 may be cleanly mapped to it (similar to how microprocessors are "universal
355 IR's", allowing many source languages to be mapped to them). By providing
356 type information, LLVM can be used as the target of optimizations: for
357 example, through pointer analysis, it can be proven that a C automatic
358 variable is never accessed outside of the current function, allowing it to
359 be promoted to a simple SSA value instead of a memory location.</p>
363 <!-- _______________________________________________________________________ -->
364 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
366 <div class="doc_text">
368 <p>It is important to note that this document describes 'well formed' LLVM
369 assembly language. There is a difference between what the parser accepts and
370 what is considered 'well formed'. For example, the following instruction is
371 syntactically okay, but not well formed:</p>
373 <pre class="doc_code">
374 %x = <a href="#i_add">add</a> i32 1, %x
377 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
378 LLVM infrastructure provides a verification pass that may be used to verify
379 that an LLVM module is well formed. This pass is automatically run by the
380 parser after parsing input assembly and by the optimizer before it outputs
381 bitcode. The violations pointed out by the verifier pass indicate bugs in
382 transformation passes or input to the parser.</p>
386 <!-- Describe the typesetting conventions here. -->
388 <!-- *********************************************************************** -->
389 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
390 <!-- *********************************************************************** -->
392 <div class="doc_text">
394 <p>LLVM identifiers come in two basic types: global and local. Global
395 identifiers (functions, global variables) begin with the <tt>'@'</tt>
396 character. Local identifiers (register names, types) begin with
397 the <tt>'%'</tt> character. Additionally, there are three different formats
398 for identifiers, for different purposes:</p>
401 <li>Named values are represented as a string of characters with their prefix.
402 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
403 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
404 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
405 other characters in their names can be surrounded with quotes. Special
406 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
407 ASCII code for the character in hexadecimal. In this way, any character
408 can be used in a name value, even quotes themselves.</li>
410 <li>Unnamed values are represented as an unsigned numeric value with their
411 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
413 <li>Constants, which are described in a <a href="#constants">section about
414 constants</a>, below.</li>
417 <p>LLVM requires that values start with a prefix for two reasons: Compilers
418 don't need to worry about name clashes with reserved words, and the set of
419 reserved words may be expanded in the future without penalty. Additionally,
420 unnamed identifiers allow a compiler to quickly come up with a temporary
421 variable without having to avoid symbol table conflicts.</p>
423 <p>Reserved words in LLVM are very similar to reserved words in other
424 languages. There are keywords for different opcodes
425 ('<tt><a href="#i_add">add</a></tt>',
426 '<tt><a href="#i_bitcast">bitcast</a></tt>',
427 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
428 ('<tt><a href="#t_void">void</a></tt>',
429 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
430 reserved words cannot conflict with variable names, because none of them
431 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
433 <p>Here is an example of LLVM code to multiply the integer variable
434 '<tt>%X</tt>' by 8:</p>
438 <pre class="doc_code">
439 %result = <a href="#i_mul">mul</a> i32 %X, 8
442 <p>After strength reduction:</p>
444 <pre class="doc_code">
445 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
448 <p>And the hard way:</p>
450 <pre class="doc_code">
451 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
452 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
453 %result = <a href="#i_add">add</a> i32 %1, %1
456 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
457 lexical features of LLVM:</p>
460 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
463 <li>Unnamed temporaries are created when the result of a computation is not
464 assigned to a named value.</li>
466 <li>Unnamed temporaries are numbered sequentially</li>
469 <p>It also shows a convention that we follow in this document. When
470 demonstrating instructions, we will follow an instruction with a comment that
471 defines the type and name of value produced. Comments are shown in italic
476 <!-- *********************************************************************** -->
477 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
478 <!-- *********************************************************************** -->
480 <!-- ======================================================================= -->
481 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
484 <div class="doc_text">
486 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
487 of the input programs. Each module consists of functions, global variables,
488 and symbol table entries. Modules may be combined together with the LLVM
489 linker, which merges function (and global variable) definitions, resolves
490 forward declarations, and merges symbol table entries. Here is an example of
491 the "hello world" module:</p>
493 <pre class="doc_code">
494 <i>; Declare the string constant as a global constant.</i>
495 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
497 <i>; External declaration of the puts function</i>
498 <a href="#functionstructure">declare</a> i32 @puts(i8*) <i>; i32 (i8*)* </i>
500 <i>; Definition of main function</i>
501 define i32 @main() { <i>; i32()* </i>
502 <i>; Convert [13 x i8]* to i8 *...</i>
503 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8*</i>
505 <i>; Call puts function to write out the string to stdout.</i>
506 <a href="#i_call">call</a> i32 @puts(i8* %cast210) <i>; i32</i>
507 <a href="#i_ret">ret</a> i32 0
510 <i>; Named metadata</i>
511 !1 = metadata !{i32 41}
515 <p>This example is made up of a <a href="#globalvars">global variable</a> named
516 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
517 a <a href="#functionstructure">function definition</a> for
518 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
521 <p>In general, a module is made up of a list of global values, where both
522 functions and global variables are global values. Global values are
523 represented by a pointer to a memory location (in this case, a pointer to an
524 array of char, and a pointer to a function), and have one of the
525 following <a href="#linkage">linkage types</a>.</p>
529 <!-- ======================================================================= -->
530 <div class="doc_subsection">
531 <a name="linkage">Linkage Types</a>
534 <div class="doc_text">
536 <p>All Global Variables and Functions have one of the following types of
540 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
541 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
542 by objects in the current module. In particular, linking code into a
543 module with an private global value may cause the private to be renamed as
544 necessary to avoid collisions. Because the symbol is private to the
545 module, all references can be updated. This doesn't show up in any symbol
546 table in the object file.</dd>
548 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
549 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
550 assembler and evaluated by the linker. Unlike normal strong symbols, they
551 are removed by the linker from the final linked image (executable or
552 dynamic library).</dd>
554 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
555 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
556 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
557 linker. The symbols are removed by the linker from the final linked image
558 (executable or dynamic library).</dd>
560 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
561 <dd>Similar to "<tt>linker_private_weak</tt>", but it's known that the address
562 of the object is not taken. For instance, functions that had an inline
563 definition, but the compiler decided not to inline it. Note,
564 unlike <tt>linker_private</tt> and <tt>linker_private_weak</tt>,
565 <tt>linker_private_weak_def_auto</tt> may have only <tt>default</tt>
566 visibility. The symbols are removed by the linker from the final linked
567 image (executable or dynamic library).</dd>
569 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
570 <dd>Similar to private, but the value shows as a local symbol
571 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
572 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
574 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
575 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
576 into the object file corresponding to the LLVM module. They exist to
577 allow inlining and other optimizations to take place given knowledge of
578 the definition of the global, which is known to be somewhere outside the
579 module. Globals with <tt>available_externally</tt> linkage are allowed to
580 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
581 This linkage type is only allowed on definitions, not declarations.</dd>
583 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
584 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
585 the same name when linkage occurs. This can be used to implement
586 some forms of inline functions, templates, or other code which must be
587 generated in each translation unit that uses it, but where the body may
588 be overridden with a more definitive definition later. Unreferenced
589 <tt>linkonce</tt> globals are allowed to be discarded. Note that
590 <tt>linkonce</tt> linkage does not actually allow the optimizer to
591 inline the body of this function into callers because it doesn't know if
592 this definition of the function is the definitive definition within the
593 program or whether it will be overridden by a stronger definition.
594 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
597 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
598 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
599 <tt>linkonce</tt> linkage, except that unreferenced globals with
600 <tt>weak</tt> linkage may not be discarded. This is used for globals that
601 are declared "weak" in C source code.</dd>
603 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
604 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
605 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
607 Symbols with "<tt>common</tt>" linkage are merged in the same way as
608 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
609 <tt>common</tt> symbols may not have an explicit section,
610 must have a zero initializer, and may not be marked '<a
611 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
612 have common linkage.</dd>
615 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
616 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
617 pointer to array type. When two global variables with appending linkage
618 are linked together, the two global arrays are appended together. This is
619 the LLVM, typesafe, equivalent of having the system linker append together
620 "sections" with identical names when .o files are linked.</dd>
622 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
623 <dd>The semantics of this linkage follow the ELF object file model: the symbol
624 is weak until linked, if not linked, the symbol becomes null instead of
625 being an undefined reference.</dd>
627 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
628 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
629 <dd>Some languages allow differing globals to be merged, such as two functions
630 with different semantics. Other languages, such as <tt>C++</tt>, ensure
631 that only equivalent globals are ever merged (the "one definition rule"
632 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
633 and <tt>weak_odr</tt> linkage types to indicate that the global will only
634 be merged with equivalent globals. These linkage types are otherwise the
635 same as their non-<tt>odr</tt> versions.</dd>
637 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
638 <dd>If none of the above identifiers are used, the global is externally
639 visible, meaning that it participates in linkage and can be used to
640 resolve external symbol references.</dd>
643 <p>The next two types of linkage are targeted for Microsoft Windows platform
644 only. They are designed to support importing (exporting) symbols from (to)
645 DLLs (Dynamic Link Libraries).</p>
648 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
649 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
650 or variable via a global pointer to a pointer that is set up by the DLL
651 exporting the symbol. On Microsoft Windows targets, the pointer name is
652 formed by combining <code>__imp_</code> and the function or variable
655 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
656 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
657 pointer to a pointer in a DLL, so that it can be referenced with the
658 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
659 name is formed by combining <code>__imp_</code> and the function or
663 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
664 another module defined a "<tt>.LC0</tt>" variable and was linked with this
665 one, one of the two would be renamed, preventing a collision. Since
666 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
667 declarations), they are accessible outside of the current module.</p>
669 <p>It is illegal for a function <i>declaration</i> to have any linkage type
670 other than "externally visible", <tt>dllimport</tt>
671 or <tt>extern_weak</tt>.</p>
673 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
674 or <tt>weak_odr</tt> linkages.</p>
678 <!-- ======================================================================= -->
679 <div class="doc_subsection">
680 <a name="callingconv">Calling Conventions</a>
683 <div class="doc_text">
685 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
686 and <a href="#i_invoke">invokes</a> can all have an optional calling
687 convention specified for the call. The calling convention of any pair of
688 dynamic caller/callee must match, or the behavior of the program is
689 undefined. The following calling conventions are supported by LLVM, and more
690 may be added in the future:</p>
693 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
694 <dd>This calling convention (the default if no other calling convention is
695 specified) matches the target C calling conventions. This calling
696 convention supports varargs function calls and tolerates some mismatch in
697 the declared prototype and implemented declaration of the function (as
700 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
701 <dd>This calling convention attempts to make calls as fast as possible
702 (e.g. by passing things in registers). This calling convention allows the
703 target to use whatever tricks it wants to produce fast code for the
704 target, without having to conform to an externally specified ABI
705 (Application Binary Interface).
706 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
707 when this or the GHC convention is used.</a> This calling convention
708 does not support varargs and requires the prototype of all callees to
709 exactly match the prototype of the function definition.</dd>
711 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
712 <dd>This calling convention attempts to make code in the caller as efficient
713 as possible under the assumption that the call is not commonly executed.
714 As such, these calls often preserve all registers so that the call does
715 not break any live ranges in the caller side. This calling convention
716 does not support varargs and requires the prototype of all callees to
717 exactly match the prototype of the function definition.</dd>
719 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
720 <dd>This calling convention has been implemented specifically for use by the
721 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
722 It passes everything in registers, going to extremes to achieve this by
723 disabling callee save registers. This calling convention should not be
724 used lightly but only for specific situations such as an alternative to
725 the <em>register pinning</em> performance technique often used when
726 implementing functional programming languages.At the moment only X86
727 supports this convention and it has the following limitations:
729 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
730 floating point types are supported.</li>
731 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
732 6 floating point parameters.</li>
734 This calling convention supports
735 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
736 requires both the caller and callee are using it.
739 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
740 <dd>Any calling convention may be specified by number, allowing
741 target-specific calling conventions to be used. Target specific calling
742 conventions start at 64.</dd>
745 <p>More calling conventions can be added/defined on an as-needed basis, to
746 support Pascal conventions or any other well-known target-independent
751 <!-- ======================================================================= -->
752 <div class="doc_subsection">
753 <a name="visibility">Visibility Styles</a>
756 <div class="doc_text">
758 <p>All Global Variables and Functions have one of the following visibility
762 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
763 <dd>On targets that use the ELF object file format, default visibility means
764 that the declaration is visible to other modules and, in shared libraries,
765 means that the declared entity may be overridden. On Darwin, default
766 visibility means that the declaration is visible to other modules. Default
767 visibility corresponds to "external linkage" in the language.</dd>
769 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
770 <dd>Two declarations of an object with hidden visibility refer to the same
771 object if they are in the same shared object. Usually, hidden visibility
772 indicates that the symbol will not be placed into the dynamic symbol
773 table, so no other module (executable or shared library) can reference it
776 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
777 <dd>On ELF, protected visibility indicates that the symbol will be placed in
778 the dynamic symbol table, but that references within the defining module
779 will bind to the local symbol. That is, the symbol cannot be overridden by
785 <!-- ======================================================================= -->
786 <div class="doc_subsection">
787 <a name="namedtypes">Named Types</a>
790 <div class="doc_text">
792 <p>LLVM IR allows you to specify name aliases for certain types. This can make
793 it easier to read the IR and make the IR more condensed (particularly when
794 recursive types are involved). An example of a name specification is:</p>
796 <pre class="doc_code">
797 %mytype = type { %mytype*, i32 }
800 <p>You may give a name to any <a href="#typesystem">type</a> except
801 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
802 is expected with the syntax "%mytype".</p>
804 <p>Note that type names are aliases for the structural type that they indicate,
805 and that you can therefore specify multiple names for the same type. This
806 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
807 uses structural typing, the name is not part of the type. When printing out
808 LLVM IR, the printer will pick <em>one name</em> to render all types of a
809 particular shape. This means that if you have code where two different
810 source types end up having the same LLVM type, that the dumper will sometimes
811 print the "wrong" or unexpected type. This is an important design point and
812 isn't going to change.</p>
816 <!-- ======================================================================= -->
817 <div class="doc_subsection">
818 <a name="globalvars">Global Variables</a>
821 <div class="doc_text">
823 <p>Global variables define regions of memory allocated at compilation time
824 instead of run-time. Global variables may optionally be initialized, may
825 have an explicit section to be placed in, and may have an optional explicit
826 alignment specified. A variable may be defined as "thread_local", which
827 means that it will not be shared by threads (each thread will have a
828 separated copy of the variable). A variable may be defined as a global
829 "constant," which indicates that the contents of the variable
830 will <b>never</b> be modified (enabling better optimization, allowing the
831 global data to be placed in the read-only section of an executable, etc).
832 Note that variables that need runtime initialization cannot be marked
833 "constant" as there is a store to the variable.</p>
835 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
836 constant, even if the final definition of the global is not. This capability
837 can be used to enable slightly better optimization of the program, but
838 requires the language definition to guarantee that optimizations based on the
839 'constantness' are valid for the translation units that do not include the
842 <p>As SSA values, global variables define pointer values that are in scope
843 (i.e. they dominate) all basic blocks in the program. Global variables
844 always define a pointer to their "content" type because they describe a
845 region of memory, and all memory objects in LLVM are accessed through
848 <p>A global variable may be declared to reside in a target-specific numbered
849 address space. For targets that support them, address spaces may affect how
850 optimizations are performed and/or what target instructions are used to
851 access the variable. The default address space is zero. The address space
852 qualifier must precede any other attributes.</p>
854 <p>LLVM allows an explicit section to be specified for globals. If the target
855 supports it, it will emit globals to the section specified.</p>
857 <p>An explicit alignment may be specified for a global, which must be a power
858 of 2. If not present, or if the alignment is set to zero, the alignment of
859 the global is set by the target to whatever it feels convenient. If an
860 explicit alignment is specified, the global is forced to have exactly that
861 alignment. Targets and optimizers are not allowed to over-align the global
862 if the global has an assigned section. In this case, the extra alignment
863 could be observable: for example, code could assume that the globals are
864 densely packed in their section and try to iterate over them as an array,
865 alignment padding would break this iteration.</p>
867 <p>For example, the following defines a global in a numbered address space with
868 an initializer, section, and alignment:</p>
870 <pre class="doc_code">
871 @G = addrspace(5) constant float 1.0, section "foo", align 4
877 <!-- ======================================================================= -->
878 <div class="doc_subsection">
879 <a name="functionstructure">Functions</a>
882 <div class="doc_text">
884 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
885 optional <a href="#linkage">linkage type</a>, an optional
886 <a href="#visibility">visibility style</a>, an optional
887 <a href="#callingconv">calling convention</a>, a return type, an optional
888 <a href="#paramattrs">parameter attribute</a> for the return type, a function
889 name, a (possibly empty) argument list (each with optional
890 <a href="#paramattrs">parameter attributes</a>), optional
891 <a href="#fnattrs">function attributes</a>, an optional section, an optional
892 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
893 curly brace, a list of basic blocks, and a closing curly brace.</p>
895 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
896 optional <a href="#linkage">linkage type</a>, an optional
897 <a href="#visibility">visibility style</a>, an optional
898 <a href="#callingconv">calling convention</a>, a return type, an optional
899 <a href="#paramattrs">parameter attribute</a> for the return type, a function
900 name, a possibly empty list of arguments, an optional alignment, and an
901 optional <a href="#gc">garbage collector name</a>.</p>
903 <p>A function definition contains a list of basic blocks, forming the CFG
904 (Control Flow Graph) for the function. Each basic block may optionally start
905 with a label (giving the basic block a symbol table entry), contains a list
906 of instructions, and ends with a <a href="#terminators">terminator</a>
907 instruction (such as a branch or function return).</p>
909 <p>The first basic block in a function is special in two ways: it is immediately
910 executed on entrance to the function, and it is not allowed to have
911 predecessor basic blocks (i.e. there can not be any branches to the entry
912 block of a function). Because the block can have no predecessors, it also
913 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
915 <p>LLVM allows an explicit section to be specified for functions. If the target
916 supports it, it will emit functions to the section specified.</p>
918 <p>An explicit alignment may be specified for a function. If not present, or if
919 the alignment is set to zero, the alignment of the function is set by the
920 target to whatever it feels convenient. If an explicit alignment is
921 specified, the function is forced to have at least that much alignment. All
922 alignments must be a power of 2.</p>
925 <pre class="doc_code">
926 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
927 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
928 <ResultType> @<FunctionName> ([argument list])
929 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
930 [<a href="#gc">gc</a>] { ... }
935 <!-- ======================================================================= -->
936 <div class="doc_subsection">
937 <a name="aliasstructure">Aliases</a>
940 <div class="doc_text">
942 <p>Aliases act as "second name" for the aliasee value (which can be either
943 function, global variable, another alias or bitcast of global value). Aliases
944 may have an optional <a href="#linkage">linkage type</a>, and an
945 optional <a href="#visibility">visibility style</a>.</p>
948 <pre class="doc_code">
949 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
954 <!-- ======================================================================= -->
955 <div class="doc_subsection">
956 <a name="namedmetadatastructure">Named Metadata</a>
959 <div class="doc_text">
961 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
962 nodes</a> (but not metadata strings) are the only valid operands for
963 a named metadata.</p>
966 <pre class="doc_code">
967 ; Some unnamed metadata nodes, which are referenced by the named metadata.
968 !0 = metadata !{metadata !"zero"}
969 !1 = metadata !{metadata !"one"}
970 !2 = metadata !{metadata !"two"}
972 !name = !{!0, !1, !2}
977 <!-- ======================================================================= -->
978 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
980 <div class="doc_text">
982 <p>The return type and each parameter of a function type may have a set of
983 <i>parameter attributes</i> associated with them. Parameter attributes are
984 used to communicate additional information about the result or parameters of
985 a function. Parameter attributes are considered to be part of the function,
986 not of the function type, so functions with different parameter attributes
987 can have the same function type.</p>
989 <p>Parameter attributes are simple keywords that follow the type specified. If
990 multiple parameter attributes are needed, they are space separated. For
993 <pre class="doc_code">
994 declare i32 @printf(i8* noalias nocapture, ...)
995 declare i32 @atoi(i8 zeroext)
996 declare signext i8 @returns_signed_char()
999 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1000 <tt>readonly</tt>) come immediately after the argument list.</p>
1002 <p>Currently, only the following parameter attributes are defined:</p>
1005 <dt><tt><b>zeroext</b></tt></dt>
1006 <dd>This indicates to the code generator that the parameter or return value
1007 should be zero-extended to a 32-bit value by the caller (for a parameter)
1008 or the callee (for a return value).</dd>
1010 <dt><tt><b>signext</b></tt></dt>
1011 <dd>This indicates to the code generator that the parameter or return value
1012 should be sign-extended to a 32-bit value by the caller (for a parameter)
1013 or the callee (for a return value).</dd>
1015 <dt><tt><b>inreg</b></tt></dt>
1016 <dd>This indicates that this parameter or return value should be treated in a
1017 special target-dependent fashion during while emitting code for a function
1018 call or return (usually, by putting it in a register as opposed to memory,
1019 though some targets use it to distinguish between two different kinds of
1020 registers). Use of this attribute is target-specific.</dd>
1022 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1023 <dd>This indicates that the pointer parameter should really be passed by value
1024 to the function. The attribute implies that a hidden copy of the pointee
1025 is made between the caller and the callee, so the callee is unable to
1026 modify the value in the callee. This attribute is only valid on LLVM
1027 pointer arguments. It is generally used to pass structs and arrays by
1028 value, but is also valid on pointers to scalars. The copy is considered
1029 to belong to the caller not the callee (for example,
1030 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1031 <tt>byval</tt> parameters). This is not a valid attribute for return
1032 values. The byval attribute also supports specifying an alignment with
1033 the align attribute. This has a target-specific effect on the code
1034 generator that usually indicates a desired alignment for the synthesized
1037 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1038 <dd>This indicates that the pointer parameter specifies the address of a
1039 structure that is the return value of the function in the source program.
1040 This pointer must be guaranteed by the caller to be valid: loads and
1041 stores to the structure may be assumed by the callee to not to trap. This
1042 may only be applied to the first parameter. This is not a valid attribute
1043 for return values. </dd>
1045 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1046 <dd>This indicates that pointer values
1047 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1048 value do not alias pointer values which are not <i>based</i> on it,
1049 ignoring certain "irrelevant" dependencies.
1050 For a call to the parent function, dependencies between memory
1051 references from before or after the call and from those during the call
1052 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1053 return value used in that call.
1054 The caller shares the responsibility with the callee for ensuring that
1055 these requirements are met.
1056 For further details, please see the discussion of the NoAlias response in
1057 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1059 Note that this definition of <tt>noalias</tt> is intentionally
1060 similar to the definition of <tt>restrict</tt> in C99 for function
1061 arguments, though it is slightly weaker.
1063 For function return values, C99's <tt>restrict</tt> is not meaningful,
1064 while LLVM's <tt>noalias</tt> is.
1067 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1068 <dd>This indicates that the callee does not make any copies of the pointer
1069 that outlive the callee itself. This is not a valid attribute for return
1072 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1073 <dd>This indicates that the pointer parameter can be excised using the
1074 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1075 attribute for return values.</dd>
1080 <!-- ======================================================================= -->
1081 <div class="doc_subsection">
1082 <a name="gc">Garbage Collector Names</a>
1085 <div class="doc_text">
1087 <p>Each function may specify a garbage collector name, which is simply a
1090 <pre class="doc_code">
1091 define void @f() gc "name" { ... }
1094 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1095 collector which will cause the compiler to alter its output in order to
1096 support the named garbage collection algorithm.</p>
1100 <!-- ======================================================================= -->
1101 <div class="doc_subsection">
1102 <a name="fnattrs">Function Attributes</a>
1105 <div class="doc_text">
1107 <p>Function attributes are set to communicate additional information about a
1108 function. Function attributes are considered to be part of the function, not
1109 of the function type, so functions with different parameter attributes can
1110 have the same function type.</p>
1112 <p>Function attributes are simple keywords that follow the type specified. If
1113 multiple attributes are needed, they are space separated. For example:</p>
1115 <pre class="doc_code">
1116 define void @f() noinline { ... }
1117 define void @f() alwaysinline { ... }
1118 define void @f() alwaysinline optsize { ... }
1119 define void @f() optsize { ... }
1123 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1124 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1125 the backend should forcibly align the stack pointer. Specify the
1126 desired alignment, which must be a power of two, in parentheses.
1128 <dt><tt><b>alwaysinline</b></tt></dt>
1129 <dd>This attribute indicates that the inliner should attempt to inline this
1130 function into callers whenever possible, ignoring any active inlining size
1131 threshold for this caller.</dd>
1133 <dt><tt><b>inlinehint</b></tt></dt>
1134 <dd>This attribute indicates that the source code contained a hint that inlining
1135 this function is desirable (such as the "inline" keyword in C/C++). It
1136 is just a hint; it imposes no requirements on the inliner.</dd>
1138 <dt><tt><b>naked</b></tt></dt>
1139 <dd>This attribute disables prologue / epilogue emission for the function.
1140 This can have very system-specific consequences.</dd>
1142 <dt><tt><b>noimplicitfloat</b></tt></dt>
1143 <dd>This attributes disables implicit floating point instructions.</dd>
1145 <dt><tt><b>noinline</b></tt></dt>
1146 <dd>This attribute indicates that the inliner should never inline this
1147 function in any situation. This attribute may not be used together with
1148 the <tt>alwaysinline</tt> attribute.</dd>
1150 <dt><tt><b>noredzone</b></tt></dt>
1151 <dd>This attribute indicates that the code generator should not use a red
1152 zone, even if the target-specific ABI normally permits it.</dd>
1154 <dt><tt><b>noreturn</b></tt></dt>
1155 <dd>This function attribute indicates that the function never returns
1156 normally. This produces undefined behavior at runtime if the function
1157 ever does dynamically return.</dd>
1159 <dt><tt><b>nounwind</b></tt></dt>
1160 <dd>This function attribute indicates that the function never returns with an
1161 unwind or exceptional control flow. If the function does unwind, its
1162 runtime behavior is undefined.</dd>
1164 <dt><tt><b>optsize</b></tt></dt>
1165 <dd>This attribute suggests that optimization passes and code generator passes
1166 make choices that keep the code size of this function low, and otherwise
1167 do optimizations specifically to reduce code size.</dd>
1169 <dt><tt><b>readnone</b></tt></dt>
1170 <dd>This attribute indicates that the function computes its result (or decides
1171 to unwind an exception) based strictly on its arguments, without
1172 dereferencing any pointer arguments or otherwise accessing any mutable
1173 state (e.g. memory, control registers, etc) visible to caller functions.
1174 It does not write through any pointer arguments
1175 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1176 changes any state visible to callers. This means that it cannot unwind
1177 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1178 could use the <tt>unwind</tt> instruction.</dd>
1180 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1181 <dd>This attribute indicates that the function does not write through any
1182 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1183 arguments) or otherwise modify any state (e.g. memory, control registers,
1184 etc) visible to caller functions. It may dereference pointer arguments
1185 and read state that may be set in the caller. A readonly function always
1186 returns the same value (or unwinds an exception identically) when called
1187 with the same set of arguments and global state. It cannot unwind an
1188 exception by calling the <tt>C++</tt> exception throwing methods, but may
1189 use the <tt>unwind</tt> instruction.</dd>
1191 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1192 <dd>This attribute indicates that the function should emit a stack smashing
1193 protector. It is in the form of a "canary"—a random value placed on
1194 the stack before the local variables that's checked upon return from the
1195 function to see if it has been overwritten. A heuristic is used to
1196 determine if a function needs stack protectors or not.<br>
1198 If a function that has an <tt>ssp</tt> attribute is inlined into a
1199 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1200 function will have an <tt>ssp</tt> attribute.</dd>
1202 <dt><tt><b>sspreq</b></tt></dt>
1203 <dd>This attribute indicates that the function should <em>always</em> emit a
1204 stack smashing protector. This overrides
1205 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1207 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1208 function that doesn't have an <tt>sspreq</tt> attribute or which has
1209 an <tt>ssp</tt> attribute, then the resulting function will have
1210 an <tt>sspreq</tt> attribute.</dd>
1215 <!-- ======================================================================= -->
1216 <div class="doc_subsection">
1217 <a name="moduleasm">Module-Level Inline Assembly</a>
1220 <div class="doc_text">
1222 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1223 the GCC "file scope inline asm" blocks. These blocks are internally
1224 concatenated by LLVM and treated as a single unit, but may be separated in
1225 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1227 <pre class="doc_code">
1228 module asm "inline asm code goes here"
1229 module asm "more can go here"
1232 <p>The strings can contain any character by escaping non-printable characters.
1233 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1236 <p>The inline asm code is simply printed to the machine code .s file when
1237 assembly code is generated.</p>
1241 <!-- ======================================================================= -->
1242 <div class="doc_subsection">
1243 <a name="datalayout">Data Layout</a>
1246 <div class="doc_text">
1248 <p>A module may specify a target specific data layout string that specifies how
1249 data is to be laid out in memory. The syntax for the data layout is
1252 <pre class="doc_code">
1253 target datalayout = "<i>layout specification</i>"
1256 <p>The <i>layout specification</i> consists of a list of specifications
1257 separated by the minus sign character ('-'). Each specification starts with
1258 a letter and may include other information after the letter to define some
1259 aspect of the data layout. The specifications accepted are as follows:</p>
1263 <dd>Specifies that the target lays out data in big-endian form. That is, the
1264 bits with the most significance have the lowest address location.</dd>
1267 <dd>Specifies that the target lays out data in little-endian form. That is,
1268 the bits with the least significance have the lowest address
1271 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1272 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1273 <i>preferred</i> alignments. All sizes are in bits. Specifying
1274 the <i>pref</i> alignment is optional. If omitted, the
1275 preceding <tt>:</tt> should be omitted too.</dd>
1277 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1278 <dd>This specifies the alignment for an integer type of a given bit
1279 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1281 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1282 <dd>This specifies the alignment for a vector type of a given bit
1285 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1286 <dd>This specifies the alignment for a floating point type of a given bit
1287 <i>size</i>. Only values of <i>size</i> that are supported by the target
1288 will work. 32 (float) and 64 (double) are supported on all targets;
1289 80 or 128 (different flavors of long double) are also supported on some
1292 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1293 <dd>This specifies the alignment for an aggregate type of a given bit
1296 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1297 <dd>This specifies the alignment for a stack object of a given bit
1300 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1301 <dd>This specifies a set of native integer widths for the target CPU
1302 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1303 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1304 this set are considered to support most general arithmetic
1305 operations efficiently.</dd>
1308 <p>When constructing the data layout for a given target, LLVM starts with a
1309 default set of specifications which are then (possibly) overridden by the
1310 specifications in the <tt>datalayout</tt> keyword. The default specifications
1311 are given in this list:</p>
1314 <li><tt>E</tt> - big endian</li>
1315 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1316 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1317 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1318 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1319 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1320 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1321 alignment of 64-bits</li>
1322 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1323 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1324 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1325 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1326 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1327 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1330 <p>When LLVM is determining the alignment for a given type, it uses the
1331 following rules:</p>
1334 <li>If the type sought is an exact match for one of the specifications, that
1335 specification is used.</li>
1337 <li>If no match is found, and the type sought is an integer type, then the
1338 smallest integer type that is larger than the bitwidth of the sought type
1339 is used. If none of the specifications are larger than the bitwidth then
1340 the the largest integer type is used. For example, given the default
1341 specifications above, the i7 type will use the alignment of i8 (next
1342 largest) while both i65 and i256 will use the alignment of i64 (largest
1345 <li>If no match is found, and the type sought is a vector type, then the
1346 largest vector type that is smaller than the sought vector type will be
1347 used as a fall back. This happens because <128 x double> can be
1348 implemented in terms of 64 <2 x double>, for example.</li>
1353 <!-- ======================================================================= -->
1354 <div class="doc_subsection">
1355 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1358 <div class="doc_text">
1360 <p>Any memory access must be done through a pointer value associated
1361 with an address range of the memory access, otherwise the behavior
1362 is undefined. Pointer values are associated with address ranges
1363 according to the following rules:</p>
1366 <li>A pointer value is associated with the addresses associated with
1367 any value it is <i>based</i> on.
1368 <li>An address of a global variable is associated with the address
1369 range of the variable's storage.</li>
1370 <li>The result value of an allocation instruction is associated with
1371 the address range of the allocated storage.</li>
1372 <li>A null pointer in the default address-space is associated with
1374 <li>An integer constant other than zero or a pointer value returned
1375 from a function not defined within LLVM may be associated with address
1376 ranges allocated through mechanisms other than those provided by
1377 LLVM. Such ranges shall not overlap with any ranges of addresses
1378 allocated by mechanisms provided by LLVM.</li>
1381 <p>A pointer value is <i>based</i> on another pointer value according
1382 to the following rules:</p>
1385 <li>A pointer value formed from a
1386 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1387 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1388 <li>The result value of a
1389 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1390 of the <tt>bitcast</tt>.</li>
1391 <li>A pointer value formed by an
1392 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1393 pointer values that contribute (directly or indirectly) to the
1394 computation of the pointer's value.</li>
1395 <li>The "<i>based</i> on" relationship is transitive.</li>
1398 <p>Note that this definition of <i>"based"</i> is intentionally
1399 similar to the definition of <i>"based"</i> in C99, though it is
1400 slightly weaker.</p>
1402 <p>LLVM IR does not associate types with memory. The result type of a
1403 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1404 alignment of the memory from which to load, as well as the
1405 interpretation of the value. The first operand type of a
1406 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1407 and alignment of the store.</p>
1409 <p>Consequently, type-based alias analysis, aka TBAA, aka
1410 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1411 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1412 additional information which specialized optimization passes may use
1413 to implement type-based alias analysis.</p>
1417 <!-- ======================================================================= -->
1418 <div class="doc_subsection">
1419 <a name="volatile">Volatile Memory Accesses</a>
1422 <div class="doc_text">
1424 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1425 href="#i_store"><tt>store</tt></a>s, and <a
1426 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1427 The optimizers must not change the number of volatile operations or change their
1428 order of execution relative to other volatile operations. The optimizers
1429 <i>may</i> change the order of volatile operations relative to non-volatile
1430 operations. This is not Java's "volatile" and has no cross-thread
1431 synchronization behavior.</p>
1435 <!-- *********************************************************************** -->
1436 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1437 <!-- *********************************************************************** -->
1439 <div class="doc_text">
1441 <p>The LLVM type system is one of the most important features of the
1442 intermediate representation. Being typed enables a number of optimizations
1443 to be performed on the intermediate representation directly, without having
1444 to do extra analyses on the side before the transformation. A strong type
1445 system makes it easier to read the generated code and enables novel analyses
1446 and transformations that are not feasible to perform on normal three address
1447 code representations.</p>
1451 <!-- ======================================================================= -->
1452 <div class="doc_subsection"> <a name="t_classifications">Type
1453 Classifications</a> </div>
1455 <div class="doc_text">
1457 <p>The types fall into a few useful classifications:</p>
1459 <table border="1" cellspacing="0" cellpadding="4">
1461 <tr><th>Classification</th><th>Types</th></tr>
1463 <td><a href="#t_integer">integer</a></td>
1464 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1467 <td><a href="#t_floating">floating point</a></td>
1468 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1471 <td><a name="t_firstclass">first class</a></td>
1472 <td><a href="#t_integer">integer</a>,
1473 <a href="#t_floating">floating point</a>,
1474 <a href="#t_pointer">pointer</a>,
1475 <a href="#t_vector">vector</a>,
1476 <a href="#t_struct">structure</a>,
1477 <a href="#t_array">array</a>,
1478 <a href="#t_label">label</a>,
1479 <a href="#t_metadata">metadata</a>.
1483 <td><a href="#t_primitive">primitive</a></td>
1484 <td><a href="#t_label">label</a>,
1485 <a href="#t_void">void</a>,
1486 <a href="#t_floating">floating point</a>,
1487 <a href="#t_metadata">metadata</a>.</td>
1490 <td><a href="#t_derived">derived</a></td>
1491 <td><a href="#t_array">array</a>,
1492 <a href="#t_function">function</a>,
1493 <a href="#t_pointer">pointer</a>,
1494 <a href="#t_struct">structure</a>,
1495 <a href="#t_pstruct">packed structure</a>,
1496 <a href="#t_vector">vector</a>,
1497 <a href="#t_opaque">opaque</a>.
1503 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1504 important. Values of these types are the only ones which can be produced by
1509 <!-- ======================================================================= -->
1510 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1512 <div class="doc_text">
1514 <p>The primitive types are the fundamental building blocks of the LLVM
1519 <!-- _______________________________________________________________________ -->
1520 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1522 <div class="doc_text">
1525 <p>The integer type is a very simple type that simply specifies an arbitrary
1526 bit width for the integer type desired. Any bit width from 1 bit to
1527 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1534 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1538 <table class="layout">
1540 <td class="left"><tt>i1</tt></td>
1541 <td class="left">a single-bit integer.</td>
1544 <td class="left"><tt>i32</tt></td>
1545 <td class="left">a 32-bit integer.</td>
1548 <td class="left"><tt>i1942652</tt></td>
1549 <td class="left">a really big integer of over 1 million bits.</td>
1555 <!-- _______________________________________________________________________ -->
1556 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1558 <div class="doc_text">
1562 <tr><th>Type</th><th>Description</th></tr>
1563 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1564 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1565 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1566 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1567 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1573 <!-- _______________________________________________________________________ -->
1574 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1576 <div class="doc_text">
1579 <p>The void type does not represent any value and has no size.</p>
1588 <!-- _______________________________________________________________________ -->
1589 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1591 <div class="doc_text">
1594 <p>The label type represents code labels.</p>
1603 <!-- _______________________________________________________________________ -->
1604 <div class="doc_subsubsection"> <a name="t_metadata">Metadata Type</a> </div>
1606 <div class="doc_text">
1609 <p>The metadata type represents embedded metadata. No derived types may be
1610 created from metadata except for <a href="#t_function">function</a>
1621 <!-- ======================================================================= -->
1622 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1624 <div class="doc_text">
1626 <p>The real power in LLVM comes from the derived types in the system. This is
1627 what allows a programmer to represent arrays, functions, pointers, and other
1628 useful types. Each of these types contain one or more element types which
1629 may be a primitive type, or another derived type. For example, it is
1630 possible to have a two dimensional array, using an array as the element type
1631 of another array.</p>
1636 <!-- _______________________________________________________________________ -->
1637 <div class="doc_subsubsection"> <a name="t_aggregate">Aggregate Types</a> </div>
1639 <div class="doc_text">
1641 <p>Aggregate Types are a subset of derived types that can contain multiple
1642 member types. <a href="#t_array">Arrays</a>,
1643 <a href="#t_struct">structs</a>, and <a href="#t_vector">vectors</a> are
1644 aggregate types.</p>
1648 <!-- _______________________________________________________________________ -->
1649 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1651 <div class="doc_text">
1654 <p>The array type is a very simple derived type that arranges elements
1655 sequentially in memory. The array type requires a size (number of elements)
1656 and an underlying data type.</p>
1660 [<# elements> x <elementtype>]
1663 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1664 be any type with a size.</p>
1667 <table class="layout">
1669 <td class="left"><tt>[40 x i32]</tt></td>
1670 <td class="left">Array of 40 32-bit integer values.</td>
1673 <td class="left"><tt>[41 x i32]</tt></td>
1674 <td class="left">Array of 41 32-bit integer values.</td>
1677 <td class="left"><tt>[4 x i8]</tt></td>
1678 <td class="left">Array of 4 8-bit integer values.</td>
1681 <p>Here are some examples of multidimensional arrays:</p>
1682 <table class="layout">
1684 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1685 <td class="left">3x4 array of 32-bit integer values.</td>
1688 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1689 <td class="left">12x10 array of single precision floating point values.</td>
1692 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1693 <td class="left">2x3x4 array of 16-bit integer values.</td>
1697 <p>There is no restriction on indexing beyond the end of the array implied by
1698 a static type (though there are restrictions on indexing beyond the bounds
1699 of an allocated object in some cases). This means that single-dimension
1700 'variable sized array' addressing can be implemented in LLVM with a zero
1701 length array type. An implementation of 'pascal style arrays' in LLVM could
1702 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1706 <!-- _______________________________________________________________________ -->
1707 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1709 <div class="doc_text">
1712 <p>The function type can be thought of as a function signature. It consists of
1713 a return type and a list of formal parameter types. The return type of a
1714 function type is a first class type or a void type.</p>
1718 <returntype> (<parameter list>)
1721 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1722 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1723 which indicates that the function takes a variable number of arguments.
1724 Variable argument functions can access their arguments with
1725 the <a href="#int_varargs">variable argument handling intrinsic</a>
1726 functions. '<tt><returntype></tt>' is any type except
1727 <a href="#t_label">label</a>.</p>
1730 <table class="layout">
1732 <td class="left"><tt>i32 (i32)</tt></td>
1733 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1735 </tr><tr class="layout">
1736 <td class="left"><tt>float (i16, i32 *) *
1738 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1739 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
1740 returning <tt>float</tt>.
1742 </tr><tr class="layout">
1743 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1744 <td class="left">A vararg function that takes at least one
1745 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1746 which returns an integer. This is the signature for <tt>printf</tt> in
1749 </tr><tr class="layout">
1750 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1751 <td class="left">A function taking an <tt>i32</tt>, returning a
1752 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
1759 <!-- _______________________________________________________________________ -->
1760 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1762 <div class="doc_text">
1765 <p>The structure type is used to represent a collection of data members together
1766 in memory. The packing of the field types is defined to match the ABI of the
1767 underlying processor. The elements of a structure may be any type that has a
1770 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
1771 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
1772 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
1773 Structures in registers are accessed using the
1774 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
1775 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
1778 { <type list> }
1782 <table class="layout">
1784 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1785 <td class="left">A triple of three <tt>i32</tt> values</td>
1786 </tr><tr class="layout">
1787 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1788 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1789 second element is a <a href="#t_pointer">pointer</a> to a
1790 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1791 an <tt>i32</tt>.</td>
1797 <!-- _______________________________________________________________________ -->
1798 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1801 <div class="doc_text">
1804 <p>The packed structure type is used to represent a collection of data members
1805 together in memory. There is no padding between fields. Further, the
1806 alignment of a packed structure is 1 byte. The elements of a packed
1807 structure may be any type that has a size.</p>
1809 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1810 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1811 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1815 < { <type list> } >
1819 <table class="layout">
1821 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1822 <td class="left">A triple of three <tt>i32</tt> values</td>
1823 </tr><tr class="layout">
1825 <tt>< { float, i32 (i32)* } ></tt></td>
1826 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1827 second element is a <a href="#t_pointer">pointer</a> to a
1828 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1829 an <tt>i32</tt>.</td>
1835 <!-- _______________________________________________________________________ -->
1836 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1838 <div class="doc_text">
1841 <p>The pointer type is used to specify memory locations.
1842 Pointers are commonly used to reference objects in memory.</p>
1844 <p>Pointer types may have an optional address space attribute defining the
1845 numbered address space where the pointed-to object resides. The default
1846 address space is number zero. The semantics of non-zero address
1847 spaces are target-specific.</p>
1849 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
1850 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
1858 <table class="layout">
1860 <td class="left"><tt>[4 x i32]*</tt></td>
1861 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1862 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1865 <td class="left"><tt>i32 (i32*) *</tt></td>
1866 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1867 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1871 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1872 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1873 that resides in address space #5.</td>
1879 <!-- _______________________________________________________________________ -->
1880 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1882 <div class="doc_text">
1885 <p>A vector type is a simple derived type that represents a vector of elements.
1886 Vector types are used when multiple primitive data are operated in parallel
1887 using a single instruction (SIMD). A vector type requires a size (number of
1888 elements) and an underlying primitive data type. Vector types are considered
1889 <a href="#t_firstclass">first class</a>.</p>
1893 < <# elements> x <elementtype> >
1896 <p>The number of elements is a constant integer value; elementtype may be any
1897 integer or floating point type.</p>
1900 <table class="layout">
1902 <td class="left"><tt><4 x i32></tt></td>
1903 <td class="left">Vector of 4 32-bit integer values.</td>
1906 <td class="left"><tt><8 x float></tt></td>
1907 <td class="left">Vector of 8 32-bit floating-point values.</td>
1910 <td class="left"><tt><2 x i64></tt></td>
1911 <td class="left">Vector of 2 64-bit integer values.</td>
1917 <!-- _______________________________________________________________________ -->
1918 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1919 <div class="doc_text">
1922 <p>Opaque types are used to represent unknown types in the system. This
1923 corresponds (for example) to the C notion of a forward declared structure
1924 type. In LLVM, opaque types can eventually be resolved to any type (not just
1925 a structure type).</p>
1933 <table class="layout">
1935 <td class="left"><tt>opaque</tt></td>
1936 <td class="left">An opaque type.</td>
1942 <!-- ======================================================================= -->
1943 <div class="doc_subsection">
1944 <a name="t_uprefs">Type Up-references</a>
1947 <div class="doc_text">
1950 <p>An "up reference" allows you to refer to a lexically enclosing type without
1951 requiring it to have a name. For instance, a structure declaration may
1952 contain a pointer to any of the types it is lexically a member of. Example
1953 of up references (with their equivalent as named type declarations)
1957 { \2 * } %x = type { %x* }
1958 { \2 }* %y = type { %y }*
1962 <p>An up reference is needed by the asmprinter for printing out cyclic types
1963 when there is no declared name for a type in the cycle. Because the
1964 asmprinter does not want to print out an infinite type string, it needs a
1965 syntax to handle recursive types that have no names (all names are optional
1973 <p>The level is the count of the lexical type that is being referred to.</p>
1976 <table class="layout">
1978 <td class="left"><tt>\1*</tt></td>
1979 <td class="left">Self-referential pointer.</td>
1982 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1983 <td class="left">Recursive structure where the upref refers to the out-most
1990 <!-- *********************************************************************** -->
1991 <div class="doc_section"> <a name="constants">Constants</a> </div>
1992 <!-- *********************************************************************** -->
1994 <div class="doc_text">
1996 <p>LLVM has several different basic types of constants. This section describes
1997 them all and their syntax.</p>
2001 <!-- ======================================================================= -->
2002 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
2004 <div class="doc_text">
2007 <dt><b>Boolean constants</b></dt>
2008 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2009 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2011 <dt><b>Integer constants</b></dt>
2012 <dd>Standard integers (such as '4') are constants of
2013 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2014 with integer types.</dd>
2016 <dt><b>Floating point constants</b></dt>
2017 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2018 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2019 notation (see below). The assembler requires the exact decimal value of a
2020 floating-point constant. For example, the assembler accepts 1.25 but
2021 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2022 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2024 <dt><b>Null pointer constants</b></dt>
2025 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2026 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2029 <p>The one non-intuitive notation for constants is the hexadecimal form of
2030 floating point constants. For example, the form '<tt>double
2031 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2032 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2033 constants are required (and the only time that they are generated by the
2034 disassembler) is when a floating point constant must be emitted but it cannot
2035 be represented as a decimal floating point number in a reasonable number of
2036 digits. For example, NaN's, infinities, and other special values are
2037 represented in their IEEE hexadecimal format so that assembly and disassembly
2038 do not cause any bits to change in the constants.</p>
2040 <p>When using the hexadecimal form, constants of types float and double are
2041 represented using the 16-digit form shown above (which matches the IEEE754
2042 representation for double); float values must, however, be exactly
2043 representable as IEE754 single precision. Hexadecimal format is always used
2044 for long double, and there are three forms of long double. The 80-bit format
2045 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2046 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2047 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2048 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2049 currently supported target uses this format. Long doubles will only work if
2050 they match the long double format on your target. All hexadecimal formats
2051 are big-endian (sign bit at the left).</p>
2055 <!-- ======================================================================= -->
2056 <div class="doc_subsection">
2057 <a name="aggregateconstants"></a> <!-- old anchor -->
2058 <a name="complexconstants">Complex Constants</a>
2061 <div class="doc_text">
2063 <p>Complex constants are a (potentially recursive) combination of simple
2064 constants and smaller complex constants.</p>
2067 <dt><b>Structure constants</b></dt>
2068 <dd>Structure constants are represented with notation similar to structure
2069 type definitions (a comma separated list of elements, surrounded by braces
2070 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2071 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2072 Structure constants must have <a href="#t_struct">structure type</a>, and
2073 the number and types of elements must match those specified by the
2076 <dt><b>Array constants</b></dt>
2077 <dd>Array constants are represented with notation similar to array type
2078 definitions (a comma separated list of elements, surrounded by square
2079 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2080 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2081 the number and types of elements must match those specified by the
2084 <dt><b>Vector constants</b></dt>
2085 <dd>Vector constants are represented with notation similar to vector type
2086 definitions (a comma separated list of elements, surrounded by
2087 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2088 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2089 have <a href="#t_vector">vector type</a>, and the number and types of
2090 elements must match those specified by the type.</dd>
2092 <dt><b>Zero initialization</b></dt>
2093 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2094 value to zero of <em>any</em> type, including scalar and
2095 <a href="#t_aggregate">aggregate</a> types.
2096 This is often used to avoid having to print large zero initializers
2097 (e.g. for large arrays) and is always exactly equivalent to using explicit
2098 zero initializers.</dd>
2100 <dt><b>Metadata node</b></dt>
2101 <dd>A metadata node is a structure-like constant with
2102 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2103 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2104 be interpreted as part of the instruction stream, metadata is a place to
2105 attach additional information such as debug info.</dd>
2110 <!-- ======================================================================= -->
2111 <div class="doc_subsection">
2112 <a name="globalconstants">Global Variable and Function Addresses</a>
2115 <div class="doc_text">
2117 <p>The addresses of <a href="#globalvars">global variables</a>
2118 and <a href="#functionstructure">functions</a> are always implicitly valid
2119 (link-time) constants. These constants are explicitly referenced when
2120 the <a href="#identifiers">identifier for the global</a> is used and always
2121 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2122 legal LLVM file:</p>
2124 <pre class="doc_code">
2127 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2132 <!-- ======================================================================= -->
2133 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
2134 <div class="doc_text">
2136 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2137 indicates that the user of the value may receive an unspecified bit-pattern.
2138 Undefined values may be of any type (other than label or void) and be used
2139 anywhere a constant is permitted.</p>
2141 <p>Undefined values are useful because they indicate to the compiler that the
2142 program is well defined no matter what value is used. This gives the
2143 compiler more freedom to optimize. Here are some examples of (potentially
2144 surprising) transformations that are valid (in pseudo IR):</p>
2147 <pre class="doc_code">
2157 <p>This is safe because all of the output bits are affected by the undef bits.
2158 Any output bit can have a zero or one depending on the input bits.</p>
2160 <pre class="doc_code">
2171 <p>These logical operations have bits that are not always affected by the input.
2172 For example, if "%X" has a zero bit, then the output of the 'and' operation will
2173 always be a zero, no matter what the corresponding bit from the undef is. As
2174 such, it is unsafe to optimize or assume that the result of the and is undef.
2175 However, it is safe to assume that all bits of the undef could be 0, and
2176 optimize the and to 0. Likewise, it is safe to assume that all the bits of
2177 the undef operand to the or could be set, allowing the or to be folded to
2180 <pre class="doc_code">
2181 %A = select undef, %X, %Y
2182 %B = select undef, 42, %Y
2183 %C = select %X, %Y, undef
2194 <p>This set of examples show that undefined select (and conditional branch)
2195 conditions can go "either way" but they have to come from one of the two
2196 operands. In the %A example, if %X and %Y were both known to have a clear low
2197 bit, then %A would have to have a cleared low bit. However, in the %C example,
2198 the optimizer is allowed to assume that the undef operand could be the same as
2199 %Y, allowing the whole select to be eliminated.</p>
2202 <pre class="doc_code">
2203 %A = xor undef, undef
2221 <p>This example points out that two undef operands are not necessarily the same.
2222 This can be surprising to people (and also matches C semantics) where they
2223 assume that "X^X" is always zero, even if X is undef. This isn't true for a
2224 number of reasons, but the short answer is that an undef "variable" can
2225 arbitrarily change its value over its "live range". This is true because the
2226 "variable" doesn't actually <em>have a live range</em>. Instead, the value is
2227 logically read from arbitrary registers that happen to be around when needed,
2228 so the value is not necessarily consistent over time. In fact, %A and %C need
2229 to have the same semantics or the core LLVM "replace all uses with" concept
2232 <pre class="doc_code">
2240 <p>These examples show the crucial difference between an <em>undefined
2241 value</em> and <em>undefined behavior</em>. An undefined value (like undef) is
2242 allowed to have an arbitrary bit-pattern. This means that the %A operation
2243 can be constant folded to undef because the undef could be an SNaN, and fdiv is
2244 not (currently) defined on SNaN's. However, in the second example, we can make
2245 a more aggressive assumption: because the undef is allowed to be an arbitrary
2246 value, we are allowed to assume that it could be zero. Since a divide by zero
2247 has <em>undefined behavior</em>, we are allowed to assume that the operation
2248 does not execute at all. This allows us to delete the divide and all code after
2249 it: since the undefined operation "can't happen", the optimizer can assume that
2250 it occurs in dead code.
2253 <pre class="doc_code">
2254 a: store undef -> %X
2255 b: store %X -> undef
2261 <p>These examples reiterate the fdiv example: a store "of" an undefined value
2262 can be assumed to not have any effect: we can assume that the value is
2263 overwritten with bits that happen to match what was already there. However, a
2264 store "to" an undefined location could clobber arbitrary memory, therefore, it
2265 has undefined behavior.</p>
2269 <!-- ======================================================================= -->
2270 <div class="doc_subsection"><a name="trapvalues">Trap Values</a></div>
2271 <div class="doc_text">
2273 <p>Trap values are similar to <a href="#undefvalues">undef values</a>, however
2274 instead of representing an unspecified bit pattern, they represent the
2275 fact that an instruction or constant expression which cannot evoke side
2276 effects has nevertheless detected a condition which results in undefined
2279 <p>There is currently no way of representing a trap value in the IR; they
2280 only exist when produced by operations such as
2281 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2283 <p>Trap value behavior is defined in terms of value <i>dependence</i>:</p>
2286 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2287 their operands.</li>
2289 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2290 to their dynamic predecessor basic block.</li>
2292 <li>Function arguments depend on the corresponding actual argument values in
2293 the dynamic callers of their functions.</li>
2295 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2296 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2297 control back to them.</li>
2299 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2300 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_unwind"><tt>unwind</tt></a>,
2301 or exception-throwing call instructions that dynamically transfer control
2304 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2305 referenced memory addresses, following the order in the IR
2306 (including loads and stores implied by intrinsics such as
2307 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2309 <!-- TODO: In the case of multiple threads, this only applies if the store
2310 "happens-before" the load or store. -->
2312 <!-- TODO: floating-point exception state -->
2314 <li>An instruction with externally visible side effects depends on the most
2315 recent preceding instruction with externally visible side effects, following
2316 the order in the IR. (This includes
2317 <a href="#volatile">volatile operations</a>.)</li>
2319 <li>An instruction <i>control-depends</i> on a
2320 <a href="#terminators">terminator instruction</a>
2321 if the terminator instruction has multiple successors and the instruction
2322 is always executed when control transfers to one of the successors, and
2323 may not be executed when control is transfered to another.</li>
2325 <li>Dependence is transitive.</li>
2329 <p>Whenever a trap value is generated, all values which depend on it evaluate
2330 to trap. If they have side effects, the evoke their side effects as if each
2331 operand with a trap value were undef. If they have externally-visible side
2332 effects, the behavior is undefined.</p>
2334 <p>Here are some examples:</p>
2336 <pre class="doc_code">
2338 %trap = sub nuw i32 0, 1 ; Results in a trap value.
2339 %still_trap = and i32 %trap, 0 ; Whereas (and i32 undef, 0) would return 0.
2340 %trap_yet_again = getelementptr i32* @h, i32 %still_trap
2341 store i32 0, i32* %trap_yet_again ; undefined behavior
2343 store i32 %trap, i32* @g ; Trap value conceptually stored to memory.
2344 %trap2 = load i32* @g ; Returns a trap value, not just undef.
2346 volatile store i32 %trap, i32* @g ; External observation; undefined behavior.
2348 %narrowaddr = bitcast i32* @g to i16*
2349 %wideaddr = bitcast i32* @g to i64*
2350 %trap3 = load 16* %narrowaddr ; Returns a trap value.
2351 %trap4 = load i64* %widaddr ; Returns a trap value.
2353 %cmp = icmp i32 slt %trap, 0 ; Returns a trap value.
2354 %br i1 %cmp, %true, %end ; Branch to either destination.
2357 volatile store i32 0, i32* @g ; This is control-dependent on %cmp, so
2358 ; it has undefined behavior.
2362 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2363 ; Both edges into this PHI are
2364 ; control-dependent on %cmp, so this
2365 ; always results in a trap value.
2367 volatile store i32 0, i32* @g ; %end is control-equivalent to %entry
2368 ; so this is defined (ignoring earlier
2369 ; undefined behavior in this example).
2374 <!-- ======================================================================= -->
2375 <div class="doc_subsection"><a name="blockaddress">Addresses of Basic
2377 <div class="doc_text">
2379 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2381 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2382 basic block in the specified function, and always has an i8* type. Taking
2383 the address of the entry block is illegal.</p>
2385 <p>This value only has defined behavior when used as an operand to the
2386 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction or for comparisons
2387 against null. Pointer equality tests between labels addresses is undefined
2388 behavior - though, again, comparison against null is ok, and no label is
2389 equal to the null pointer. This may also be passed around as an opaque
2390 pointer sized value as long as the bits are not inspected. This allows
2391 <tt>ptrtoint</tt> and arithmetic to be performed on these values so long as
2392 the original value is reconstituted before the <tt>indirectbr</tt>.</p>
2394 <p>Finally, some targets may provide defined semantics when
2395 using the value as the operand to an inline assembly, but that is target
2402 <!-- ======================================================================= -->
2403 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
2406 <div class="doc_text">
2408 <p>Constant expressions are used to allow expressions involving other constants
2409 to be used as constants. Constant expressions may be of
2410 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2411 operation that does not have side effects (e.g. load and call are not
2412 supported). The following is the syntax for constant expressions:</p>
2415 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2416 <dd>Truncate a constant to another type. The bit size of CST must be larger
2417 than the bit size of TYPE. Both types must be integers.</dd>
2419 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2420 <dd>Zero extend a constant to another type. The bit size of CST must be
2421 smaller than the bit size of TYPE. Both types must be integers.</dd>
2423 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2424 <dd>Sign extend a constant to another type. The bit size of CST must be
2425 smaller than the bit size of TYPE. Both types must be integers.</dd>
2427 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2428 <dd>Truncate a floating point constant to another floating point type. The
2429 size of CST must be larger than the size of TYPE. Both types must be
2430 floating point.</dd>
2432 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2433 <dd>Floating point extend a constant to another type. The size of CST must be
2434 smaller or equal to the size of TYPE. Both types must be floating
2437 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2438 <dd>Convert a floating point constant to the corresponding unsigned integer
2439 constant. TYPE must be a scalar or vector integer type. CST must be of
2440 scalar or vector floating point type. Both CST and TYPE must be scalars,
2441 or vectors of the same number of elements. If the value won't fit in the
2442 integer type, the results are undefined.</dd>
2444 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2445 <dd>Convert a floating point constant to the corresponding signed integer
2446 constant. TYPE must be a scalar or vector integer type. CST must be of
2447 scalar or vector floating point type. Both CST and TYPE must be scalars,
2448 or vectors of the same number of elements. If the value won't fit in the
2449 integer type, the results are undefined.</dd>
2451 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2452 <dd>Convert an unsigned integer constant to the corresponding floating point
2453 constant. TYPE must be a scalar or vector floating point type. CST must be
2454 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2455 vectors of the same number of elements. If the value won't fit in the
2456 floating point type, the results are undefined.</dd>
2458 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2459 <dd>Convert a signed integer constant to the corresponding floating point
2460 constant. TYPE must be a scalar or vector floating point type. CST must be
2461 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2462 vectors of the same number of elements. If the value won't fit in the
2463 floating point type, the results are undefined.</dd>
2465 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2466 <dd>Convert a pointer typed constant to the corresponding integer constant
2467 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2468 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2469 make it fit in <tt>TYPE</tt>.</dd>
2471 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2472 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2473 type. CST must be of integer type. The CST value is zero extended,
2474 truncated, or unchanged to make it fit in a pointer size. This one is
2475 <i>really</i> dangerous!</dd>
2477 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2478 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2479 are the same as those for the <a href="#i_bitcast">bitcast
2480 instruction</a>.</dd>
2482 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2483 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2484 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2485 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2486 instruction, the index list may have zero or more indexes, which are
2487 required to make sense for the type of "CSTPTR".</dd>
2489 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2490 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2492 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2493 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2495 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2496 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2498 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2499 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2502 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2503 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2506 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2507 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2510 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2511 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2512 constants. The index list is interpreted in a similar manner as indices in
2513 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2514 index value must be specified.</dd>
2516 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2517 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2518 constants. The index list is interpreted in a similar manner as indices in
2519 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2520 index value must be specified.</dd>
2522 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2523 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2524 be any of the <a href="#binaryops">binary</a>
2525 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2526 on operands are the same as those for the corresponding instruction
2527 (e.g. no bitwise operations on floating point values are allowed).</dd>
2532 <!-- *********************************************************************** -->
2533 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2534 <!-- *********************************************************************** -->
2536 <!-- ======================================================================= -->
2537 <div class="doc_subsection">
2538 <a name="inlineasm">Inline Assembler Expressions</a>
2541 <div class="doc_text">
2543 <p>LLVM supports inline assembler expressions (as opposed
2544 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2545 a special value. This value represents the inline assembler as a string
2546 (containing the instructions to emit), a list of operand constraints (stored
2547 as a string), a flag that indicates whether or not the inline asm
2548 expression has side effects, and a flag indicating whether the function
2549 containing the asm needs to align its stack conservatively. An example
2550 inline assembler expression is:</p>
2552 <pre class="doc_code">
2553 i32 (i32) asm "bswap $0", "=r,r"
2556 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2557 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2560 <pre class="doc_code">
2561 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2564 <p>Inline asms with side effects not visible in the constraint list must be
2565 marked as having side effects. This is done through the use of the
2566 '<tt>sideeffect</tt>' keyword, like so:</p>
2568 <pre class="doc_code">
2569 call void asm sideeffect "eieio", ""()
2572 <p>In some cases inline asms will contain code that will not work unless the
2573 stack is aligned in some way, such as calls or SSE instructions on x86,
2574 yet will not contain code that does that alignment within the asm.
2575 The compiler should make conservative assumptions about what the asm might
2576 contain and should generate its usual stack alignment code in the prologue
2577 if the '<tt>alignstack</tt>' keyword is present:</p>
2579 <pre class="doc_code">
2580 call void asm alignstack "eieio", ""()
2583 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2586 <p>TODO: The format of the asm and constraints string still need to be
2587 documented here. Constraints on what can be done (e.g. duplication, moving,
2588 etc need to be documented). This is probably best done by reference to
2589 another document that covers inline asm from a holistic perspective.</p>
2592 <div class="doc_subsubsection">
2593 <a name="inlineasm_md">Inline Asm Metadata</a>
2596 <div class="doc_text">
2598 <p>The call instructions that wrap inline asm nodes may have a "!srcloc" MDNode
2599 attached to it that contains a constant integer. If present, the code
2600 generator will use the integer as the location cookie value when report
2601 errors through the LLVMContext error reporting mechanisms. This allows a
2602 front-end to correlate backend errors that occur with inline asm back to the
2603 source code that produced it. For example:</p>
2605 <pre class="doc_code">
2606 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2608 !42 = !{ i32 1234567 }
2611 <p>It is up to the front-end to make sense of the magic numbers it places in the
2616 <!-- ======================================================================= -->
2617 <div class="doc_subsection"><a name="metadata">Metadata Nodes and Metadata
2621 <div class="doc_text">
2623 <p>LLVM IR allows metadata to be attached to instructions in the program that
2624 can convey extra information about the code to the optimizers and code
2625 generator. One example application of metadata is source-level debug
2626 information. There are two metadata primitives: strings and nodes. All
2627 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2628 preceding exclamation point ('<tt>!</tt>').</p>
2630 <p>A metadata string is a string surrounded by double quotes. It can contain
2631 any character by escaping non-printable characters with "\xx" where "xx" is
2632 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2634 <p>Metadata nodes are represented with notation similar to structure constants
2635 (a comma separated list of elements, surrounded by braces and preceded by an
2636 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2637 10}</tt>". Metadata nodes can have any values as their operand.</p>
2639 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2640 metadata nodes, which can be looked up in the module symbol table. For
2641 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2643 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2644 function is using two metadata arguments.</p>
2646 <pre class="doc_code">
2647 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2650 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2651 attached with <tt>add</tt> instruction using <tt>!dbg</tt> identifier.</p>
2653 <pre class="doc_code">
2654 %indvar.next = add i64 %indvar, 1, !dbg !21
2659 <!-- *********************************************************************** -->
2660 <div class="doc_section">
2661 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2663 <!-- *********************************************************************** -->
2665 <p>LLVM has a number of "magic" global variables that contain data that affect
2666 code generation or other IR semantics. These are documented here. All globals
2667 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2668 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2671 <!-- ======================================================================= -->
2672 <div class="doc_subsection">
2673 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2676 <div class="doc_text">
2678 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2679 href="#linkage_appending">appending linkage</a>. This array contains a list of
2680 pointers to global variables and functions which may optionally have a pointer
2681 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2687 @llvm.used = appending global [2 x i8*] [
2689 i8* bitcast (i32* @Y to i8*)
2690 ], section "llvm.metadata"
2693 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2694 compiler, assembler, and linker are required to treat the symbol as if there is
2695 a reference to the global that it cannot see. For example, if a variable has
2696 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2697 list, it cannot be deleted. This is commonly used to represent references from
2698 inline asms and other things the compiler cannot "see", and corresponds to
2699 "attribute((used))" in GNU C.</p>
2701 <p>On some targets, the code generator must emit a directive to the assembler or
2702 object file to prevent the assembler and linker from molesting the symbol.</p>
2706 <!-- ======================================================================= -->
2707 <div class="doc_subsection">
2708 <a name="intg_compiler_used">The '<tt>llvm.compiler.used</tt>' Global Variable</a>
2711 <div class="doc_text">
2713 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2714 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2715 touching the symbol. On targets that support it, this allows an intelligent
2716 linker to optimize references to the symbol without being impeded as it would be
2717 by <tt>@llvm.used</tt>.</p>
2719 <p>This is a rare construct that should only be used in rare circumstances, and
2720 should not be exposed to source languages.</p>
2724 <!-- ======================================================================= -->
2725 <div class="doc_subsection">
2726 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2729 <div class="doc_text">
2731 %0 = type { i32, void ()* }
2732 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2734 <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.
2739 <!-- ======================================================================= -->
2740 <div class="doc_subsection">
2741 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
2744 <div class="doc_text">
2746 %0 = type { i32, void ()* }
2747 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
2750 <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.
2756 <!-- *********************************************************************** -->
2757 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2758 <!-- *********************************************************************** -->
2760 <div class="doc_text">
2762 <p>The LLVM instruction set consists of several different classifications of
2763 instructions: <a href="#terminators">terminator
2764 instructions</a>, <a href="#binaryops">binary instructions</a>,
2765 <a href="#bitwiseops">bitwise binary instructions</a>,
2766 <a href="#memoryops">memory instructions</a>, and
2767 <a href="#otherops">other instructions</a>.</p>
2771 <!-- ======================================================================= -->
2772 <div class="doc_subsection"> <a name="terminators">Terminator
2773 Instructions</a> </div>
2775 <div class="doc_text">
2777 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
2778 in a program ends with a "Terminator" instruction, which indicates which
2779 block should be executed after the current block is finished. These
2780 terminator instructions typically yield a '<tt>void</tt>' value: they produce
2781 control flow, not values (the one exception being the
2782 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2784 <p>There are seven different terminator instructions: the
2785 '<a href="#i_ret"><tt>ret</tt></a>' instruction, the
2786 '<a href="#i_br"><tt>br</tt></a>' instruction, the
2787 '<a href="#i_switch"><tt>switch</tt></a>' instruction, the
2788 '<a href="#i_indirectbr">'<tt>indirectbr</tt></a>' Instruction, the
2789 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the
2790 '<a href="#i_unwind"><tt>unwind</tt></a>' instruction, and the
2791 '<a href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2795 <!-- _______________________________________________________________________ -->
2796 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2797 Instruction</a> </div>
2799 <div class="doc_text">
2803 ret <type> <value> <i>; Return a value from a non-void function</i>
2804 ret void <i>; Return from void function</i>
2808 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
2809 a value) from a function back to the caller.</p>
2811 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
2812 value and then causes control flow, and one that just causes control flow to
2816 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
2817 return value. The type of the return value must be a
2818 '<a href="#t_firstclass">first class</a>' type.</p>
2820 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
2821 non-void return type and contains a '<tt>ret</tt>' instruction with no return
2822 value or a return value with a type that does not match its type, or if it
2823 has a void return type and contains a '<tt>ret</tt>' instruction with a
2827 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
2828 the calling function's context. If the caller is a
2829 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
2830 instruction after the call. If the caller was an
2831 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
2832 the beginning of the "normal" destination block. If the instruction returns
2833 a value, that value shall set the call or invoke instruction's return
2838 ret i32 5 <i>; Return an integer value of 5</i>
2839 ret void <i>; Return from a void function</i>
2840 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
2844 <!-- _______________________________________________________________________ -->
2845 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2847 <div class="doc_text">
2851 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2855 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
2856 different basic block in the current function. There are two forms of this
2857 instruction, corresponding to a conditional branch and an unconditional
2861 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
2862 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
2863 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
2867 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2868 argument is evaluated. If the value is <tt>true</tt>, control flows to the
2869 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2870 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2875 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
2876 br i1 %cond, label %IfEqual, label %IfUnequal
2878 <a href="#i_ret">ret</a> i32 1
2880 <a href="#i_ret">ret</a> i32 0
2885 <!-- _______________________________________________________________________ -->
2886 <div class="doc_subsubsection">
2887 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2890 <div class="doc_text">
2894 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2898 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2899 several different places. It is a generalization of the '<tt>br</tt>'
2900 instruction, allowing a branch to occur to one of many possible
2904 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2905 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
2906 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
2907 The table is not allowed to contain duplicate constant entries.</p>
2910 <p>The <tt>switch</tt> instruction specifies a table of values and
2911 destinations. When the '<tt>switch</tt>' instruction is executed, this table
2912 is searched for the given value. If the value is found, control flow is
2913 transferred to the corresponding destination; otherwise, control flow is
2914 transferred to the default destination.</p>
2916 <h5>Implementation:</h5>
2917 <p>Depending on properties of the target machine and the particular
2918 <tt>switch</tt> instruction, this instruction may be code generated in
2919 different ways. For example, it could be generated as a series of chained
2920 conditional branches or with a lookup table.</p>
2924 <i>; Emulate a conditional br instruction</i>
2925 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2926 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2928 <i>; Emulate an unconditional br instruction</i>
2929 switch i32 0, label %dest [ ]
2931 <i>; Implement a jump table:</i>
2932 switch i32 %val, label %otherwise [ i32 0, label %onzero
2934 i32 2, label %ontwo ]
2940 <!-- _______________________________________________________________________ -->
2941 <div class="doc_subsubsection">
2942 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
2945 <div class="doc_text">
2949 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
2954 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
2955 within the current function, whose address is specified by
2956 "<tt>address</tt>". Address must be derived from a <a
2957 href="#blockaddress">blockaddress</a> constant.</p>
2961 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
2962 rest of the arguments indicate the full set of possible destinations that the
2963 address may point to. Blocks are allowed to occur multiple times in the
2964 destination list, though this isn't particularly useful.</p>
2966 <p>This destination list is required so that dataflow analysis has an accurate
2967 understanding of the CFG.</p>
2971 <p>Control transfers to the block specified in the address argument. All
2972 possible destination blocks must be listed in the label list, otherwise this
2973 instruction has undefined behavior. This implies that jumps to labels
2974 defined in other functions have undefined behavior as well.</p>
2976 <h5>Implementation:</h5>
2978 <p>This is typically implemented with a jump through a register.</p>
2982 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
2988 <!-- _______________________________________________________________________ -->
2989 <div class="doc_subsubsection">
2990 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2993 <div class="doc_text">
2997 <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>]
2998 to label <normal label> unwind label <exception label>
3002 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3003 function, with the possibility of control flow transfer to either the
3004 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3005 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3006 control flow will return to the "normal" label. If the callee (or any
3007 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
3008 instruction, control is interrupted and continued at the dynamically nearest
3009 "exception" label.</p>
3012 <p>This instruction requires several arguments:</p>
3015 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3016 convention</a> the call should use. If none is specified, the call
3017 defaults to using C calling conventions.</li>
3019 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3020 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3021 '<tt>inreg</tt>' attributes are valid here.</li>
3023 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3024 function value being invoked. In most cases, this is a direct function
3025 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3026 off an arbitrary pointer to function value.</li>
3028 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3029 function to be invoked. </li>
3031 <li>'<tt>function args</tt>': argument list whose types match the function
3032 signature argument types and parameter attributes. All arguments must be
3033 of <a href="#t_firstclass">first class</a> type. If the function
3034 signature indicates the function accepts a variable number of arguments,
3035 the extra arguments can be specified.</li>
3037 <li>'<tt>normal label</tt>': the label reached when the called function
3038 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3040 <li>'<tt>exception label</tt>': the label reached when a callee returns with
3041 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
3043 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3044 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3045 '<tt>readnone</tt>' attributes are valid here.</li>
3049 <p>This instruction is designed to operate as a standard
3050 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3051 primary difference is that it establishes an association with a label, which
3052 is used by the runtime library to unwind the stack.</p>
3054 <p>This instruction is used in languages with destructors to ensure that proper
3055 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3056 exception. Additionally, this is important for implementation of
3057 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3059 <p>For the purposes of the SSA form, the definition of the value returned by the
3060 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3061 block to the "normal" label. If the callee unwinds then no return value is
3064 <p>Note that the code generator does not yet completely support unwind, and
3065 that the invoke/unwind semantics are likely to change in future versions.</p>
3069 %retval = invoke i32 @Test(i32 15) to label %Continue
3070 unwind label %TestCleanup <i>; {i32}:retval set</i>
3071 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3072 unwind label %TestCleanup <i>; {i32}:retval set</i>
3077 <!-- _______________________________________________________________________ -->
3079 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
3080 Instruction</a> </div>
3082 <div class="doc_text">
3090 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
3091 at the first callee in the dynamic call stack which used
3092 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
3093 This is primarily used to implement exception handling.</p>
3096 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
3097 immediately halt. The dynamic call stack is then searched for the
3098 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
3099 Once found, execution continues at the "exceptional" destination block
3100 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
3101 instruction in the dynamic call chain, undefined behavior results.</p>
3103 <p>Note that the code generator does not yet completely support unwind, and
3104 that the invoke/unwind semantics are likely to change in future versions.</p>
3108 <!-- _______________________________________________________________________ -->
3110 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
3111 Instruction</a> </div>
3113 <div class="doc_text">
3121 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3122 instruction is used to inform the optimizer that a particular portion of the
3123 code is not reachable. This can be used to indicate that the code after a
3124 no-return function cannot be reached, and other facts.</p>
3127 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3131 <!-- ======================================================================= -->
3132 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
3134 <div class="doc_text">
3136 <p>Binary operators are used to do most of the computation in a program. They
3137 require two operands of the same type, execute an operation on them, and
3138 produce a single value. The operands might represent multiple data, as is
3139 the case with the <a href="#t_vector">vector</a> data type. The result value
3140 has the same type as its operands.</p>
3142 <p>There are several different binary operators:</p>
3146 <!-- _______________________________________________________________________ -->
3147 <div class="doc_subsubsection">
3148 <a name="i_add">'<tt>add</tt>' Instruction</a>
3151 <div class="doc_text">
3155 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3156 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3157 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3158 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3162 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3165 <p>The two arguments to the '<tt>add</tt>' instruction must
3166 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3167 integer values. Both arguments must have identical types.</p>
3170 <p>The value produced is the integer sum of the two operands.</p>
3172 <p>If the sum has unsigned overflow, the result returned is the mathematical
3173 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3175 <p>Because LLVM integers use a two's complement representation, this instruction
3176 is appropriate for both signed and unsigned integers.</p>
3178 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3179 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3180 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3181 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3182 respectively, occurs.</p>
3186 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3191 <!-- _______________________________________________________________________ -->
3192 <div class="doc_subsubsection">
3193 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3196 <div class="doc_text">
3200 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3204 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3207 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3208 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3209 floating point values. Both arguments must have identical types.</p>
3212 <p>The value produced is the floating point sum of the two operands.</p>
3216 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3221 <!-- _______________________________________________________________________ -->
3222 <div class="doc_subsubsection">
3223 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3226 <div class="doc_text">
3230 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3231 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3232 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3233 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3237 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3240 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3241 '<tt>neg</tt>' instruction present in most other intermediate
3242 representations.</p>
3245 <p>The two arguments to the '<tt>sub</tt>' instruction must
3246 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3247 integer values. Both arguments must have identical types.</p>
3250 <p>The value produced is the integer difference of the two operands.</p>
3252 <p>If the difference has unsigned overflow, the result returned is the
3253 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3256 <p>Because LLVM integers use a two's complement representation, this instruction
3257 is appropriate for both signed and unsigned integers.</p>
3259 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3260 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3261 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3262 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3263 respectively, occurs.</p>
3267 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3268 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3273 <!-- _______________________________________________________________________ -->
3274 <div class="doc_subsubsection">
3275 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3278 <div class="doc_text">
3282 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3286 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3289 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3290 '<tt>fneg</tt>' instruction present in most other intermediate
3291 representations.</p>
3294 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3295 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3296 floating point values. Both arguments must have identical types.</p>
3299 <p>The value produced is the floating point difference of the two operands.</p>
3303 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3304 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3309 <!-- _______________________________________________________________________ -->
3310 <div class="doc_subsubsection">
3311 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3314 <div class="doc_text">
3318 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3319 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3320 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3321 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3325 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3328 <p>The two arguments to the '<tt>mul</tt>' instruction must
3329 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3330 integer values. Both arguments must have identical types.</p>
3333 <p>The value produced is the integer product of the two operands.</p>
3335 <p>If the result of the multiplication has unsigned overflow, the result
3336 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3337 width of the result.</p>
3339 <p>Because LLVM integers use a two's complement representation, and the result
3340 is the same width as the operands, this instruction returns the correct
3341 result for both signed and unsigned integers. If a full product
3342 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3343 be sign-extended or zero-extended as appropriate to the width of the full
3346 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3347 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3348 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3349 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3350 respectively, occurs.</p>
3354 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3359 <!-- _______________________________________________________________________ -->
3360 <div class="doc_subsubsection">
3361 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3364 <div class="doc_text">
3368 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3372 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3375 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3376 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3377 floating point values. Both arguments must have identical types.</p>
3380 <p>The value produced is the floating point product of the two operands.</p>
3384 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3389 <!-- _______________________________________________________________________ -->
3390 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
3393 <div class="doc_text">
3397 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3401 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3404 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3405 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3406 values. Both arguments must have identical types.</p>
3409 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3411 <p>Note that unsigned integer division and signed integer division are distinct
3412 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3414 <p>Division by zero leads to undefined behavior.</p>
3418 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3423 <!-- _______________________________________________________________________ -->
3424 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
3427 <div class="doc_text">
3431 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3432 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3436 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3439 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3440 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3441 values. Both arguments must have identical types.</p>
3444 <p>The value produced is the signed integer quotient of the two operands rounded
3447 <p>Note that signed integer division and unsigned integer division are distinct
3448 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3450 <p>Division by zero leads to undefined behavior. Overflow also leads to
3451 undefined behavior; this is a rare case, but can occur, for example, by doing
3452 a 32-bit division of -2147483648 by -1.</p>
3454 <p>If the <tt>exact</tt> keyword is present, the result value of the
3455 <tt>sdiv</tt> is a <a href="#trapvalues">trap value</a> if the result would
3460 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3465 <!-- _______________________________________________________________________ -->
3466 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
3467 Instruction</a> </div>
3469 <div class="doc_text">
3473 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3477 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3480 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3481 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3482 floating point values. Both arguments must have identical types.</p>
3485 <p>The value produced is the floating point quotient of the two operands.</p>
3489 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3494 <!-- _______________________________________________________________________ -->
3495 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3498 <div class="doc_text">
3502 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3506 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3507 division of its two arguments.</p>
3510 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3511 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3512 values. Both arguments must have identical types.</p>
3515 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3516 This instruction always performs an unsigned division to get the
3519 <p>Note that unsigned integer remainder and signed integer remainder are
3520 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3522 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3526 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3531 <!-- _______________________________________________________________________ -->
3532 <div class="doc_subsubsection">
3533 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3536 <div class="doc_text">
3540 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3544 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3545 division of its two operands. This instruction can also take
3546 <a href="#t_vector">vector</a> versions of the values in which case the
3547 elements must be integers.</p>
3550 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3551 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3552 values. Both arguments must have identical types.</p>
3555 <p>This instruction returns the <i>remainder</i> of a division (where the result
3556 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
3557 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
3558 a value. For more information about the difference,
3559 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3560 Math Forum</a>. For a table of how this is implemented in various languages,
3561 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3562 Wikipedia: modulo operation</a>.</p>
3564 <p>Note that signed integer remainder and unsigned integer remainder are
3565 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3567 <p>Taking the remainder of a division by zero leads to undefined behavior.
3568 Overflow also leads to undefined behavior; this is a rare case, but can
3569 occur, for example, by taking the remainder of a 32-bit division of
3570 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3571 lets srem be implemented using instructions that return both the result of
3572 the division and the remainder.)</p>
3576 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3581 <!-- _______________________________________________________________________ -->
3582 <div class="doc_subsubsection">
3583 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
3585 <div class="doc_text">
3589 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3593 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3594 its two operands.</p>
3597 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3598 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3599 floating point values. Both arguments must have identical types.</p>
3602 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3603 has the same sign as the dividend.</p>
3607 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3612 <!-- ======================================================================= -->
3613 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
3614 Operations</a> </div>
3616 <div class="doc_text">
3618 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3619 program. They are generally very efficient instructions and can commonly be
3620 strength reduced from other instructions. They require two operands of the
3621 same type, execute an operation on them, and produce a single value. The
3622 resulting value is the same type as its operands.</p>
3626 <!-- _______________________________________________________________________ -->
3627 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
3628 Instruction</a> </div>
3630 <div class="doc_text">
3634 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3638 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3639 a specified number of bits.</p>
3642 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3643 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3644 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3647 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3648 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3649 is (statically or dynamically) negative or equal to or larger than the number
3650 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3651 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3652 shift amount in <tt>op2</tt>.</p>
3656 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3657 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3658 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3659 <result> = shl i32 1, 32 <i>; undefined</i>
3660 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3665 <!-- _______________________________________________________________________ -->
3666 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
3667 Instruction</a> </div>
3669 <div class="doc_text">
3673 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3677 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
3678 operand shifted to the right a specified number of bits with zero fill.</p>
3681 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
3682 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3683 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3686 <p>This instruction always performs a logical shift right operation. The most
3687 significant bits of the result will be filled with zero bits after the shift.
3688 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
3689 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3690 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3691 shift amount in <tt>op2</tt>.</p>
3695 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
3696 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
3697 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
3698 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
3699 <result> = lshr i32 1, 32 <i>; undefined</i>
3700 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3705 <!-- _______________________________________________________________________ -->
3706 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
3707 Instruction</a> </div>
3708 <div class="doc_text">
3712 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3716 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
3717 operand shifted to the right a specified number of bits with sign
3721 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
3722 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3723 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3726 <p>This instruction always performs an arithmetic shift right operation, The
3727 most significant bits of the result will be filled with the sign bit
3728 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
3729 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
3730 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
3731 the corresponding shift amount in <tt>op2</tt>.</p>
3735 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
3736 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
3737 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
3738 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
3739 <result> = ashr i32 1, 32 <i>; undefined</i>
3740 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3745 <!-- _______________________________________________________________________ -->
3746 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
3747 Instruction</a> </div>
3749 <div class="doc_text">
3753 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3757 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
3761 <p>The two arguments to the '<tt>and</tt>' instruction must be
3762 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3763 values. Both arguments must have identical types.</p>
3766 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
3768 <table border="1" cellspacing="0" cellpadding="4">
3800 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
3801 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
3802 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
3805 <!-- _______________________________________________________________________ -->
3806 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
3808 <div class="doc_text">
3812 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3816 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
3820 <p>The two arguments to the '<tt>or</tt>' instruction must be
3821 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3822 values. Both arguments must have identical types.</p>
3825 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
3827 <table border="1" cellspacing="0" cellpadding="4">
3859 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
3860 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
3861 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3866 <!-- _______________________________________________________________________ -->
3867 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3868 Instruction</a> </div>
3870 <div class="doc_text">
3874 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3878 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
3879 its two operands. The <tt>xor</tt> is used to implement the "one's
3880 complement" operation, which is the "~" operator in C.</p>
3883 <p>The two arguments to the '<tt>xor</tt>' instruction must be
3884 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3885 values. Both arguments must have identical types.</p>
3888 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3890 <table border="1" cellspacing="0" cellpadding="4">
3922 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3923 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3924 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3925 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3930 <!-- ======================================================================= -->
3931 <div class="doc_subsection">
3932 <a name="vectorops">Vector Operations</a>
3935 <div class="doc_text">
3937 <p>LLVM supports several instructions to represent vector operations in a
3938 target-independent manner. These instructions cover the element-access and
3939 vector-specific operations needed to process vectors effectively. While LLVM
3940 does directly support these vector operations, many sophisticated algorithms
3941 will want to use target-specific intrinsics to take full advantage of a
3942 specific target.</p>
3946 <!-- _______________________________________________________________________ -->
3947 <div class="doc_subsubsection">
3948 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3951 <div class="doc_text">
3955 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3959 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
3960 from a vector at a specified index.</p>
3964 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
3965 of <a href="#t_vector">vector</a> type. The second operand is an index
3966 indicating the position from which to extract the element. The index may be
3970 <p>The result is a scalar of the same type as the element type of
3971 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3972 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3973 results are undefined.</p>
3977 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
3982 <!-- _______________________________________________________________________ -->
3983 <div class="doc_subsubsection">
3984 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3987 <div class="doc_text">
3991 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3995 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
3996 vector at a specified index.</p>
3999 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4000 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4001 whose type must equal the element type of the first operand. The third
4002 operand is an index indicating the position at which to insert the value.
4003 The index may be a variable.</p>
4006 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4007 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4008 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4009 results are undefined.</p>
4013 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4018 <!-- _______________________________________________________________________ -->
4019 <div class="doc_subsubsection">
4020 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4023 <div class="doc_text">
4027 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4031 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4032 from two input vectors, returning a vector with the same element type as the
4033 input and length that is the same as the shuffle mask.</p>
4036 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4037 with types that match each other. The third argument is a shuffle mask whose
4038 element type is always 'i32'. The result of the instruction is a vector
4039 whose length is the same as the shuffle mask and whose element type is the
4040 same as the element type of the first two operands.</p>
4042 <p>The shuffle mask operand is required to be a constant vector with either
4043 constant integer or undef values.</p>
4046 <p>The elements of the two input vectors are numbered from left to right across
4047 both of the vectors. The shuffle mask operand specifies, for each element of
4048 the result vector, which element of the two input vectors the result element
4049 gets. The element selector may be undef (meaning "don't care") and the
4050 second operand may be undef if performing a shuffle from only one vector.</p>
4054 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4055 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4056 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4057 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4058 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4059 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4060 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4061 <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>
4066 <!-- ======================================================================= -->
4067 <div class="doc_subsection">
4068 <a name="aggregateops">Aggregate Operations</a>
4071 <div class="doc_text">
4073 <p>LLVM supports several instructions for working with
4074 <a href="#t_aggregate">aggregate</a> values.</p>
4078 <!-- _______________________________________________________________________ -->
4079 <div class="doc_subsubsection">
4080 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4083 <div class="doc_text">
4087 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4091 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4092 from an <a href="#t_aggregate">aggregate</a> value.</p>
4095 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4096 of <a href="#t_struct">struct</a> or
4097 <a href="#t_array">array</a> type. The operands are constant indices to
4098 specify which value to extract in a similar manner as indices in a
4099 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4102 <p>The result is the value at the position in the aggregate specified by the
4107 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4112 <!-- _______________________________________________________________________ -->
4113 <div class="doc_subsubsection">
4114 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4117 <div class="doc_text">
4121 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx> <i>; yields <aggregate type></i>
4125 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4126 in an <a href="#t_aggregate">aggregate</a> value.</p>
4129 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4130 of <a href="#t_struct">struct</a> or
4131 <a href="#t_array">array</a> type. The second operand is a first-class
4132 value to insert. The following operands are constant indices indicating
4133 the position at which to insert the value in a similar manner as indices in a
4134 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction. The
4135 value to insert must have the same type as the value identified by the
4139 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4140 that of <tt>val</tt> except that the value at the position specified by the
4141 indices is that of <tt>elt</tt>.</p>
4145 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4146 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4152 <!-- ======================================================================= -->
4153 <div class="doc_subsection">
4154 <a name="memoryops">Memory Access and Addressing Operations</a>
4157 <div class="doc_text">
4159 <p>A key design point of an SSA-based representation is how it represents
4160 memory. In LLVM, no memory locations are in SSA form, which makes things
4161 very simple. This section describes how to read, write, and allocate
4166 <!-- _______________________________________________________________________ -->
4167 <div class="doc_subsubsection">
4168 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4171 <div class="doc_text">
4175 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4179 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4180 currently executing function, to be automatically released when this function
4181 returns to its caller. The object is always allocated in the generic address
4182 space (address space zero).</p>
4185 <p>The '<tt>alloca</tt>' instruction
4186 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4187 runtime stack, returning a pointer of the appropriate type to the program.
4188 If "NumElements" is specified, it is the number of elements allocated,
4189 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4190 specified, the value result of the allocation is guaranteed to be aligned to
4191 at least that boundary. If not specified, or if zero, the target can choose
4192 to align the allocation on any convenient boundary compatible with the
4195 <p>'<tt>type</tt>' may be any sized type.</p>
4198 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4199 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4200 memory is automatically released when the function returns. The
4201 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4202 variables that must have an address available. When the function returns
4203 (either with the <tt><a href="#i_ret">ret</a></tt>
4204 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
4205 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4209 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4210 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4211 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4212 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4217 <!-- _______________________________________________________________________ -->
4218 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
4219 Instruction</a> </div>
4221 <div class="doc_text">
4225 <result> = load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4226 <result> = volatile load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4227 !<index> = !{ i32 1 }
4231 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4234 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4235 from which to load. The pointer must point to
4236 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4237 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4238 number or order of execution of this <tt>load</tt> with other <a
4239 href="#volatile">volatile operations</a>.</p>
4241 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4242 operation (that is, the alignment of the memory address). A value of 0 or an
4243 omitted <tt>align</tt> argument means that the operation has the preferential
4244 alignment for the target. It is the responsibility of the code emitter to
4245 ensure that the alignment information is correct. Overestimating the
4246 alignment results in undefined behavior. Underestimating the alignment may
4247 produce less efficient code. An alignment of 1 is always safe.</p>
4249 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4250 metatadata name <index> corresponding to a metadata node with
4251 one <tt>i32</tt> entry of value 1. The existence of
4252 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4253 and code generator that this load is not expected to be reused in the cache.
4254 The code generator may select special instructions to save cache bandwidth,
4255 such as the <tt>MOVNT</tt> instruction on x86.</p>
4258 <p>The location of memory pointed to is loaded. If the value being loaded is of
4259 scalar type then the number of bytes read does not exceed the minimum number
4260 of bytes needed to hold all bits of the type. For example, loading an
4261 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4262 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4263 is undefined if the value was not originally written using a store of the
4268 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4269 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4270 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4275 <!-- _______________________________________________________________________ -->
4276 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
4277 Instruction</a> </div>
4279 <div class="doc_text">
4283 store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4284 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4288 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4291 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4292 and an address at which to store it. The type of the
4293 '<tt><pointer></tt>' operand must be a pointer to
4294 the <a href="#t_firstclass">first class</a> type of the
4295 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
4296 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
4297 order of execution of this <tt>store</tt> with other <a
4298 href="#volatile">volatile operations</a>.</p>
4300 <p>The optional constant "align" argument specifies the alignment of the
4301 operation (that is, the alignment of the memory address). A value of 0 or an
4302 omitted "align" argument means that the operation has the preferential
4303 alignment for the target. It is the responsibility of the code emitter to
4304 ensure that the alignment information is correct. Overestimating the
4305 alignment results in an undefined behavior. Underestimating the alignment may
4306 produce less efficient code. An alignment of 1 is always safe.</p>
4308 <p>The optional !nontemporal metadata must reference a single metatadata
4309 name <index> corresponding to a metadata node with one i32 entry of
4310 value 1. The existence of the !nontemporal metatadata on the
4311 instruction tells the optimizer and code generator that this load is
4312 not expected to be reused in the cache. The code generator may
4313 select special instructions to save cache bandwidth, such as the
4314 MOVNT instruction on x86.</p>
4318 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4319 location specified by the '<tt><pointer></tt>' operand. If
4320 '<tt><value></tt>' is of scalar type then the number of bytes written
4321 does not exceed the minimum number of bytes needed to hold all bits of the
4322 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4323 writing a value of a type like <tt>i20</tt> with a size that is not an
4324 integral number of bytes, it is unspecified what happens to the extra bits
4325 that do not belong to the type, but they will typically be overwritten.</p>
4329 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4330 store i32 3, i32* %ptr <i>; yields {void}</i>
4331 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4336 <!-- _______________________________________________________________________ -->
4337 <div class="doc_subsubsection">
4338 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
4341 <div class="doc_text">
4345 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4346 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4350 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
4351 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
4352 It performs address calculation only and does not access memory.</p>
4355 <p>The first argument is always a pointer, and forms the basis of the
4356 calculation. The remaining arguments are indices that indicate which of the
4357 elements of the aggregate object are indexed. The interpretation of each
4358 index is dependent on the type being indexed into. The first index always
4359 indexes the pointer value given as the first argument, the second index
4360 indexes a value of the type pointed to (not necessarily the value directly
4361 pointed to, since the first index can be non-zero), etc. The first type
4362 indexed into must be a pointer value, subsequent types can be arrays,
4363 vectors, and structs. Note that subsequent types being indexed into
4364 can never be pointers, since that would require loading the pointer before
4365 continuing calculation.</p>
4367 <p>The type of each index argument depends on the type it is indexing into.
4368 When indexing into a (optionally packed) structure, only <tt>i32</tt>
4369 integer <b>constants</b> are allowed. When indexing into an array, pointer
4370 or vector, integers of any width are allowed, and they are not required to be
4373 <p>For example, let's consider a C code fragment and how it gets compiled to
4376 <pre class="doc_code">
4388 int *foo(struct ST *s) {
4389 return &s[1].Z.B[5][13];
4393 <p>The LLVM code generated by the GCC frontend is:</p>
4395 <pre class="doc_code">
4396 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
4397 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
4399 define i32* @foo(%ST* %s) {
4401 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
4407 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
4408 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
4409 }</tt>' type, a structure. The second index indexes into the third element
4410 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
4411 i8 }</tt>' type, another structure. The third index indexes into the second
4412 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
4413 array. The two dimensions of the array are subscripted into, yielding an
4414 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
4415 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
4417 <p>Note that it is perfectly legal to index partially through a structure,
4418 returning a pointer to an inner element. Because of this, the LLVM code for
4419 the given testcase is equivalent to:</p>
4422 define i32* @foo(%ST* %s) {
4423 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
4424 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
4425 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
4426 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
4427 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
4432 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
4433 <tt>getelementptr</tt> is a <a href="#trapvalues">trap value</a> if the
4434 base pointer is not an <i>in bounds</i> address of an allocated object,
4435 or if any of the addresses that would be formed by successive addition of
4436 the offsets implied by the indices to the base address with infinitely
4437 precise arithmetic are not an <i>in bounds</i> address of that allocated
4438 object. The <i>in bounds</i> addresses for an allocated object are all
4439 the addresses that point into the object, plus the address one byte past
4442 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
4443 the base address with silently-wrapping two's complement arithmetic, and
4444 the result value of the <tt>getelementptr</tt> may be outside the object
4445 pointed to by the base pointer. The result value may not necessarily be
4446 used to access memory though, even if it happens to point into allocated
4447 storage. See the <a href="#pointeraliasing">Pointer Aliasing Rules</a>
4448 section for more information.</p>
4450 <p>The getelementptr instruction is often confusing. For some more insight into
4451 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
4455 <i>; yields [12 x i8]*:aptr</i>
4456 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4457 <i>; yields i8*:vptr</i>
4458 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4459 <i>; yields i8*:eptr</i>
4460 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4461 <i>; yields i32*:iptr</i>
4462 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4467 <!-- ======================================================================= -->
4468 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
4471 <div class="doc_text">
4473 <p>The instructions in this category are the conversion instructions (casting)
4474 which all take a single operand and a type. They perform various bit
4475 conversions on the operand.</p>
4479 <!-- _______________________________________________________________________ -->
4480 <div class="doc_subsubsection">
4481 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
4483 <div class="doc_text">
4487 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
4491 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
4492 type <tt>ty2</tt>.</p>
4495 <p>The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
4496 be an <a href="#t_integer">integer</a> type, and a type that specifies the
4497 size and type of the result, which must be
4498 an <a href="#t_integer">integer</a> type. The bit size of <tt>value</tt> must
4499 be larger than the bit size of <tt>ty2</tt>. Equal sized types are not
4503 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
4504 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
4505 source size must be larger than the destination size, <tt>trunc</tt> cannot
4506 be a <i>no-op cast</i>. It will always truncate bits.</p>
4510 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
4511 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
4512 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
4517 <!-- _______________________________________________________________________ -->
4518 <div class="doc_subsubsection">
4519 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
4521 <div class="doc_text">
4525 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
4529 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
4534 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
4535 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4536 also be of <a href="#t_integer">integer</a> type. The bit size of the
4537 <tt>value</tt> must be smaller than the bit size of the destination type,
4541 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
4542 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
4544 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
4548 %X = zext i32 257 to i64 <i>; yields i64:257</i>
4549 %Y = zext i1 true to i32 <i>; yields i32:1</i>
4554 <!-- _______________________________________________________________________ -->
4555 <div class="doc_subsubsection">
4556 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
4558 <div class="doc_text">
4562 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
4566 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
4569 <p>The '<tt>sext</tt>' instruction takes a value to cast, which must be of
4570 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4571 also be of <a href="#t_integer">integer</a> type. The bit size of the
4572 <tt>value</tt> must be smaller than the bit size of the destination type,
4576 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
4577 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
4578 of the type <tt>ty2</tt>.</p>
4580 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
4584 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
4585 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
4590 <!-- _______________________________________________________________________ -->
4591 <div class="doc_subsubsection">
4592 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
4595 <div class="doc_text">
4599 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
4603 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
4607 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
4608 point</a> value to cast and a <a href="#t_floating">floating point</a> type
4609 to cast it to. The size of <tt>value</tt> must be larger than the size of
4610 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
4611 <i>no-op cast</i>.</p>
4614 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
4615 <a href="#t_floating">floating point</a> type to a smaller
4616 <a href="#t_floating">floating point</a> type. If the value cannot fit
4617 within the destination type, <tt>ty2</tt>, then the results are
4622 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
4623 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
4628 <!-- _______________________________________________________________________ -->
4629 <div class="doc_subsubsection">
4630 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
4632 <div class="doc_text">
4636 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
4640 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
4641 floating point value.</p>
4644 <p>The '<tt>fpext</tt>' instruction takes a
4645 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
4646 a <a href="#t_floating">floating point</a> type to cast it to. The source
4647 type must be smaller than the destination type.</p>
4650 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
4651 <a href="#t_floating">floating point</a> type to a larger
4652 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
4653 used to make a <i>no-op cast</i> because it always changes bits. Use
4654 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
4658 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
4659 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
4664 <!-- _______________________________________________________________________ -->
4665 <div class="doc_subsubsection">
4666 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
4668 <div class="doc_text">
4672 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
4676 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
4677 unsigned integer equivalent of type <tt>ty2</tt>.</p>
4680 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
4681 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4682 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4683 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4684 vector integer type with the same number of elements as <tt>ty</tt></p>
4687 <p>The '<tt>fptoui</tt>' instruction converts its
4688 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4689 towards zero) unsigned integer value. If the value cannot fit
4690 in <tt>ty2</tt>, the results are undefined.</p>
4694 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
4695 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
4696 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
4701 <!-- _______________________________________________________________________ -->
4702 <div class="doc_subsubsection">
4703 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
4705 <div class="doc_text">
4709 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
4713 <p>The '<tt>fptosi</tt>' instruction converts
4714 <a href="#t_floating">floating point</a> <tt>value</tt> to
4715 type <tt>ty2</tt>.</p>
4718 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
4719 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4720 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4721 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4722 vector integer type with the same number of elements as <tt>ty</tt></p>
4725 <p>The '<tt>fptosi</tt>' instruction converts its
4726 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4727 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
4728 the results are undefined.</p>
4732 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
4733 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
4734 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
4739 <!-- _______________________________________________________________________ -->
4740 <div class="doc_subsubsection">
4741 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
4743 <div class="doc_text">
4747 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4751 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
4752 integer and converts that value to the <tt>ty2</tt> type.</p>
4755 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
4756 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4757 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4758 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4759 floating point type with the same number of elements as <tt>ty</tt></p>
4762 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4763 integer quantity and converts it to the corresponding floating point
4764 value. If the value cannot fit in the floating point value, the results are
4769 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
4770 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
4775 <!-- _______________________________________________________________________ -->
4776 <div class="doc_subsubsection">
4777 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4779 <div class="doc_text">
4783 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4787 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
4788 and converts that value to the <tt>ty2</tt> type.</p>
4791 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4792 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4793 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4794 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4795 floating point type with the same number of elements as <tt>ty</tt></p>
4798 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
4799 quantity and converts it to the corresponding floating point value. If the
4800 value cannot fit in the floating point value, the results are undefined.</p>
4804 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
4805 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
4810 <!-- _______________________________________________________________________ -->
4811 <div class="doc_subsubsection">
4812 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4814 <div class="doc_text">
4818 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
4822 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4823 the integer type <tt>ty2</tt>.</p>
4826 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4827 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
4828 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
4831 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4832 <tt>ty2</tt> by interpreting the pointer value as an integer and either
4833 truncating or zero extending that value to the size of the integer type. If
4834 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4835 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4836 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4841 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4842 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4847 <!-- _______________________________________________________________________ -->
4848 <div class="doc_subsubsection">
4849 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4851 <div class="doc_text">
4855 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4859 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
4860 pointer type, <tt>ty2</tt>.</p>
4863 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4864 value to cast, and a type to cast it to, which must be a
4865 <a href="#t_pointer">pointer</a> type.</p>
4868 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4869 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4870 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4871 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
4872 than the size of a pointer then a zero extension is done. If they are the
4873 same size, nothing is done (<i>no-op cast</i>).</p>
4877 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4878 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4879 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4884 <!-- _______________________________________________________________________ -->
4885 <div class="doc_subsubsection">
4886 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4888 <div class="doc_text">
4892 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4896 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4897 <tt>ty2</tt> without changing any bits.</p>
4900 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
4901 non-aggregate first class value, and a type to cast it to, which must also be
4902 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
4903 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
4904 identical. If the source type is a pointer, the destination type must also be
4905 a pointer. This instruction supports bitwise conversion of vectors to
4906 integers and to vectors of other types (as long as they have the same
4910 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4911 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4912 this conversion. The conversion is done as if the <tt>value</tt> had been
4913 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
4914 be converted to other pointer types with this instruction. To convert
4915 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
4916 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4920 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4921 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4922 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4927 <!-- ======================================================================= -->
4928 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4930 <div class="doc_text">
4932 <p>The instructions in this category are the "miscellaneous" instructions, which
4933 defy better classification.</p>
4937 <!-- _______________________________________________________________________ -->
4938 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4941 <div class="doc_text">
4945 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4949 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
4950 boolean values based on comparison of its two integer, integer vector, or
4951 pointer operands.</p>
4954 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4955 the condition code indicating the kind of comparison to perform. It is not a
4956 value, just a keyword. The possible condition code are:</p>
4959 <li><tt>eq</tt>: equal</li>
4960 <li><tt>ne</tt>: not equal </li>
4961 <li><tt>ugt</tt>: unsigned greater than</li>
4962 <li><tt>uge</tt>: unsigned greater or equal</li>
4963 <li><tt>ult</tt>: unsigned less than</li>
4964 <li><tt>ule</tt>: unsigned less or equal</li>
4965 <li><tt>sgt</tt>: signed greater than</li>
4966 <li><tt>sge</tt>: signed greater or equal</li>
4967 <li><tt>slt</tt>: signed less than</li>
4968 <li><tt>sle</tt>: signed less or equal</li>
4971 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4972 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
4973 typed. They must also be identical types.</p>
4976 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
4977 condition code given as <tt>cond</tt>. The comparison performed always yields
4978 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
4979 result, as follows:</p>
4982 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4983 <tt>false</tt> otherwise. No sign interpretation is necessary or
4986 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4987 <tt>false</tt> otherwise. No sign interpretation is necessary or
4990 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4991 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4993 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4994 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4995 to <tt>op2</tt>.</li>
4997 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4998 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5000 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
5001 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5003 <li><tt>sgt</tt>: interprets the operands as signed values and yields
5004 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5006 <li><tt>sge</tt>: interprets the operands as signed values and yields
5007 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5008 to <tt>op2</tt>.</li>
5010 <li><tt>slt</tt>: interprets the operands as signed values and yields
5011 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5013 <li><tt>sle</tt>: interprets the operands as signed values and yields
5014 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5017 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
5018 values are compared as if they were integers.</p>
5020 <p>If the operands are integer vectors, then they are compared element by
5021 element. The result is an <tt>i1</tt> vector with the same number of elements
5022 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
5026 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
5027 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
5028 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
5029 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
5030 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
5031 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
5034 <p>Note that the code generator does not yet support vector types with
5035 the <tt>icmp</tt> instruction.</p>
5039 <!-- _______________________________________________________________________ -->
5040 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
5043 <div class="doc_text">
5047 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5051 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
5052 values based on comparison of its operands.</p>
5054 <p>If the operands are floating point scalars, then the result type is a boolean
5055 (<a href="#t_integer"><tt>i1</tt></a>).</p>
5057 <p>If the operands are floating point vectors, then the result type is a vector
5058 of boolean with the same number of elements as the operands being
5062 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
5063 the condition code indicating the kind of comparison to perform. It is not a
5064 value, just a keyword. The possible condition code are:</p>
5067 <li><tt>false</tt>: no comparison, always returns false</li>
5068 <li><tt>oeq</tt>: ordered and equal</li>
5069 <li><tt>ogt</tt>: ordered and greater than </li>
5070 <li><tt>oge</tt>: ordered and greater than or equal</li>
5071 <li><tt>olt</tt>: ordered and less than </li>
5072 <li><tt>ole</tt>: ordered and less than or equal</li>
5073 <li><tt>one</tt>: ordered and not equal</li>
5074 <li><tt>ord</tt>: ordered (no nans)</li>
5075 <li><tt>ueq</tt>: unordered or equal</li>
5076 <li><tt>ugt</tt>: unordered or greater than </li>
5077 <li><tt>uge</tt>: unordered or greater than or equal</li>
5078 <li><tt>ult</tt>: unordered or less than </li>
5079 <li><tt>ule</tt>: unordered or less than or equal</li>
5080 <li><tt>une</tt>: unordered or not equal</li>
5081 <li><tt>uno</tt>: unordered (either nans)</li>
5082 <li><tt>true</tt>: no comparison, always returns true</li>
5085 <p><i>Ordered</i> means that neither operand is a QNAN while
5086 <i>unordered</i> means that either operand may be a QNAN.</p>
5088 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
5089 a <a href="#t_floating">floating point</a> type or
5090 a <a href="#t_vector">vector</a> of floating point type. They must have
5091 identical types.</p>
5094 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
5095 according to the condition code given as <tt>cond</tt>. If the operands are
5096 vectors, then the vectors are compared element by element. Each comparison
5097 performed always yields an <a href="#t_integer">i1</a> result, as
5101 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
5103 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5104 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5106 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5107 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5109 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5110 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5112 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5113 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5115 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5116 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5118 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5119 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5121 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
5123 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
5124 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5126 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
5127 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5129 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
5130 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5132 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
5133 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5135 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
5136 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5138 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
5139 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5141 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
5143 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
5148 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
5149 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
5150 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
5151 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
5154 <p>Note that the code generator does not yet support vector types with
5155 the <tt>fcmp</tt> instruction.</p>
5159 <!-- _______________________________________________________________________ -->
5160 <div class="doc_subsubsection">
5161 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
5164 <div class="doc_text">
5168 <result> = phi <ty> [ <val0>, <label0>], ...
5172 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
5173 SSA graph representing the function.</p>
5176 <p>The type of the incoming values is specified with the first type field. After
5177 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
5178 one pair for each predecessor basic block of the current block. Only values
5179 of <a href="#t_firstclass">first class</a> type may be used as the value
5180 arguments to the PHI node. Only labels may be used as the label
5183 <p>There must be no non-phi instructions between the start of a basic block and
5184 the PHI instructions: i.e. PHI instructions must be first in a basic
5187 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5188 occur on the edge from the corresponding predecessor block to the current
5189 block (but after any definition of an '<tt>invoke</tt>' instruction's return
5190 value on the same edge).</p>
5193 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
5194 specified by the pair corresponding to the predecessor basic block that
5195 executed just prior to the current block.</p>
5199 Loop: ; Infinite loop that counts from 0 on up...
5200 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5201 %nextindvar = add i32 %indvar, 1
5207 <!-- _______________________________________________________________________ -->
5208 <div class="doc_subsubsection">
5209 <a name="i_select">'<tt>select</tt>' Instruction</a>
5212 <div class="doc_text">
5216 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5218 <i>selty</i> is either i1 or {<N x i1>}
5222 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
5223 condition, without branching.</p>
5227 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
5228 values indicating the condition, and two values of the
5229 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
5230 vectors and the condition is a scalar, then entire vectors are selected, not
5231 individual elements.</p>
5234 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
5235 first value argument; otherwise, it returns the second value argument.</p>
5237 <p>If the condition is a vector of i1, then the value arguments must be vectors
5238 of the same size, and the selection is done element by element.</p>
5242 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
5245 <p>Note that the code generator does not yet support conditions
5246 with vector type.</p>
5250 <!-- _______________________________________________________________________ -->
5251 <div class="doc_subsubsection">
5252 <a name="i_call">'<tt>call</tt>' Instruction</a>
5255 <div class="doc_text">
5259 <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>]
5263 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
5266 <p>This instruction requires several arguments:</p>
5269 <li>The optional "tail" marker indicates that the callee function does not
5270 access any allocas or varargs in the caller. Note that calls may be
5271 marked "tail" even if they do not occur before
5272 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
5273 present, the function call is eligible for tail call optimization,
5274 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
5275 optimized into a jump</a>. The code generator may optimize calls marked
5276 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
5277 sibling call optimization</a> when the caller and callee have
5278 matching signatures, or 2) forced tail call optimization when the
5279 following extra requirements are met:
5281 <li>Caller and callee both have the calling
5282 convention <tt>fastcc</tt>.</li>
5283 <li>The call is in tail position (ret immediately follows call and ret
5284 uses value of call or is void).</li>
5285 <li>Option <tt>-tailcallopt</tt> is enabled,
5286 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
5287 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
5288 constraints are met.</a></li>
5292 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
5293 convention</a> the call should use. If none is specified, the call
5294 defaults to using C calling conventions. The calling convention of the
5295 call must match the calling convention of the target function, or else the
5296 behavior is undefined.</li>
5298 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
5299 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
5300 '<tt>inreg</tt>' attributes are valid here.</li>
5302 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
5303 type of the return value. Functions that return no value are marked
5304 <tt><a href="#t_void">void</a></tt>.</li>
5306 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5307 being invoked. The argument types must match the types implied by this
5308 signature. This type can be omitted if the function is not varargs and if
5309 the function type does not return a pointer to a function.</li>
5311 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5312 be invoked. In most cases, this is a direct function invocation, but
5313 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5314 to function value.</li>
5316 <li>'<tt>function args</tt>': argument list whose types match the function
5317 signature argument types and parameter attributes. All arguments must be
5318 of <a href="#t_firstclass">first class</a> type. If the function
5319 signature indicates the function accepts a variable number of arguments,
5320 the extra arguments can be specified.</li>
5322 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
5323 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
5324 '<tt>readnone</tt>' attributes are valid here.</li>
5328 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
5329 a specified function, with its incoming arguments bound to the specified
5330 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
5331 function, control flow continues with the instruction after the function
5332 call, and the return value of the function is bound to the result
5337 %retval = call i32 @test(i32 %argc)
5338 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
5339 %X = tail call i32 @foo() <i>; yields i32</i>
5340 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
5341 call void %foo(i8 97 signext)
5343 %struct.A = type { i32, i8 }
5344 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
5345 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
5346 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
5347 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
5348 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
5351 <p>llvm treats calls to some functions with names and arguments that match the
5352 standard C99 library as being the C99 library functions, and may perform
5353 optimizations or generate code for them under that assumption. This is
5354 something we'd like to change in the future to provide better support for
5355 freestanding environments and non-C-based languages.</p>
5359 <!-- _______________________________________________________________________ -->
5360 <div class="doc_subsubsection">
5361 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
5364 <div class="doc_text">
5368 <resultval> = va_arg <va_list*> <arglist>, <argty>
5372 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
5373 the "variable argument" area of a function call. It is used to implement the
5374 <tt>va_arg</tt> macro in C.</p>
5377 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
5378 argument. It returns a value of the specified argument type and increments
5379 the <tt>va_list</tt> to point to the next argument. The actual type
5380 of <tt>va_list</tt> is target specific.</p>
5383 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
5384 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
5385 to the next argument. For more information, see the variable argument
5386 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
5388 <p>It is legal for this instruction to be called in a function which does not
5389 take a variable number of arguments, for example, the <tt>vfprintf</tt>
5392 <p><tt>va_arg</tt> is an LLVM instruction instead of
5393 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
5397 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
5399 <p>Note that the code generator does not yet fully support va_arg on many
5400 targets. Also, it does not currently support va_arg with aggregate types on
5405 <!-- *********************************************************************** -->
5406 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
5407 <!-- *********************************************************************** -->
5409 <div class="doc_text">
5411 <p>LLVM supports the notion of an "intrinsic function". These functions have
5412 well known names and semantics and are required to follow certain
5413 restrictions. Overall, these intrinsics represent an extension mechanism for
5414 the LLVM language that does not require changing all of the transformations
5415 in LLVM when adding to the language (or the bitcode reader/writer, the
5416 parser, etc...).</p>
5418 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
5419 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5420 begin with this prefix. Intrinsic functions must always be external
5421 functions: you cannot define the body of intrinsic functions. Intrinsic
5422 functions may only be used in call or invoke instructions: it is illegal to
5423 take the address of an intrinsic function. Additionally, because intrinsic
5424 functions are part of the LLVM language, it is required if any are added that
5425 they be documented here.</p>
5427 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5428 family of functions that perform the same operation but on different data
5429 types. Because LLVM can represent over 8 million different integer types,
5430 overloading is used commonly to allow an intrinsic function to operate on any
5431 integer type. One or more of the argument types or the result type can be
5432 overloaded to accept any integer type. Argument types may also be defined as
5433 exactly matching a previous argument's type or the result type. This allows
5434 an intrinsic function which accepts multiple arguments, but needs all of them
5435 to be of the same type, to only be overloaded with respect to a single
5436 argument or the result.</p>
5438 <p>Overloaded intrinsics will have the names of its overloaded argument types
5439 encoded into its function name, each preceded by a period. Only those types
5440 which are overloaded result in a name suffix. Arguments whose type is matched
5441 against another type do not. For example, the <tt>llvm.ctpop</tt> function
5442 can take an integer of any width and returns an integer of exactly the same
5443 integer width. This leads to a family of functions such as
5444 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
5445 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5446 suffix is required. Because the argument's type is matched against the return
5447 type, it does not require its own name suffix.</p>
5449 <p>To learn how to add an intrinsic function, please see the
5450 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
5454 <!-- ======================================================================= -->
5455 <div class="doc_subsection">
5456 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
5459 <div class="doc_text">
5461 <p>Variable argument support is defined in LLVM with
5462 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
5463 intrinsic functions. These functions are related to the similarly named
5464 macros defined in the <tt><stdarg.h></tt> header file.</p>
5466 <p>All of these functions operate on arguments that use a target-specific value
5467 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5468 not define what this type is, so all transformations should be prepared to
5469 handle these functions regardless of the type used.</p>
5471 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
5472 instruction and the variable argument handling intrinsic functions are
5475 <pre class="doc_code">
5476 define i32 @test(i32 %X, ...) {
5477 ; Initialize variable argument processing
5479 %ap2 = bitcast i8** %ap to i8*
5480 call void @llvm.va_start(i8* %ap2)
5482 ; Read a single integer argument
5483 %tmp = va_arg i8** %ap, i32
5485 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5487 %aq2 = bitcast i8** %aq to i8*
5488 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5489 call void @llvm.va_end(i8* %aq2)
5491 ; Stop processing of arguments.
5492 call void @llvm.va_end(i8* %ap2)
5496 declare void @llvm.va_start(i8*)
5497 declare void @llvm.va_copy(i8*, i8*)
5498 declare void @llvm.va_end(i8*)
5503 <!-- _______________________________________________________________________ -->
5504 <div class="doc_subsubsection">
5505 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
5509 <div class="doc_text">
5513 declare void %llvm.va_start(i8* <arglist>)
5517 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
5518 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
5521 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
5524 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
5525 macro available in C. In a target-dependent way, it initializes
5526 the <tt>va_list</tt> element to which the argument points, so that the next
5527 call to <tt>va_arg</tt> will produce the first variable argument passed to
5528 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
5529 need to know the last argument of the function as the compiler can figure
5534 <!-- _______________________________________________________________________ -->
5535 <div class="doc_subsubsection">
5536 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
5539 <div class="doc_text">
5543 declare void @llvm.va_end(i8* <arglist>)
5547 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
5548 which has been initialized previously
5549 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
5550 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
5553 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
5556 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
5557 macro available in C. In a target-dependent way, it destroys
5558 the <tt>va_list</tt> element to which the argument points. Calls
5559 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
5560 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
5561 with calls to <tt>llvm.va_end</tt>.</p>
5565 <!-- _______________________________________________________________________ -->
5566 <div class="doc_subsubsection">
5567 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
5570 <div class="doc_text">
5574 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
5578 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
5579 from the source argument list to the destination argument list.</p>
5582 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
5583 The second argument is a pointer to a <tt>va_list</tt> element to copy
5587 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
5588 macro available in C. In a target-dependent way, it copies the
5589 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
5590 element. This intrinsic is necessary because
5591 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
5592 arbitrarily complex and require, for example, memory allocation.</p>
5596 <!-- ======================================================================= -->
5597 <div class="doc_subsection">
5598 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
5601 <div class="doc_text">
5603 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
5604 Collection</a> (GC) requires the implementation and generation of these
5605 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
5606 roots on the stack</a>, as well as garbage collector implementations that
5607 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
5608 barriers. Front-ends for type-safe garbage collected languages should generate
5609 these intrinsics to make use of the LLVM garbage collectors. For more details,
5610 see <a href="GarbageCollection.html">Accurate Garbage Collection with
5613 <p>The garbage collection intrinsics only operate on objects in the generic
5614 address space (address space zero).</p>
5618 <!-- _______________________________________________________________________ -->
5619 <div class="doc_subsubsection">
5620 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
5623 <div class="doc_text">
5627 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5631 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
5632 the code generator, and allows some metadata to be associated with it.</p>
5635 <p>The first argument specifies the address of a stack object that contains the
5636 root pointer. The second pointer (which must be either a constant or a
5637 global value address) contains the meta-data to be associated with the
5641 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
5642 location. At compile-time, the code generator generates information to allow
5643 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
5644 intrinsic may only be used in a function which <a href="#gc">specifies a GC
5649 <!-- _______________________________________________________________________ -->
5650 <div class="doc_subsubsection">
5651 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
5654 <div class="doc_text">
5658 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5662 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
5663 locations, allowing garbage collector implementations that require read
5667 <p>The second argument is the address to read from, which should be an address
5668 allocated from the garbage collector. The first object is a pointer to the
5669 start of the referenced object, if needed by the language runtime (otherwise
5673 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
5674 instruction, but may be replaced with substantially more complex code by the
5675 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
5676 may only be used in a function which <a href="#gc">specifies a GC
5681 <!-- _______________________________________________________________________ -->
5682 <div class="doc_subsubsection">
5683 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
5686 <div class="doc_text">
5690 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5694 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5695 locations, allowing garbage collector implementations that require write
5696 barriers (such as generational or reference counting collectors).</p>
5699 <p>The first argument is the reference to store, the second is the start of the
5700 object to store it to, and the third is the address of the field of Obj to
5701 store to. If the runtime does not require a pointer to the object, Obj may
5705 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5706 instruction, but may be replaced with substantially more complex code by the
5707 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5708 may only be used in a function which <a href="#gc">specifies a GC
5713 <!-- ======================================================================= -->
5714 <div class="doc_subsection">
5715 <a name="int_codegen">Code Generator Intrinsics</a>
5718 <div class="doc_text">
5720 <p>These intrinsics are provided by LLVM to expose special features that may
5721 only be implemented with code generator support.</p>
5725 <!-- _______________________________________________________________________ -->
5726 <div class="doc_subsubsection">
5727 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5730 <div class="doc_text">
5734 declare i8 *@llvm.returnaddress(i32 <level>)
5738 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5739 target-specific value indicating the return address of the current function
5740 or one of its callers.</p>
5743 <p>The argument to this intrinsic indicates which function to return the address
5744 for. Zero indicates the calling function, one indicates its caller, etc.
5745 The argument is <b>required</b> to be a constant integer value.</p>
5748 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
5749 indicating the return address of the specified call frame, or zero if it
5750 cannot be identified. The value returned by this intrinsic is likely to be
5751 incorrect or 0 for arguments other than zero, so it should only be used for
5752 debugging purposes.</p>
5754 <p>Note that calling this intrinsic does not prevent function inlining or other
5755 aggressive transformations, so the value returned may not be that of the
5756 obvious source-language caller.</p>
5760 <!-- _______________________________________________________________________ -->
5761 <div class="doc_subsubsection">
5762 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5765 <div class="doc_text">
5769 declare i8* @llvm.frameaddress(i32 <level>)
5773 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5774 target-specific frame pointer value for the specified stack frame.</p>
5777 <p>The argument to this intrinsic indicates which function to return the frame
5778 pointer for. Zero indicates the calling function, one indicates its caller,
5779 etc. The argument is <b>required</b> to be a constant integer value.</p>
5782 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
5783 indicating the frame address of the specified call frame, or zero if it
5784 cannot be identified. The value returned by this intrinsic is likely to be
5785 incorrect or 0 for arguments other than zero, so it should only be used for
5786 debugging purposes.</p>
5788 <p>Note that calling this intrinsic does not prevent function inlining or other
5789 aggressive transformations, so the value returned may not be that of the
5790 obvious source-language caller.</p>
5794 <!-- _______________________________________________________________________ -->
5795 <div class="doc_subsubsection">
5796 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5799 <div class="doc_text">
5803 declare i8* @llvm.stacksave()
5807 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
5808 of the function stack, for use
5809 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
5810 useful for implementing language features like scoped automatic variable
5811 sized arrays in C99.</p>
5814 <p>This intrinsic returns a opaque pointer value that can be passed
5815 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
5816 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
5817 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
5818 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
5819 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
5820 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
5824 <!-- _______________________________________________________________________ -->
5825 <div class="doc_subsubsection">
5826 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5829 <div class="doc_text">
5833 declare void @llvm.stackrestore(i8* %ptr)
5837 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5838 the function stack to the state it was in when the
5839 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
5840 executed. This is useful for implementing language features like scoped
5841 automatic variable sized arrays in C99.</p>
5844 <p>See the description
5845 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
5849 <!-- _______________________________________________________________________ -->
5850 <div class="doc_subsubsection">
5851 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5854 <div class="doc_text">
5858 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5862 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
5863 insert a prefetch instruction if supported; otherwise, it is a noop.
5864 Prefetches have no effect on the behavior of the program but can change its
5865 performance characteristics.</p>
5868 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
5869 specifier determining if the fetch should be for a read (0) or write (1),
5870 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5871 locality, to (3) - extremely local keep in cache. The <tt>rw</tt>
5872 and <tt>locality</tt> arguments must be constant integers.</p>
5875 <p>This intrinsic does not modify the behavior of the program. In particular,
5876 prefetches cannot trap and do not produce a value. On targets that support
5877 this intrinsic, the prefetch can provide hints to the processor cache for
5878 better performance.</p>
5882 <!-- _______________________________________________________________________ -->
5883 <div class="doc_subsubsection">
5884 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5887 <div class="doc_text">
5891 declare void @llvm.pcmarker(i32 <id>)
5895 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
5896 Counter (PC) in a region of code to simulators and other tools. The method
5897 is target specific, but it is expected that the marker will use exported
5898 symbols to transmit the PC of the marker. The marker makes no guarantees
5899 that it will remain with any specific instruction after optimizations. It is
5900 possible that the presence of a marker will inhibit optimizations. The
5901 intended use is to be inserted after optimizations to allow correlations of
5902 simulation runs.</p>
5905 <p><tt>id</tt> is a numerical id identifying the marker.</p>
5908 <p>This intrinsic does not modify the behavior of the program. Backends that do
5909 not support this intrinsic may ignore it.</p>
5913 <!-- _______________________________________________________________________ -->
5914 <div class="doc_subsubsection">
5915 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5918 <div class="doc_text">
5922 declare i64 @llvm.readcyclecounter()
5926 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5927 counter register (or similar low latency, high accuracy clocks) on those
5928 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5929 should map to RPCC. As the backing counters overflow quickly (on the order
5930 of 9 seconds on alpha), this should only be used for small timings.</p>
5933 <p>When directly supported, reading the cycle counter should not modify any
5934 memory. Implementations are allowed to either return a application specific
5935 value or a system wide value. On backends without support, this is lowered
5936 to a constant 0.</p>
5940 <!-- ======================================================================= -->
5941 <div class="doc_subsection">
5942 <a name="int_libc">Standard C Library Intrinsics</a>
5945 <div class="doc_text">
5947 <p>LLVM provides intrinsics for a few important standard C library functions.
5948 These intrinsics allow source-language front-ends to pass information about
5949 the alignment of the pointer arguments to the code generator, providing
5950 opportunity for more efficient code generation.</p>
5954 <!-- _______________________________________________________________________ -->
5955 <div class="doc_subsubsection">
5956 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5959 <div class="doc_text">
5962 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
5963 integer bit width and for different address spaces. Not all targets support
5964 all bit widths however.</p>
5967 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
5968 i32 <len>, i32 <align>, i1 <isvolatile>)
5969 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
5970 i64 <len>, i32 <align>, i1 <isvolatile>)
5974 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5975 source location to the destination location.</p>
5977 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5978 intrinsics do not return a value, takes extra alignment/isvolatile arguments
5979 and the pointers can be in specified address spaces.</p>
5983 <p>The first argument is a pointer to the destination, the second is a pointer
5984 to the source. The third argument is an integer argument specifying the
5985 number of bytes to copy, the fourth argument is the alignment of the
5986 source and destination locations, and the fifth is a boolean indicating a
5987 volatile access.</p>
5989 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
5990 then the caller guarantees that both the source and destination pointers are
5991 aligned to that boundary.</p>
5993 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
5994 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
5995 The detailed access behavior is not very cleanly specified and it is unwise
5996 to depend on it.</p>
6000 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6001 source location to the destination location, which are not allowed to
6002 overlap. It copies "len" bytes of memory over. If the argument is known to
6003 be aligned to some boundary, this can be specified as the fourth argument,
6004 otherwise it should be set to 0 or 1.</p>
6008 <!-- _______________________________________________________________________ -->
6009 <div class="doc_subsubsection">
6010 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
6013 <div class="doc_text">
6016 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
6017 width and for different address space. Not all targets support all bit
6021 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6022 i32 <len>, i32 <align>, i1 <isvolatile>)
6023 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6024 i64 <len>, i32 <align>, i1 <isvolatile>)
6028 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
6029 source location to the destination location. It is similar to the
6030 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
6033 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
6034 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6035 and the pointers can be in specified address spaces.</p>
6039 <p>The first argument is a pointer to the destination, the second is a pointer
6040 to the source. The third argument is an integer argument specifying the
6041 number of bytes to copy, the fourth argument is the alignment of the
6042 source and destination locations, and the fifth is a boolean indicating a
6043 volatile access.</p>
6045 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6046 then the caller guarantees that the source and destination pointers are
6047 aligned to that boundary.</p>
6049 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6050 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
6051 The detailed access behavior is not very cleanly specified and it is unwise
6052 to depend on it.</p>
6056 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
6057 source location to the destination location, which may overlap. It copies
6058 "len" bytes of memory over. If the argument is known to be aligned to some
6059 boundary, this can be specified as the fourth argument, otherwise it should
6060 be set to 0 or 1.</p>
6064 <!-- _______________________________________________________________________ -->
6065 <div class="doc_subsubsection">
6066 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
6069 <div class="doc_text">
6072 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6073 width and for different address spaces. However, not all targets support all
6077 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6078 i32 <len>, i32 <align>, i1 <isvolatile>)
6079 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6080 i64 <len>, i32 <align>, i1 <isvolatile>)
6084 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
6085 particular byte value.</p>
6087 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
6088 intrinsic does not return a value and takes extra alignment/volatile
6089 arguments. Also, the destination can be in an arbitrary address space.</p>
6092 <p>The first argument is a pointer to the destination to fill, the second is the
6093 byte value with which to fill it, the third argument is an integer argument
6094 specifying the number of bytes to fill, and the fourth argument is the known
6095 alignment of the destination location.</p>
6097 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6098 then the caller guarantees that the destination pointer is aligned to that
6101 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6102 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
6103 The detailed access behavior is not very cleanly specified and it is unwise
6104 to depend on it.</p>
6107 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
6108 at the destination location. If the argument is known to be aligned to some
6109 boundary, this can be specified as the fourth argument, otherwise it should
6110 be set to 0 or 1.</p>
6114 <!-- _______________________________________________________________________ -->
6115 <div class="doc_subsubsection">
6116 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
6119 <div class="doc_text">
6122 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
6123 floating point or vector of floating point type. Not all targets support all
6127 declare float @llvm.sqrt.f32(float %Val)
6128 declare double @llvm.sqrt.f64(double %Val)
6129 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6130 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6131 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6135 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
6136 returning the same value as the libm '<tt>sqrt</tt>' functions would.
6137 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
6138 behavior for negative numbers other than -0.0 (which allows for better
6139 optimization, because there is no need to worry about errno being
6140 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
6143 <p>The argument and return value are floating point numbers of the same
6147 <p>This function returns the sqrt of the specified operand if it is a
6148 nonnegative floating point number.</p>
6152 <!-- _______________________________________________________________________ -->
6153 <div class="doc_subsubsection">
6154 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
6157 <div class="doc_text">
6160 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
6161 floating point or vector of floating point type. Not all targets support all
6165 declare float @llvm.powi.f32(float %Val, i32 %power)
6166 declare double @llvm.powi.f64(double %Val, i32 %power)
6167 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6168 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6169 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6173 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
6174 specified (positive or negative) power. The order of evaluation of
6175 multiplications is not defined. When a vector of floating point type is
6176 used, the second argument remains a scalar integer value.</p>
6179 <p>The second argument is an integer power, and the first is a value to raise to
6183 <p>This function returns the first value raised to the second power with an
6184 unspecified sequence of rounding operations.</p>
6188 <!-- _______________________________________________________________________ -->
6189 <div class="doc_subsubsection">
6190 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
6193 <div class="doc_text">
6196 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
6197 floating point or vector of floating point type. Not all targets support all
6201 declare float @llvm.sin.f32(float %Val)
6202 declare double @llvm.sin.f64(double %Val)
6203 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6204 declare fp128 @llvm.sin.f128(fp128 %Val)
6205 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6209 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
6212 <p>The argument and return value are floating point numbers of the same
6216 <p>This function returns the sine of the specified operand, returning the same
6217 values as the libm <tt>sin</tt> functions would, and handles error conditions
6218 in the same way.</p>
6222 <!-- _______________________________________________________________________ -->
6223 <div class="doc_subsubsection">
6224 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
6227 <div class="doc_text">
6230 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
6231 floating point or vector of floating point type. Not all targets support all
6235 declare float @llvm.cos.f32(float %Val)
6236 declare double @llvm.cos.f64(double %Val)
6237 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6238 declare fp128 @llvm.cos.f128(fp128 %Val)
6239 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6243 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
6246 <p>The argument and return value are floating point numbers of the same
6250 <p>This function returns the cosine of the specified operand, returning the same
6251 values as the libm <tt>cos</tt> functions would, and handles error conditions
6252 in the same way.</p>
6256 <!-- _______________________________________________________________________ -->
6257 <div class="doc_subsubsection">
6258 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
6261 <div class="doc_text">
6264 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
6265 floating point or vector of floating point type. Not all targets support all
6269 declare float @llvm.pow.f32(float %Val, float %Power)
6270 declare double @llvm.pow.f64(double %Val, double %Power)
6271 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6272 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6273 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6277 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
6278 specified (positive or negative) power.</p>
6281 <p>The second argument is a floating point power, and the first is a value to
6282 raise to that power.</p>
6285 <p>This function returns the first value raised to the second power, returning
6286 the same values as the libm <tt>pow</tt> functions would, and handles error
6287 conditions in the same way.</p>
6291 <!-- ======================================================================= -->
6292 <div class="doc_subsection">
6293 <a name="int_manip">Bit Manipulation Intrinsics</a>
6296 <div class="doc_text">
6298 <p>LLVM provides intrinsics for a few important bit manipulation operations.
6299 These allow efficient code generation for some algorithms.</p>
6303 <!-- _______________________________________________________________________ -->
6304 <div class="doc_subsubsection">
6305 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
6308 <div class="doc_text">
6311 <p>This is an overloaded intrinsic function. You can use bswap on any integer
6312 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
6315 declare i16 @llvm.bswap.i16(i16 <id>)
6316 declare i32 @llvm.bswap.i32(i32 <id>)
6317 declare i64 @llvm.bswap.i64(i64 <id>)
6321 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
6322 values with an even number of bytes (positive multiple of 16 bits). These
6323 are useful for performing operations on data that is not in the target's
6324 native byte order.</p>
6327 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
6328 and low byte of the input i16 swapped. Similarly,
6329 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
6330 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
6331 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
6332 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
6333 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
6334 more, respectively).</p>
6338 <!-- _______________________________________________________________________ -->
6339 <div class="doc_subsubsection">
6340 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
6343 <div class="doc_text">
6346 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
6347 width. Not all targets support all bit widths however.</p>
6350 declare i8 @llvm.ctpop.i8(i8 <src>)
6351 declare i16 @llvm.ctpop.i16(i16 <src>)
6352 declare i32 @llvm.ctpop.i32(i32 <src>)
6353 declare i64 @llvm.ctpop.i64(i64 <src>)
6354 declare i256 @llvm.ctpop.i256(i256 <src>)
6358 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
6362 <p>The only argument is the value to be counted. The argument may be of any
6363 integer type. The return type must match the argument type.</p>
6366 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.</p>
6370 <!-- _______________________________________________________________________ -->
6371 <div class="doc_subsubsection">
6372 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
6375 <div class="doc_text">
6378 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
6379 integer bit width. Not all targets support all bit widths however.</p>
6382 declare i8 @llvm.ctlz.i8 (i8 <src>)
6383 declare i16 @llvm.ctlz.i16(i16 <src>)
6384 declare i32 @llvm.ctlz.i32(i32 <src>)
6385 declare i64 @llvm.ctlz.i64(i64 <src>)
6386 declare i256 @llvm.ctlz.i256(i256 <src>)
6390 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
6391 leading zeros in a variable.</p>
6394 <p>The only argument is the value to be counted. The argument may be of any
6395 integer type. The return type must match the argument type.</p>
6398 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
6399 zeros in a variable. If the src == 0 then the result is the size in bits of
6400 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
6404 <!-- _______________________________________________________________________ -->
6405 <div class="doc_subsubsection">
6406 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
6409 <div class="doc_text">
6412 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
6413 integer bit width. Not all targets support all bit widths however.</p>
6416 declare i8 @llvm.cttz.i8 (i8 <src>)
6417 declare i16 @llvm.cttz.i16(i16 <src>)
6418 declare i32 @llvm.cttz.i32(i32 <src>)
6419 declare i64 @llvm.cttz.i64(i64 <src>)
6420 declare i256 @llvm.cttz.i256(i256 <src>)
6424 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
6428 <p>The only argument is the value to be counted. The argument may be of any
6429 integer type. The return type must match the argument type.</p>
6432 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
6433 zeros in a variable. If the src == 0 then the result is the size in bits of
6434 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
6438 <!-- ======================================================================= -->
6439 <div class="doc_subsection">
6440 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
6443 <div class="doc_text">
6445 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
6449 <!-- _______________________________________________________________________ -->
6450 <div class="doc_subsubsection">
6451 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
6454 <div class="doc_text">
6457 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
6458 on any integer bit width.</p>
6461 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6462 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6463 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6467 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6468 a signed addition of the two arguments, and indicate whether an overflow
6469 occurred during the signed summation.</p>
6472 <p>The arguments (%a and %b) and the first element of the result structure may
6473 be of integer types of any bit width, but they must have the same bit
6474 width. The second element of the result structure must be of
6475 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6476 undergo signed addition.</p>
6479 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6480 a signed addition of the two variables. They return a structure — the
6481 first element of which is the signed summation, and the second element of
6482 which is a bit specifying if the signed summation resulted in an
6487 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6488 %sum = extractvalue {i32, i1} %res, 0
6489 %obit = extractvalue {i32, i1} %res, 1
6490 br i1 %obit, label %overflow, label %normal
6495 <!-- _______________________________________________________________________ -->
6496 <div class="doc_subsubsection">
6497 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
6500 <div class="doc_text">
6503 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
6504 on any integer bit width.</p>
6507 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6508 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6509 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6513 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6514 an unsigned addition of the two arguments, and indicate whether a carry
6515 occurred during the unsigned summation.</p>
6518 <p>The arguments (%a and %b) and the first element of the result structure may
6519 be of integer types of any bit width, but they must have the same bit
6520 width. The second element of the result structure must be of
6521 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6522 undergo unsigned addition.</p>
6525 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6526 an unsigned addition of the two arguments. They return a structure —
6527 the first element of which is the sum, and the second element of which is a
6528 bit specifying if the unsigned summation resulted in a carry.</p>
6532 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6533 %sum = extractvalue {i32, i1} %res, 0
6534 %obit = extractvalue {i32, i1} %res, 1
6535 br i1 %obit, label %carry, label %normal
6540 <!-- _______________________________________________________________________ -->
6541 <div class="doc_subsubsection">
6542 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
6545 <div class="doc_text">
6548 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
6549 on any integer bit width.</p>
6552 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6553 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6554 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6558 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6559 a signed subtraction of the two arguments, and indicate whether an overflow
6560 occurred during the signed subtraction.</p>
6563 <p>The arguments (%a and %b) and the first element of the result structure may
6564 be of integer types of any bit width, but they must have the same bit
6565 width. The second element of the result structure must be of
6566 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6567 undergo signed subtraction.</p>
6570 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6571 a signed subtraction of the two arguments. They return a structure —
6572 the first element of which is the subtraction, and the second element of
6573 which is a bit specifying if the signed subtraction resulted in an
6578 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6579 %sum = extractvalue {i32, i1} %res, 0
6580 %obit = extractvalue {i32, i1} %res, 1
6581 br i1 %obit, label %overflow, label %normal
6586 <!-- _______________________________________________________________________ -->
6587 <div class="doc_subsubsection">
6588 <a name="int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6591 <div class="doc_text">
6594 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6595 on any integer bit width.</p>
6598 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6599 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6600 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6604 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6605 an unsigned subtraction of the two arguments, and indicate whether an
6606 overflow occurred during the unsigned subtraction.</p>
6609 <p>The arguments (%a and %b) and the first element of the result structure may
6610 be of integer types of any bit width, but they must have the same bit
6611 width. The second element of the result structure must be of
6612 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6613 undergo unsigned subtraction.</p>
6616 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6617 an unsigned subtraction of the two arguments. They return a structure —
6618 the first element of which is the subtraction, and the second element of
6619 which is a bit specifying if the unsigned subtraction resulted in an
6624 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6625 %sum = extractvalue {i32, i1} %res, 0
6626 %obit = extractvalue {i32, i1} %res, 1
6627 br i1 %obit, label %overflow, label %normal
6632 <!-- _______________________________________________________________________ -->
6633 <div class="doc_subsubsection">
6634 <a name="int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6637 <div class="doc_text">
6640 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6641 on any integer bit width.</p>
6644 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6645 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6646 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6651 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6652 a signed multiplication of the two arguments, and indicate whether an
6653 overflow occurred during the signed multiplication.</p>
6656 <p>The arguments (%a and %b) and the first element of the result structure may
6657 be of integer types of any bit width, but they must have the same bit
6658 width. The second element of the result structure must be of
6659 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6660 undergo signed multiplication.</p>
6663 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6664 a signed multiplication of the two arguments. They return a structure —
6665 the first element of which is the multiplication, and the second element of
6666 which is a bit specifying if the signed multiplication resulted in an
6671 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6672 %sum = extractvalue {i32, i1} %res, 0
6673 %obit = extractvalue {i32, i1} %res, 1
6674 br i1 %obit, label %overflow, label %normal
6679 <!-- _______________________________________________________________________ -->
6680 <div class="doc_subsubsection">
6681 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6684 <div class="doc_text">
6687 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6688 on any integer bit width.</p>
6691 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6692 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6693 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6697 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6698 a unsigned multiplication of the two arguments, and indicate whether an
6699 overflow occurred during the unsigned multiplication.</p>
6702 <p>The arguments (%a and %b) and the first element of the result structure may
6703 be of integer types of any bit width, but they must have the same bit
6704 width. The second element of the result structure must be of
6705 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6706 undergo unsigned multiplication.</p>
6709 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6710 an unsigned multiplication of the two arguments. They return a structure
6711 — the first element of which is the multiplication, and the second
6712 element of which is a bit specifying if the unsigned multiplication resulted
6717 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6718 %sum = extractvalue {i32, i1} %res, 0
6719 %obit = extractvalue {i32, i1} %res, 1
6720 br i1 %obit, label %overflow, label %normal
6725 <!-- ======================================================================= -->
6726 <div class="doc_subsection">
6727 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
6730 <div class="doc_text">
6732 <p>Half precision floating point is a storage-only format. This means that it is
6733 a dense encoding (in memory) but does not support computation in the
6736 <p>This means that code must first load the half-precision floating point
6737 value as an i16, then convert it to float with <a
6738 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
6739 Computation can then be performed on the float value (including extending to
6740 double etc). To store the value back to memory, it is first converted to
6741 float if needed, then converted to i16 with
6742 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
6743 storing as an i16 value.</p>
6746 <!-- _______________________________________________________________________ -->
6747 <div class="doc_subsubsection">
6748 <a name="int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a>
6751 <div class="doc_text">
6755 declare i16 @llvm.convert.to.fp16(f32 %a)
6759 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
6760 a conversion from single precision floating point format to half precision
6761 floating point format.</p>
6764 <p>The intrinsic function contains single argument - the value to be
6768 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
6769 a conversion from single precision floating point format to half precision
6770 floating point format. The return value is an <tt>i16</tt> which
6771 contains the converted number.</p>
6775 %res = call i16 @llvm.convert.to.fp16(f32 %a)
6776 store i16 %res, i16* @x, align 2
6781 <!-- _______________________________________________________________________ -->
6782 <div class="doc_subsubsection">
6783 <a name="int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a>
6786 <div class="doc_text">
6790 declare f32 @llvm.convert.from.fp16(i16 %a)
6794 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
6795 a conversion from half precision floating point format to single precision
6796 floating point format.</p>
6799 <p>The intrinsic function contains single argument - the value to be
6803 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
6804 conversion from half single precision floating point format to single
6805 precision floating point format. The input half-float value is represented by
6806 an <tt>i16</tt> value.</p>
6810 %a = load i16* @x, align 2
6811 %res = call f32 @llvm.convert.from.fp16(i16 %a)
6816 <!-- ======================================================================= -->
6817 <div class="doc_subsection">
6818 <a name="int_debugger">Debugger Intrinsics</a>
6821 <div class="doc_text">
6823 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
6824 prefix), are described in
6825 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
6826 Level Debugging</a> document.</p>
6830 <!-- ======================================================================= -->
6831 <div class="doc_subsection">
6832 <a name="int_eh">Exception Handling Intrinsics</a>
6835 <div class="doc_text">
6837 <p>The LLVM exception handling intrinsics (which all start with
6838 <tt>llvm.eh.</tt> prefix), are described in
6839 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6840 Handling</a> document.</p>
6844 <!-- ======================================================================= -->
6845 <div class="doc_subsection">
6846 <a name="int_trampoline">Trampoline Intrinsic</a>
6849 <div class="doc_text">
6851 <p>This intrinsic makes it possible to excise one parameter, marked with
6852 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
6853 The result is a callable
6854 function pointer lacking the nest parameter - the caller does not need to
6855 provide a value for it. Instead, the value to use is stored in advance in a
6856 "trampoline", a block of memory usually allocated on the stack, which also
6857 contains code to splice the nest value into the argument list. This is used
6858 to implement the GCC nested function address extension.</p>
6860 <p>For example, if the function is
6861 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6862 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
6865 <pre class="doc_code">
6866 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6867 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6868 %p = call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval)
6869 %fp = bitcast i8* %p to i32 (i32, i32)*
6872 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
6873 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
6877 <!-- _______________________________________________________________________ -->
6878 <div class="doc_subsubsection">
6879 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6882 <div class="doc_text">
6886 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6890 <p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
6891 function pointer suitable for executing it.</p>
6894 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6895 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
6896 sufficiently aligned block of memory; this memory is written to by the
6897 intrinsic. Note that the size and the alignment are target-specific - LLVM
6898 currently provides no portable way of determining them, so a front-end that
6899 generates this intrinsic needs to have some target-specific knowledge.
6900 The <tt>func</tt> argument must hold a function bitcast to
6901 an <tt>i8*</tt>.</p>
6904 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
6905 dependent code, turning it into a function. A pointer to this function is
6906 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
6907 function pointer type</a> before being called. The new function's signature
6908 is the same as that of <tt>func</tt> with any arguments marked with
6909 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
6910 is allowed, and it must be of pointer type. Calling the new function is
6911 equivalent to calling <tt>func</tt> with the same argument list, but
6912 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
6913 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
6914 by <tt>tramp</tt> is modified, then the effect of any later call to the
6915 returned function pointer is undefined.</p>
6919 <!-- ======================================================================= -->
6920 <div class="doc_subsection">
6921 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6924 <div class="doc_text">
6926 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
6927 hardware constructs for atomic operations and memory synchronization. This
6928 provides an interface to the hardware, not an interface to the programmer. It
6929 is aimed at a low enough level to allow any programming models or APIs
6930 (Application Programming Interfaces) which need atomic behaviors to map
6931 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6932 hardware provides a "universal IR" for source languages, it also provides a
6933 starting point for developing a "universal" atomic operation and
6934 synchronization IR.</p>
6936 <p>These do <em>not</em> form an API such as high-level threading libraries,
6937 software transaction memory systems, atomic primitives, and intrinsic
6938 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6939 application libraries. The hardware interface provided by LLVM should allow
6940 a clean implementation of all of these APIs and parallel programming models.
6941 No one model or paradigm should be selected above others unless the hardware
6942 itself ubiquitously does so.</p>
6946 <!-- _______________________________________________________________________ -->
6947 <div class="doc_subsubsection">
6948 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6950 <div class="doc_text">
6953 declare void @llvm.memory.barrier(i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device>)
6957 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6958 specific pairs of memory access types.</p>
6961 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6962 The first four arguments enables a specific barrier as listed below. The
6963 fifth argument specifies that the barrier applies to io or device or uncached
6967 <li><tt>ll</tt>: load-load barrier</li>
6968 <li><tt>ls</tt>: load-store barrier</li>
6969 <li><tt>sl</tt>: store-load barrier</li>
6970 <li><tt>ss</tt>: store-store barrier</li>
6971 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6975 <p>This intrinsic causes the system to enforce some ordering constraints upon
6976 the loads and stores of the program. This barrier does not
6977 indicate <em>when</em> any events will occur, it only enforces
6978 an <em>order</em> in which they occur. For any of the specified pairs of load
6979 and store operations (f.ex. load-load, or store-load), all of the first
6980 operations preceding the barrier will complete before any of the second
6981 operations succeeding the barrier begin. Specifically the semantics for each
6982 pairing is as follows:</p>
6985 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6986 after the barrier begins.</li>
6987 <li><tt>ls</tt>: All loads before the barrier must complete before any
6988 store after the barrier begins.</li>
6989 <li><tt>ss</tt>: All stores before the barrier must complete before any
6990 store after the barrier begins.</li>
6991 <li><tt>sl</tt>: All stores before the barrier must complete before any
6992 load after the barrier begins.</li>
6995 <p>These semantics are applied with a logical "and" behavior when more than one
6996 is enabled in a single memory barrier intrinsic.</p>
6998 <p>Backends may implement stronger barriers than those requested when they do
6999 not support as fine grained a barrier as requested. Some architectures do
7000 not need all types of barriers and on such architectures, these become
7005 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7006 %ptr = bitcast i8* %mallocP to i32*
7009 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
7010 call void @llvm.memory.barrier(i1 false, i1 true, i1 false, i1 false)
7011 <i>; guarantee the above finishes</i>
7012 store i32 8, %ptr <i>; before this begins</i>
7017 <!-- _______________________________________________________________________ -->
7018 <div class="doc_subsubsection">
7019 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
7022 <div class="doc_text">
7025 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
7026 any integer bit width and for different address spaces. Not all targets
7027 support all bit widths however.</p>
7030 declare i8 @llvm.atomic.cmp.swap.i8.p0i8(i8* <ptr>, i8 <cmp>, i8 <val>)
7031 declare i16 @llvm.atomic.cmp.swap.i16.p0i16(i16* <ptr>, i16 <cmp>, i16 <val>)
7032 declare i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* <ptr>, i32 <cmp>, i32 <val>)
7033 declare i64 @llvm.atomic.cmp.swap.i64.p0i64(i64* <ptr>, i64 <cmp>, i64 <val>)
7037 <p>This loads a value in memory and compares it to a given value. If they are
7038 equal, it stores a new value into the memory.</p>
7041 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
7042 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
7043 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
7044 this integer type. While any bit width integer may be used, targets may only
7045 lower representations they support in hardware.</p>
7048 <p>This entire intrinsic must be executed atomically. It first loads the value
7049 in memory pointed to by <tt>ptr</tt> and compares it with the
7050 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
7051 memory. The loaded value is yielded in all cases. This provides the
7052 equivalent of an atomic compare-and-swap operation within the SSA
7057 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7058 %ptr = bitcast i8* %mallocP to i32*
7061 %val1 = add i32 4, 4
7062 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 4, %val1)
7063 <i>; yields {i32}:result1 = 4</i>
7064 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7065 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7067 %val2 = add i32 1, 1
7068 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 5, %val2)
7069 <i>; yields {i32}:result2 = 8</i>
7070 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
7072 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
7077 <!-- _______________________________________________________________________ -->
7078 <div class="doc_subsubsection">
7079 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
7081 <div class="doc_text">
7084 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
7085 integer bit width. Not all targets support all bit widths however.</p>
7088 declare i8 @llvm.atomic.swap.i8.p0i8(i8* <ptr>, i8 <val>)
7089 declare i16 @llvm.atomic.swap.i16.p0i16(i16* <ptr>, i16 <val>)
7090 declare i32 @llvm.atomic.swap.i32.p0i32(i32* <ptr>, i32 <val>)
7091 declare i64 @llvm.atomic.swap.i64.p0i64(i64* <ptr>, i64 <val>)
7095 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
7096 the value from memory. It then stores the value in <tt>val</tt> in the memory
7097 at <tt>ptr</tt>.</p>
7100 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
7101 the <tt>val</tt> argument and the result must be integers of the same bit
7102 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
7103 integer type. The targets may only lower integer representations they
7107 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
7108 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
7109 equivalent of an atomic swap operation within the SSA framework.</p>
7113 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7114 %ptr = bitcast i8* %mallocP to i32*
7117 %val1 = add i32 4, 4
7118 %result1 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val1)
7119 <i>; yields {i32}:result1 = 4</i>
7120 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7121 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7123 %val2 = add i32 1, 1
7124 %result2 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val2)
7125 <i>; yields {i32}:result2 = 8</i>
7127 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
7128 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
7133 <!-- _______________________________________________________________________ -->
7134 <div class="doc_subsubsection">
7135 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
7139 <div class="doc_text">
7142 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
7143 any integer bit width. Not all targets support all bit widths however.</p>
7146 declare i8 @llvm.atomic.load.add.i8.p0i8(i8* <ptr>, i8 <delta>)
7147 declare i16 @llvm.atomic.load.add.i16.p0i16(i16* <ptr>, i16 <delta>)
7148 declare i32 @llvm.atomic.load.add.i32.p0i32(i32* <ptr>, i32 <delta>)
7149 declare i64 @llvm.atomic.load.add.i64.p0i64(i64* <ptr>, i64 <delta>)
7153 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
7154 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
7157 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7158 and the second an integer value. The result is also an integer value. These
7159 integer types can have any bit width, but they must all have the same bit
7160 width. The targets may only lower integer representations they support.</p>
7163 <p>This intrinsic does a series of operations atomically. It first loads the
7164 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
7165 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
7169 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7170 %ptr = bitcast i8* %mallocP to i32*
7172 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 4)
7173 <i>; yields {i32}:result1 = 4</i>
7174 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 2)
7175 <i>; yields {i32}:result2 = 8</i>
7176 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 5)
7177 <i>; yields {i32}:result3 = 10</i>
7178 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
7183 <!-- _______________________________________________________________________ -->
7184 <div class="doc_subsubsection">
7185 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
7189 <div class="doc_text">
7192 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
7193 any integer bit width and for different address spaces. Not all targets
7194 support all bit widths however.</p>
7197 declare i8 @llvm.atomic.load.sub.i8.p0i32(i8* <ptr>, i8 <delta>)
7198 declare i16 @llvm.atomic.load.sub.i16.p0i32(i16* <ptr>, i16 <delta>)
7199 declare i32 @llvm.atomic.load.sub.i32.p0i32(i32* <ptr>, i32 <delta>)
7200 declare i64 @llvm.atomic.load.sub.i64.p0i32(i64* <ptr>, i64 <delta>)
7204 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
7205 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
7208 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7209 and the second an integer value. The result is also an integer value. These
7210 integer types can have any bit width, but they must all have the same bit
7211 width. The targets may only lower integer representations they support.</p>
7214 <p>This intrinsic does a series of operations atomically. It first loads the
7215 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
7216 result to <tt>ptr</tt>. It yields the original value stored
7217 at <tt>ptr</tt>.</p>
7221 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7222 %ptr = bitcast i8* %mallocP to i32*
7224 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 4)
7225 <i>; yields {i32}:result1 = 8</i>
7226 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 2)
7227 <i>; yields {i32}:result2 = 4</i>
7228 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 5)
7229 <i>; yields {i32}:result3 = 2</i>
7230 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
7235 <!-- _______________________________________________________________________ -->
7236 <div class="doc_subsubsection">
7237 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
7238 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
7239 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
7240 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
7243 <div class="doc_text">
7246 <p>These are overloaded intrinsics. You can
7247 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
7248 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
7249 bit width and for different address spaces. Not all targets support all bit
7253 declare i8 @llvm.atomic.load.and.i8.p0i8(i8* <ptr>, i8 <delta>)
7254 declare i16 @llvm.atomic.load.and.i16.p0i16(i16* <ptr>, i16 <delta>)
7255 declare i32 @llvm.atomic.load.and.i32.p0i32(i32* <ptr>, i32 <delta>)
7256 declare i64 @llvm.atomic.load.and.i64.p0i64(i64* <ptr>, i64 <delta>)
7260 declare i8 @llvm.atomic.load.or.i8.p0i8(i8* <ptr>, i8 <delta>)
7261 declare i16 @llvm.atomic.load.or.i16.p0i16(i16* <ptr>, i16 <delta>)
7262 declare i32 @llvm.atomic.load.or.i32.p0i32(i32* <ptr>, i32 <delta>)
7263 declare i64 @llvm.atomic.load.or.i64.p0i64(i64* <ptr>, i64 <delta>)
7267 declare i8 @llvm.atomic.load.nand.i8.p0i32(i8* <ptr>, i8 <delta>)
7268 declare i16 @llvm.atomic.load.nand.i16.p0i32(i16* <ptr>, i16 <delta>)
7269 declare i32 @llvm.atomic.load.nand.i32.p0i32(i32* <ptr>, i32 <delta>)
7270 declare i64 @llvm.atomic.load.nand.i64.p0i32(i64* <ptr>, i64 <delta>)
7274 declare i8 @llvm.atomic.load.xor.i8.p0i32(i8* <ptr>, i8 <delta>)
7275 declare i16 @llvm.atomic.load.xor.i16.p0i32(i16* <ptr>, i16 <delta>)
7276 declare i32 @llvm.atomic.load.xor.i32.p0i32(i32* <ptr>, i32 <delta>)
7277 declare i64 @llvm.atomic.load.xor.i64.p0i32(i64* <ptr>, i64 <delta>)
7281 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
7282 the value stored in memory at <tt>ptr</tt>. It yields the original value
7283 at <tt>ptr</tt>.</p>
7286 <p>These intrinsics take two arguments, the first a pointer to an integer value
7287 and the second an integer value. The result is also an integer value. These
7288 integer types can have any bit width, but they must all have the same bit
7289 width. The targets may only lower integer representations they support.</p>
7292 <p>These intrinsics does a series of operations atomically. They first load the
7293 value stored at <tt>ptr</tt>. They then do the bitwise
7294 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
7295 original value stored at <tt>ptr</tt>.</p>
7299 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7300 %ptr = bitcast i8* %mallocP to i32*
7301 store i32 0x0F0F, %ptr
7302 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32(i32* %ptr, i32 0xFF)
7303 <i>; yields {i32}:result0 = 0x0F0F</i>
7304 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32(i32* %ptr, i32 0xFF)
7305 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
7306 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32(i32* %ptr, i32 0F)
7307 <i>; yields {i32}:result2 = 0xF0</i>
7308 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32(i32* %ptr, i32 0F)
7309 <i>; yields {i32}:result3 = FF</i>
7310 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
7315 <!-- _______________________________________________________________________ -->
7316 <div class="doc_subsubsection">
7317 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
7318 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
7319 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
7320 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
7323 <div class="doc_text">
7326 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
7327 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
7328 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
7329 address spaces. Not all targets support all bit widths however.</p>
7332 declare i8 @llvm.atomic.load.max.i8.p0i8(i8* <ptr>, i8 <delta>)
7333 declare i16 @llvm.atomic.load.max.i16.p0i16(i16* <ptr>, i16 <delta>)
7334 declare i32 @llvm.atomic.load.max.i32.p0i32(i32* <ptr>, i32 <delta>)
7335 declare i64 @llvm.atomic.load.max.i64.p0i64(i64* <ptr>, i64 <delta>)
7339 declare i8 @llvm.atomic.load.min.i8.p0i8(i8* <ptr>, i8 <delta>)
7340 declare i16 @llvm.atomic.load.min.i16.p0i16(i16* <ptr>, i16 <delta>)
7341 declare i32 @llvm.atomic.load.min.i32.p0i32(i32* <ptr>, i32 <delta>)
7342 declare i64 @llvm.atomic.load.min.i64.p0i64(i64* <ptr>, i64 <delta>)
7346 declare i8 @llvm.atomic.load.umax.i8.p0i8(i8* <ptr>, i8 <delta>)
7347 declare i16 @llvm.atomic.load.umax.i16.p0i16(i16* <ptr>, i16 <delta>)
7348 declare i32 @llvm.atomic.load.umax.i32.p0i32(i32* <ptr>, i32 <delta>)
7349 declare i64 @llvm.atomic.load.umax.i64.p0i64(i64* <ptr>, i64 <delta>)
7353 declare i8 @llvm.atomic.load.umin.i8.p0i8(i8* <ptr>, i8 <delta>)
7354 declare i16 @llvm.atomic.load.umin.i16.p0i16(i16* <ptr>, i16 <delta>)
7355 declare i32 @llvm.atomic.load.umin.i32.p0i32(i32* <ptr>, i32 <delta>)
7356 declare i64 @llvm.atomic.load.umin.i64.p0i64(i64* <ptr>, i64 <delta>)
7360 <p>These intrinsics takes the signed or unsigned minimum or maximum of
7361 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
7362 original value at <tt>ptr</tt>.</p>
7365 <p>These intrinsics take two arguments, the first a pointer to an integer value
7366 and the second an integer value. The result is also an integer value. These
7367 integer types can have any bit width, but they must all have the same bit
7368 width. The targets may only lower integer representations they support.</p>
7371 <p>These intrinsics does a series of operations atomically. They first load the
7372 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
7373 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
7374 yield the original value stored at <tt>ptr</tt>.</p>
7378 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7379 %ptr = bitcast i8* %mallocP to i32*
7381 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32(i32* %ptr, i32 -2)
7382 <i>; yields {i32}:result0 = 7</i>
7383 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32(i32* %ptr, i32 8)
7384 <i>; yields {i32}:result1 = -2</i>
7385 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32(i32* %ptr, i32 10)
7386 <i>; yields {i32}:result2 = 8</i>
7387 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32(i32* %ptr, i32 30)
7388 <i>; yields {i32}:result3 = 8</i>
7389 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
7395 <!-- ======================================================================= -->
7396 <div class="doc_subsection">
7397 <a name="int_memorymarkers">Memory Use Markers</a>
7400 <div class="doc_text">
7402 <p>This class of intrinsics exists to information about the lifetime of memory
7403 objects and ranges where variables are immutable.</p>
7407 <!-- _______________________________________________________________________ -->
7408 <div class="doc_subsubsection">
7409 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
7412 <div class="doc_text">
7416 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
7420 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
7421 object's lifetime.</p>
7424 <p>The first argument is a constant integer representing the size of the
7425 object, or -1 if it is variable sized. The second argument is a pointer to
7429 <p>This intrinsic indicates that before this point in the code, the value of the
7430 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7431 never be used and has an undefined value. A load from the pointer that
7432 precedes this intrinsic can be replaced with
7433 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
7437 <!-- _______________________________________________________________________ -->
7438 <div class="doc_subsubsection">
7439 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
7442 <div class="doc_text">
7446 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
7450 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
7451 object's lifetime.</p>
7454 <p>The first argument is a constant integer representing the size of the
7455 object, or -1 if it is variable sized. The second argument is a pointer to
7459 <p>This intrinsic indicates that after this point in the code, the value of the
7460 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7461 never be used and has an undefined value. Any stores into the memory object
7462 following this intrinsic may be removed as dead.
7466 <!-- _______________________________________________________________________ -->
7467 <div class="doc_subsubsection">
7468 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
7471 <div class="doc_text">
7475 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>) readonly
7479 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
7480 a memory object will not change.</p>
7483 <p>The first argument is a constant integer representing the size of the
7484 object, or -1 if it is variable sized. The second argument is a pointer to
7488 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
7489 the return value, the referenced memory location is constant and
7494 <!-- _______________________________________________________________________ -->
7495 <div class="doc_subsubsection">
7496 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
7499 <div class="doc_text">
7503 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7507 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
7508 a memory object are mutable.</p>
7511 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
7512 The second argument is a constant integer representing the size of the
7513 object, or -1 if it is variable sized and the third argument is a pointer
7517 <p>This intrinsic indicates that the memory is mutable again.</p>
7521 <!-- ======================================================================= -->
7522 <div class="doc_subsection">
7523 <a name="int_general">General Intrinsics</a>
7526 <div class="doc_text">
7528 <p>This class of intrinsics is designed to be generic and has no specific
7533 <!-- _______________________________________________________________________ -->
7534 <div class="doc_subsubsection">
7535 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
7538 <div class="doc_text">
7542 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
7546 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
7549 <p>The first argument is a pointer to a value, the second is a pointer to a
7550 global string, the third is a pointer to a global string which is the source
7551 file name, and the last argument is the line number.</p>
7554 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7555 This can be useful for special purpose optimizations that want to look for
7556 these annotations. These have no other defined use, they are ignored by code
7557 generation and optimization.</p>
7561 <!-- _______________________________________________________________________ -->
7562 <div class="doc_subsubsection">
7563 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
7566 <div class="doc_text">
7569 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
7570 any integer bit width.</p>
7573 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
7574 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
7575 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
7576 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
7577 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
7581 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
7584 <p>The first argument is an integer value (result of some expression), the
7585 second is a pointer to a global string, the third is a pointer to a global
7586 string which is the source file name, and the last argument is the line
7587 number. It returns the value of the first argument.</p>
7590 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7591 arbitrary strings. This can be useful for special purpose optimizations that
7592 want to look for these annotations. These have no other defined use, they
7593 are ignored by code generation and optimization.</p>
7597 <!-- _______________________________________________________________________ -->
7598 <div class="doc_subsubsection">
7599 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
7602 <div class="doc_text">
7606 declare void @llvm.trap()
7610 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
7616 <p>This intrinsics is lowered to the target dependent trap instruction. If the
7617 target does not have a trap instruction, this intrinsic will be lowered to
7618 the call of the <tt>abort()</tt> function.</p>
7622 <!-- _______________________________________________________________________ -->
7623 <div class="doc_subsubsection">
7624 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
7627 <div class="doc_text">
7631 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
7635 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
7636 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
7637 ensure that it is placed on the stack before local variables.</p>
7640 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
7641 arguments. The first argument is the value loaded from the stack
7642 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
7643 that has enough space to hold the value of the guard.</p>
7646 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
7647 the <tt>AllocaInst</tt> stack slot to be before local variables on the
7648 stack. This is to ensure that if a local variable on the stack is
7649 overwritten, it will destroy the value of the guard. When the function exits,
7650 the guard on the stack is checked against the original guard. If they're
7651 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
7656 <!-- _______________________________________________________________________ -->
7657 <div class="doc_subsubsection">
7658 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
7661 <div class="doc_text">
7665 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <type>)
7666 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <type>)
7670 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information
7671 to the optimizers to discover at compile time either a) when an
7672 operation like memcpy will either overflow a buffer that corresponds to
7673 an object, or b) to determine that a runtime check for overflow isn't
7674 necessary. An object in this context means an allocation of a
7675 specific class, structure, array, or other object.</p>
7678 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
7679 argument is a pointer to or into the <tt>object</tt>. The second argument
7680 is a boolean 0 or 1. This argument determines whether you want the
7681 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
7682 1, variables are not allowed.</p>
7685 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
7686 representing the size of the object concerned or <tt>i32/i64 -1 or 0</tt>
7687 (depending on the <tt>type</tt> argument if the size cannot be determined
7688 at compile time.</p>
7692 <!-- *********************************************************************** -->
7695 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
7696 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
7697 <a href="http://validator.w3.org/check/referer"><img
7698 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
7700 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
7701 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
7702 Last modified: $Date$