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="_static/llvm.css" type="text/css">
15 <h1>LLVM Language Reference Manual</h1>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a>
25 <li><a href="#linkage_private">'<tt>private</tt>' Linkage</a></li>
26 <li><a href="#linkage_linker_private">'<tt>linker_private</tt>' Linkage</a></li>
27 <li><a href="#linkage_linker_private_weak">'<tt>linker_private_weak</tt>' Linkage</a></li>
28 <li><a href="#linkage_internal">'<tt>internal</tt>' Linkage</a></li>
29 <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
30 <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
31 <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
32 <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
33 <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
34 <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
35 <li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
36 <li><a href="#linkage_linkonce_odr_auto_hide">'<tt>linkonce_odr_auto_hide</tt>' Linkage</a></li>
37 <li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
38 <li><a href="#linkage_external">'<tt>external</tt>' Linkage</a></li>
39 <li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
40 <li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
43 <li><a href="#callingconv">Calling Conventions</a></li>
44 <li><a href="#namedtypes">Named Types</a></li>
45 <li><a href="#globalvars">Global Variables</a></li>
46 <li><a href="#functionstructure">Functions</a></li>
47 <li><a href="#aliasstructure">Aliases</a></li>
48 <li><a href="#namedmetadatastructure">Named Metadata</a></li>
49 <li><a href="#paramattrs">Parameter Attributes</a></li>
50 <li><a href="#fnattrs">Function Attributes</a></li>
51 <li><a href="#gc">Garbage Collector Names</a></li>
52 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
53 <li><a href="#datalayout">Data Layout</a></li>
54 <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
55 <li><a href="#volatile">Volatile Memory Accesses</a></li>
56 <li><a href="#memmodel">Memory Model for Concurrent Operations</a></li>
57 <li><a href="#ordering">Atomic Memory Ordering Constraints</a></li>
60 <li><a href="#typesystem">Type System</a>
62 <li><a href="#t_classifications">Type Classifications</a></li>
63 <li><a href="#t_primitive">Primitive Types</a>
65 <li><a href="#t_integer">Integer Type</a></li>
66 <li><a href="#t_floating">Floating Point Types</a></li>
67 <li><a href="#t_x86mmx">X86mmx Type</a></li>
68 <li><a href="#t_void">Void Type</a></li>
69 <li><a href="#t_label">Label Type</a></li>
70 <li><a href="#t_metadata">Metadata Type</a></li>
73 <li><a href="#t_derived">Derived Types</a>
75 <li><a href="#t_aggregate">Aggregate Types</a>
77 <li><a href="#t_array">Array Type</a></li>
78 <li><a href="#t_struct">Structure Type</a></li>
79 <li><a href="#t_opaque">Opaque Structure Types</a></li>
80 <li><a href="#t_vector">Vector Type</a></li>
83 <li><a href="#t_function">Function Type</a></li>
84 <li><a href="#t_pointer">Pointer Type</a></li>
89 <li><a href="#constants">Constants</a>
91 <li><a href="#simpleconstants">Simple Constants</a></li>
92 <li><a href="#complexconstants">Complex Constants</a></li>
93 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
94 <li><a href="#undefvalues">Undefined Values</a></li>
95 <li><a href="#poisonvalues">Poison Values</a></li>
96 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
97 <li><a href="#constantexprs">Constant Expressions</a></li>
100 <li><a href="#othervalues">Other Values</a>
102 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
103 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a>
105 <li><a href="#tbaa">'<tt>tbaa</tt>' Metadata</a></li>
106 <li><a href="#tbaa.struct">'<tt>tbaa.struct</tt>' Metadata</a></li>
107 <li><a href="#fpmath">'<tt>fpmath</tt>' Metadata</a></li>
108 <li><a href="#range">'<tt>range</tt>' Metadata</a></li>
113 <li><a href="#module_flags">Module Flags Metadata</a>
115 <li><a href="#objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a></li>
118 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
120 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
121 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
122 Global Variable</a></li>
123 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
124 Global Variable</a></li>
125 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
126 Global Variable</a></li>
129 <li><a href="#instref">Instruction Reference</a>
131 <li><a href="#terminators">Terminator Instructions</a>
133 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
134 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
135 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
136 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
137 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
138 <li><a href="#i_resume">'<tt>resume</tt>' Instruction</a></li>
139 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
142 <li><a href="#binaryops">Binary Operations</a>
144 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
145 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
146 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
147 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
148 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
149 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
150 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
151 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
152 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
153 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
154 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
155 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
158 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
160 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
161 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
162 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
163 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
164 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
165 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
168 <li><a href="#vectorops">Vector Operations</a>
170 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
171 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
172 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
175 <li><a href="#aggregateops">Aggregate Operations</a>
177 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
178 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
181 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
183 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
184 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
185 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
186 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
187 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
188 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
189 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
192 <li><a href="#convertops">Conversion Operations</a>
194 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
195 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
196 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
197 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
198 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
199 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
200 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
201 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
202 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
203 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
204 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
205 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
208 <li><a href="#otherops">Other Operations</a>
210 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
211 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
212 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
213 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
214 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
215 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
216 <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
221 <li><a href="#intrinsics">Intrinsic Functions</a>
223 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
225 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
226 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
227 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
230 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
232 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
233 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
234 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
237 <li><a href="#int_codegen">Code Generator Intrinsics</a>
239 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
240 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
241 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
242 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
243 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
244 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
245 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
248 <li><a href="#int_libc">Standard C Library Intrinsics</a>
250 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
251 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
252 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
253 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
254 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
255 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
256 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
257 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
258 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
259 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
260 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
261 <li><a href="#int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a></li>
262 <li><a href="#int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a></li>
263 <li><a href="#int_ceil">'<tt>llvm.ceil.*</tt>' Intrinsic</a></li>
264 <li><a href="#int_trunc">'<tt>llvm.trunc.*</tt>' Intrinsic</a></li>
265 <li><a href="#int_rint">'<tt>llvm.rint.*</tt>' Intrinsic</a></li>
266 <li><a href="#int_nearbyint">'<tt>llvm.nearbyint.*</tt>' Intrinsic</a></li>
269 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
271 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
272 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
273 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
274 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
277 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
279 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
280 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
281 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
282 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
283 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
284 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
287 <li><a href="#spec_arithmetic">Specialised Arithmetic Intrinsics</a>
289 <li><a href="#fmuladd">'<tt>llvm.fmuladd</tt> Intrinsic</a></li>
292 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
294 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
295 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
298 <li><a href="#int_debugger">Debugger intrinsics</a></li>
299 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
300 <li><a href="#int_trampoline">Trampoline Intrinsics</a>
302 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
303 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
306 <li><a href="#int_memorymarkers">Memory Use Markers</a>
308 <li><a href="#int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a></li>
309 <li><a href="#int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a></li>
310 <li><a href="#int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a></li>
311 <li><a href="#int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a></li>
314 <li><a href="#int_general">General intrinsics</a>
316 <li><a href="#int_var_annotation">
317 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
318 <li><a href="#int_annotation">
319 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
320 <li><a href="#int_trap">
321 '<tt>llvm.trap</tt>' Intrinsic</a></li>
322 <li><a href="#int_debugtrap">
323 '<tt>llvm.debugtrap</tt>' Intrinsic</a></li>
324 <li><a href="#int_stackprotector">
325 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
326 <li><a href="#int_objectsize">
327 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
328 <li><a href="#int_expect">
329 '<tt>llvm.expect</tt>' Intrinsic</a></li>
330 <li><a href="#int_donothing">
331 '<tt>llvm.donothing</tt>' Intrinsic</a></li>
338 <div class="doc_author">
339 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
340 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
343 <!-- *********************************************************************** -->
344 <h2><a name="abstract">Abstract</a></h2>
345 <!-- *********************************************************************** -->
349 <p>This document is a reference manual for the LLVM assembly language. LLVM is
350 a Static Single Assignment (SSA) based representation that provides type
351 safety, low-level operations, flexibility, and the capability of representing
352 'all' high-level languages cleanly. It is the common code representation
353 used throughout all phases of the LLVM compilation strategy.</p>
357 <!-- *********************************************************************** -->
358 <h2><a name="introduction">Introduction</a></h2>
359 <!-- *********************************************************************** -->
363 <p>The LLVM code representation is designed to be used in three different forms:
364 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
365 for fast loading by a Just-In-Time compiler), and as a human readable
366 assembly language representation. This allows LLVM to provide a powerful
367 intermediate representation for efficient compiler transformations and
368 analysis, while providing a natural means to debug and visualize the
369 transformations. The three different forms of LLVM are all equivalent. This
370 document describes the human readable representation and notation.</p>
372 <p>The LLVM representation aims to be light-weight and low-level while being
373 expressive, typed, and extensible at the same time. It aims to be a
374 "universal IR" of sorts, by being at a low enough level that high-level ideas
375 may be cleanly mapped to it (similar to how microprocessors are "universal
376 IR's", allowing many source languages to be mapped to them). By providing
377 type information, LLVM can be used as the target of optimizations: for
378 example, through pointer analysis, it can be proven that a C automatic
379 variable is never accessed outside of the current function, allowing it to
380 be promoted to a simple SSA value instead of a memory location.</p>
382 <!-- _______________________________________________________________________ -->
384 <a name="wellformed">Well-Formedness</a>
389 <p>It is important to note that this document describes 'well formed' LLVM
390 assembly language. There is a difference between what the parser accepts and
391 what is considered 'well formed'. For example, the following instruction is
392 syntactically okay, but not well formed:</p>
394 <pre class="doc_code">
395 %x = <a href="#i_add">add</a> i32 1, %x
398 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
399 LLVM infrastructure provides a verification pass that may be used to verify
400 that an LLVM module is well formed. This pass is automatically run by the
401 parser after parsing input assembly and by the optimizer before it outputs
402 bitcode. The violations pointed out by the verifier pass indicate bugs in
403 transformation passes or input to the parser.</p>
409 <!-- Describe the typesetting conventions here. -->
411 <!-- *********************************************************************** -->
412 <h2><a name="identifiers">Identifiers</a></h2>
413 <!-- *********************************************************************** -->
417 <p>LLVM identifiers come in two basic types: global and local. Global
418 identifiers (functions, global variables) begin with the <tt>'@'</tt>
419 character. Local identifiers (register names, types) begin with
420 the <tt>'%'</tt> character. Additionally, there are three different formats
421 for identifiers, for different purposes:</p>
424 <li>Named values are represented as a string of characters with their prefix.
425 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
426 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
427 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
428 other characters in their names can be surrounded with quotes. Special
429 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
430 ASCII code for the character in hexadecimal. In this way, any character
431 can be used in a name value, even quotes themselves.</li>
433 <li>Unnamed values are represented as an unsigned numeric value with their
434 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
436 <li>Constants, which are described in a <a href="#constants">section about
437 constants</a>, below.</li>
440 <p>LLVM requires that values start with a prefix for two reasons: Compilers
441 don't need to worry about name clashes with reserved words, and the set of
442 reserved words may be expanded in the future without penalty. Additionally,
443 unnamed identifiers allow a compiler to quickly come up with a temporary
444 variable without having to avoid symbol table conflicts.</p>
446 <p>Reserved words in LLVM are very similar to reserved words in other
447 languages. There are keywords for different opcodes
448 ('<tt><a href="#i_add">add</a></tt>',
449 '<tt><a href="#i_bitcast">bitcast</a></tt>',
450 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
451 ('<tt><a href="#t_void">void</a></tt>',
452 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
453 reserved words cannot conflict with variable names, because none of them
454 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
456 <p>Here is an example of LLVM code to multiply the integer variable
457 '<tt>%X</tt>' by 8:</p>
461 <pre class="doc_code">
462 %result = <a href="#i_mul">mul</a> i32 %X, 8
465 <p>After strength reduction:</p>
467 <pre class="doc_code">
468 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
471 <p>And the hard way:</p>
473 <pre class="doc_code">
474 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
475 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
476 %result = <a href="#i_add">add</a> i32 %1, %1
479 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
480 lexical features of LLVM:</p>
483 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
486 <li>Unnamed temporaries are created when the result of a computation is not
487 assigned to a named value.</li>
489 <li>Unnamed temporaries are numbered sequentially</li>
492 <p>It also shows a convention that we follow in this document. When
493 demonstrating instructions, we will follow an instruction with a comment that
494 defines the type and name of value produced. Comments are shown in italic
499 <!-- *********************************************************************** -->
500 <h2><a name="highlevel">High Level Structure</a></h2>
501 <!-- *********************************************************************** -->
503 <!-- ======================================================================= -->
505 <a name="modulestructure">Module Structure</a>
510 <p>LLVM programs are composed of <tt>Module</tt>s, each of which is a
511 translation unit of the input programs. Each module consists of functions,
512 global variables, and symbol table entries. Modules may be combined together
513 with the LLVM linker, which merges function (and global variable)
514 definitions, resolves forward declarations, and merges symbol table
515 entries. Here is an example of the "hello world" module:</p>
517 <pre class="doc_code">
518 <i>; Declare the string constant as a global constant.</i>
519 <a href="#identifiers">@.str</a> = <a href="#linkage_private">private</a> <a href="#globalvars">unnamed_addr</a> <a href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00"
521 <i>; External declaration of the puts function</i>
522 <a href="#functionstructure">declare</a> i32 @puts(i8* <a href="#nocapture">nocapture</a>) <a href="#fnattrs">nounwind</a>
524 <i>; Definition of main function</i>
525 define i32 @main() { <i>; i32()* </i>
526 <i>; Convert [13 x i8]* to i8 *...</i>
527 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.str, i64 0, i64 0
529 <i>; Call puts function to write out the string to stdout.</i>
530 <a href="#i_call">call</a> i32 @puts(i8* %cast210)
531 <a href="#i_ret">ret</a> i32 0
534 <i>; Named metadata</i>
535 !1 = metadata !{i32 42}
539 <p>This example is made up of a <a href="#globalvars">global variable</a> named
540 "<tt>.str</tt>", an external declaration of the "<tt>puts</tt>" function,
541 a <a href="#functionstructure">function definition</a> for
542 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
545 <p>In general, a module is made up of a list of global values (where both
546 functions and global variables are global values). Global values are
547 represented by a pointer to a memory location (in this case, a pointer to an
548 array of char, and a pointer to a function), and have one of the
549 following <a href="#linkage">linkage types</a>.</p>
553 <!-- ======================================================================= -->
555 <a name="linkage">Linkage Types</a>
560 <p>All Global Variables and Functions have one of the following types of
564 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
565 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
566 by objects in the current module. In particular, linking code into a
567 module with an private global value may cause the private to be renamed as
568 necessary to avoid collisions. Because the symbol is private to the
569 module, all references can be updated. This doesn't show up in any symbol
570 table in the object file.</dd>
572 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
573 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
574 assembler and evaluated by the linker. Unlike normal strong symbols, they
575 are removed by the linker from the final linked image (executable or
576 dynamic library).</dd>
578 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
579 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
580 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
581 linker. The symbols are removed by the linker from the final linked image
582 (executable or dynamic library).</dd>
584 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
585 <dd>Similar to private, but the value shows as a local symbol
586 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
587 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
589 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
590 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
591 into the object file corresponding to the LLVM module. They exist to
592 allow inlining and other optimizations to take place given knowledge of
593 the definition of the global, which is known to be somewhere outside the
594 module. Globals with <tt>available_externally</tt> linkage are allowed to
595 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
596 This linkage type is only allowed on definitions, not declarations.</dd>
598 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
599 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
600 the same name when linkage occurs. This can be used to implement
601 some forms of inline functions, templates, or other code which must be
602 generated in each translation unit that uses it, but where the body may
603 be overridden with a more definitive definition later. Unreferenced
604 <tt>linkonce</tt> globals are allowed to be discarded. Note that
605 <tt>linkonce</tt> linkage does not actually allow the optimizer to
606 inline the body of this function into callers because it doesn't know if
607 this definition of the function is the definitive definition within the
608 program or whether it will be overridden by a stronger definition.
609 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
612 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
613 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
614 <tt>linkonce</tt> linkage, except that unreferenced globals with
615 <tt>weak</tt> linkage may not be discarded. This is used for globals that
616 are declared "weak" in C source code.</dd>
618 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
619 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
620 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
622 Symbols with "<tt>common</tt>" linkage are merged in the same way as
623 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
624 <tt>common</tt> symbols may not have an explicit section,
625 must have a zero initializer, and may not be marked '<a
626 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
627 have common linkage.</dd>
630 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
631 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
632 pointer to array type. When two global variables with appending linkage
633 are linked together, the two global arrays are appended together. This is
634 the LLVM, typesafe, equivalent of having the system linker append together
635 "sections" with identical names when .o files are linked.</dd>
637 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
638 <dd>The semantics of this linkage follow the ELF object file model: the symbol
639 is weak until linked, if not linked, the symbol becomes null instead of
640 being an undefined reference.</dd>
642 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
643 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
644 <dd>Some languages allow differing globals to be merged, such as two functions
645 with different semantics. Other languages, such as <tt>C++</tt>, ensure
646 that only equivalent globals are ever merged (the "one definition rule"
647 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
648 and <tt>weak_odr</tt> linkage types to indicate that the global will only
649 be merged with equivalent globals. These linkage types are otherwise the
650 same as their non-<tt>odr</tt> versions.</dd>
652 <dt><tt><b><a name="linkage_linkonce_odr_auto_hide">linkonce_odr_auto_hide</a></b></tt></dt>
653 <dd>Similar to "<tt>linkonce_odr</tt>", but nothing in the translation unit
654 takes the address of this definition. For instance, functions that had an
655 inline definition, but the compiler decided not to inline it.
656 <tt>linkonce_odr_auto_hide</tt> may have only <tt>default</tt> visibility.
657 The symbols are removed by the linker from the final linked image
658 (executable or dynamic library).</dd>
660 <dt><tt><b><a name="linkage_external">external</a></b></tt></dt>
661 <dd>If none of the above identifiers are used, the global is externally
662 visible, meaning that it participates in linkage and can be used to
663 resolve external symbol references.</dd>
666 <p>The next two types of linkage are targeted for Microsoft Windows platform
667 only. They are designed to support importing (exporting) symbols from (to)
668 DLLs (Dynamic Link Libraries).</p>
671 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
672 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
673 or variable via a global pointer to a pointer that is set up by the DLL
674 exporting the symbol. On Microsoft Windows targets, the pointer name is
675 formed by combining <code>__imp_</code> and the function or variable
678 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
679 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
680 pointer to a pointer in a DLL, so that it can be referenced with the
681 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
682 name is formed by combining <code>__imp_</code> and the function or
686 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
687 another module defined a "<tt>.LC0</tt>" variable and was linked with this
688 one, one of the two would be renamed, preventing a collision. Since
689 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
690 declarations), they are accessible outside of the current module.</p>
692 <p>It is illegal for a function <i>declaration</i> to have any linkage type
693 other than <tt>external</tt>, <tt>dllimport</tt>
694 or <tt>extern_weak</tt>.</p>
696 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
697 or <tt>weak_odr</tt> linkages.</p>
701 <!-- ======================================================================= -->
703 <a name="callingconv">Calling Conventions</a>
708 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
709 and <a href="#i_invoke">invokes</a> can all have an optional calling
710 convention specified for the call. The calling convention of any pair of
711 dynamic caller/callee must match, or the behavior of the program is
712 undefined. The following calling conventions are supported by LLVM, and more
713 may be added in the future:</p>
716 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
717 <dd>This calling convention (the default if no other calling convention is
718 specified) matches the target C calling conventions. This calling
719 convention supports varargs function calls and tolerates some mismatch in
720 the declared prototype and implemented declaration of the function (as
723 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
724 <dd>This calling convention attempts to make calls as fast as possible
725 (e.g. by passing things in registers). This calling convention allows the
726 target to use whatever tricks it wants to produce fast code for the
727 target, without having to conform to an externally specified ABI
728 (Application Binary Interface).
729 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
730 when this or the GHC convention is used.</a> This calling convention
731 does not support varargs and requires the prototype of all callees to
732 exactly match the prototype of the function definition.</dd>
734 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
735 <dd>This calling convention attempts to make code in the caller as efficient
736 as possible under the assumption that the call is not commonly executed.
737 As such, these calls often preserve all registers so that the call does
738 not break any live ranges in the caller side. This calling convention
739 does not support varargs and requires the prototype of all callees to
740 exactly match the prototype of the function definition.</dd>
742 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
743 <dd>This calling convention has been implemented specifically for use by the
744 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
745 It passes everything in registers, going to extremes to achieve this by
746 disabling callee save registers. This calling convention should not be
747 used lightly but only for specific situations such as an alternative to
748 the <em>register pinning</em> performance technique often used when
749 implementing functional programming languages.At the moment only X86
750 supports this convention and it has the following limitations:
752 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
753 floating point types are supported.</li>
754 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
755 6 floating point parameters.</li>
757 This calling convention supports
758 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
759 requires both the caller and callee are using it.
762 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
763 <dd>Any calling convention may be specified by number, allowing
764 target-specific calling conventions to be used. Target specific calling
765 conventions start at 64.</dd>
768 <p>More calling conventions can be added/defined on an as-needed basis, to
769 support Pascal conventions or any other well-known target-independent
774 <!-- ======================================================================= -->
776 <a name="visibility">Visibility Styles</a>
781 <p>All Global Variables and Functions have one of the following visibility
785 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
786 <dd>On targets that use the ELF object file format, default visibility means
787 that the declaration is visible to other modules and, in shared libraries,
788 means that the declared entity may be overridden. On Darwin, default
789 visibility means that the declaration is visible to other modules. Default
790 visibility corresponds to "external linkage" in the language.</dd>
792 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
793 <dd>Two declarations of an object with hidden visibility refer to the same
794 object if they are in the same shared object. Usually, hidden visibility
795 indicates that the symbol will not be placed into the dynamic symbol
796 table, so no other module (executable or shared library) can reference it
799 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
800 <dd>On ELF, protected visibility indicates that the symbol will be placed in
801 the dynamic symbol table, but that references within the defining module
802 will bind to the local symbol. That is, the symbol cannot be overridden by
808 <!-- ======================================================================= -->
810 <a name="namedtypes">Named Types</a>
815 <p>LLVM IR allows you to specify name aliases for certain types. This can make
816 it easier to read the IR and make the IR more condensed (particularly when
817 recursive types are involved). An example of a name specification is:</p>
819 <pre class="doc_code">
820 %mytype = type { %mytype*, i32 }
823 <p>You may give a name to any <a href="#typesystem">type</a> except
824 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
825 is expected with the syntax "%mytype".</p>
827 <p>Note that type names are aliases for the structural type that they indicate,
828 and that you can therefore specify multiple names for the same type. This
829 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
830 uses structural typing, the name is not part of the type. When printing out
831 LLVM IR, the printer will pick <em>one name</em> to render all types of a
832 particular shape. This means that if you have code where two different
833 source types end up having the same LLVM type, that the dumper will sometimes
834 print the "wrong" or unexpected type. This is an important design point and
835 isn't going to change.</p>
839 <!-- ======================================================================= -->
841 <a name="globalvars">Global Variables</a>
846 <p>Global variables define regions of memory allocated at compilation time
847 instead of run-time. Global variables may optionally be initialized, may
848 have an explicit section to be placed in, and may have an optional explicit
849 alignment specified.</p>
851 <p>A variable may be defined as <tt>thread_local</tt>, which
852 means that it will not be shared by threads (each thread will have a
853 separated copy of the variable). Not all targets support thread-local
854 variables. Optionally, a TLS model may be specified:</p>
857 <dt><b><tt>localdynamic</tt></b>:</dt>
858 <dd>For variables that are only used within the current shared library.</dd>
860 <dt><b><tt>initialexec</tt></b>:</dt>
861 <dd>For variables in modules that will not be loaded dynamically.</dd>
863 <dt><b><tt>localexec</tt></b>:</dt>
864 <dd>For variables defined in the executable and only used within it.</dd>
867 <p>The models correspond to the ELF TLS models; see
868 <a href="http://people.redhat.com/drepper/tls.pdf">ELF
869 Handling For Thread-Local Storage</a> for more information on under which
870 circumstances the different models may be used. The target may choose a
871 different TLS model if the specified model is not supported, or if a better
872 choice of model can be made.</p>
874 <p>A variable may be defined as a global
875 "constant," which indicates that the contents of the variable
876 will <b>never</b> be modified (enabling better optimization, allowing the
877 global data to be placed in the read-only section of an executable, etc).
878 Note that variables that need runtime initialization cannot be marked
879 "constant" as there is a store to the variable.</p>
881 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
882 constant, even if the final definition of the global is not. This capability
883 can be used to enable slightly better optimization of the program, but
884 requires the language definition to guarantee that optimizations based on the
885 'constantness' are valid for the translation units that do not include the
888 <p>As SSA values, global variables define pointer values that are in scope
889 (i.e. they dominate) all basic blocks in the program. Global variables
890 always define a pointer to their "content" type because they describe a
891 region of memory, and all memory objects in LLVM are accessed through
894 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
895 that the address is not significant, only the content. Constants marked
896 like this can be merged with other constants if they have the same
897 initializer. Note that a constant with significant address <em>can</em>
898 be merged with a <tt>unnamed_addr</tt> constant, the result being a
899 constant whose address is significant.</p>
901 <p>A global variable may be declared to reside in a target-specific numbered
902 address space. For targets that support them, address spaces may affect how
903 optimizations are performed and/or what target instructions are used to
904 access the variable. The default address space is zero. The address space
905 qualifier must precede any other attributes.</p>
907 <p>LLVM allows an explicit section to be specified for globals. If the target
908 supports it, it will emit globals to the section specified.</p>
910 <p>An explicit alignment may be specified for a global, which must be a power
911 of 2. If not present, or if the alignment is set to zero, the alignment of
912 the global is set by the target to whatever it feels convenient. If an
913 explicit alignment is specified, the global is forced to have exactly that
914 alignment. Targets and optimizers are not allowed to over-align the global
915 if the global has an assigned section. In this case, the extra alignment
916 could be observable: for example, code could assume that the globals are
917 densely packed in their section and try to iterate over them as an array,
918 alignment padding would break this iteration.</p>
920 <p>For example, the following defines a global in a numbered address space with
921 an initializer, section, and alignment:</p>
923 <pre class="doc_code">
924 @G = addrspace(5) constant float 1.0, section "foo", align 4
927 <p>The following example defines a thread-local global with
928 the <tt>initialexec</tt> TLS model:</p>
930 <pre class="doc_code">
931 @G = thread_local(initialexec) global i32 0, align 4
937 <!-- ======================================================================= -->
939 <a name="functionstructure">Functions</a>
944 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
945 optional <a href="#linkage">linkage type</a>, an optional
946 <a href="#visibility">visibility style</a>, an optional
947 <a href="#callingconv">calling convention</a>,
948 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
949 <a href="#paramattrs">parameter attribute</a> for the return type, a function
950 name, a (possibly empty) argument list (each with optional
951 <a href="#paramattrs">parameter attributes</a>), optional
952 <a href="#fnattrs">function attributes</a>, an optional section, an optional
953 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
954 curly brace, a list of basic blocks, and a closing curly brace.</p>
956 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
957 optional <a href="#linkage">linkage type</a>, an optional
958 <a href="#visibility">visibility style</a>, an optional
959 <a href="#callingconv">calling convention</a>,
960 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
961 <a href="#paramattrs">parameter attribute</a> for the return type, a function
962 name, a possibly empty list of arguments, an optional alignment, and an
963 optional <a href="#gc">garbage collector name</a>.</p>
965 <p>A function definition contains a list of basic blocks, forming the CFG
966 (Control Flow Graph) for the function. Each basic block may optionally start
967 with a label (giving the basic block a symbol table entry), contains a list
968 of instructions, and ends with a <a href="#terminators">terminator</a>
969 instruction (such as a branch or function return).</p>
971 <p>The first basic block in a function is special in two ways: it is immediately
972 executed on entrance to the function, and it is not allowed to have
973 predecessor basic blocks (i.e. there can not be any branches to the entry
974 block of a function). Because the block can have no predecessors, it also
975 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
977 <p>LLVM allows an explicit section to be specified for functions. If the target
978 supports it, it will emit functions to the section specified.</p>
980 <p>An explicit alignment may be specified for a function. If not present, or if
981 the alignment is set to zero, the alignment of the function is set by the
982 target to whatever it feels convenient. If an explicit alignment is
983 specified, the function is forced to have at least that much alignment. All
984 alignments must be a power of 2.</p>
986 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
987 be significant and two identical functions can be merged.</p>
990 <pre class="doc_code">
991 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
992 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
993 <ResultType> @<FunctionName> ([argument list])
994 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
995 [<a href="#gc">gc</a>] { ... }
1000 <!-- ======================================================================= -->
1002 <a name="aliasstructure">Aliases</a>
1007 <p>Aliases act as "second name" for the aliasee value (which can be either
1008 function, global variable, another alias or bitcast of global value). Aliases
1009 may have an optional <a href="#linkage">linkage type</a>, and an
1010 optional <a href="#visibility">visibility style</a>.</p>
1013 <pre class="doc_code">
1014 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
1019 <!-- ======================================================================= -->
1021 <a name="namedmetadatastructure">Named Metadata</a>
1026 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
1027 nodes</a> (but not metadata strings) are the only valid operands for
1028 a named metadata.</p>
1031 <pre class="doc_code">
1032 ; Some unnamed metadata nodes, which are referenced by the named metadata.
1033 !0 = metadata !{metadata !"zero"}
1034 !1 = metadata !{metadata !"one"}
1035 !2 = metadata !{metadata !"two"}
1037 !name = !{!0, !1, !2}
1042 <!-- ======================================================================= -->
1044 <a name="paramattrs">Parameter Attributes</a>
1049 <p>The return type and each parameter of a function type may have a set of
1050 <i>parameter attributes</i> associated with them. Parameter attributes are
1051 used to communicate additional information about the result or parameters of
1052 a function. Parameter attributes are considered to be part of the function,
1053 not of the function type, so functions with different parameter attributes
1054 can have the same function type.</p>
1056 <p>Parameter attributes are simple keywords that follow the type specified. If
1057 multiple parameter attributes are needed, they are space separated. For
1060 <pre class="doc_code">
1061 declare i32 @printf(i8* noalias nocapture, ...)
1062 declare i32 @atoi(i8 zeroext)
1063 declare signext i8 @returns_signed_char()
1066 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1067 <tt>readonly</tt>) come immediately after the argument list.</p>
1069 <p>Currently, only the following parameter attributes are defined:</p>
1072 <dt><tt><b>zeroext</b></tt></dt>
1073 <dd>This indicates to the code generator that the parameter or return value
1074 should be zero-extended to the extent required by the target's ABI (which
1075 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1076 parameter) or the callee (for a return value).</dd>
1078 <dt><tt><b>signext</b></tt></dt>
1079 <dd>This indicates to the code generator that the parameter or return value
1080 should be sign-extended to the extent required by the target's ABI (which
1081 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1084 <dt><tt><b>inreg</b></tt></dt>
1085 <dd>This indicates that this parameter or return value should be treated in a
1086 special target-dependent fashion during while emitting code for a function
1087 call or return (usually, by putting it in a register as opposed to memory,
1088 though some targets use it to distinguish between two different kinds of
1089 registers). Use of this attribute is target-specific.</dd>
1091 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1092 <dd><p>This indicates that the pointer parameter should really be passed by
1093 value to the function. The attribute implies that a hidden copy of the
1095 is made between the caller and the callee, so the callee is unable to
1096 modify the value in the caller. This attribute is only valid on LLVM
1097 pointer arguments. It is generally used to pass structs and arrays by
1098 value, but is also valid on pointers to scalars. The copy is considered
1099 to belong to the caller not the callee (for example,
1100 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1101 <tt>byval</tt> parameters). This is not a valid attribute for return
1104 <p>The byval attribute also supports specifying an alignment with
1105 the align attribute. It indicates the alignment of the stack slot to
1106 form and the known alignment of the pointer specified to the call site. If
1107 the alignment is not specified, then the code generator makes a
1108 target-specific assumption.</p></dd>
1110 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1111 <dd>This indicates that the pointer parameter specifies the address of a
1112 structure that is the return value of the function in the source program.
1113 This pointer must be guaranteed by the caller to be valid: loads and
1114 stores to the structure may be assumed by the callee to not to trap and
1115 to be properly aligned. This may only be applied to the first parameter.
1116 This is not a valid attribute for return values. </dd>
1118 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1119 <dd>This indicates that pointer values
1120 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1121 value do not alias pointer values which are not <i>based</i> on it,
1122 ignoring certain "irrelevant" dependencies.
1123 For a call to the parent function, dependencies between memory
1124 references from before or after the call and from those during the call
1125 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1126 return value used in that call.
1127 The caller shares the responsibility with the callee for ensuring that
1128 these requirements are met.
1129 For further details, please see the discussion of the NoAlias response in
1130 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1132 Note that this definition of <tt>noalias</tt> is intentionally
1133 similar to the definition of <tt>restrict</tt> in C99 for function
1134 arguments, though it is slightly weaker.
1136 For function return values, C99's <tt>restrict</tt> is not meaningful,
1137 while LLVM's <tt>noalias</tt> is.
1140 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1141 <dd>This indicates that the callee does not make any copies of the pointer
1142 that outlive the callee itself. This is not a valid attribute for return
1145 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1146 <dd>This indicates that the pointer parameter can be excised using the
1147 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1148 attribute for return values.</dd>
1153 <!-- ======================================================================= -->
1155 <a name="gc">Garbage Collector Names</a>
1160 <p>Each function may specify a garbage collector name, which is simply a
1163 <pre class="doc_code">
1164 define void @f() gc "name" { ... }
1167 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1168 collector which will cause the compiler to alter its output in order to
1169 support the named garbage collection algorithm.</p>
1173 <!-- ======================================================================= -->
1175 <a name="fnattrs">Function Attributes</a>
1180 <p>Function attributes are set to communicate additional information about a
1181 function. Function attributes are considered to be part of the function, not
1182 of the function type, so functions with different parameter attributes can
1183 have the same function type.</p>
1185 <p>Function attributes are simple keywords that follow the type specified. If
1186 multiple attributes are needed, they are space separated. For example:</p>
1188 <pre class="doc_code">
1189 define void @f() noinline { ... }
1190 define void @f() alwaysinline { ... }
1191 define void @f() alwaysinline optsize { ... }
1192 define void @f() optsize { ... }
1196 <dt><tt><b>address_safety</b></tt></dt>
1197 <dd>This attribute indicates that the address safety analysis
1198 is enabled for this function. </dd>
1200 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1201 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1202 the backend should forcibly align the stack pointer. Specify the
1203 desired alignment, which must be a power of two, in parentheses.
1205 <dt><tt><b>alwaysinline</b></tt></dt>
1206 <dd>This attribute indicates that the inliner should attempt to inline this
1207 function into callers whenever possible, ignoring any active inlining size
1208 threshold for this caller.</dd>
1210 <dt><tt><b>nonlazybind</b></tt></dt>
1211 <dd>This attribute suppresses lazy symbol binding for the function. This
1212 may make calls to the function faster, at the cost of extra program
1213 startup time if the function is not called during program startup.</dd>
1215 <dt><tt><b>inlinehint</b></tt></dt>
1216 <dd>This attribute indicates that the source code contained a hint that inlining
1217 this function is desirable (such as the "inline" keyword in C/C++). It
1218 is just a hint; it imposes no requirements on the inliner.</dd>
1220 <dt><tt><b>naked</b></tt></dt>
1221 <dd>This attribute disables prologue / epilogue emission for the function.
1222 This can have very system-specific consequences.</dd>
1224 <dt><tt><b>noimplicitfloat</b></tt></dt>
1225 <dd>This attributes disables implicit floating point instructions.</dd>
1227 <dt><tt><b>noinline</b></tt></dt>
1228 <dd>This attribute indicates that the inliner should never inline this
1229 function in any situation. This attribute may not be used together with
1230 the <tt>alwaysinline</tt> attribute.</dd>
1232 <dt><tt><b>noredzone</b></tt></dt>
1233 <dd>This attribute indicates that the code generator should not use a red
1234 zone, even if the target-specific ABI normally permits it.</dd>
1236 <dt><tt><b>noreturn</b></tt></dt>
1237 <dd>This function attribute indicates that the function never returns
1238 normally. This produces undefined behavior at runtime if the function
1239 ever does dynamically return.</dd>
1241 <dt><tt><b>nounwind</b></tt></dt>
1242 <dd>This function attribute indicates that the function never returns with an
1243 unwind or exceptional control flow. If the function does unwind, its
1244 runtime behavior is undefined.</dd>
1246 <dt><tt><b>optsize</b></tt></dt>
1247 <dd>This attribute suggests that optimization passes and code generator passes
1248 make choices that keep the code size of this function low, and otherwise
1249 do optimizations specifically to reduce code size.</dd>
1251 <dt><tt><b>readnone</b></tt></dt>
1252 <dd>This attribute indicates that the function computes its result (or decides
1253 to unwind an exception) based strictly on its arguments, without
1254 dereferencing any pointer arguments or otherwise accessing any mutable
1255 state (e.g. memory, control registers, etc) visible to caller functions.
1256 It does not write through any pointer arguments
1257 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1258 changes any state visible to callers. This means that it cannot unwind
1259 exceptions by calling the <tt>C++</tt> exception throwing methods.</dd>
1261 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1262 <dd>This attribute indicates that the function does not write through any
1263 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1264 arguments) or otherwise modify any state (e.g. memory, control registers,
1265 etc) visible to caller functions. It may dereference pointer arguments
1266 and read state that may be set in the caller. A readonly function always
1267 returns the same value (or unwinds an exception identically) when called
1268 with the same set of arguments and global state. It cannot unwind an
1269 exception by calling the <tt>C++</tt> exception throwing methods.</dd>
1271 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1272 <dd>This attribute indicates that this function can return twice. The
1273 C <code>setjmp</code> is an example of such a function. The compiler
1274 disables some optimizations (like tail calls) in the caller of these
1277 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1278 <dd>This attribute indicates that the function should emit a stack smashing
1279 protector. It is in the form of a "canary"—a random value placed on
1280 the stack before the local variables that's checked upon return from the
1281 function to see if it has been overwritten. A heuristic is used to
1282 determine if a function needs stack protectors or not.<br>
1284 If a function that has an <tt>ssp</tt> attribute is inlined into a
1285 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1286 function will have an <tt>ssp</tt> attribute.</dd>
1288 <dt><tt><b>sspreq</b></tt></dt>
1289 <dd>This attribute indicates that the function should <em>always</em> emit a
1290 stack smashing protector. This overrides
1291 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1293 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1294 function that doesn't have an <tt>sspreq</tt> attribute or which has
1295 an <tt>ssp</tt> attribute, then the resulting function will have
1296 an <tt>sspreq</tt> attribute.</dd>
1298 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1299 <dd>This attribute indicates that the ABI being targeted requires that
1300 an unwind table entry be produce for this function even if we can
1301 show that no exceptions passes by it. This is normally the case for
1302 the ELF x86-64 abi, but it can be disabled for some compilation
1308 <!-- ======================================================================= -->
1310 <a name="moduleasm">Module-Level Inline Assembly</a>
1315 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1316 the GCC "file scope inline asm" blocks. These blocks are internally
1317 concatenated by LLVM and treated as a single unit, but may be separated in
1318 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1320 <pre class="doc_code">
1321 module asm "inline asm code goes here"
1322 module asm "more can go here"
1325 <p>The strings can contain any character by escaping non-printable characters.
1326 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1329 <p>The inline asm code is simply printed to the machine code .s file when
1330 assembly code is generated.</p>
1334 <!-- ======================================================================= -->
1336 <a name="datalayout">Data Layout</a>
1341 <p>A module may specify a target specific data layout string that specifies how
1342 data is to be laid out in memory. The syntax for the data layout is
1345 <pre class="doc_code">
1346 target datalayout = "<i>layout specification</i>"
1349 <p>The <i>layout specification</i> consists of a list of specifications
1350 separated by the minus sign character ('-'). Each specification starts with
1351 a letter and may include other information after the letter to define some
1352 aspect of the data layout. The specifications accepted are as follows:</p>
1356 <dd>Specifies that the target lays out data in big-endian form. That is, the
1357 bits with the most significance have the lowest address location.</dd>
1360 <dd>Specifies that the target lays out data in little-endian form. That is,
1361 the bits with the least significance have the lowest address
1364 <dt><tt>S<i>size</i></tt></dt>
1365 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1366 of stack variables is limited to the natural stack alignment to avoid
1367 dynamic stack realignment. The stack alignment must be a multiple of
1368 8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1369 which does not prevent any alignment promotions.</dd>
1371 <dt><tt>p[n]:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1372 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1373 <i>preferred</i> alignments for address space <i>n</i>. All sizes are in
1374 bits. Specifying the <i>pref</i> alignment is optional. If omitted, the
1375 preceding <tt>:</tt> should be omitted too. The address space,
1376 <i>n</i> is optional, and if not specified, denotes the default address
1377 space 0. The value of <i>n</i> must be in the range [1,2^23).</dd>
1379 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1380 <dd>This specifies the alignment for an integer type of a given bit
1381 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1383 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1384 <dd>This specifies the alignment for a vector type of a given bit
1387 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1388 <dd>This specifies the alignment for a floating point type of a given bit
1389 <i>size</i>. Only values of <i>size</i> that are supported by the target
1390 will work. 32 (float) and 64 (double) are supported on all targets;
1391 80 or 128 (different flavors of long double) are also supported on some
1394 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1395 <dd>This specifies the alignment for an aggregate type of a given bit
1398 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1399 <dd>This specifies the alignment for a stack object of a given bit
1402 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1403 <dd>This specifies a set of native integer widths for the target CPU
1404 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1405 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1406 this set are considered to support most general arithmetic
1407 operations efficiently.</dd>
1410 <p>When constructing the data layout for a given target, LLVM starts with a
1411 default set of specifications which are then (possibly) overridden by the
1412 specifications in the <tt>datalayout</tt> keyword. The default specifications
1413 are given in this list:</p>
1416 <li><tt>E</tt> - big endian</li>
1417 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1418 <li><tt>p1:32:32:32</tt> - 32-bit pointers with 32-bit alignment for
1419 address space 1</li>
1420 <li><tt>p2:16:32:32</tt> - 16-bit pointers with 32-bit alignment for
1421 address space 2</li>
1422 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1423 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1424 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1425 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1426 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1427 alignment of 64-bits</li>
1428 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1429 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1430 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1431 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1432 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1433 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1436 <p>When LLVM is determining the alignment for a given type, it uses the
1437 following rules:</p>
1440 <li>If the type sought is an exact match for one of the specifications, that
1441 specification is used.</li>
1443 <li>If no match is found, and the type sought is an integer type, then the
1444 smallest integer type that is larger than the bitwidth of the sought type
1445 is used. If none of the specifications are larger than the bitwidth then
1446 the largest integer type is used. For example, given the default
1447 specifications above, the i7 type will use the alignment of i8 (next
1448 largest) while both i65 and i256 will use the alignment of i64 (largest
1451 <li>If no match is found, and the type sought is a vector type, then the
1452 largest vector type that is smaller than the sought vector type will be
1453 used as a fall back. This happens because <128 x double> can be
1454 implemented in terms of 64 <2 x double>, for example.</li>
1457 <p>The function of the data layout string may not be what you expect. Notably,
1458 this is not a specification from the frontend of what alignment the code
1459 generator should use.</p>
1461 <p>Instead, if specified, the target data layout is required to match what the
1462 ultimate <em>code generator</em> expects. This string is used by the
1463 mid-level optimizers to
1464 improve code, and this only works if it matches what the ultimate code
1465 generator uses. If you would like to generate IR that does not embed this
1466 target-specific detail into the IR, then you don't have to specify the
1467 string. This will disable some optimizations that require precise layout
1468 information, but this also prevents those optimizations from introducing
1469 target specificity into the IR.</p>
1475 <!-- ======================================================================= -->
1477 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1482 <p>Any memory access must be done through a pointer value associated
1483 with an address range of the memory access, otherwise the behavior
1484 is undefined. Pointer values are associated with address ranges
1485 according to the following rules:</p>
1488 <li>A pointer value is associated with the addresses associated with
1489 any value it is <i>based</i> on.
1490 <li>An address of a global variable is associated with the address
1491 range of the variable's storage.</li>
1492 <li>The result value of an allocation instruction is associated with
1493 the address range of the allocated storage.</li>
1494 <li>A null pointer in the default address-space is associated with
1496 <li>An integer constant other than zero or a pointer value returned
1497 from a function not defined within LLVM may be associated with address
1498 ranges allocated through mechanisms other than those provided by
1499 LLVM. Such ranges shall not overlap with any ranges of addresses
1500 allocated by mechanisms provided by LLVM.</li>
1503 <p>A pointer value is <i>based</i> on another pointer value according
1504 to the following rules:</p>
1507 <li>A pointer value formed from a
1508 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1509 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1510 <li>The result value of a
1511 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1512 of the <tt>bitcast</tt>.</li>
1513 <li>A pointer value formed by an
1514 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1515 pointer values that contribute (directly or indirectly) to the
1516 computation of the pointer's value.</li>
1517 <li>The "<i>based</i> on" relationship is transitive.</li>
1520 <p>Note that this definition of <i>"based"</i> is intentionally
1521 similar to the definition of <i>"based"</i> in C99, though it is
1522 slightly weaker.</p>
1524 <p>LLVM IR does not associate types with memory. The result type of a
1525 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1526 alignment of the memory from which to load, as well as the
1527 interpretation of the value. The first operand type of a
1528 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1529 and alignment of the store.</p>
1531 <p>Consequently, type-based alias analysis, aka TBAA, aka
1532 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1533 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1534 additional information which specialized optimization passes may use
1535 to implement type-based alias analysis.</p>
1539 <!-- ======================================================================= -->
1541 <a name="volatile">Volatile Memory Accesses</a>
1546 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1547 href="#i_store"><tt>store</tt></a>s, and <a
1548 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1549 The optimizers must not change the number of volatile operations or change their
1550 order of execution relative to other volatile operations. The optimizers
1551 <i>may</i> change the order of volatile operations relative to non-volatile
1552 operations. This is not Java's "volatile" and has no cross-thread
1553 synchronization behavior.</p>
1557 <!-- ======================================================================= -->
1559 <a name="memmodel">Memory Model for Concurrent Operations</a>
1564 <p>The LLVM IR does not define any way to start parallel threads of execution
1565 or to register signal handlers. Nonetheless, there are platform-specific
1566 ways to create them, and we define LLVM IR's behavior in their presence. This
1567 model is inspired by the C++0x memory model.</p>
1569 <p>For a more informal introduction to this model, see the
1570 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1572 <p>We define a <i>happens-before</i> partial order as the least partial order
1575 <li>Is a superset of single-thread program order, and</li>
1576 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1577 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1578 by platform-specific techniques, like pthread locks, thread
1579 creation, thread joining, etc., and by atomic instructions.
1580 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1584 <p>Note that program order does not introduce <i>happens-before</i> edges
1585 between a thread and signals executing inside that thread.</p>
1587 <p>Every (defined) read operation (load instructions, memcpy, atomic
1588 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1589 (defined) write operations (store instructions, atomic
1590 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1591 initialized globals are considered to have a write of the initializer which is
1592 atomic and happens before any other read or write of the memory in question.
1593 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1594 any write to the same byte, except:</p>
1597 <li>If <var>write<sub>1</sub></var> happens before
1598 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1599 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1600 does not see <var>write<sub>1</sub></var>.
1601 <li>If <var>R<sub>byte</sub></var> happens before
1602 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1603 see <var>write<sub>3</sub></var>.
1606 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1608 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1609 is supposed to give guarantees which can support
1610 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1611 addresses which do not behave like normal memory. It does not generally
1612 provide cross-thread synchronization.)
1613 <li>Otherwise, if there is no write to the same byte that happens before
1614 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1615 <tt>undef</tt> for that byte.
1616 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1617 <var>R<sub>byte</sub></var> returns the value written by that
1619 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1620 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1621 values written. See the <a href="#ordering">Atomic Memory Ordering
1622 Constraints</a> section for additional constraints on how the choice
1624 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1627 <p><var>R</var> returns the value composed of the series of bytes it read.
1628 This implies that some bytes within the value may be <tt>undef</tt>
1629 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1630 defines the semantics of the operation; it doesn't mean that targets will
1631 emit more than one instruction to read the series of bytes.</p>
1633 <p>Note that in cases where none of the atomic intrinsics are used, this model
1634 places only one restriction on IR transformations on top of what is required
1635 for single-threaded execution: introducing a store to a byte which might not
1636 otherwise be stored is not allowed in general. (Specifically, in the case
1637 where another thread might write to and read from an address, introducing a
1638 store can change a load that may see exactly one write into a load that may
1639 see multiple writes.)</p>
1641 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1642 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1643 none of the backends currently in the tree fall into this category; however,
1644 there might be targets which care. If there are, we want a paragraph
1647 Targets may specify that stores narrower than a certain width are not
1648 available; on such a target, for the purposes of this model, treat any
1649 non-atomic write with an alignment or width less than the minimum width
1650 as if it writes to the relevant surrounding bytes.
1655 <!-- ======================================================================= -->
1657 <a name="ordering">Atomic Memory Ordering Constraints</a>
1662 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1663 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1664 <a href="#i_fence"><code>fence</code></a>,
1665 <a href="#i_load"><code>atomic load</code></a>, and
1666 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1667 that determines which other atomic instructions on the same address they
1668 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1669 but are somewhat more colloquial. If these descriptions aren't precise enough,
1670 check those specs (see spec references in the
1671 <a href="Atomics.html#introduction">atomics guide</a>).
1672 <a href="#i_fence"><code>fence</code></a> instructions
1673 treat these orderings somewhat differently since they don't take an address.
1674 See that instruction's documentation for details.</p>
1676 <p>For a simpler introduction to the ordering constraints, see the
1677 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1680 <dt><code>unordered</code></dt>
1681 <dd>The set of values that can be read is governed by the happens-before
1682 partial order. A value cannot be read unless some operation wrote it.
1683 This is intended to provide a guarantee strong enough to model Java's
1684 non-volatile shared variables. This ordering cannot be specified for
1685 read-modify-write operations; it is not strong enough to make them atomic
1686 in any interesting way.</dd>
1687 <dt><code>monotonic</code></dt>
1688 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1689 total order for modifications by <code>monotonic</code> operations on each
1690 address. All modification orders must be compatible with the happens-before
1691 order. There is no guarantee that the modification orders can be combined to
1692 a global total order for the whole program (and this often will not be
1693 possible). The read in an atomic read-modify-write operation
1694 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1695 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1696 reads the value in the modification order immediately before the value it
1697 writes. If one atomic read happens before another atomic read of the same
1698 address, the later read must see the same value or a later value in the
1699 address's modification order. This disallows reordering of
1700 <code>monotonic</code> (or stronger) operations on the same address. If an
1701 address is written <code>monotonic</code>ally by one thread, and other threads
1702 <code>monotonic</code>ally read that address repeatedly, the other threads must
1703 eventually see the write. This corresponds to the C++0x/C1x
1704 <code>memory_order_relaxed</code>.</dd>
1705 <dt><code>acquire</code></dt>
1706 <dd>In addition to the guarantees of <code>monotonic</code>,
1707 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1708 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1709 <dt><code>release</code></dt>
1710 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1711 writes a value which is subsequently read by an <code>acquire</code> operation,
1712 it <i>synchronizes-with</i> that operation. (This isn't a complete
1713 description; see the C++0x definition of a release sequence.) This corresponds
1714 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1715 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1716 <code>acquire</code> and <code>release</code> operation on its address.
1717 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1718 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1719 <dd>In addition to the guarantees of <code>acq_rel</code>
1720 (<code>acquire</code> for an operation which only reads, <code>release</code>
1721 for an operation which only writes), there is a global total order on all
1722 sequentially-consistent operations on all addresses, which is consistent with
1723 the <i>happens-before</i> partial order and with the modification orders of
1724 all the affected addresses. Each sequentially-consistent read sees the last
1725 preceding write to the same address in this global order. This corresponds
1726 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1729 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1730 it only <i>synchronizes with</i> or participates in modification and seq_cst
1731 total orderings with other operations running in the same thread (for example,
1732 in signal handlers).</p>
1738 <!-- *********************************************************************** -->
1739 <h2><a name="typesystem">Type System</a></h2>
1740 <!-- *********************************************************************** -->
1744 <p>The LLVM type system is one of the most important features of the
1745 intermediate representation. Being typed enables a number of optimizations
1746 to be performed on the intermediate representation directly, without having
1747 to do extra analyses on the side before the transformation. A strong type
1748 system makes it easier to read the generated code and enables novel analyses
1749 and transformations that are not feasible to perform on normal three address
1750 code representations.</p>
1752 <!-- ======================================================================= -->
1754 <a name="t_classifications">Type Classifications</a>
1759 <p>The types fall into a few useful classifications:</p>
1761 <table border="1" cellspacing="0" cellpadding="4">
1763 <tr><th>Classification</th><th>Types</th></tr>
1765 <td><a href="#t_integer">integer</a></td>
1766 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1769 <td><a href="#t_floating">floating point</a></td>
1770 <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1773 <td><a name="t_firstclass">first class</a></td>
1774 <td><a href="#t_integer">integer</a>,
1775 <a href="#t_floating">floating point</a>,
1776 <a href="#t_pointer">pointer</a>,
1777 <a href="#t_vector">vector</a>,
1778 <a href="#t_struct">structure</a>,
1779 <a href="#t_array">array</a>,
1780 <a href="#t_label">label</a>,
1781 <a href="#t_metadata">metadata</a>.
1785 <td><a href="#t_primitive">primitive</a></td>
1786 <td><a href="#t_label">label</a>,
1787 <a href="#t_void">void</a>,
1788 <a href="#t_integer">integer</a>,
1789 <a href="#t_floating">floating point</a>,
1790 <a href="#t_x86mmx">x86mmx</a>,
1791 <a href="#t_metadata">metadata</a>.</td>
1794 <td><a href="#t_derived">derived</a></td>
1795 <td><a href="#t_array">array</a>,
1796 <a href="#t_function">function</a>,
1797 <a href="#t_pointer">pointer</a>,
1798 <a href="#t_struct">structure</a>,
1799 <a href="#t_vector">vector</a>,
1800 <a href="#t_opaque">opaque</a>.
1806 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1807 important. Values of these types are the only ones which can be produced by
1812 <!-- ======================================================================= -->
1814 <a name="t_primitive">Primitive Types</a>
1819 <p>The primitive types are the fundamental building blocks of the LLVM
1822 <!-- _______________________________________________________________________ -->
1824 <a name="t_integer">Integer Type</a>
1830 <p>The integer type is a very simple type that simply specifies an arbitrary
1831 bit width for the integer type desired. Any bit width from 1 bit to
1832 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1839 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1843 <table class="layout">
1845 <td class="left"><tt>i1</tt></td>
1846 <td class="left">a single-bit integer.</td>
1849 <td class="left"><tt>i32</tt></td>
1850 <td class="left">a 32-bit integer.</td>
1853 <td class="left"><tt>i1942652</tt></td>
1854 <td class="left">a really big integer of over 1 million bits.</td>
1860 <!-- _______________________________________________________________________ -->
1862 <a name="t_floating">Floating Point Types</a>
1869 <tr><th>Type</th><th>Description</th></tr>
1870 <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr>
1871 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1872 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1873 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1874 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1875 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1881 <!-- _______________________________________________________________________ -->
1883 <a name="t_x86mmx">X86mmx Type</a>
1889 <p>The x86mmx type represents a value held in an MMX register on an x86 machine. The operations allowed on it are quite limited: parameters and return values, load and store, and bitcast. User-specified MMX instructions are represented as intrinsic or asm calls with arguments and/or results of this type. There are no arrays, vectors or constants of this type.</p>
1898 <!-- _______________________________________________________________________ -->
1900 <a name="t_void">Void Type</a>
1906 <p>The void type does not represent any value and has no size.</p>
1915 <!-- _______________________________________________________________________ -->
1917 <a name="t_label">Label Type</a>
1923 <p>The label type represents code labels.</p>
1932 <!-- _______________________________________________________________________ -->
1934 <a name="t_metadata">Metadata Type</a>
1940 <p>The metadata type represents embedded metadata. No derived types may be
1941 created from metadata except for <a href="#t_function">function</a>
1953 <!-- ======================================================================= -->
1955 <a name="t_derived">Derived Types</a>
1960 <p>The real power in LLVM comes from the derived types in the system. This is
1961 what allows a programmer to represent arrays, functions, pointers, and other
1962 useful types. Each of these types contain one or more element types which
1963 may be a primitive type, or another derived type. For example, it is
1964 possible to have a two dimensional array, using an array as the element type
1965 of another array.</p>
1967 <!-- _______________________________________________________________________ -->
1969 <a name="t_aggregate">Aggregate Types</a>
1974 <p>Aggregate Types are a subset of derived types that can contain multiple
1975 member types. <a href="#t_array">Arrays</a> and
1976 <a href="#t_struct">structs</a> are aggregate types.
1977 <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p>
1981 <!-- _______________________________________________________________________ -->
1983 <a name="t_array">Array Type</a>
1989 <p>The array type is a very simple derived type that arranges elements
1990 sequentially in memory. The array type requires a size (number of elements)
1991 and an underlying data type.</p>
1995 [<# elements> x <elementtype>]
1998 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1999 be any type with a size.</p>
2002 <table class="layout">
2004 <td class="left"><tt>[40 x i32]</tt></td>
2005 <td class="left">Array of 40 32-bit integer values.</td>
2008 <td class="left"><tt>[41 x i32]</tt></td>
2009 <td class="left">Array of 41 32-bit integer values.</td>
2012 <td class="left"><tt>[4 x i8]</tt></td>
2013 <td class="left">Array of 4 8-bit integer values.</td>
2016 <p>Here are some examples of multidimensional arrays:</p>
2017 <table class="layout">
2019 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
2020 <td class="left">3x4 array of 32-bit integer values.</td>
2023 <td class="left"><tt>[12 x [10 x float]]</tt></td>
2024 <td class="left">12x10 array of single precision floating point values.</td>
2027 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
2028 <td class="left">2x3x4 array of 16-bit integer values.</td>
2032 <p>There is no restriction on indexing beyond the end of the array implied by
2033 a static type (though there are restrictions on indexing beyond the bounds
2034 of an allocated object in some cases). This means that single-dimension
2035 'variable sized array' addressing can be implemented in LLVM with a zero
2036 length array type. An implementation of 'pascal style arrays' in LLVM could
2037 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
2041 <!-- _______________________________________________________________________ -->
2043 <a name="t_function">Function Type</a>
2049 <p>The function type can be thought of as a function signature. It consists of
2050 a return type and a list of formal parameter types. The return type of a
2051 function type is a first class type or a void type.</p>
2055 <returntype> (<parameter list>)
2058 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
2059 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
2060 which indicates that the function takes a variable number of arguments.
2061 Variable argument functions can access their arguments with
2062 the <a href="#int_varargs">variable argument handling intrinsic</a>
2063 functions. '<tt><returntype></tt>' is any type except
2064 <a href="#t_label">label</a>.</p>
2067 <table class="layout">
2069 <td class="left"><tt>i32 (i32)</tt></td>
2070 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2072 </tr><tr class="layout">
2073 <td class="left"><tt>float (i16, i32 *) *
2075 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2076 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2077 returning <tt>float</tt>.
2079 </tr><tr class="layout">
2080 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2081 <td class="left">A vararg function that takes at least one
2082 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2083 which returns an integer. This is the signature for <tt>printf</tt> in
2086 </tr><tr class="layout">
2087 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2088 <td class="left">A function taking an <tt>i32</tt>, returning a
2089 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2096 <!-- _______________________________________________________________________ -->
2098 <a name="t_struct">Structure Type</a>
2104 <p>The structure type is used to represent a collection of data members together
2105 in memory. The elements of a structure may be any type that has a size.</p>
2107 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2108 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2109 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2110 Structures in registers are accessed using the
2111 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2112 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2114 <p>Structures may optionally be "packed" structures, which indicate that the
2115 alignment of the struct is one byte, and that there is no padding between
2116 the elements. In non-packed structs, padding between field types is inserted
2117 as defined by the DataLayout string in the module, which is required to match
2118 what the underlying code generator expects.</p>
2120 <p>Structures can either be "literal" or "identified". A literal structure is
2121 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2122 types are always defined at the top level with a name. Literal types are
2123 uniqued by their contents and can never be recursive or opaque since there is
2124 no way to write one. Identified types can be recursive, can be opaqued, and are
2130 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2131 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2135 <table class="layout">
2137 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2138 <td class="left">A triple of three <tt>i32</tt> values</td>
2141 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2142 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2143 second element is a <a href="#t_pointer">pointer</a> to a
2144 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2145 an <tt>i32</tt>.</td>
2148 <td class="left"><tt><{ i8, i32 }></tt></td>
2149 <td class="left">A packed struct known to be 5 bytes in size.</td>
2155 <!-- _______________________________________________________________________ -->
2157 <a name="t_opaque">Opaque Structure Types</a>
2163 <p>Opaque structure types are used to represent named structure types that do
2164 not have a body specified. This corresponds (for example) to the C notion of
2165 a forward declared structure.</p>
2174 <table class="layout">
2176 <td class="left"><tt>opaque</tt></td>
2177 <td class="left">An opaque type.</td>
2185 <!-- _______________________________________________________________________ -->
2187 <a name="t_pointer">Pointer Type</a>
2193 <p>The pointer type is used to specify memory locations.
2194 Pointers are commonly used to reference objects in memory.</p>
2196 <p>Pointer types may have an optional address space attribute defining the
2197 numbered address space where the pointed-to object resides. The default
2198 address space is number zero. The semantics of non-zero address
2199 spaces are target-specific.</p>
2201 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2202 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2210 <table class="layout">
2212 <td class="left"><tt>[4 x i32]*</tt></td>
2213 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2214 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2217 <td class="left"><tt>i32 (i32*) *</tt></td>
2218 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2219 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2223 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2224 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2225 that resides in address space #5.</td>
2231 <!-- _______________________________________________________________________ -->
2233 <a name="t_vector">Vector Type</a>
2239 <p>A vector type is a simple derived type that represents a vector of elements.
2240 Vector types are used when multiple primitive data are operated in parallel
2241 using a single instruction (SIMD). A vector type requires a size (number of
2242 elements) and an underlying primitive data type. Vector types are considered
2243 <a href="#t_firstclass">first class</a>.</p>
2247 < <# elements> x <elementtype> >
2250 <p>The number of elements is a constant integer value larger than 0; elementtype
2251 may be any integer or floating point type, or a pointer to these types.
2252 Vectors of size zero are not allowed. </p>
2255 <table class="layout">
2257 <td class="left"><tt><4 x i32></tt></td>
2258 <td class="left">Vector of 4 32-bit integer values.</td>
2261 <td class="left"><tt><8 x float></tt></td>
2262 <td class="left">Vector of 8 32-bit floating-point values.</td>
2265 <td class="left"><tt><2 x i64></tt></td>
2266 <td class="left">Vector of 2 64-bit integer values.</td>
2269 <td class="left"><tt><4 x i64*></tt></td>
2270 <td class="left">Vector of 4 pointers to 64-bit integer values.</td>
2280 <!-- *********************************************************************** -->
2281 <h2><a name="constants">Constants</a></h2>
2282 <!-- *********************************************************************** -->
2286 <p>LLVM has several different basic types of constants. This section describes
2287 them all and their syntax.</p>
2289 <!-- ======================================================================= -->
2291 <a name="simpleconstants">Simple Constants</a>
2297 <dt><b>Boolean constants</b></dt>
2298 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2299 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2301 <dt><b>Integer constants</b></dt>
2302 <dd>Standard integers (such as '4') are constants of
2303 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2304 with integer types.</dd>
2306 <dt><b>Floating point constants</b></dt>
2307 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2308 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2309 notation (see below). The assembler requires the exact decimal value of a
2310 floating-point constant. For example, the assembler accepts 1.25 but
2311 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2312 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2314 <dt><b>Null pointer constants</b></dt>
2315 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2316 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2319 <p>The one non-intuitive notation for constants is the hexadecimal form of
2320 floating point constants. For example, the form '<tt>double
2321 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2322 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2323 constants are required (and the only time that they are generated by the
2324 disassembler) is when a floating point constant must be emitted but it cannot
2325 be represented as a decimal floating point number in a reasonable number of
2326 digits. For example, NaN's, infinities, and other special values are
2327 represented in their IEEE hexadecimal format so that assembly and disassembly
2328 do not cause any bits to change in the constants.</p>
2330 <p>When using the hexadecimal form, constants of types half, float, and double are
2331 represented using the 16-digit form shown above (which matches the IEEE754
2332 representation for double); half and float values must, however, be exactly
2333 representable as IEE754 half and single precision, respectively.
2334 Hexadecimal format is always used
2335 for long double, and there are three forms of long double. The 80-bit format
2336 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2337 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2338 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2339 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2340 currently supported target uses this format. Long doubles will only work if
2341 they match the long double format on your target. The IEEE 16-bit format
2342 (half precision) is represented by <tt>0xH</tt> followed by 4 hexadecimal
2343 digits. All hexadecimal formats are big-endian (sign bit at the left).</p>
2345 <p>There are no constants of type x86mmx.</p>
2348 <!-- ======================================================================= -->
2350 <a name="aggregateconstants"></a> <!-- old anchor -->
2351 <a name="complexconstants">Complex Constants</a>
2356 <p>Complex constants are a (potentially recursive) combination of simple
2357 constants and smaller complex constants.</p>
2360 <dt><b>Structure constants</b></dt>
2361 <dd>Structure constants are represented with notation similar to structure
2362 type definitions (a comma separated list of elements, surrounded by braces
2363 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2364 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2365 Structure constants must have <a href="#t_struct">structure type</a>, and
2366 the number and types of elements must match those specified by the
2369 <dt><b>Array constants</b></dt>
2370 <dd>Array constants are represented with notation similar to array type
2371 definitions (a comma separated list of elements, surrounded by square
2372 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2373 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2374 the number and types of elements must match those specified by the
2377 <dt><b>Vector constants</b></dt>
2378 <dd>Vector constants are represented with notation similar to vector type
2379 definitions (a comma separated list of elements, surrounded by
2380 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2381 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2382 have <a href="#t_vector">vector type</a>, and the number and types of
2383 elements must match those specified by the type.</dd>
2385 <dt><b>Zero initialization</b></dt>
2386 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2387 value to zero of <em>any</em> type, including scalar and
2388 <a href="#t_aggregate">aggregate</a> types.
2389 This is often used to avoid having to print large zero initializers
2390 (e.g. for large arrays) and is always exactly equivalent to using explicit
2391 zero initializers.</dd>
2393 <dt><b>Metadata node</b></dt>
2394 <dd>A metadata node is a structure-like constant with
2395 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2396 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2397 be interpreted as part of the instruction stream, metadata is a place to
2398 attach additional information such as debug info.</dd>
2403 <!-- ======================================================================= -->
2405 <a name="globalconstants">Global Variable and Function Addresses</a>
2410 <p>The addresses of <a href="#globalvars">global variables</a>
2411 and <a href="#functionstructure">functions</a> are always implicitly valid
2412 (link-time) constants. These constants are explicitly referenced when
2413 the <a href="#identifiers">identifier for the global</a> is used and always
2414 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2415 legal LLVM file:</p>
2417 <pre class="doc_code">
2420 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2425 <!-- ======================================================================= -->
2427 <a name="undefvalues">Undefined Values</a>
2432 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2433 indicates that the user of the value may receive an unspecified bit-pattern.
2434 Undefined values may be of any type (other than '<tt>label</tt>'
2435 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2437 <p>Undefined values are useful because they indicate to the compiler that the
2438 program is well defined no matter what value is used. This gives the
2439 compiler more freedom to optimize. Here are some examples of (potentially
2440 surprising) transformations that are valid (in pseudo IR):</p>
2443 <pre class="doc_code">
2453 <p>This is safe because all of the output bits are affected by the undef bits.
2454 Any output bit can have a zero or one depending on the input bits.</p>
2456 <pre class="doc_code">
2467 <p>These logical operations have bits that are not always affected by the input.
2468 For example, if <tt>%X</tt> has a zero bit, then the output of the
2469 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2470 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2471 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2472 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2473 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2474 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2475 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2477 <pre class="doc_code">
2478 %A = select undef, %X, %Y
2479 %B = select undef, 42, %Y
2480 %C = select %X, %Y, undef
2491 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2492 branch) conditions can go <em>either way</em>, but they have to come from one
2493 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2494 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2495 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2496 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2497 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2500 <pre class="doc_code">
2501 %A = xor undef, undef
2519 <p>This example points out that two '<tt>undef</tt>' operands are not
2520 necessarily the same. This can be surprising to people (and also matches C
2521 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2522 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2523 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2524 its value over its "live range". This is true because the variable doesn't
2525 actually <em>have a live range</em>. Instead, the value is logically read
2526 from arbitrary registers that happen to be around when needed, so the value
2527 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2528 need to have the same semantics or the core LLVM "replace all uses with"
2529 concept would not hold.</p>
2531 <pre class="doc_code">
2539 <p>These examples show the crucial difference between an <em>undefined
2540 value</em> and <em>undefined behavior</em>. An undefined value (like
2541 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2542 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2543 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2544 defined on SNaN's. However, in the second example, we can make a more
2545 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2546 arbitrary value, we are allowed to assume that it could be zero. Since a
2547 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2548 the operation does not execute at all. This allows us to delete the divide and
2549 all code after it. Because the undefined operation "can't happen", the
2550 optimizer can assume that it occurs in dead code.</p>
2552 <pre class="doc_code">
2553 a: store undef -> %X
2554 b: store %X -> undef
2560 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2561 undefined value can be assumed to not have any effect; we can assume that the
2562 value is overwritten with bits that happen to match what was already there.
2563 However, a store <em>to</em> an undefined location could clobber arbitrary
2564 memory, therefore, it has undefined behavior.</p>
2568 <!-- ======================================================================= -->
2570 <a name="poisonvalues">Poison Values</a>
2575 <p>Poison values are similar to <a href="#undefvalues">undef values</a>, however
2576 they also represent the fact that an instruction or constant expression which
2577 cannot evoke side effects has nevertheless detected a condition which results
2578 in undefined behavior.</p>
2580 <p>There is currently no way of representing a poison value in the IR; they
2581 only exist when produced by operations such as
2582 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2584 <p>Poison value behavior is defined in terms of value <i>dependence</i>:</p>
2587 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2588 their operands.</li>
2590 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2591 to their dynamic predecessor basic block.</li>
2593 <li>Function arguments depend on the corresponding actual argument values in
2594 the dynamic callers of their functions.</li>
2596 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2597 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2598 control back to them.</li>
2600 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2601 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>,
2602 or exception-throwing call instructions that dynamically transfer control
2605 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2606 referenced memory addresses, following the order in the IR
2607 (including loads and stores implied by intrinsics such as
2608 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2610 <!-- TODO: In the case of multiple threads, this only applies if the store
2611 "happens-before" the load or store. -->
2613 <!-- TODO: floating-point exception state -->
2615 <li>An instruction with externally visible side effects depends on the most
2616 recent preceding instruction with externally visible side effects, following
2617 the order in the IR. (This includes
2618 <a href="#volatile">volatile operations</a>.)</li>
2620 <li>An instruction <i>control-depends</i> on a
2621 <a href="#terminators">terminator instruction</a>
2622 if the terminator instruction has multiple successors and the instruction
2623 is always executed when control transfers to one of the successors, and
2624 may not be executed when control is transferred to another.</li>
2626 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2627 instruction if the set of instructions it otherwise depends on would be
2628 different if the terminator had transferred control to a different
2631 <li>Dependence is transitive.</li>
2635 <p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>,
2636 with the additional affect that any instruction which has a <i>dependence</i>
2637 on a poison value has undefined behavior.</p>
2639 <p>Here are some examples:</p>
2641 <pre class="doc_code">
2643 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2644 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2645 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2646 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2648 store i32 %poison, i32* @g ; Poison value stored to memory.
2649 %poison2 = load i32* @g ; Poison value loaded back from memory.
2651 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2653 %narrowaddr = bitcast i32* @g to i16*
2654 %wideaddr = bitcast i32* @g to i64*
2655 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2656 %poison4 = load i64* %wideaddr ; Returns a poison value.
2658 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2659 br i1 %cmp, label %true, label %end ; Branch to either destination.
2662 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2663 ; it has undefined behavior.
2667 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2668 ; Both edges into this PHI are
2669 ; control-dependent on %cmp, so this
2670 ; always results in a poison value.
2672 store volatile i32 0, i32* @g ; This would depend on the store in %true
2673 ; if %cmp is true, or the store in %entry
2674 ; otherwise, so this is undefined behavior.
2676 br i1 %cmp, label %second_true, label %second_end
2677 ; The same branch again, but this time the
2678 ; true block doesn't have side effects.
2685 store volatile i32 0, i32* @g ; This time, the instruction always depends
2686 ; on the store in %end. Also, it is
2687 ; control-equivalent to %end, so this is
2688 ; well-defined (ignoring earlier undefined
2689 ; behavior in this example).
2694 <!-- ======================================================================= -->
2696 <a name="blockaddress">Addresses of Basic Blocks</a>
2701 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2703 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2704 basic block in the specified function, and always has an i8* type. Taking
2705 the address of the entry block is illegal.</p>
2707 <p>This value only has defined behavior when used as an operand to the
2708 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2709 comparisons against null. Pointer equality tests between labels addresses
2710 results in undefined behavior — though, again, comparison against null
2711 is ok, and no label is equal to the null pointer. This may be passed around
2712 as an opaque pointer sized value as long as the bits are not inspected. This
2713 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2714 long as the original value is reconstituted before the <tt>indirectbr</tt>
2717 <p>Finally, some targets may provide defined semantics when using the value as
2718 the operand to an inline assembly, but that is target specific.</p>
2723 <!-- ======================================================================= -->
2725 <a name="constantexprs">Constant Expressions</a>
2730 <p>Constant expressions are used to allow expressions involving other constants
2731 to be used as constants. Constant expressions may be of
2732 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2733 operation that does not have side effects (e.g. load and call are not
2734 supported). The following is the syntax for constant expressions:</p>
2737 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2738 <dd>Truncate a constant to another type. The bit size of CST must be larger
2739 than the bit size of TYPE. Both types must be integers.</dd>
2741 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2742 <dd>Zero extend a constant to another type. The bit size of CST must be
2743 smaller than the bit size of TYPE. Both types must be integers.</dd>
2745 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2746 <dd>Sign extend a constant to another type. The bit size of CST must be
2747 smaller than the bit size of TYPE. Both types must be integers.</dd>
2749 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2750 <dd>Truncate a floating point constant to another floating point type. The
2751 size of CST must be larger than the size of TYPE. Both types must be
2752 floating point.</dd>
2754 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2755 <dd>Floating point extend a constant to another type. The size of CST must be
2756 smaller or equal to the size of TYPE. Both types must be floating
2759 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2760 <dd>Convert a floating point constant to the corresponding unsigned integer
2761 constant. TYPE must be a scalar or vector integer type. CST must be of
2762 scalar or vector floating point type. Both CST and TYPE must be scalars,
2763 or vectors of the same number of elements. If the value won't fit in the
2764 integer type, the results are undefined.</dd>
2766 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2767 <dd>Convert a floating point constant to the corresponding signed integer
2768 constant. TYPE must be a scalar or vector integer type. CST must be of
2769 scalar or vector floating point type. Both CST and TYPE must be scalars,
2770 or vectors of the same number of elements. If the value won't fit in the
2771 integer type, the results are undefined.</dd>
2773 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2774 <dd>Convert an unsigned integer constant to the corresponding floating point
2775 constant. TYPE must be a scalar or vector floating point type. CST must be
2776 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2777 vectors of the same number of elements. If the value won't fit in the
2778 floating point type, the results are undefined.</dd>
2780 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2781 <dd>Convert a signed integer constant to the corresponding floating point
2782 constant. TYPE must be a scalar or vector floating point type. CST must be
2783 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2784 vectors of the same number of elements. If the value won't fit in the
2785 floating point type, the results are undefined.</dd>
2787 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2788 <dd>Convert a pointer typed constant to the corresponding integer constant
2789 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2790 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2791 make it fit in <tt>TYPE</tt>.</dd>
2793 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2794 <dd>Convert an integer constant to a pointer constant. TYPE must be a pointer
2795 type. CST must be of integer type. The CST value is zero extended,
2796 truncated, or unchanged to make it fit in a pointer size. This one is
2797 <i>really</i> dangerous!</dd>
2799 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2800 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2801 are the same as those for the <a href="#i_bitcast">bitcast
2802 instruction</a>.</dd>
2804 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2805 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2806 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2807 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2808 instruction, the index list may have zero or more indexes, which are
2809 required to make sense for the type of "CSTPTR".</dd>
2811 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2812 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2814 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2815 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2817 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2818 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2820 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2821 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2824 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2825 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2828 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2829 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2832 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2833 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2834 constants. The index list is interpreted in a similar manner as indices in
2835 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2836 index value must be specified.</dd>
2838 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2839 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2840 constants. The index list is interpreted in a similar manner as indices in
2841 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2842 index value must be specified.</dd>
2844 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2845 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2846 be any of the <a href="#binaryops">binary</a>
2847 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2848 on operands are the same as those for the corresponding instruction
2849 (e.g. no bitwise operations on floating point values are allowed).</dd>
2856 <!-- *********************************************************************** -->
2857 <h2><a name="othervalues">Other Values</a></h2>
2858 <!-- *********************************************************************** -->
2860 <!-- ======================================================================= -->
2862 <a name="inlineasm">Inline Assembler Expressions</a>
2867 <p>LLVM supports inline assembler expressions (as opposed
2868 to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of
2869 a special value. This value represents the inline assembler as a string
2870 (containing the instructions to emit), a list of operand constraints (stored
2871 as a string), a flag that indicates whether or not the inline asm
2872 expression has side effects, and a flag indicating whether the function
2873 containing the asm needs to align its stack conservatively. An example
2874 inline assembler expression is:</p>
2876 <pre class="doc_code">
2877 i32 (i32) asm "bswap $0", "=r,r"
2880 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2881 a <a href="#i_call"><tt>call</tt></a> or an
2882 <a href="#i_invoke"><tt>invoke</tt></a> instruction.
2883 Thus, typically we have:</p>
2885 <pre class="doc_code">
2886 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2889 <p>Inline asms with side effects not visible in the constraint list must be
2890 marked as having side effects. This is done through the use of the
2891 '<tt>sideeffect</tt>' keyword, like so:</p>
2893 <pre class="doc_code">
2894 call void asm sideeffect "eieio", ""()
2897 <p>In some cases inline asms will contain code that will not work unless the
2898 stack is aligned in some way, such as calls or SSE instructions on x86,
2899 yet will not contain code that does that alignment within the asm.
2900 The compiler should make conservative assumptions about what the asm might
2901 contain and should generate its usual stack alignment code in the prologue
2902 if the '<tt>alignstack</tt>' keyword is present:</p>
2904 <pre class="doc_code">
2905 call void asm alignstack "eieio", ""()
2908 <p>Inline asms also support using non-standard assembly dialects. The assumed
2909 dialect is ATT. When the '<tt>inteldialect</tt>' keyword is present, the
2910 inline asm is using the Intel dialect. Currently, ATT and Intel are the
2911 only supported dialects. An example is:</p>
2913 <pre class="doc_code">
2914 call void asm inteldialect "eieio", ""()
2917 <p>If multiple keywords appear the '<tt>sideeffect</tt>' keyword must come
2918 first, the '<tt>alignstack</tt>' keyword second and the
2919 '<tt>inteldialect</tt>' keyword last.</p>
2922 <p>TODO: The format of the asm and constraints string still need to be
2923 documented here. Constraints on what can be done (e.g. duplication, moving,
2924 etc need to be documented). This is probably best done by reference to
2925 another document that covers inline asm from a holistic perspective.</p>
2928 <!-- _______________________________________________________________________ -->
2930 <a name="inlineasm_md">Inline Asm Metadata</a>
2935 <p>The call instructions that wrap inline asm nodes may have a
2936 "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant
2937 integers. If present, the code generator will use the integer as the
2938 location cookie value when report errors through the <tt>LLVMContext</tt>
2939 error reporting mechanisms. This allows a front-end to correlate backend
2940 errors that occur with inline asm back to the source code that produced it.
2943 <pre class="doc_code">
2944 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2946 !42 = !{ i32 1234567 }
2949 <p>It is up to the front-end to make sense of the magic numbers it places in the
2950 IR. If the MDNode contains multiple constants, the code generator will use
2951 the one that corresponds to the line of the asm that the error occurs on.</p>
2957 <!-- ======================================================================= -->
2959 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2964 <p>LLVM IR allows metadata to be attached to instructions in the program that
2965 can convey extra information about the code to the optimizers and code
2966 generator. One example application of metadata is source-level debug
2967 information. There are two metadata primitives: strings and nodes. All
2968 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2969 preceding exclamation point ('<tt>!</tt>').</p>
2971 <p>A metadata string is a string surrounded by double quotes. It can contain
2972 any character by escaping non-printable characters with "<tt>\xx</tt>" where
2973 "<tt>xx</tt>" is the two digit hex code. For example:
2974 "<tt>!"test\00"</tt>".</p>
2976 <p>Metadata nodes are represented with notation similar to structure constants
2977 (a comma separated list of elements, surrounded by braces and preceded by an
2978 exclamation point). Metadata nodes can have any values as their operand. For
2981 <div class="doc_code">
2983 !{ metadata !"test\00", i32 10}
2987 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2988 metadata nodes, which can be looked up in the module symbol table. For
2991 <div class="doc_code">
2993 !foo = metadata !{!4, !3}
2997 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2998 function is using two metadata arguments:</p>
3000 <div class="doc_code">
3002 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3006 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
3007 attached to the <tt>add</tt> instruction using the <tt>!dbg</tt>
3010 <div class="doc_code">
3012 %indvar.next = add i64 %indvar, 1, !dbg !21
3016 <p>More information about specific metadata nodes recognized by the optimizers
3017 and code generator is found below.</p>
3019 <!-- _______________________________________________________________________ -->
3021 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
3026 <p>In LLVM IR, memory does not have types, so LLVM's own type system is not
3027 suitable for doing TBAA. Instead, metadata is added to the IR to describe
3028 a type system of a higher level language. This can be used to implement
3029 typical C/C++ TBAA, but it can also be used to implement custom alias
3030 analysis behavior for other languages.</p>
3032 <p>The current metadata format is very simple. TBAA metadata nodes have up to
3033 three fields, e.g.:</p>
3035 <div class="doc_code">
3037 !0 = metadata !{ metadata !"an example type tree" }
3038 !1 = metadata !{ metadata !"int", metadata !0 }
3039 !2 = metadata !{ metadata !"float", metadata !0 }
3040 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
3044 <p>The first field is an identity field. It can be any value, usually
3045 a metadata string, which uniquely identifies the type. The most important
3046 name in the tree is the name of the root node. Two trees with
3047 different root node names are entirely disjoint, even if they
3048 have leaves with common names.</p>
3050 <p>The second field identifies the type's parent node in the tree, or
3051 is null or omitted for a root node. A type is considered to alias
3052 all of its descendants and all of its ancestors in the tree. Also,
3053 a type is considered to alias all types in other trees, so that
3054 bitcode produced from multiple front-ends is handled conservatively.</p>
3056 <p>If the third field is present, it's an integer which if equal to 1
3057 indicates that the type is "constant" (meaning
3058 <tt>pointsToConstantMemory</tt> should return true; see
3059 <a href="AliasAnalysis.html#OtherItfs">other useful
3060 <tt>AliasAnalysis</tt> methods</a>).</p>
3064 <!-- _______________________________________________________________________ -->
3066 <a name="tbaa.struct">'<tt>tbaa.struct</tt>' Metadata</a>
3071 <p>The <a href="#int_memcpy"><tt>llvm.memcpy</tt></a> is often used to implement
3072 aggregate assignment operations in C and similar languages, however it is
3073 defined to copy a contiguous region of memory, which is more than strictly
3074 necessary for aggregate types which contain holes due to padding. Also, it
3075 doesn't contain any TBAA information about the fields of the aggregate.</p>
3077 <p><tt>!tbaa.struct</tt> metadata can describe which memory subregions in a memcpy
3078 are padding and what the TBAA tags of the struct are.</p>
3080 <p>The current metadata format is very simple. <tt>!tbaa.struct</tt> metadata nodes
3081 are a list of operands which are in conceptual groups of three. For each
3082 group of three, the first operand gives the byte offset of a field in bytes,
3083 the second gives its size in bytes, and the third gives its
3086 <div class="doc_code">
3088 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
3092 <p>This describes a struct with two fields. The first is at offset 0 bytes
3093 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
3094 and has size 4 bytes and has tbaa tag !2.</p>
3096 <p>Note that the fields need not be contiguous. In this example, there is a
3097 4 byte gap between the two fields. This gap represents padding which
3098 does not carry useful data and need not be preserved.</p>
3102 <!-- _______________________________________________________________________ -->
3104 <a name="fpmath">'<tt>fpmath</tt>' Metadata</a>
3109 <p><tt>fpmath</tt> metadata may be attached to any instruction of floating point
3110 type. It can be used to express the maximum acceptable error in the result of
3111 that instruction, in ULPs, thus potentially allowing the compiler to use a
3112 more efficient but less accurate method of computing it. ULP is defined as
3117 <p>If <tt>x</tt> is a real number that lies between two finite consecutive
3118 floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one
3119 of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the
3120 distance between the two non-equal finite floating-point numbers nearest
3121 <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p>
3125 <p>The metadata node shall consist of a single positive floating point number
3126 representing the maximum relative error, for example:</p>
3128 <div class="doc_code">
3130 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3136 <!-- _______________________________________________________________________ -->
3138 <a name="range">'<tt>range</tt>' Metadata</a>
3142 <p><tt>range</tt> metadata may be attached only to loads of integer types. It
3143 expresses the possible ranges the loaded value is in. The ranges are
3144 represented with a flattened list of integers. The loaded value is known to
3145 be in the union of the ranges defined by each consecutive pair. Each pair
3146 has the following properties:</p>
3148 <li>The type must match the type loaded by the instruction.</li>
3149 <li>The pair <tt>a,b</tt> represents the range <tt>[a,b)</tt>.</li>
3150 <li>Both <tt>a</tt> and <tt>b</tt> are constants.</li>
3151 <li>The range is allowed to wrap.</li>
3152 <li>The range should not represent the full or empty set. That is,
3153 <tt>a!=b</tt>. </li>
3155 <p> In addition, the pairs must be in signed order of the lower bound and
3156 they must be non-contiguous.</p>
3159 <div class="doc_code">
3161 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
3162 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3163 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
3164 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
3166 !0 = metadata !{ i8 0, i8 2 }
3167 !1 = metadata !{ i8 255, i8 2 }
3168 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
3169 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
3177 <!-- *********************************************************************** -->
3179 <a name="module_flags">Module Flags Metadata</a>
3181 <!-- *********************************************************************** -->
3185 <p>Information about the module as a whole is difficult to convey to LLVM's
3186 subsystems. The LLVM IR isn't sufficient to transmit this
3187 information. The <tt>llvm.module.flags</tt> named metadata exists in order to
3188 facilitate this. These flags are in the form of key / value pairs —
3189 much like a dictionary — making it easy for any subsystem who cares
3190 about a flag to look it up.</p>
3192 <p>The <tt>llvm.module.flags</tt> metadata contains a list of metadata
3193 triplets. Each triplet has the following form:</p>
3196 <li>The first element is a <i>behavior</i> flag, which specifies the behavior
3197 when two (or more) modules are merged together, and it encounters two (or
3198 more) metadata with the same ID. The supported behaviors are described
3201 <li>The second element is a metadata string that is a unique ID for the
3202 metadata. How each ID is interpreted is documented below.</li>
3204 <li>The third element is the value of the flag.</li>
3207 <p>When two (or more) modules are merged together, the resulting
3208 <tt>llvm.module.flags</tt> metadata is the union of the
3209 modules' <tt>llvm.module.flags</tt> metadata. The only exception being a flag
3210 with the <i>Override</i> behavior, which may override another flag's value
3213 <p>The following behaviors are supported:</p>
3215 <table border="1" cellspacing="0" cellpadding="4">
3225 <dt><b>Error</b></dt>
3226 <dd>Emits an error if two values disagree. It is an error to have an ID
3227 with both an Error and a Warning behavior.</dd>
3235 <dt><b>Warning</b></dt>
3236 <dd>Emits a warning if two values disagree.</dd>
3244 <dt><b>Require</b></dt>
3245 <dd>Emits an error when the specified value is not present or doesn't
3246 have the specified value. It is an error for two (or more)
3247 <tt>llvm.module.flags</tt> with the same ID to have the Require
3248 behavior but different values. There may be multiple Require flags
3257 <dt><b>Override</b></dt>
3258 <dd>Uses the specified value if the two values disagree. It is an
3259 error for two (or more) <tt>llvm.module.flags</tt> with the same
3260 ID to have the Override behavior but different values.</dd>
3267 <p>An example of module flags:</p>
3269 <pre class="doc_code">
3270 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3271 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3272 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3273 !3 = metadata !{ i32 3, metadata !"qux",
3275 metadata !"foo", i32 1
3278 !llvm.module.flags = !{ !0, !1, !2, !3 }
3282 <li><p>Metadata <tt>!0</tt> has the ID <tt>!"foo"</tt> and the value '1'. The
3283 behavior if two or more <tt>!"foo"</tt> flags are seen is to emit an
3284 error if their values are not equal.</p></li>
3286 <li><p>Metadata <tt>!1</tt> has the ID <tt>!"bar"</tt> and the value '37'. The
3287 behavior if two or more <tt>!"bar"</tt> flags are seen is to use the
3288 value '37' if their values are not equal.</p></li>
3290 <li><p>Metadata <tt>!2</tt> has the ID <tt>!"qux"</tt> and the value '42'. The
3291 behavior if two or more <tt>!"qux"</tt> flags are seen is to emit a
3292 warning if their values are not equal.</p></li>
3294 <li><p>Metadata <tt>!3</tt> has the ID <tt>!"qux"</tt> and the value:</p>
3296 <pre class="doc_code">
3297 metadata !{ metadata !"foo", i32 1 }
3300 <p>The behavior is to emit an error if the <tt>llvm.module.flags</tt> does
3301 not contain a flag with the ID <tt>!"foo"</tt> that has the value
3302 '1'. If two or more <tt>!"qux"</tt> flags exist, then they must have
3303 the same value or an error will be issued.</p></li>
3307 <!-- ======================================================================= -->
3309 <a name="objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a>
3314 <p>On the Mach-O platform, Objective-C stores metadata about garbage collection
3315 in a special section called "image info". The metadata consists of a version
3316 number and a bitmask specifying what types of garbage collection are
3317 supported (if any) by the file. If two or more modules are linked together
3318 their garbage collection metadata needs to be merged rather than appended
3321 <p>The Objective-C garbage collection module flags metadata consists of the
3322 following key-value pairs:</p>
3324 <table border="1" cellspacing="0" cellpadding="4">
3332 <td><tt>Objective-C Version</tt></td>
3333 <td align="left"><b>[Required]</b> — The Objective-C ABI
3334 version. Valid values are 1 and 2.</td>
3337 <td><tt>Objective-C Image Info Version</tt></td>
3338 <td align="left"><b>[Required]</b> — The version of the image info
3339 section. Currently always 0.</td>
3342 <td><tt>Objective-C Image Info Section</tt></td>
3343 <td align="left"><b>[Required]</b> — The section to place the
3344 metadata. Valid values are <tt>"__OBJC, __image_info, regular"</tt> for
3345 Objective-C ABI version 1, and <tt>"__DATA,__objc_imageinfo, regular,
3346 no_dead_strip"</tt> for Objective-C ABI version 2.</td>
3349 <td><tt>Objective-C Garbage Collection</tt></td>
3350 <td align="left"><b>[Required]</b> — Specifies whether garbage
3351 collection is supported or not. Valid values are 0, for no garbage
3352 collection, and 2, for garbage collection supported.</td>
3355 <td><tt>Objective-C GC Only</tt></td>
3356 <td align="left"><b>[Optional]</b> — Specifies that only garbage
3357 collection is supported. If present, its value must be 6. This flag
3358 requires that the <tt>Objective-C Garbage Collection</tt> flag have the
3364 <p>Some important flag interactions:</p>
3367 <li>If a module with <tt>Objective-C Garbage Collection</tt> set to 0 is
3368 merged with a module with <tt>Objective-C Garbage Collection</tt> set to
3369 2, then the resulting module has the <tt>Objective-C Garbage
3370 Collection</tt> flag set to 0.</li>
3372 <li>A module with <tt>Objective-C Garbage Collection</tt> set to 0 cannot be
3373 merged with a module with <tt>Objective-C GC Only</tt> set to 6.</li>
3380 <!-- *********************************************************************** -->
3382 <a name="intrinsic_globals">Intrinsic Global Variables</a>
3384 <!-- *********************************************************************** -->
3386 <p>LLVM has a number of "magic" global variables that contain data that affect
3387 code generation or other IR semantics. These are documented here. All globals
3388 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
3389 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
3392 <!-- ======================================================================= -->
3394 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
3399 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
3400 href="#linkage_appending">appending linkage</a>. This array contains a list of
3401 pointers to global variables and functions which may optionally have a pointer
3402 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
3404 <div class="doc_code">
3409 @llvm.used = appending global [2 x i8*] [
3411 i8* bitcast (i32* @Y to i8*)
3412 ], section "llvm.metadata"
3416 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
3417 compiler, assembler, and linker are required to treat the symbol as if there
3418 is a reference to the global that it cannot see. For example, if a variable
3419 has internal linkage and no references other than that from
3420 the <tt>@llvm.used</tt> list, it cannot be deleted. This is commonly used to
3421 represent references from inline asms and other things the compiler cannot
3422 "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p>
3424 <p>On some targets, the code generator must emit a directive to the assembler or
3425 object file to prevent the assembler and linker from molesting the
3430 <!-- ======================================================================= -->
3432 <a name="intg_compiler_used">
3433 The '<tt>llvm.compiler.used</tt>' Global Variable
3439 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
3440 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
3441 touching the symbol. On targets that support it, this allows an intelligent
3442 linker to optimize references to the symbol without being impeded as it would
3443 be by <tt>@llvm.used</tt>.</p>
3445 <p>This is a rare construct that should only be used in rare circumstances, and
3446 should not be exposed to source languages.</p>
3450 <!-- ======================================================================= -->
3452 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
3457 <div class="doc_code">
3459 %0 = type { i32, void ()* }
3460 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3464 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor
3465 functions and associated priorities. The functions referenced by this array
3466 will be called in ascending order of priority (i.e. lowest first) when the
3467 module is loaded. The order of functions with the same priority is not
3472 <!-- ======================================================================= -->
3474 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3479 <div class="doc_code">
3481 %0 = type { i32, void ()* }
3482 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3486 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions
3487 and associated priorities. The functions referenced by this array will be
3488 called in descending order of priority (i.e. highest first) when the module
3489 is loaded. The order of functions with the same priority is not defined.</p>
3495 <!-- *********************************************************************** -->
3496 <h2><a name="instref">Instruction Reference</a></h2>
3497 <!-- *********************************************************************** -->
3501 <p>The LLVM instruction set consists of several different classifications of
3502 instructions: <a href="#terminators">terminator
3503 instructions</a>, <a href="#binaryops">binary instructions</a>,
3504 <a href="#bitwiseops">bitwise binary instructions</a>,
3505 <a href="#memoryops">memory instructions</a>, and
3506 <a href="#otherops">other instructions</a>.</p>
3508 <!-- ======================================================================= -->
3510 <a name="terminators">Terminator Instructions</a>
3515 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3516 in a program ends with a "Terminator" instruction, which indicates which
3517 block should be executed after the current block is finished. These
3518 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3519 control flow, not values (the one exception being the
3520 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3522 <p>The terminator instructions are:
3523 '<a href="#i_ret"><tt>ret</tt></a>',
3524 '<a href="#i_br"><tt>br</tt></a>',
3525 '<a href="#i_switch"><tt>switch</tt></a>',
3526 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3527 '<a href="#i_invoke"><tt>invoke</tt></a>',
3528 '<a href="#i_resume"><tt>resume</tt></a>', and
3529 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3531 <!-- _______________________________________________________________________ -->
3533 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3540 ret <type> <value> <i>; Return a value from a non-void function</i>
3541 ret void <i>; Return from void function</i>
3545 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3546 a value) from a function back to the caller.</p>
3548 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3549 value and then causes control flow, and one that just causes control flow to
3553 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3554 return value. The type of the return value must be a
3555 '<a href="#t_firstclass">first class</a>' type.</p>
3557 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3558 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3559 value or a return value with a type that does not match its type, or if it
3560 has a void return type and contains a '<tt>ret</tt>' instruction with a
3564 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3565 the calling function's context. If the caller is a
3566 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3567 instruction after the call. If the caller was an
3568 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3569 the beginning of the "normal" destination block. If the instruction returns
3570 a value, that value shall set the call or invoke instruction's return
3575 ret i32 5 <i>; Return an integer value of 5</i>
3576 ret void <i>; Return from a void function</i>
3577 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3581 <!-- _______________________________________________________________________ -->
3583 <a name="i_br">'<tt>br</tt>' Instruction</a>
3590 br i1 <cond>, label <iftrue>, label <iffalse>
3591 br label <dest> <i>; Unconditional branch</i>
3595 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3596 different basic block in the current function. There are two forms of this
3597 instruction, corresponding to a conditional branch and an unconditional
3601 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3602 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3603 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3607 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3608 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3609 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3610 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3615 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3616 br i1 %cond, label %IfEqual, label %IfUnequal
3618 <a href="#i_ret">ret</a> i32 1
3620 <a href="#i_ret">ret</a> i32 0
3625 <!-- _______________________________________________________________________ -->
3627 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3634 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3638 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3639 several different places. It is a generalization of the '<tt>br</tt>'
3640 instruction, allowing a branch to occur to one of many possible
3644 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3645 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3646 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3647 The table is not allowed to contain duplicate constant entries.</p>
3650 <p>The <tt>switch</tt> instruction specifies a table of values and
3651 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3652 is searched for the given value. If the value is found, control flow is
3653 transferred to the corresponding destination; otherwise, control flow is
3654 transferred to the default destination.</p>
3656 <h5>Implementation:</h5>
3657 <p>Depending on properties of the target machine and the particular
3658 <tt>switch</tt> instruction, this instruction may be code generated in
3659 different ways. For example, it could be generated as a series of chained
3660 conditional branches or with a lookup table.</p>
3664 <i>; Emulate a conditional br instruction</i>
3665 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3666 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3668 <i>; Emulate an unconditional br instruction</i>
3669 switch i32 0, label %dest [ ]
3671 <i>; Implement a jump table:</i>
3672 switch i32 %val, label %otherwise [ i32 0, label %onzero
3674 i32 2, label %ontwo ]
3680 <!-- _______________________________________________________________________ -->
3682 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3689 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3694 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3695 within the current function, whose address is specified by
3696 "<tt>address</tt>". Address must be derived from a <a
3697 href="#blockaddress">blockaddress</a> constant.</p>
3701 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3702 rest of the arguments indicate the full set of possible destinations that the
3703 address may point to. Blocks are allowed to occur multiple times in the
3704 destination list, though this isn't particularly useful.</p>
3706 <p>This destination list is required so that dataflow analysis has an accurate
3707 understanding of the CFG.</p>
3711 <p>Control transfers to the block specified in the address argument. All
3712 possible destination blocks must be listed in the label list, otherwise this
3713 instruction has undefined behavior. This implies that jumps to labels
3714 defined in other functions have undefined behavior as well.</p>
3716 <h5>Implementation:</h5>
3718 <p>This is typically implemented with a jump through a register.</p>
3722 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3728 <!-- _______________________________________________________________________ -->
3730 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3737 <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>]
3738 to label <normal label> unwind label <exception label>
3742 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3743 function, with the possibility of control flow transfer to either the
3744 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3745 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3746 control flow will return to the "normal" label. If the callee (or any
3747 indirect callees) returns via the "<a href="#i_resume"><tt>resume</tt></a>"
3748 instruction or other exception handling mechanism, control is interrupted and
3749 continued at the dynamically nearest "exception" label.</p>
3751 <p>The '<tt>exception</tt>' label is a
3752 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3753 exception. As such, '<tt>exception</tt>' label is required to have the
3754 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3755 the information about the behavior of the program after unwinding
3756 happens, as its first non-PHI instruction. The restrictions on the
3757 "<tt>landingpad</tt>" instruction's tightly couples it to the
3758 "<tt>invoke</tt>" instruction, so that the important information contained
3759 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3763 <p>This instruction requires several arguments:</p>
3766 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3767 convention</a> the call should use. If none is specified, the call
3768 defaults to using C calling conventions.</li>
3770 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3771 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3772 '<tt>inreg</tt>' attributes are valid here.</li>
3774 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3775 function value being invoked. In most cases, this is a direct function
3776 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3777 off an arbitrary pointer to function value.</li>
3779 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3780 function to be invoked. </li>
3782 <li>'<tt>function args</tt>': argument list whose types match the function
3783 signature argument types and parameter attributes. All arguments must be
3784 of <a href="#t_firstclass">first class</a> type. If the function
3785 signature indicates the function accepts a variable number of arguments,
3786 the extra arguments can be specified.</li>
3788 <li>'<tt>normal label</tt>': the label reached when the called function
3789 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3791 <li>'<tt>exception label</tt>': the label reached when a callee returns via
3792 the <a href="#i_resume"><tt>resume</tt></a> instruction or other exception
3793 handling mechanism.</li>
3795 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3796 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3797 '<tt>readnone</tt>' attributes are valid here.</li>
3801 <p>This instruction is designed to operate as a standard
3802 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3803 primary difference is that it establishes an association with a label, which
3804 is used by the runtime library to unwind the stack.</p>
3806 <p>This instruction is used in languages with destructors to ensure that proper
3807 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3808 exception. Additionally, this is important for implementation of
3809 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3811 <p>For the purposes of the SSA form, the definition of the value returned by the
3812 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3813 block to the "normal" label. If the callee unwinds then no return value is
3818 %retval = invoke i32 @Test(i32 15) to label %Continue
3819 unwind label %TestCleanup <i>; {i32}:retval set</i>
3820 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3821 unwind label %TestCleanup <i>; {i32}:retval set</i>
3826 <!-- _______________________________________________________________________ -->
3829 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3836 resume <type> <value>
3840 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3844 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3845 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3849 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3850 (in-flight) exception whose unwinding was interrupted with
3851 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3855 resume { i8*, i32 } %exn
3860 <!-- _______________________________________________________________________ -->
3863 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3874 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3875 instruction is used to inform the optimizer that a particular portion of the
3876 code is not reachable. This can be used to indicate that the code after a
3877 no-return function cannot be reached, and other facts.</p>
3880 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3886 <!-- ======================================================================= -->
3888 <a name="binaryops">Binary Operations</a>
3893 <p>Binary operators are used to do most of the computation in a program. They
3894 require two operands of the same type, execute an operation on them, and
3895 produce a single value. The operands might represent multiple data, as is
3896 the case with the <a href="#t_vector">vector</a> data type. The result value
3897 has the same type as its operands.</p>
3899 <p>There are several different binary operators:</p>
3901 <!-- _______________________________________________________________________ -->
3903 <a name="i_add">'<tt>add</tt>' Instruction</a>
3910 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3911 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3912 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3913 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3917 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3920 <p>The two arguments to the '<tt>add</tt>' instruction must
3921 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3922 integer values. Both arguments must have identical types.</p>
3925 <p>The value produced is the integer sum of the two operands.</p>
3927 <p>If the sum has unsigned overflow, the result returned is the mathematical
3928 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3930 <p>Because LLVM integers use a two's complement representation, this instruction
3931 is appropriate for both signed and unsigned integers.</p>
3933 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3934 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3935 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3936 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3937 respectively, occurs.</p>
3941 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3946 <!-- _______________________________________________________________________ -->
3948 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3955 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3959 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3962 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3963 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3964 floating point values. Both arguments must have identical types.</p>
3967 <p>The value produced is the floating point sum of the two operands.</p>
3971 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3976 <!-- _______________________________________________________________________ -->
3978 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3985 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3986 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3987 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3988 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3992 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3995 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3996 '<tt>neg</tt>' instruction present in most other intermediate
3997 representations.</p>
4000 <p>The two arguments to the '<tt>sub</tt>' instruction must
4001 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4002 integer values. Both arguments must have identical types.</p>
4005 <p>The value produced is the integer difference of the two operands.</p>
4007 <p>If the difference has unsigned overflow, the result returned is the
4008 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
4011 <p>Because LLVM integers use a two's complement representation, this instruction
4012 is appropriate for both signed and unsigned integers.</p>
4014 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
4015 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
4016 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
4017 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4018 respectively, occurs.</p>
4022 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
4023 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
4028 <!-- _______________________________________________________________________ -->
4030 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
4037 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4041 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
4044 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
4045 '<tt>fneg</tt>' instruction present in most other intermediate
4046 representations.</p>
4049 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
4050 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4051 floating point values. Both arguments must have identical types.</p>
4054 <p>The value produced is the floating point difference of the two operands.</p>
4058 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
4059 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
4064 <!-- _______________________________________________________________________ -->
4066 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
4073 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4074 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4075 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4076 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4080 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
4083 <p>The two arguments to the '<tt>mul</tt>' instruction must
4084 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4085 integer values. Both arguments must have identical types.</p>
4088 <p>The value produced is the integer product of the two operands.</p>
4090 <p>If the result of the multiplication has unsigned overflow, the result
4091 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
4092 width of the result.</p>
4094 <p>Because LLVM integers use a two's complement representation, and the result
4095 is the same width as the operands, this instruction returns the correct
4096 result for both signed and unsigned integers. If a full product
4097 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
4098 be sign-extended or zero-extended as appropriate to the width of the full
4101 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
4102 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
4103 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
4104 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4105 respectively, occurs.</p>
4109 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
4114 <!-- _______________________________________________________________________ -->
4116 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
4123 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4127 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
4130 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
4131 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4132 floating point values. Both arguments must have identical types.</p>
4135 <p>The value produced is the floating point product of the two operands.</p>
4139 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
4144 <!-- _______________________________________________________________________ -->
4146 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
4153 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4154 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4158 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
4161 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
4162 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4163 values. Both arguments must have identical types.</p>
4166 <p>The value produced is the unsigned integer quotient of the two operands.</p>
4168 <p>Note that unsigned integer division and signed integer division are distinct
4169 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
4171 <p>Division by zero leads to undefined behavior.</p>
4173 <p>If the <tt>exact</tt> keyword is present, the result value of the
4174 <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a
4175 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
4180 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4185 <!-- _______________________________________________________________________ -->
4187 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
4194 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4195 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4199 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
4202 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
4203 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4204 values. Both arguments must have identical types.</p>
4207 <p>The value produced is the signed integer quotient of the two operands rounded
4210 <p>Note that signed integer division and unsigned integer division are distinct
4211 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
4213 <p>Division by zero leads to undefined behavior. Overflow also leads to
4214 undefined behavior; this is a rare case, but can occur, for example, by doing
4215 a 32-bit division of -2147483648 by -1.</p>
4217 <p>If the <tt>exact</tt> keyword is present, the result value of the
4218 <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would
4223 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4228 <!-- _______________________________________________________________________ -->
4230 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
4237 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4241 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
4244 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
4245 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4246 floating point values. Both arguments must have identical types.</p>
4249 <p>The value produced is the floating point quotient of the two operands.</p>
4253 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
4258 <!-- _______________________________________________________________________ -->
4260 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
4267 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4271 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
4272 division of its two arguments.</p>
4275 <p>The two arguments to the '<tt>urem</tt>' instruction must be
4276 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4277 values. Both arguments must have identical types.</p>
4280 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
4281 This instruction always performs an unsigned division to get the
4284 <p>Note that unsigned integer remainder and signed integer remainder are
4285 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
4287 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
4291 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4296 <!-- _______________________________________________________________________ -->
4298 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
4305 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4309 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
4310 division of its two operands. This instruction can also take
4311 <a href="#t_vector">vector</a> versions of the values in which case the
4312 elements must be integers.</p>
4315 <p>The two arguments to the '<tt>srem</tt>' instruction must be
4316 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4317 values. Both arguments must have identical types.</p>
4320 <p>This instruction returns the <i>remainder</i> of a division (where the result
4321 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
4322 <i>modulo</i> operator (where the result is either zero or has the same sign
4323 as the divisor, <tt>op2</tt>) of a value.
4324 For more information about the difference,
4325 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
4326 Math Forum</a>. For a table of how this is implemented in various languages,
4327 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
4328 Wikipedia: modulo operation</a>.</p>
4330 <p>Note that signed integer remainder and unsigned integer remainder are
4331 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
4333 <p>Taking the remainder of a division by zero leads to undefined behavior.
4334 Overflow also leads to undefined behavior; this is a rare case, but can
4335 occur, for example, by taking the remainder of a 32-bit division of
4336 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
4337 lets srem be implemented using instructions that return both the result of
4338 the division and the remainder.)</p>
4342 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4347 <!-- _______________________________________________________________________ -->
4349 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
4356 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4360 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
4361 its two operands.</p>
4364 <p>The two arguments to the '<tt>frem</tt>' instruction must be
4365 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4366 floating point values. Both arguments must have identical types.</p>
4369 <p>This instruction returns the <i>remainder</i> of a division. The remainder
4370 has the same sign as the dividend.</p>
4374 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
4381 <!-- ======================================================================= -->
4383 <a name="bitwiseops">Bitwise Binary Operations</a>
4388 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
4389 program. They are generally very efficient instructions and can commonly be
4390 strength reduced from other instructions. They require two operands of the
4391 same type, execute an operation on them, and produce a single value. The
4392 resulting value is the same type as its operands.</p>
4394 <!-- _______________________________________________________________________ -->
4396 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
4403 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4404 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4405 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4406 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4410 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
4411 a specified number of bits.</p>
4414 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
4415 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4416 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4419 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
4420 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
4421 is (statically or dynamically) negative or equal to or larger than the number
4422 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4423 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4424 shift amount in <tt>op2</tt>.</p>
4426 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
4427 <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits. If
4428 the <tt>nsw</tt> keyword is present, then the shift produces a
4429 <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree
4430 with the resultant sign bit. As such, NUW/NSW have the same semantics as
4431 they would if the shift were expressed as a mul instruction with the same
4432 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
4436 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
4437 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
4438 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
4439 <result> = shl i32 1, 32 <i>; undefined</i>
4440 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4445 <!-- _______________________________________________________________________ -->
4447 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4454 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4455 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4459 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4460 operand shifted to the right a specified number of bits with zero fill.</p>
4463 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4464 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4465 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4468 <p>This instruction always performs a logical shift right operation. The most
4469 significant bits of the result will be filled with zero bits after the shift.
4470 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4471 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4472 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4473 shift amount in <tt>op2</tt>.</p>
4475 <p>If the <tt>exact</tt> keyword is present, the result value of the
4476 <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4477 shifted out are non-zero.</p>
4482 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4483 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4484 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4485 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4486 <result> = lshr i32 1, 32 <i>; undefined</i>
4487 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4492 <!-- _______________________________________________________________________ -->
4494 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4501 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4502 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4506 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4507 operand shifted to the right a specified number of bits with sign
4511 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4512 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4513 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4516 <p>This instruction always performs an arithmetic shift right operation, The
4517 most significant bits of the result will be filled with the sign bit
4518 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4519 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4520 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4521 the corresponding shift amount in <tt>op2</tt>.</p>
4523 <p>If the <tt>exact</tt> keyword is present, the result value of the
4524 <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4525 shifted out are non-zero.</p>
4529 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4530 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4531 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4532 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4533 <result> = ashr i32 1, 32 <i>; undefined</i>
4534 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4539 <!-- _______________________________________________________________________ -->
4541 <a name="i_and">'<tt>and</tt>' Instruction</a>
4548 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4552 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4556 <p>The two arguments to the '<tt>and</tt>' instruction must be
4557 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4558 values. Both arguments must have identical types.</p>
4561 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4563 <table border="1" cellspacing="0" cellpadding="4">
4595 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4596 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4597 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4600 <!-- _______________________________________________________________________ -->
4602 <a name="i_or">'<tt>or</tt>' Instruction</a>
4609 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4613 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4617 <p>The two arguments to the '<tt>or</tt>' instruction must be
4618 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4619 values. Both arguments must have identical types.</p>
4622 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4624 <table border="1" cellspacing="0" cellpadding="4">
4656 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4657 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4658 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4663 <!-- _______________________________________________________________________ -->
4665 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4672 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4676 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4677 its two operands. The <tt>xor</tt> is used to implement the "one's
4678 complement" operation, which is the "~" operator in C.</p>
4681 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4682 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4683 values. Both arguments must have identical types.</p>
4686 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4688 <table border="1" cellspacing="0" cellpadding="4">
4720 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4721 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4722 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4723 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4730 <!-- ======================================================================= -->
4732 <a name="vectorops">Vector Operations</a>
4737 <p>LLVM supports several instructions to represent vector operations in a
4738 target-independent manner. These instructions cover the element-access and
4739 vector-specific operations needed to process vectors effectively. While LLVM
4740 does directly support these vector operations, many sophisticated algorithms
4741 will want to use target-specific intrinsics to take full advantage of a
4742 specific target.</p>
4744 <!-- _______________________________________________________________________ -->
4746 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4753 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4757 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4758 from a vector at a specified index.</p>
4762 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4763 of <a href="#t_vector">vector</a> type. The second operand is an index
4764 indicating the position from which to extract the element. The index may be
4768 <p>The result is a scalar of the same type as the element type of
4769 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4770 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4771 results are undefined.</p>
4775 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4780 <!-- _______________________________________________________________________ -->
4782 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4789 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4793 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4794 vector at a specified index.</p>
4797 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4798 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4799 whose type must equal the element type of the first operand. The third
4800 operand is an index indicating the position at which to insert the value.
4801 The index may be a variable.</p>
4804 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4805 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4806 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4807 results are undefined.</p>
4811 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4816 <!-- _______________________________________________________________________ -->
4818 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4825 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4829 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4830 from two input vectors, returning a vector with the same element type as the
4831 input and length that is the same as the shuffle mask.</p>
4834 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4835 with the same type. The third argument is a shuffle mask whose
4836 element type is always 'i32'. The result of the instruction is a vector
4837 whose length is the same as the shuffle mask and whose element type is the
4838 same as the element type of the first two operands.</p>
4840 <p>The shuffle mask operand is required to be a constant vector with either
4841 constant integer or undef values.</p>
4844 <p>The elements of the two input vectors are numbered from left to right across
4845 both of the vectors. The shuffle mask operand specifies, for each element of
4846 the result vector, which element of the two input vectors the result element
4847 gets. The element selector may be undef (meaning "don't care") and the
4848 second operand may be undef if performing a shuffle from only one vector.</p>
4852 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4853 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4854 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4855 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4856 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4857 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4858 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4859 <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>
4866 <!-- ======================================================================= -->
4868 <a name="aggregateops">Aggregate Operations</a>
4873 <p>LLVM supports several instructions for working with
4874 <a href="#t_aggregate">aggregate</a> values.</p>
4876 <!-- _______________________________________________________________________ -->
4878 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4885 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4889 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4890 from an <a href="#t_aggregate">aggregate</a> value.</p>
4893 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4894 of <a href="#t_struct">struct</a> or
4895 <a href="#t_array">array</a> type. The operands are constant indices to
4896 specify which value to extract in a similar manner as indices in a
4897 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4898 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4900 <li>Since the value being indexed is not a pointer, the first index is
4901 omitted and assumed to be zero.</li>
4902 <li>At least one index must be specified.</li>
4903 <li>Not only struct indices but also array indices must be in
4908 <p>The result is the value at the position in the aggregate specified by the
4913 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4918 <!-- _______________________________________________________________________ -->
4920 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4927 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4931 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4932 in an <a href="#t_aggregate">aggregate</a> value.</p>
4935 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4936 of <a href="#t_struct">struct</a> or
4937 <a href="#t_array">array</a> type. The second operand is a first-class
4938 value to insert. The following operands are constant indices indicating
4939 the position at which to insert the value in a similar manner as indices in a
4940 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4941 value to insert must have the same type as the value identified by the
4945 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4946 that of <tt>val</tt> except that the value at the position specified by the
4947 indices is that of <tt>elt</tt>.</p>
4951 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4952 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4953 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4960 <!-- ======================================================================= -->
4962 <a name="memoryops">Memory Access and Addressing Operations</a>
4967 <p>A key design point of an SSA-based representation is how it represents
4968 memory. In LLVM, no memory locations are in SSA form, which makes things
4969 very simple. This section describes how to read, write, and allocate
4972 <!-- _______________________________________________________________________ -->
4974 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4981 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4985 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4986 currently executing function, to be automatically released when this function
4987 returns to its caller. The object is always allocated in the generic address
4988 space (address space zero).</p>
4991 <p>The '<tt>alloca</tt>' instruction
4992 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4993 runtime stack, returning a pointer of the appropriate type to the program.
4994 If "NumElements" is specified, it is the number of elements allocated,
4995 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4996 specified, the value result of the allocation is guaranteed to be aligned to
4997 at least that boundary. If not specified, or if zero, the target can choose
4998 to align the allocation on any convenient boundary compatible with the
5001 <p>'<tt>type</tt>' may be any sized type.</p>
5004 <p>Memory is allocated; a pointer is returned. The operation is undefined if
5005 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
5006 memory is automatically released when the function returns. The
5007 '<tt>alloca</tt>' instruction is commonly used to represent automatic
5008 variables that must have an address available. When the function returns
5009 (either with the <tt><a href="#i_ret">ret</a></tt>
5010 or <tt><a href="#i_resume">resume</a></tt> instructions), the memory is
5011 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5012 The order in which memory is allocated (ie., which way the stack grows) is
5019 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
5020 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
5021 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
5022 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
5027 <!-- _______________________________________________________________________ -->
5029 <a name="i_load">'<tt>load</tt>' Instruction</a>
5036 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
5037 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5038 !<index> = !{ i32 1 }
5042 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
5045 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
5046 from which to load. The pointer must point to
5047 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
5048 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
5049 number or order of execution of this <tt>load</tt> with other <a
5050 href="#volatile">volatile operations</a>.</p>
5052 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
5053 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5054 argument. The <code>release</code> and <code>acq_rel</code> orderings are
5055 not valid on <code>load</code> instructions. Atomic loads produce <a
5056 href="#memorymodel">defined</a> results when they may see multiple atomic
5057 stores. The type of the pointee must be an integer type whose bit width
5058 is a power of two greater than or equal to eight and less than or equal
5059 to a target-specific size limit. <code>align</code> must be explicitly
5060 specified on atomic loads, and the load has undefined behavior if the
5061 alignment is not set to a value which is at least the size in bytes of
5062 the pointee. <code>!nontemporal</code> does not have any defined semantics
5063 for atomic loads.</p>
5065 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
5066 operation (that is, the alignment of the memory address). A value of 0 or an
5067 omitted <tt>align</tt> argument means that the operation has the abi
5068 alignment for the target. It is the responsibility of the code emitter to
5069 ensure that the alignment information is correct. Overestimating the
5070 alignment results in undefined behavior. Underestimating the alignment may
5071 produce less efficient code. An alignment of 1 is always safe.</p>
5073 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
5074 metatadata name <index> corresponding to a metadata node with
5075 one <tt>i32</tt> entry of value 1. The existence of
5076 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
5077 and code generator that this load is not expected to be reused in the cache.
5078 The code generator may select special instructions to save cache bandwidth,
5079 such as the <tt>MOVNT</tt> instruction on x86.</p>
5081 <p>The optional <tt>!invariant.load</tt> metadata must reference a single
5082 metatadata name <index> corresponding to a metadata node with no
5083 entries. The existence of the <tt>!invariant.load</tt> metatadata on the
5084 instruction tells the optimizer and code generator that this load address
5085 points to memory which does not change value during program execution.
5086 The optimizer may then move this load around, for example, by hoisting it
5087 out of loops using loop invariant code motion.</p>
5090 <p>The location of memory pointed to is loaded. If the value being loaded is of
5091 scalar type then the number of bytes read does not exceed the minimum number
5092 of bytes needed to hold all bits of the type. For example, loading an
5093 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
5094 <tt>i20</tt> with a size that is not an integral number of bytes, the result
5095 is undefined if the value was not originally written using a store of the
5100 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5101 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
5102 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
5107 <!-- _______________________________________________________________________ -->
5109 <a name="i_store">'<tt>store</tt>' Instruction</a>
5116 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
5117 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
5121 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
5124 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
5125 and an address at which to store it. The type of the
5126 '<tt><pointer></tt>' operand must be a pointer to
5127 the <a href="#t_firstclass">first class</a> type of the
5128 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
5129 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
5130 order of execution of this <tt>store</tt> with other <a
5131 href="#volatile">volatile operations</a>.</p>
5133 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
5134 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5135 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
5136 valid on <code>store</code> instructions. Atomic loads produce <a
5137 href="#memorymodel">defined</a> results when they may see multiple atomic
5138 stores. The type of the pointee must be an integer type whose bit width
5139 is a power of two greater than or equal to eight and less than or equal
5140 to a target-specific size limit. <code>align</code> must be explicitly
5141 specified on atomic stores, and the store has undefined behavior if the
5142 alignment is not set to a value which is at least the size in bytes of
5143 the pointee. <code>!nontemporal</code> does not have any defined semantics
5144 for atomic stores.</p>
5146 <p>The optional constant "align" argument specifies the alignment of the
5147 operation (that is, the alignment of the memory address). A value of 0 or an
5148 omitted "align" argument means that the operation has the abi
5149 alignment for the target. It is the responsibility of the code emitter to
5150 ensure that the alignment information is correct. Overestimating the
5151 alignment results in an undefined behavior. Underestimating the alignment may
5152 produce less efficient code. An alignment of 1 is always safe.</p>
5154 <p>The optional !nontemporal metadata must reference a single metatadata
5155 name <index> corresponding to a metadata node with one i32 entry of
5156 value 1. The existence of the !nontemporal metatadata on the
5157 instruction tells the optimizer and code generator that this load is
5158 not expected to be reused in the cache. The code generator may
5159 select special instructions to save cache bandwidth, such as the
5160 MOVNT instruction on x86.</p>
5164 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
5165 location specified by the '<tt><pointer></tt>' operand. If
5166 '<tt><value></tt>' is of scalar type then the number of bytes written
5167 does not exceed the minimum number of bytes needed to hold all bits of the
5168 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
5169 writing a value of a type like <tt>i20</tt> with a size that is not an
5170 integral number of bytes, it is unspecified what happens to the extra bits
5171 that do not belong to the type, but they will typically be overwritten.</p>
5175 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5176 store i32 3, i32* %ptr <i>; yields {void}</i>
5177 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
5182 <!-- _______________________________________________________________________ -->
5184 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
5191 fence [singlethread] <ordering> <i>; yields {void}</i>
5195 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
5196 between operations.</p>
5198 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
5199 href="#ordering">ordering</a> argument which defines what
5200 <i>synchronizes-with</i> edges they add. They can only be given
5201 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
5202 <code>seq_cst</code> orderings.</p>
5205 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
5206 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
5207 <code>acquire</code> ordering semantics if and only if there exist atomic
5208 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
5209 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
5210 <var>X</var> modifies <var>M</var> (either directly or through some side effect
5211 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
5212 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
5213 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
5214 than an explicit <code>fence</code>, one (but not both) of the atomic operations
5215 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
5216 <code>acquire</code> (resp.) ordering constraint and still
5217 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
5218 <i>happens-before</i> edge.</p>
5220 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
5221 having both <code>acquire</code> and <code>release</code> semantics specified
5222 above, participates in the global program order of other <code>seq_cst</code>
5223 operations and/or fences.</p>
5225 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
5226 specifies that the fence only synchronizes with other fences in the same
5227 thread. (This is useful for interacting with signal handlers.)</p>
5231 fence acquire <i>; yields {void}</i>
5232 fence singlethread seq_cst <i>; yields {void}</i>
5237 <!-- _______________________________________________________________________ -->
5239 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
5246 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
5250 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
5251 It loads a value in memory and compares it to a given value. If they are
5252 equal, it stores a new value into the memory.</p>
5255 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
5256 address to operate on, a value to compare to the value currently be at that
5257 address, and a new value to place at that address if the compared values are
5258 equal. The type of '<var><cmp></var>' must be an integer type whose
5259 bit width is a power of two greater than or equal to eight and less than
5260 or equal to a target-specific size limit. '<var><cmp></var>' and
5261 '<var><new></var>' must have the same type, and the type of
5262 '<var><pointer></var>' must be a pointer to that type. If the
5263 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
5264 optimizer is not allowed to modify the number or order of execution
5265 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
5268 <!-- FIXME: Extend allowed types. -->
5270 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
5271 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
5273 <p>The optional "<code>singlethread</code>" argument declares that the
5274 <code>cmpxchg</code> is only atomic with respect to code (usually signal
5275 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
5276 cmpxchg is atomic with respect to all other code in the system.</p>
5278 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
5279 the size in memory of the operand.
5282 <p>The contents of memory at the location specified by the
5283 '<tt><pointer></tt>' operand is read and compared to
5284 '<tt><cmp></tt>'; if the read value is the equal,
5285 '<tt><new></tt>' is written. The original value at the location
5288 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
5289 purpose of identifying <a href="#release_sequence">release sequences</a>. A
5290 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
5291 parameter determined by dropping any <code>release</code> part of the
5292 <code>cmpxchg</code>'s ordering.</p>
5295 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
5296 optimization work on ARM.)
5298 FIXME: Is a weaker ordering constraint on failure helpful in practice?
5304 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
5305 <a href="#i_br">br</a> label %loop
5308 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
5309 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
5310 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
5311 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
5312 <a href="#i_br">br</a> i1 %success, label %done, label %loop
5320 <!-- _______________________________________________________________________ -->
5322 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
5329 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
5333 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
5336 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
5337 operation to apply, an address whose value to modify, an argument to the
5338 operation. The operation must be one of the following keywords:</p>
5353 <p>The type of '<var><value></var>' must be an integer type whose
5354 bit width is a power of two greater than or equal to eight and less than
5355 or equal to a target-specific size limit. The type of the
5356 '<code><pointer></code>' operand must be a pointer to that type.
5357 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
5358 optimizer is not allowed to modify the number or order of execution of this
5359 <code>atomicrmw</code> with other <a href="#volatile">volatile
5362 <!-- FIXME: Extend allowed types. -->
5365 <p>The contents of memory at the location specified by the
5366 '<tt><pointer></tt>' operand are atomically read, modified, and written
5367 back. The original value at the location is returned. The modification is
5368 specified by the <var>operation</var> argument:</p>
5371 <li>xchg: <code>*ptr = val</code></li>
5372 <li>add: <code>*ptr = *ptr + val</code></li>
5373 <li>sub: <code>*ptr = *ptr - val</code></li>
5374 <li>and: <code>*ptr = *ptr & val</code></li>
5375 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
5376 <li>or: <code>*ptr = *ptr | val</code></li>
5377 <li>xor: <code>*ptr = *ptr ^ val</code></li>
5378 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
5379 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
5380 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
5381 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
5386 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
5391 <!-- _______________________________________________________________________ -->
5393 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
5400 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5401 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5402 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5406 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
5407 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
5408 It performs address calculation only and does not access memory.</p>
5411 <p>The first argument is always a pointer or a vector of pointers,
5412 and forms the basis of the
5413 calculation. The remaining arguments are indices that indicate which of the
5414 elements of the aggregate object are indexed. The interpretation of each
5415 index is dependent on the type being indexed into. The first index always
5416 indexes the pointer value given as the first argument, the second index
5417 indexes a value of the type pointed to (not necessarily the value directly
5418 pointed to, since the first index can be non-zero), etc. The first type
5419 indexed into must be a pointer value, subsequent types can be arrays,
5420 vectors, and structs. Note that subsequent types being indexed into
5421 can never be pointers, since that would require loading the pointer before
5422 continuing calculation.</p>
5424 <p>The type of each index argument depends on the type it is indexing into.
5425 When indexing into a (optionally packed) structure, only <tt>i32</tt>
5426 integer <b>constants</b> are allowed (when using a vector of indices they
5427 must all be the <b>same</b> <tt>i32</tt> integer constant). When indexing
5428 into an array, pointer or vector, integers of any width are allowed, and
5429 they are not required to be constant. These integers are treated as signed
5430 values where relevant.</p>
5432 <p>For example, let's consider a C code fragment and how it gets compiled to
5435 <pre class="doc_code">
5447 int *foo(struct ST *s) {
5448 return &s[1].Z.B[5][13];
5452 <p>The LLVM code generated by Clang is:</p>
5454 <pre class="doc_code">
5455 %struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 }
5456 %struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT }
5458 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5460 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5466 <p>In the example above, the first index is indexing into the
5467 '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a
5468 '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a
5469 structure. The second index indexes into the third element of the structure,
5470 yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>'
5471 type, another structure. The third index indexes into the second element of
5472 the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The
5473 two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>'
5474 type. The '<tt>getelementptr</tt>' instruction returns a pointer to this
5475 element, thus computing a value of '<tt>i32*</tt>' type.</p>
5477 <p>Note that it is perfectly legal to index partially through a structure,
5478 returning a pointer to an inner element. Because of this, the LLVM code for
5479 the given testcase is equivalent to:</p>
5481 <pre class="doc_code">
5482 define i32* @foo(%struct.ST* %s) {
5483 %t1 = getelementptr %struct.ST* %s, i32 1 <i>; yields %struct.ST*:%t1</i>
5484 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 <i>; yields %struct.RT*:%t2</i>
5485 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5486 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5487 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5492 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5493 <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the
5494 base pointer is not an <i>in bounds</i> address of an allocated object,
5495 or if any of the addresses that would be formed by successive addition of
5496 the offsets implied by the indices to the base address with infinitely
5497 precise signed arithmetic are not an <i>in bounds</i> address of that
5498 allocated object. The <i>in bounds</i> addresses for an allocated object
5499 are all the addresses that point into the object, plus the address one
5501 In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword
5502 applies to each of the computations element-wise. </p>
5504 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5505 the base address with silently-wrapping two's complement arithmetic. If the
5506 offsets have a different width from the pointer, they are sign-extended or
5507 truncated to the width of the pointer. The result value of the
5508 <tt>getelementptr</tt> may be outside the object pointed to by the base
5509 pointer. The result value may not necessarily be used to access memory
5510 though, even if it happens to point into allocated storage. See the
5511 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5514 <p>The getelementptr instruction is often confusing. For some more insight into
5515 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5519 <i>; yields [12 x i8]*:aptr</i>
5520 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5521 <i>; yields i8*:vptr</i>
5522 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5523 <i>; yields i8*:eptr</i>
5524 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5525 <i>; yields i32*:iptr</i>
5526 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5529 <p>In cases where the pointer argument is a vector of pointers, each index must
5530 be a vector with the same number of elements. For example: </p>
5531 <pre class="doc_code">
5532 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5539 <!-- ======================================================================= -->
5541 <a name="convertops">Conversion Operations</a>
5546 <p>The instructions in this category are the conversion instructions (casting)
5547 which all take a single operand and a type. They perform various bit
5548 conversions on the operand.</p>
5550 <!-- _______________________________________________________________________ -->
5552 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5559 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5563 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5564 type <tt>ty2</tt>.</p>
5567 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5568 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5569 of the same number of integers.
5570 The bit size of the <tt>value</tt> must be larger than
5571 the bit size of the destination type, <tt>ty2</tt>.
5572 Equal sized types are not allowed.</p>
5575 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5576 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5577 source size must be larger than the destination size, <tt>trunc</tt> cannot
5578 be a <i>no-op cast</i>. It will always truncate bits.</p>
5582 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5583 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5584 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5585 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5590 <!-- _______________________________________________________________________ -->
5592 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5599 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5603 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5608 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5609 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5610 of the same number of integers.
5611 The bit size of the <tt>value</tt> must be smaller than
5612 the bit size of the destination type,
5616 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5617 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5619 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5623 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5624 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5625 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5630 <!-- _______________________________________________________________________ -->
5632 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5639 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5643 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5646 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5647 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5648 of the same number of integers.
5649 The bit size of the <tt>value</tt> must be smaller than
5650 the bit size of the destination type,
5654 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5655 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5656 of the type <tt>ty2</tt>.</p>
5658 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5662 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5663 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5664 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5669 <!-- _______________________________________________________________________ -->
5671 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5678 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5682 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5686 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5687 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5688 to cast it to. The size of <tt>value</tt> must be larger than the size of
5689 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5690 <i>no-op cast</i>.</p>
5693 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5694 <a href="#t_floating">floating point</a> type to a smaller
5695 <a href="#t_floating">floating point</a> type. If the value cannot fit
5696 within the destination type, <tt>ty2</tt>, then the results are
5701 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5702 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5707 <!-- _______________________________________________________________________ -->
5709 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5716 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5720 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5721 floating point value.</p>
5724 <p>The '<tt>fpext</tt>' instruction takes a
5725 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5726 a <a href="#t_floating">floating point</a> type to cast it to. The source
5727 type must be smaller than the destination type.</p>
5730 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5731 <a href="#t_floating">floating point</a> type to a larger
5732 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5733 used to make a <i>no-op cast</i> because it always changes bits. Use
5734 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5738 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5739 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5744 <!-- _______________________________________________________________________ -->
5746 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5753 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5757 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5758 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5761 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5762 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5763 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5764 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5765 vector integer type with the same number of elements as <tt>ty</tt></p>
5768 <p>The '<tt>fptoui</tt>' instruction converts its
5769 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5770 towards zero) unsigned integer value. If the value cannot fit
5771 in <tt>ty2</tt>, the results are undefined.</p>
5775 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5776 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5777 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5782 <!-- _______________________________________________________________________ -->
5784 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5791 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5795 <p>The '<tt>fptosi</tt>' instruction converts
5796 <a href="#t_floating">floating point</a> <tt>value</tt> to
5797 type <tt>ty2</tt>.</p>
5800 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5801 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5802 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5803 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5804 vector integer type with the same number of elements as <tt>ty</tt></p>
5807 <p>The '<tt>fptosi</tt>' instruction converts its
5808 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5809 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5810 the results are undefined.</p>
5814 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5815 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5816 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5821 <!-- _______________________________________________________________________ -->
5823 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5830 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5834 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5835 integer and converts that value to the <tt>ty2</tt> type.</p>
5838 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5839 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5840 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5841 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5842 floating point type with the same number of elements as <tt>ty</tt></p>
5845 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5846 integer quantity and converts it to the corresponding floating point
5847 value. If the value cannot fit in the floating point value, the results are
5852 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5853 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5858 <!-- _______________________________________________________________________ -->
5860 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5867 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5871 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5872 and converts that value to the <tt>ty2</tt> type.</p>
5875 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5876 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5877 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5878 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5879 floating point type with the same number of elements as <tt>ty</tt></p>
5882 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5883 quantity and converts it to the corresponding floating point value. If the
5884 value cannot fit in the floating point value, the results are undefined.</p>
5888 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5889 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5894 <!-- _______________________________________________________________________ -->
5896 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5903 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5907 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of
5908 pointers <tt>value</tt> to
5909 the integer (or vector of integers) type <tt>ty2</tt>.</p>
5912 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5913 must be a a value of type <a href="#t_pointer">pointer</a> or a vector of
5914 pointers, and a type to cast it to
5915 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector
5916 of integers type.</p>
5919 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5920 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5921 truncating or zero extending that value to the size of the integer type. If
5922 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5923 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5924 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5929 %X = ptrtoint i32* %P to i8 <i>; yields truncation on 32-bit architecture</i>
5930 %Y = ptrtoint i32* %P to i64 <i>; yields zero extension on 32-bit architecture</i>
5931 %Z = ptrtoint <4 x i32*> %P to <4 x i64><i>; yields vector zero extension for a vector of addresses on 32-bit architecture</i>
5936 <!-- _______________________________________________________________________ -->
5938 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5945 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5949 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5950 pointer type, <tt>ty2</tt>.</p>
5953 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5954 value to cast, and a type to cast it to, which must be a
5955 <a href="#t_pointer">pointer</a> type.</p>
5958 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5959 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5960 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5961 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5962 than the size of a pointer then a zero extension is done. If they are the
5963 same size, nothing is done (<i>no-op cast</i>).</p>
5967 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5968 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5969 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5970 %Z = inttoptr <4 x i32> %G to <4 x i8*><i>; yields truncation of vector G to four pointers</i>
5975 <!-- _______________________________________________________________________ -->
5977 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5984 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5988 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5989 <tt>ty2</tt> without changing any bits.</p>
5992 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5993 non-aggregate first class value, and a type to cast it to, which must also be
5994 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5995 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5996 identical. If the source type is a pointer, the destination type must also be
5997 a pointer. This instruction supports bitwise conversion of vectors to
5998 integers and to vectors of other types (as long as they have the same
6002 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
6003 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
6004 this conversion. The conversion is done as if the <tt>value</tt> had been
6005 stored to memory and read back as type <tt>ty2</tt>.
6006 Pointer (or vector of pointers) types may only be converted to other pointer
6007 (or vector of pointers) types with this instruction. To convert
6008 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
6009 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
6013 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
6014 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
6015 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
6016 %Z = bitcast <2 x i32*> %V to <2 x i64*> <i>; yields <2 x i64*></i>
6023 <!-- ======================================================================= -->
6025 <a name="otherops">Other Operations</a>
6030 <p>The instructions in this category are the "miscellaneous" instructions, which
6031 defy better classification.</p>
6033 <!-- _______________________________________________________________________ -->
6035 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
6042 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
6046 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
6047 boolean values based on comparison of its two integer, integer vector,
6048 pointer, or pointer vector operands.</p>
6051 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
6052 the condition code indicating the kind of comparison to perform. It is not a
6053 value, just a keyword. The possible condition code are:</p>
6056 <li><tt>eq</tt>: equal</li>
6057 <li><tt>ne</tt>: not equal </li>
6058 <li><tt>ugt</tt>: unsigned greater than</li>
6059 <li><tt>uge</tt>: unsigned greater or equal</li>
6060 <li><tt>ult</tt>: unsigned less than</li>
6061 <li><tt>ule</tt>: unsigned less or equal</li>
6062 <li><tt>sgt</tt>: signed greater than</li>
6063 <li><tt>sge</tt>: signed greater or equal</li>
6064 <li><tt>slt</tt>: signed less than</li>
6065 <li><tt>sle</tt>: signed less or equal</li>
6068 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
6069 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
6070 typed. They must also be identical types.</p>
6073 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
6074 condition code given as <tt>cond</tt>. The comparison performed always yields
6075 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
6076 result, as follows:</p>
6079 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
6080 <tt>false</tt> otherwise. No sign interpretation is necessary or
6083 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
6084 <tt>false</tt> otherwise. No sign interpretation is necessary or
6087 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
6088 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6090 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
6091 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6092 to <tt>op2</tt>.</li>
6094 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
6095 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6097 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
6098 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6100 <li><tt>sgt</tt>: interprets the operands as signed values and yields
6101 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6103 <li><tt>sge</tt>: interprets the operands as signed values and yields
6104 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6105 to <tt>op2</tt>.</li>
6107 <li><tt>slt</tt>: interprets the operands as signed values and yields
6108 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6110 <li><tt>sle</tt>: interprets the operands as signed values and yields
6111 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6114 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
6115 values are compared as if they were integers.</p>
6117 <p>If the operands are integer vectors, then they are compared element by
6118 element. The result is an <tt>i1</tt> vector with the same number of elements
6119 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
6123 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
6124 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
6125 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
6126 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
6127 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
6128 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
6131 <p>Note that the code generator does not yet support vector types with
6132 the <tt>icmp</tt> instruction.</p>
6136 <!-- _______________________________________________________________________ -->
6138 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
6145 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
6149 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
6150 values based on comparison of its operands.</p>
6152 <p>If the operands are floating point scalars, then the result type is a boolean
6153 (<a href="#t_integer"><tt>i1</tt></a>).</p>
6155 <p>If the operands are floating point vectors, then the result type is a vector
6156 of boolean with the same number of elements as the operands being
6160 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
6161 the condition code indicating the kind of comparison to perform. It is not a
6162 value, just a keyword. The possible condition code are:</p>
6165 <li><tt>false</tt>: no comparison, always returns false</li>
6166 <li><tt>oeq</tt>: ordered and equal</li>
6167 <li><tt>ogt</tt>: ordered and greater than </li>
6168 <li><tt>oge</tt>: ordered and greater than or equal</li>
6169 <li><tt>olt</tt>: ordered and less than </li>
6170 <li><tt>ole</tt>: ordered and less than or equal</li>
6171 <li><tt>one</tt>: ordered and not equal</li>
6172 <li><tt>ord</tt>: ordered (no nans)</li>
6173 <li><tt>ueq</tt>: unordered or equal</li>
6174 <li><tt>ugt</tt>: unordered or greater than </li>
6175 <li><tt>uge</tt>: unordered or greater than or equal</li>
6176 <li><tt>ult</tt>: unordered or less than </li>
6177 <li><tt>ule</tt>: unordered or less than or equal</li>
6178 <li><tt>une</tt>: unordered or not equal</li>
6179 <li><tt>uno</tt>: unordered (either nans)</li>
6180 <li><tt>true</tt>: no comparison, always returns true</li>
6183 <p><i>Ordered</i> means that neither operand is a QNAN while
6184 <i>unordered</i> means that either operand may be a QNAN.</p>
6186 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
6187 a <a href="#t_floating">floating point</a> type or
6188 a <a href="#t_vector">vector</a> of floating point type. They must have
6189 identical types.</p>
6192 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
6193 according to the condition code given as <tt>cond</tt>. If the operands are
6194 vectors, then the vectors are compared element by element. Each comparison
6195 performed always yields an <a href="#t_integer">i1</a> result, as
6199 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
6201 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6202 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6204 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6205 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6207 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6208 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6210 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6211 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6213 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6214 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6216 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6217 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6219 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
6221 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
6222 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6224 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
6225 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6227 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
6228 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6230 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
6231 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6233 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
6234 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6236 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
6237 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6239 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
6241 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
6246 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
6247 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
6248 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
6249 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
6252 <p>Note that the code generator does not yet support vector types with
6253 the <tt>fcmp</tt> instruction.</p>
6257 <!-- _______________________________________________________________________ -->
6259 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
6266 <result> = phi <ty> [ <val0>, <label0>], ...
6270 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
6271 SSA graph representing the function.</p>
6274 <p>The type of the incoming values is specified with the first type field. After
6275 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
6276 one pair for each predecessor basic block of the current block. Only values
6277 of <a href="#t_firstclass">first class</a> type may be used as the value
6278 arguments to the PHI node. Only labels may be used as the label
6281 <p>There must be no non-phi instructions between the start of a basic block and
6282 the PHI instructions: i.e. PHI instructions must be first in a basic
6285 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
6286 occur on the edge from the corresponding predecessor block to the current
6287 block (but after any definition of an '<tt>invoke</tt>' instruction's return
6288 value on the same edge).</p>
6291 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
6292 specified by the pair corresponding to the predecessor basic block that
6293 executed just prior to the current block.</p>
6297 Loop: ; Infinite loop that counts from 0 on up...
6298 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6299 %nextindvar = add i32 %indvar, 1
6305 <!-- _______________________________________________________________________ -->
6307 <a name="i_select">'<tt>select</tt>' Instruction</a>
6314 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
6316 <i>selty</i> is either i1 or {<N x i1>}
6320 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
6321 condition, without branching.</p>
6325 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
6326 values indicating the condition, and two values of the
6327 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
6328 vectors and the condition is a scalar, then entire vectors are selected, not
6329 individual elements.</p>
6332 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
6333 first value argument; otherwise, it returns the second value argument.</p>
6335 <p>If the condition is a vector of i1, then the value arguments must be vectors
6336 of the same size, and the selection is done element by element.</p>
6340 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
6345 <!-- _______________________________________________________________________ -->
6347 <a name="i_call">'<tt>call</tt>' Instruction</a>
6354 <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>]
6358 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
6361 <p>This instruction requires several arguments:</p>
6364 <li>The optional "tail" marker indicates that the callee function does not
6365 access any allocas or varargs in the caller. Note that calls may be
6366 marked "tail" even if they do not occur before
6367 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
6368 present, the function call is eligible for tail call optimization,
6369 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
6370 optimized into a jump</a>. The code generator may optimize calls marked
6371 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
6372 sibling call optimization</a> when the caller and callee have
6373 matching signatures, or 2) forced tail call optimization when the
6374 following extra requirements are met:
6376 <li>Caller and callee both have the calling
6377 convention <tt>fastcc</tt>.</li>
6378 <li>The call is in tail position (ret immediately follows call and ret
6379 uses value of call or is void).</li>
6380 <li>Option <tt>-tailcallopt</tt> is enabled,
6381 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
6382 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
6383 constraints are met.</a></li>
6387 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
6388 convention</a> the call should use. If none is specified, the call
6389 defaults to using C calling conventions. The calling convention of the
6390 call must match the calling convention of the target function, or else the
6391 behavior is undefined.</li>
6393 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
6394 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
6395 '<tt>inreg</tt>' attributes are valid here.</li>
6397 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
6398 type of the return value. Functions that return no value are marked
6399 <tt><a href="#t_void">void</a></tt>.</li>
6401 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
6402 being invoked. The argument types must match the types implied by this
6403 signature. This type can be omitted if the function is not varargs and if
6404 the function type does not return a pointer to a function.</li>
6406 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
6407 be invoked. In most cases, this is a direct function invocation, but
6408 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
6409 to function value.</li>
6411 <li>'<tt>function args</tt>': argument list whose types match the function
6412 signature argument types and parameter attributes. All arguments must be
6413 of <a href="#t_firstclass">first class</a> type. If the function
6414 signature indicates the function accepts a variable number of arguments,
6415 the extra arguments can be specified.</li>
6417 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
6418 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
6419 '<tt>readnone</tt>' attributes are valid here.</li>
6423 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
6424 a specified function, with its incoming arguments bound to the specified
6425 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
6426 function, control flow continues with the instruction after the function
6427 call, and the return value of the function is bound to the result
6432 %retval = call i32 @test(i32 %argc)
6433 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
6434 %X = tail call i32 @foo() <i>; yields i32</i>
6435 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
6436 call void %foo(i8 97 signext)
6438 %struct.A = type { i32, i8 }
6439 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
6440 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
6441 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
6442 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
6443 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
6446 <p>llvm treats calls to some functions with names and arguments that match the
6447 standard C99 library as being the C99 library functions, and may perform
6448 optimizations or generate code for them under that assumption. This is
6449 something we'd like to change in the future to provide better support for
6450 freestanding environments and non-C-based languages.</p>
6454 <!-- _______________________________________________________________________ -->
6456 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
6463 <resultval> = va_arg <va_list*> <arglist>, <argty>
6467 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6468 the "variable argument" area of a function call. It is used to implement the
6469 <tt>va_arg</tt> macro in C.</p>
6472 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6473 argument. It returns a value of the specified argument type and increments
6474 the <tt>va_list</tt> to point to the next argument. The actual type
6475 of <tt>va_list</tt> is target specific.</p>
6478 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6479 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6480 to the next argument. For more information, see the variable argument
6481 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6483 <p>It is legal for this instruction to be called in a function which does not
6484 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6487 <p><tt>va_arg</tt> is an LLVM instruction instead of
6488 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6492 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6494 <p>Note that the code generator does not yet fully support va_arg on many
6495 targets. Also, it does not currently support va_arg with aggregate types on
6500 <!-- _______________________________________________________________________ -->
6502 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6509 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6510 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6512 <clause> := catch <type> <value>
6513 <clause> := filter <array constant type> <array constant>
6517 <p>The '<tt>landingpad</tt>' instruction is used by
6518 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6519 system</a> to specify that a basic block is a landing pad — one where
6520 the exception lands, and corresponds to the code found in the
6521 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6522 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6523 re-entry to the function. The <tt>resultval</tt> has the
6524 type <tt>resultty</tt>.</p>
6527 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6528 function associated with the unwinding mechanism. The optional
6529 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6531 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6532 or <tt>filter</tt> — and contains the global variable representing the
6533 "type" that may be caught or filtered respectively. Unlike the
6534 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6535 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6536 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6537 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6540 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6541 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6542 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6543 calling conventions, how the personality function results are represented in
6544 LLVM IR is target specific.</p>
6546 <p>The clauses are applied in order from top to bottom. If two
6547 <tt>landingpad</tt> instructions are merged together through inlining, the
6548 clauses from the calling function are appended to the list of clauses.
6549 When the call stack is being unwound due to an exception being thrown, the
6550 exception is compared against each <tt>clause</tt> in turn. If it doesn't
6551 match any of the clauses, and the <tt>cleanup</tt> flag is not set, then
6552 unwinding continues further up the call stack.</p>
6554 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6557 <li>A landing pad block is a basic block which is the unwind destination of an
6558 '<tt>invoke</tt>' instruction.</li>
6559 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6560 first non-PHI instruction.</li>
6561 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6563 <li>A basic block that is not a landing pad block may not include a
6564 '<tt>landingpad</tt>' instruction.</li>
6565 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6566 personality function.</li>
6571 ;; A landing pad which can catch an integer.
6572 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6574 ;; A landing pad that is a cleanup.
6575 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6577 ;; A landing pad which can catch an integer and can only throw a double.
6578 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6580 filter [1 x i8**] [@_ZTId]
6589 <!-- *********************************************************************** -->
6590 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6591 <!-- *********************************************************************** -->
6595 <p>LLVM supports the notion of an "intrinsic function". These functions have
6596 well known names and semantics and are required to follow certain
6597 restrictions. Overall, these intrinsics represent an extension mechanism for
6598 the LLVM language that does not require changing all of the transformations
6599 in LLVM when adding to the language (or the bitcode reader/writer, the
6600 parser, etc...).</p>
6602 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6603 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6604 begin with this prefix. Intrinsic functions must always be external
6605 functions: you cannot define the body of intrinsic functions. Intrinsic
6606 functions may only be used in call or invoke instructions: it is illegal to
6607 take the address of an intrinsic function. Additionally, because intrinsic
6608 functions are part of the LLVM language, it is required if any are added that
6609 they be documented here.</p>
6611 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6612 family of functions that perform the same operation but on different data
6613 types. Because LLVM can represent over 8 million different integer types,
6614 overloading is used commonly to allow an intrinsic function to operate on any
6615 integer type. One or more of the argument types or the result type can be
6616 overloaded to accept any integer type. Argument types may also be defined as
6617 exactly matching a previous argument's type or the result type. This allows
6618 an intrinsic function which accepts multiple arguments, but needs all of them
6619 to be of the same type, to only be overloaded with respect to a single
6620 argument or the result.</p>
6622 <p>Overloaded intrinsics will have the names of its overloaded argument types
6623 encoded into its function name, each preceded by a period. Only those types
6624 which are overloaded result in a name suffix. Arguments whose type is matched
6625 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6626 can take an integer of any width and returns an integer of exactly the same
6627 integer width. This leads to a family of functions such as
6628 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6629 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6630 suffix is required. Because the argument's type is matched against the return
6631 type, it does not require its own name suffix.</p>
6633 <p>To learn how to add an intrinsic function, please see the
6634 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6636 <!-- ======================================================================= -->
6638 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6643 <p>Variable argument support is defined in LLVM with
6644 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6645 intrinsic functions. These functions are related to the similarly named
6646 macros defined in the <tt><stdarg.h></tt> header file.</p>
6648 <p>All of these functions operate on arguments that use a target-specific value
6649 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6650 not define what this type is, so all transformations should be prepared to
6651 handle these functions regardless of the type used.</p>
6653 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6654 instruction and the variable argument handling intrinsic functions are
6657 <pre class="doc_code">
6658 define i32 @test(i32 %X, ...) {
6659 ; Initialize variable argument processing
6661 %ap2 = bitcast i8** %ap to i8*
6662 call void @llvm.va_start(i8* %ap2)
6664 ; Read a single integer argument
6665 %tmp = va_arg i8** %ap, i32
6667 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6669 %aq2 = bitcast i8** %aq to i8*
6670 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6671 call void @llvm.va_end(i8* %aq2)
6673 ; Stop processing of arguments.
6674 call void @llvm.va_end(i8* %ap2)
6678 declare void @llvm.va_start(i8*)
6679 declare void @llvm.va_copy(i8*, i8*)
6680 declare void @llvm.va_end(i8*)
6683 <!-- _______________________________________________________________________ -->
6685 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6693 declare void %llvm.va_start(i8* <arglist>)
6697 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6698 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6701 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6704 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6705 macro available in C. In a target-dependent way, it initializes
6706 the <tt>va_list</tt> element to which the argument points, so that the next
6707 call to <tt>va_arg</tt> will produce the first variable argument passed to
6708 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6709 need to know the last argument of the function as the compiler can figure
6714 <!-- _______________________________________________________________________ -->
6716 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6723 declare void @llvm.va_end(i8* <arglist>)
6727 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6728 which has been initialized previously
6729 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6730 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6733 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6736 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6737 macro available in C. In a target-dependent way, it destroys
6738 the <tt>va_list</tt> element to which the argument points. Calls
6739 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6740 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6741 with calls to <tt>llvm.va_end</tt>.</p>
6745 <!-- _______________________________________________________________________ -->
6747 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6754 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6758 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6759 from the source argument list to the destination argument list.</p>
6762 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6763 The second argument is a pointer to a <tt>va_list</tt> element to copy
6767 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6768 macro available in C. In a target-dependent way, it copies the
6769 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6770 element. This intrinsic is necessary because
6771 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6772 arbitrarily complex and require, for example, memory allocation.</p>
6778 <!-- ======================================================================= -->
6780 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6785 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6786 Collection</a> (GC) requires the implementation and generation of these
6787 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6788 roots on the stack</a>, as well as garbage collector implementations that
6789 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6790 barriers. Front-ends for type-safe garbage collected languages should generate
6791 these intrinsics to make use of the LLVM garbage collectors. For more details,
6792 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6795 <p>The garbage collection intrinsics only operate on objects in the generic
6796 address space (address space zero).</p>
6798 <!-- _______________________________________________________________________ -->
6800 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6807 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6811 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6812 the code generator, and allows some metadata to be associated with it.</p>
6815 <p>The first argument specifies the address of a stack object that contains the
6816 root pointer. The second pointer (which must be either a constant or a
6817 global value address) contains the meta-data to be associated with the
6821 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6822 location. At compile-time, the code generator generates information to allow
6823 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6824 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6829 <!-- _______________________________________________________________________ -->
6831 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6838 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6842 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6843 locations, allowing garbage collector implementations that require read
6847 <p>The second argument is the address to read from, which should be an address
6848 allocated from the garbage collector. The first object is a pointer to the
6849 start of the referenced object, if needed by the language runtime (otherwise
6853 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6854 instruction, but may be replaced with substantially more complex code by the
6855 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6856 may only be used in a function which <a href="#gc">specifies a GC
6861 <!-- _______________________________________________________________________ -->
6863 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6870 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6874 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6875 locations, allowing garbage collector implementations that require write
6876 barriers (such as generational or reference counting collectors).</p>
6879 <p>The first argument is the reference to store, the second is the start of the
6880 object to store it to, and the third is the address of the field of Obj to
6881 store to. If the runtime does not require a pointer to the object, Obj may
6885 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6886 instruction, but may be replaced with substantially more complex code by the
6887 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6888 may only be used in a function which <a href="#gc">specifies a GC
6895 <!-- ======================================================================= -->
6897 <a name="int_codegen">Code Generator Intrinsics</a>
6902 <p>These intrinsics are provided by LLVM to expose special features that may
6903 only be implemented with code generator support.</p>
6905 <!-- _______________________________________________________________________ -->
6907 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6914 declare i8 *@llvm.returnaddress(i32 <level>)
6918 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6919 target-specific value indicating the return address of the current function
6920 or one of its callers.</p>
6923 <p>The argument to this intrinsic indicates which function to return the address
6924 for. Zero indicates the calling function, one indicates its caller, etc.
6925 The argument is <b>required</b> to be a constant integer value.</p>
6928 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6929 indicating the return address of the specified call frame, or zero if it
6930 cannot be identified. The value returned by this intrinsic is likely to be
6931 incorrect or 0 for arguments other than zero, so it should only be used for
6932 debugging purposes.</p>
6934 <p>Note that calling this intrinsic does not prevent function inlining or other
6935 aggressive transformations, so the value returned may not be that of the
6936 obvious source-language caller.</p>
6940 <!-- _______________________________________________________________________ -->
6942 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6949 declare i8* @llvm.frameaddress(i32 <level>)
6953 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6954 target-specific frame pointer value for the specified stack frame.</p>
6957 <p>The argument to this intrinsic indicates which function to return the frame
6958 pointer for. Zero indicates the calling function, one indicates its caller,
6959 etc. The argument is <b>required</b> to be a constant integer value.</p>
6962 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6963 indicating the frame address of the specified call frame, or zero if it
6964 cannot be identified. The value returned by this intrinsic is likely to be
6965 incorrect or 0 for arguments other than zero, so it should only be used for
6966 debugging purposes.</p>
6968 <p>Note that calling this intrinsic does not prevent function inlining or other
6969 aggressive transformations, so the value returned may not be that of the
6970 obvious source-language caller.</p>
6974 <!-- _______________________________________________________________________ -->
6976 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6983 declare i8* @llvm.stacksave()
6987 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6988 of the function stack, for use
6989 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6990 useful for implementing language features like scoped automatic variable
6991 sized arrays in C99.</p>
6994 <p>This intrinsic returns a opaque pointer value that can be passed
6995 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6996 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6997 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6998 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6999 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
7000 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
7004 <!-- _______________________________________________________________________ -->
7006 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
7013 declare void @llvm.stackrestore(i8* %ptr)
7017 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
7018 the function stack to the state it was in when the
7019 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
7020 executed. This is useful for implementing language features like scoped
7021 automatic variable sized arrays in C99.</p>
7024 <p>See the description
7025 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
7029 <!-- _______________________________________________________________________ -->
7031 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
7038 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7042 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
7043 insert a prefetch instruction if supported; otherwise, it is a noop.
7044 Prefetches have no effect on the behavior of the program but can change its
7045 performance characteristics.</p>
7048 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
7049 specifier determining if the fetch should be for a read (0) or write (1),
7050 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
7051 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
7052 specifies whether the prefetch is performed on the data (1) or instruction (0)
7053 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
7054 must be constant integers.</p>
7057 <p>This intrinsic does not modify the behavior of the program. In particular,
7058 prefetches cannot trap and do not produce a value. On targets that support
7059 this intrinsic, the prefetch can provide hints to the processor cache for
7060 better performance.</p>
7064 <!-- _______________________________________________________________________ -->
7066 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
7073 declare void @llvm.pcmarker(i32 <id>)
7077 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
7078 Counter (PC) in a region of code to simulators and other tools. The method
7079 is target specific, but it is expected that the marker will use exported
7080 symbols to transmit the PC of the marker. The marker makes no guarantees
7081 that it will remain with any specific instruction after optimizations. It is
7082 possible that the presence of a marker will inhibit optimizations. The
7083 intended use is to be inserted after optimizations to allow correlations of
7084 simulation runs.</p>
7087 <p><tt>id</tt> is a numerical id identifying the marker.</p>
7090 <p>This intrinsic does not modify the behavior of the program. Backends that do
7091 not support this intrinsic may ignore it.</p>
7095 <!-- _______________________________________________________________________ -->
7097 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
7104 declare i64 @llvm.readcyclecounter()
7108 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
7109 counter register (or similar low latency, high accuracy clocks) on those
7110 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7111 should map to RPCC. As the backing counters overflow quickly (on the order
7112 of 9 seconds on alpha), this should only be used for small timings.</p>
7115 <p>When directly supported, reading the cycle counter should not modify any
7116 memory. Implementations are allowed to either return a application specific
7117 value or a system wide value. On backends without support, this is lowered
7118 to a constant 0.</p>
7124 <!-- ======================================================================= -->
7126 <a name="int_libc">Standard C Library Intrinsics</a>
7131 <p>LLVM provides intrinsics for a few important standard C library functions.
7132 These intrinsics allow source-language front-ends to pass information about
7133 the alignment of the pointer arguments to the code generator, providing
7134 opportunity for more efficient code generation.</p>
7136 <!-- _______________________________________________________________________ -->
7138 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
7144 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
7145 integer bit width and for different address spaces. Not all targets support
7146 all bit widths however.</p>
7149 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7150 i32 <len>, i32 <align>, i1 <isvolatile>)
7151 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7152 i64 <len>, i32 <align>, i1 <isvolatile>)
7156 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7157 source location to the destination location.</p>
7159 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
7160 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7161 and the pointers can be in specified address spaces.</p>
7165 <p>The first argument is a pointer to the destination, the second is a pointer
7166 to the source. The third argument is an integer argument specifying the
7167 number of bytes to copy, the fourth argument is the alignment of the
7168 source and destination locations, and the fifth is a boolean indicating a
7169 volatile access.</p>
7171 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7172 then the caller guarantees that both the source and destination pointers are
7173 aligned to that boundary.</p>
7175 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7176 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
7177 The detailed access behavior is not very cleanly specified and it is unwise
7178 to depend on it.</p>
7182 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7183 source location to the destination location, which are not allowed to
7184 overlap. It copies "len" bytes of memory over. If the argument is known to
7185 be aligned to some boundary, this can be specified as the fourth argument,
7186 otherwise it should be set to 0 or 1.</p>
7190 <!-- _______________________________________________________________________ -->
7192 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
7198 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
7199 width and for different address space. Not all targets support all bit
7203 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7204 i32 <len>, i32 <align>, i1 <isvolatile>)
7205 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7206 i64 <len>, i32 <align>, i1 <isvolatile>)
7210 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
7211 source location to the destination location. It is similar to the
7212 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
7215 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
7216 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7217 and the pointers can be in specified address spaces.</p>
7221 <p>The first argument is a pointer to the destination, the second is a pointer
7222 to the source. The third argument is an integer argument specifying the
7223 number of bytes to copy, the fourth argument is the alignment of the
7224 source and destination locations, and the fifth is a boolean indicating a
7225 volatile access.</p>
7227 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7228 then the caller guarantees that the source and destination pointers are
7229 aligned to that boundary.</p>
7231 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7232 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
7233 The detailed access behavior is not very cleanly specified and it is unwise
7234 to depend on it.</p>
7238 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
7239 source location to the destination location, which may overlap. It copies
7240 "len" bytes of memory over. If the argument is known to be aligned to some
7241 boundary, this can be specified as the fourth argument, otherwise it should
7242 be set to 0 or 1.</p>
7246 <!-- _______________________________________________________________________ -->
7248 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
7254 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
7255 width and for different address spaces. However, not all targets support all
7259 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7260 i32 <len>, i32 <align>, i1 <isvolatile>)
7261 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7262 i64 <len>, i32 <align>, i1 <isvolatile>)
7266 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
7267 particular byte value.</p>
7269 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
7270 intrinsic does not return a value and takes extra alignment/volatile
7271 arguments. Also, the destination can be in an arbitrary address space.</p>
7274 <p>The first argument is a pointer to the destination to fill, the second is the
7275 byte value with which to fill it, the third argument is an integer argument
7276 specifying the number of bytes to fill, and the fourth argument is the known
7277 alignment of the destination location.</p>
7279 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7280 then the caller guarantees that the destination pointer is aligned to that
7283 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7284 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
7285 The detailed access behavior is not very cleanly specified and it is unwise
7286 to depend on it.</p>
7289 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
7290 at the destination location. If the argument is known to be aligned to some
7291 boundary, this can be specified as the fourth argument, otherwise it should
7292 be set to 0 or 1.</p>
7296 <!-- _______________________________________________________________________ -->
7298 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
7304 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
7305 floating point or vector of floating point type. Not all targets support all
7309 declare float @llvm.sqrt.f32(float %Val)
7310 declare double @llvm.sqrt.f64(double %Val)
7311 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7312 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7313 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7317 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
7318 returning the same value as the libm '<tt>sqrt</tt>' functions would.
7319 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
7320 behavior for negative numbers other than -0.0 (which allows for better
7321 optimization, because there is no need to worry about errno being
7322 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
7325 <p>The argument and return value are floating point numbers of the same
7329 <p>This function returns the sqrt of the specified operand if it is a
7330 nonnegative floating point number.</p>
7334 <!-- _______________________________________________________________________ -->
7336 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
7342 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
7343 floating point or vector of floating point type. Not all targets support all
7347 declare float @llvm.powi.f32(float %Val, i32 %power)
7348 declare double @llvm.powi.f64(double %Val, i32 %power)
7349 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7350 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7351 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7355 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
7356 specified (positive or negative) power. The order of evaluation of
7357 multiplications is not defined. When a vector of floating point type is
7358 used, the second argument remains a scalar integer value.</p>
7361 <p>The second argument is an integer power, and the first is a value to raise to
7365 <p>This function returns the first value raised to the second power with an
7366 unspecified sequence of rounding operations.</p>
7370 <!-- _______________________________________________________________________ -->
7372 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
7378 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
7379 floating point or vector of floating point type. Not all targets support all
7383 declare float @llvm.sin.f32(float %Val)
7384 declare double @llvm.sin.f64(double %Val)
7385 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7386 declare fp128 @llvm.sin.f128(fp128 %Val)
7387 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7391 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
7394 <p>The argument and return value are floating point numbers of the same
7398 <p>This function returns the sine of the specified operand, returning the same
7399 values as the libm <tt>sin</tt> functions would, and handles error conditions
7400 in the same way.</p>
7404 <!-- _______________________________________________________________________ -->
7406 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
7412 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
7413 floating point or vector of floating point type. Not all targets support all
7417 declare float @llvm.cos.f32(float %Val)
7418 declare double @llvm.cos.f64(double %Val)
7419 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7420 declare fp128 @llvm.cos.f128(fp128 %Val)
7421 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7425 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
7428 <p>The argument and return value are floating point numbers of the same
7432 <p>This function returns the cosine of the specified operand, returning the same
7433 values as the libm <tt>cos</tt> functions would, and handles error conditions
7434 in the same way.</p>
7438 <!-- _______________________________________________________________________ -->
7440 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
7446 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
7447 floating point or vector of floating point type. Not all targets support all
7451 declare float @llvm.pow.f32(float %Val, float %Power)
7452 declare double @llvm.pow.f64(double %Val, double %Power)
7453 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7454 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7455 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7459 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
7460 specified (positive or negative) power.</p>
7463 <p>The second argument is a floating point power, and the first is a value to
7464 raise to that power.</p>
7467 <p>This function returns the first value raised to the second power, returning
7468 the same values as the libm <tt>pow</tt> functions would, and handles error
7469 conditions in the same way.</p>
7473 <!-- _______________________________________________________________________ -->
7475 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7481 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7482 floating point or vector of floating point type. Not all targets support all
7486 declare float @llvm.exp.f32(float %Val)
7487 declare double @llvm.exp.f64(double %Val)
7488 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7489 declare fp128 @llvm.exp.f128(fp128 %Val)
7490 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7494 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7497 <p>The argument and return value are floating point numbers of the same
7501 <p>This function returns the same values as the libm <tt>exp</tt> functions
7502 would, and handles error conditions in the same way.</p>
7506 <!-- _______________________________________________________________________ -->
7508 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7514 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7515 floating point or vector of floating point type. Not all targets support all
7519 declare float @llvm.log.f32(float %Val)
7520 declare double @llvm.log.f64(double %Val)
7521 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7522 declare fp128 @llvm.log.f128(fp128 %Val)
7523 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7527 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7530 <p>The argument and return value are floating point numbers of the same
7534 <p>This function returns the same values as the libm <tt>log</tt> functions
7535 would, and handles error conditions in the same way.</p>
7539 <!-- _______________________________________________________________________ -->
7541 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7547 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7548 floating point or vector of floating point type. Not all targets support all
7552 declare float @llvm.fma.f32(float %a, float %b, float %c)
7553 declare double @llvm.fma.f64(double %a, double %b, double %c)
7554 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7555 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7556 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7560 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7564 <p>The argument and return value are floating point numbers of the same
7568 <p>This function returns the same values as the libm <tt>fma</tt> functions
7573 <!-- _______________________________________________________________________ -->
7575 <a name="int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a>
7581 <p>This is an overloaded intrinsic. You can use <tt>llvm.fabs</tt> on any
7582 floating point or vector of floating point type. Not all targets support all
7586 declare float @llvm.fabs.f32(float %Val)
7587 declare double @llvm.fabs.f64(double %Val)
7588 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7589 declare fp128 @llvm.fabs.f128(fp128 %Val)
7590 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7594 <p>The '<tt>llvm.fabs.*</tt>' intrinsics return the absolute value of
7598 <p>The argument and return value are floating point numbers of the same
7602 <p>This function returns the same values as the libm <tt>fabs</tt> functions
7603 would, and handles error conditions in the same way.</p>
7607 <!-- _______________________________________________________________________ -->
7609 <a name="int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a>
7615 <p>This is an overloaded intrinsic. You can use <tt>llvm.floor</tt> on any
7616 floating point or vector of floating point type. Not all targets support all
7620 declare float @llvm.floor.f32(float %Val)
7621 declare double @llvm.floor.f64(double %Val)
7622 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7623 declare fp128 @llvm.floor.f128(fp128 %Val)
7624 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7628 <p>The '<tt>llvm.floor.*</tt>' intrinsics return the floor of
7632 <p>The argument and return value are floating point numbers of the same
7636 <p>This function returns the same values as the libm <tt>floor</tt> functions
7637 would, and handles error conditions in the same way.</p>
7641 <!-- _______________________________________________________________________ -->
7643 <a name="int_ceil">'<tt>llvm.ceil.*</tt>' Intrinsic</a>
7649 <p>This is an overloaded intrinsic. You can use <tt>llvm.ceil</tt> on any
7650 floating point or vector of floating point type. Not all targets support all
7654 declare float @llvm.ceil.f32(float %Val)
7655 declare double @llvm.ceil.f64(double %Val)
7656 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7657 declare fp128 @llvm.ceil.f128(fp128 %Val)
7658 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7662 <p>The '<tt>llvm.ceil.*</tt>' intrinsics return the ceiling of
7666 <p>The argument and return value are floating point numbers of the same
7670 <p>This function returns the same values as the libm <tt>ceil</tt> functions
7671 would, and handles error conditions in the same way.</p>
7675 <!-- _______________________________________________________________________ -->
7677 <a name="int_trunc">'<tt>llvm.trunc.*</tt>' Intrinsic</a>
7683 <p>This is an overloaded intrinsic. You can use <tt>llvm.trunc</tt> on any
7684 floating point or vector of floating point type. Not all targets support all
7688 declare float @llvm.trunc.f32(float %Val)
7689 declare double @llvm.trunc.f64(double %Val)
7690 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7691 declare fp128 @llvm.trunc.f128(fp128 %Val)
7692 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7696 <p>The '<tt>llvm.trunc.*</tt>' intrinsics returns the operand rounded to the
7697 nearest integer not larger in magnitude than the operand.</p>
7700 <p>The argument and return value are floating point numbers of the same
7704 <p>This function returns the same values as the libm <tt>trunc</tt> functions
7705 would, and handles error conditions in the same way.</p>
7709 <!-- _______________________________________________________________________ -->
7711 <a name="int_rint">'<tt>llvm.rint.*</tt>' Intrinsic</a>
7717 <p>This is an overloaded intrinsic. You can use <tt>llvm.rint</tt> on any
7718 floating point or vector of floating point type. Not all targets support all
7722 declare float @llvm.rint.f32(float %Val)
7723 declare double @llvm.rint.f64(double %Val)
7724 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7725 declare fp128 @llvm.rint.f128(fp128 %Val)
7726 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7730 <p>The '<tt>llvm.rint.*</tt>' intrinsics returns the operand rounded to the
7731 nearest integer. It may raise an inexact floating-point exception if the
7732 operand isn't an integer.</p>
7735 <p>The argument and return value are floating point numbers of the same
7739 <p>This function returns the same values as the libm <tt>rint</tt> functions
7740 would, and handles error conditions in the same way.</p>
7744 <!-- _______________________________________________________________________ -->
7746 <a name="int_nearbyint">'<tt>llvm.nearbyint.*</tt>' Intrinsic</a>
7752 <p>This is an overloaded intrinsic. You can use <tt>llvm.nearbyint</tt> on any
7753 floating point or vector of floating point type. Not all targets support all
7757 declare float @llvm.nearbyint.f32(float %Val)
7758 declare double @llvm.nearbyint.f64(double %Val)
7759 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7760 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7761 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7765 <p>The '<tt>llvm.nearbyint.*</tt>' intrinsics returns the operand rounded to the
7766 nearest integer.</p>
7769 <p>The argument and return value are floating point numbers of the same
7773 <p>This function returns the same values as the libm <tt>nearbyint</tt>
7774 functions would, and handles error conditions in the same way.</p>
7780 <!-- ======================================================================= -->
7782 <a name="int_manip">Bit Manipulation Intrinsics</a>
7787 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7788 These allow efficient code generation for some algorithms.</p>
7790 <!-- _______________________________________________________________________ -->
7792 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7798 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7799 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7802 declare i16 @llvm.bswap.i16(i16 <id>)
7803 declare i32 @llvm.bswap.i32(i32 <id>)
7804 declare i64 @llvm.bswap.i64(i64 <id>)
7808 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7809 values with an even number of bytes (positive multiple of 16 bits). These
7810 are useful for performing operations on data that is not in the target's
7811 native byte order.</p>
7814 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7815 and low byte of the input i16 swapped. Similarly,
7816 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7817 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7818 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7819 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7820 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7821 more, respectively).</p>
7825 <!-- _______________________________________________________________________ -->
7827 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7833 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7834 width, or on any vector with integer elements. Not all targets support all
7835 bit widths or vector types, however.</p>
7838 declare i8 @llvm.ctpop.i8(i8 <src>)
7839 declare i16 @llvm.ctpop.i16(i16 <src>)
7840 declare i32 @llvm.ctpop.i32(i32 <src>)
7841 declare i64 @llvm.ctpop.i64(i64 <src>)
7842 declare i256 @llvm.ctpop.i256(i256 <src>)
7843 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7847 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7851 <p>The only argument is the value to be counted. The argument may be of any
7852 integer type, or a vector with integer elements.
7853 The return type must match the argument type.</p>
7856 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7857 element of a vector.</p>
7861 <!-- _______________________________________________________________________ -->
7863 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7869 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7870 integer bit width, or any vector whose elements are integers. Not all
7871 targets support all bit widths or vector types, however.</p>
7874 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7875 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7876 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7877 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7878 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7879 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7883 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7884 leading zeros in a variable.</p>
7887 <p>The first argument is the value to be counted. This argument may be of any
7888 integer type, or a vectory with integer element type. The return type
7889 must match the first argument type.</p>
7891 <p>The second argument must be a constant and is a flag to indicate whether the
7892 intrinsic should ensure that a zero as the first argument produces a defined
7893 result. Historically some architectures did not provide a defined result for
7894 zero values as efficiently, and many algorithms are now predicated on
7895 avoiding zero-value inputs.</p>
7898 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7899 zeros in a variable, or within each element of the vector.
7900 If <tt>src == 0</tt> then the result is the size in bits of the type of
7901 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7902 For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7906 <!-- _______________________________________________________________________ -->
7908 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7914 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7915 integer bit width, or any vector of integer elements. Not all targets
7916 support all bit widths or vector types, however.</p>
7919 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7920 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7921 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7922 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7923 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7924 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7928 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7932 <p>The first argument is the value to be counted. This argument may be of any
7933 integer type, or a vectory with integer element type. The return type
7934 must match the first argument type.</p>
7936 <p>The second argument must be a constant and is a flag to indicate whether the
7937 intrinsic should ensure that a zero as the first argument produces a defined
7938 result. Historically some architectures did not provide a defined result for
7939 zero values as efficiently, and many algorithms are now predicated on
7940 avoiding zero-value inputs.</p>
7943 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7944 zeros in a variable, or within each element of a vector.
7945 If <tt>src == 0</tt> then the result is the size in bits of the type of
7946 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7947 For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7953 <!-- ======================================================================= -->
7955 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7960 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7962 <!-- _______________________________________________________________________ -->
7964 <a name="int_sadd_overflow">
7965 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7972 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7973 on any integer bit width.</p>
7976 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7977 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7978 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7982 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7983 a signed addition of the two arguments, and indicate whether an overflow
7984 occurred during the signed summation.</p>
7987 <p>The arguments (%a and %b) and the first element of the result structure may
7988 be of integer types of any bit width, but they must have the same bit
7989 width. The second element of the result structure must be of
7990 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7991 undergo signed addition.</p>
7994 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7995 a signed addition of the two variables. They return a structure — the
7996 first element of which is the signed summation, and the second element of
7997 which is a bit specifying if the signed summation resulted in an
8002 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8003 %sum = extractvalue {i32, i1} %res, 0
8004 %obit = extractvalue {i32, i1} %res, 1
8005 br i1 %obit, label %overflow, label %normal
8010 <!-- _______________________________________________________________________ -->
8012 <a name="int_uadd_overflow">
8013 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
8020 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
8021 on any integer bit width.</p>
8024 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8025 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8026 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8030 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
8031 an unsigned addition of the two arguments, and indicate whether a carry
8032 occurred during the unsigned summation.</p>
8035 <p>The arguments (%a and %b) and the first element of the result structure may
8036 be of integer types of any bit width, but they must have the same bit
8037 width. The second element of the result structure must be of
8038 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8039 undergo unsigned addition.</p>
8042 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
8043 an unsigned addition of the two arguments. They return a structure —
8044 the first element of which is the sum, and the second element of which is a
8045 bit specifying if the unsigned summation resulted in a carry.</p>
8049 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8050 %sum = extractvalue {i32, i1} %res, 0
8051 %obit = extractvalue {i32, i1} %res, 1
8052 br i1 %obit, label %carry, label %normal
8057 <!-- _______________________________________________________________________ -->
8059 <a name="int_ssub_overflow">
8060 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
8067 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
8068 on any integer bit width.</p>
8071 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8072 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8073 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8077 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
8078 a signed subtraction of the two arguments, and indicate whether an overflow
8079 occurred during the signed subtraction.</p>
8082 <p>The arguments (%a and %b) and the first element of the result structure may
8083 be of integer types of any bit width, but they must have the same bit
8084 width. The second element of the result structure must be of
8085 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8086 undergo signed subtraction.</p>
8089 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
8090 a signed subtraction of the two arguments. They return a structure —
8091 the first element of which is the subtraction, and the second element of
8092 which is a bit specifying if the signed subtraction resulted in an
8097 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8098 %sum = extractvalue {i32, i1} %res, 0
8099 %obit = extractvalue {i32, i1} %res, 1
8100 br i1 %obit, label %overflow, label %normal
8105 <!-- _______________________________________________________________________ -->
8107 <a name="int_usub_overflow">
8108 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
8115 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
8116 on any integer bit width.</p>
8119 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8120 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8121 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8125 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
8126 an unsigned subtraction of the two arguments, and indicate whether an
8127 overflow occurred during the unsigned subtraction.</p>
8130 <p>The arguments (%a and %b) and the first element of the result structure may
8131 be of integer types of any bit width, but they must have the same bit
8132 width. The second element of the result structure must be of
8133 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8134 undergo unsigned subtraction.</p>
8137 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
8138 an unsigned subtraction of the two arguments. They return a structure —
8139 the first element of which is the subtraction, and the second element of
8140 which is a bit specifying if the unsigned subtraction resulted in an
8145 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8146 %sum = extractvalue {i32, i1} %res, 0
8147 %obit = extractvalue {i32, i1} %res, 1
8148 br i1 %obit, label %overflow, label %normal
8153 <!-- _______________________________________________________________________ -->
8155 <a name="int_smul_overflow">
8156 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
8163 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
8164 on any integer bit width.</p>
8167 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8168 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8169 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8174 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
8175 a signed multiplication of the two arguments, and indicate whether an
8176 overflow occurred during the signed multiplication.</p>
8179 <p>The arguments (%a and %b) and the first element of the result structure may
8180 be of integer types of any bit width, but they must have the same bit
8181 width. The second element of the result structure must be of
8182 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8183 undergo signed multiplication.</p>
8186 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
8187 a signed multiplication of the two arguments. They return a structure —
8188 the first element of which is the multiplication, and the second element of
8189 which is a bit specifying if the signed multiplication resulted in an
8194 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8195 %sum = extractvalue {i32, i1} %res, 0
8196 %obit = extractvalue {i32, i1} %res, 1
8197 br i1 %obit, label %overflow, label %normal
8202 <!-- _______________________________________________________________________ -->
8204 <a name="int_umul_overflow">
8205 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
8212 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
8213 on any integer bit width.</p>
8216 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8217 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8218 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8222 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
8223 a unsigned multiplication of the two arguments, and indicate whether an
8224 overflow occurred during the unsigned multiplication.</p>
8227 <p>The arguments (%a and %b) and the first element of the result structure may
8228 be of integer types of any bit width, but they must have the same bit
8229 width. The second element of the result structure must be of
8230 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8231 undergo unsigned multiplication.</p>
8234 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
8235 an unsigned multiplication of the two arguments. They return a structure
8236 — the first element of which is the multiplication, and the second
8237 element of which is a bit specifying if the unsigned multiplication resulted
8242 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8243 %sum = extractvalue {i32, i1} %res, 0
8244 %obit = extractvalue {i32, i1} %res, 1
8245 br i1 %obit, label %overflow, label %normal
8252 <!-- ======================================================================= -->
8254 <a name="spec_arithmetic">Specialised Arithmetic Intrinsics</a>
8257 <!-- _______________________________________________________________________ -->
8260 <a name="fmuladd">'<tt>llvm.fmuladd.*</tt>' Intrinsic</a>
8267 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8268 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8272 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsic functions represent multiply-add
8273 expressions that can be fused if the code generator determines that the fused
8274 expression would be legal and efficient.</p>
8277 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsics each take three arguments: two
8278 multiplicands, a and b, and an addend c.</p>
8281 <p>The expression:</p>
8283 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8285 <p>is equivalent to the expression a * b + c, except that rounding will not be
8286 performed between the multiplication and addition steps if the code generator
8287 fuses the operations. Fusion is not guaranteed, even if the target platform
8288 supports it. If a fused multiply-add is required the corresponding llvm.fma.*
8289 intrinsic function should be used instead.</p>
8293 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8298 <!-- ======================================================================= -->
8300 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
8305 <p>For most target platforms, half precision floating point is a storage-only
8306 format. This means that it is
8307 a dense encoding (in memory) but does not support computation in the
8310 <p>This means that code must first load the half-precision floating point
8311 value as an i16, then convert it to float with <a
8312 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
8313 Computation can then be performed on the float value (including extending to
8314 double etc). To store the value back to memory, it is first converted to
8315 float if needed, then converted to i16 with
8316 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
8317 storing as an i16 value.</p>
8319 <!-- _______________________________________________________________________ -->
8321 <a name="int_convert_to_fp16">
8322 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
8330 declare i16 @llvm.convert.to.fp16(f32 %a)
8334 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8335 a conversion from single precision floating point format to half precision
8336 floating point format.</p>
8339 <p>The intrinsic function contains single argument - the value to be
8343 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8344 a conversion from single precision floating point format to half precision
8345 floating point format. The return value is an <tt>i16</tt> which
8346 contains the converted number.</p>
8350 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8351 store i16 %res, i16* @x, align 2
8356 <!-- _______________________________________________________________________ -->
8358 <a name="int_convert_from_fp16">
8359 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
8367 declare f32 @llvm.convert.from.fp16(i16 %a)
8371 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
8372 a conversion from half precision floating point format to single precision
8373 floating point format.</p>
8376 <p>The intrinsic function contains single argument - the value to be
8380 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
8381 conversion from half single precision floating point format to single
8382 precision floating point format. The input half-float value is represented by
8383 an <tt>i16</tt> value.</p>
8387 %a = load i16* @x, align 2
8388 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8395 <!-- ======================================================================= -->
8397 <a name="int_debugger">Debugger Intrinsics</a>
8402 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
8403 prefix), are described in
8404 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
8405 Level Debugging</a> document.</p>
8409 <!-- ======================================================================= -->
8411 <a name="int_eh">Exception Handling Intrinsics</a>
8416 <p>The LLVM exception handling intrinsics (which all start with
8417 <tt>llvm.eh.</tt> prefix), are described in
8418 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
8419 Handling</a> document.</p>
8423 <!-- ======================================================================= -->
8425 <a name="int_trampoline">Trampoline Intrinsics</a>
8430 <p>These intrinsics make it possible to excise one parameter, marked with
8431 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
8432 The result is a callable
8433 function pointer lacking the nest parameter - the caller does not need to
8434 provide a value for it. Instead, the value to use is stored in advance in a
8435 "trampoline", a block of memory usually allocated on the stack, which also
8436 contains code to splice the nest value into the argument list. This is used
8437 to implement the GCC nested function address extension.</p>
8439 <p>For example, if the function is
8440 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
8441 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
8444 <pre class="doc_code">
8445 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8446 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8447 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8448 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8449 %fp = bitcast i8* %p to i32 (i32, i32)*
8452 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
8453 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
8455 <!-- _______________________________________________________________________ -->
8458 '<tt>llvm.init.trampoline</tt>' Intrinsic
8466 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8470 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
8471 turning it into a trampoline.</p>
8474 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
8475 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
8476 sufficiently aligned block of memory; this memory is written to by the
8477 intrinsic. Note that the size and the alignment are target-specific - LLVM
8478 currently provides no portable way of determining them, so a front-end that
8479 generates this intrinsic needs to have some target-specific knowledge.
8480 The <tt>func</tt> argument must hold a function bitcast to
8481 an <tt>i8*</tt>.</p>
8484 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
8485 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
8486 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
8487 which can be <a href="#int_trampoline">bitcast (to a new function) and
8488 called</a>. The new function's signature is the same as that of
8489 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
8490 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
8491 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
8492 with the same argument list, but with <tt>nval</tt> used for the missing
8493 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
8494 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
8495 to the returned function pointer is undefined.</p>
8498 <!-- _______________________________________________________________________ -->
8501 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
8509 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8513 <p>This performs any required machine-specific adjustment to the address of a
8514 trampoline (passed as <tt>tramp</tt>).</p>
8517 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
8518 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
8522 <p>On some architectures the address of the code to be executed needs to be
8523 different to the address where the trampoline is actually stored. This
8524 intrinsic returns the executable address corresponding to <tt>tramp</tt>
8525 after performing the required machine specific adjustments.
8526 The pointer returned can then be <a href="#int_trampoline"> bitcast and
8534 <!-- ======================================================================= -->
8536 <a name="int_memorymarkers">Memory Use Markers</a>
8541 <p>This class of intrinsics exists to information about the lifetime of memory
8542 objects and ranges where variables are immutable.</p>
8544 <!-- _______________________________________________________________________ -->
8546 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8553 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8557 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8558 object's lifetime.</p>
8561 <p>The first argument is a constant integer representing the size of the
8562 object, or -1 if it is variable sized. The second argument is a pointer to
8566 <p>This intrinsic indicates that before this point in the code, the value of the
8567 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8568 never be used and has an undefined value. A load from the pointer that
8569 precedes this intrinsic can be replaced with
8570 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8574 <!-- _______________________________________________________________________ -->
8576 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8583 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8587 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8588 object's lifetime.</p>
8591 <p>The first argument is a constant integer representing the size of the
8592 object, or -1 if it is variable sized. The second argument is a pointer to
8596 <p>This intrinsic indicates that after this point in the code, the value of the
8597 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8598 never be used and has an undefined value. Any stores into the memory object
8599 following this intrinsic may be removed as dead.
8603 <!-- _______________________________________________________________________ -->
8605 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8612 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8616 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8617 a memory object will not change.</p>
8620 <p>The first argument is a constant integer representing the size of the
8621 object, or -1 if it is variable sized. The second argument is a pointer to
8625 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8626 the return value, the referenced memory location is constant and
8631 <!-- _______________________________________________________________________ -->
8633 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8640 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8644 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8645 a memory object are mutable.</p>
8648 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8649 The second argument is a constant integer representing the size of the
8650 object, or -1 if it is variable sized and the third argument is a pointer
8654 <p>This intrinsic indicates that the memory is mutable again.</p>
8660 <!-- ======================================================================= -->
8662 <a name="int_general">General Intrinsics</a>
8667 <p>This class of intrinsics is designed to be generic and has no specific
8670 <!-- _______________________________________________________________________ -->
8672 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8679 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8683 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8686 <p>The first argument is a pointer to a value, the second is a pointer to a
8687 global string, the third is a pointer to a global string which is the source
8688 file name, and the last argument is the line number.</p>
8691 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8692 This can be useful for special purpose optimizations that want to look for
8693 these annotations. These have no other defined use; they are ignored by code
8694 generation and optimization.</p>
8698 <!-- _______________________________________________________________________ -->
8700 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8706 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8707 any integer bit width.</p>
8710 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8711 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8712 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8713 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8714 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8718 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8721 <p>The first argument is an integer value (result of some expression), the
8722 second is a pointer to a global string, the third is a pointer to a global
8723 string which is the source file name, and the last argument is the line
8724 number. It returns the value of the first argument.</p>
8727 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8728 arbitrary strings. This can be useful for special purpose optimizations that
8729 want to look for these annotations. These have no other defined use; they
8730 are ignored by code generation and optimization.</p>
8734 <!-- _______________________________________________________________________ -->
8736 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8743 declare void @llvm.trap() noreturn nounwind
8747 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8753 <p>This intrinsic is lowered to the target dependent trap instruction. If the
8754 target does not have a trap instruction, this intrinsic will be lowered to
8755 a call of the <tt>abort()</tt> function.</p>
8759 <!-- _______________________________________________________________________ -->
8761 <a name="int_debugtrap">'<tt>llvm.debugtrap</tt>' Intrinsic</a>
8768 declare void @llvm.debugtrap() nounwind
8772 <p>The '<tt>llvm.debugtrap</tt>' intrinsic.</p>
8778 <p>This intrinsic is lowered to code which is intended to cause an execution
8779 trap with the intention of requesting the attention of a debugger.</p>
8783 <!-- _______________________________________________________________________ -->
8785 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8792 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8796 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8797 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8798 ensure that it is placed on the stack before local variables.</p>
8801 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8802 arguments. The first argument is the value loaded from the stack
8803 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8804 that has enough space to hold the value of the guard.</p>
8807 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8808 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8809 stack. This is to ensure that if a local variable on the stack is
8810 overwritten, it will destroy the value of the guard. When the function exits,
8811 the guard on the stack is checked against the original guard. If they are
8812 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8817 <!-- _______________________________________________________________________ -->
8819 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8826 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8827 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8831 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8832 the optimizers to determine at compile time whether a) an operation (like
8833 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8834 runtime check for overflow isn't necessary. An object in this context means
8835 an allocation of a specific class, structure, array, or other object.</p>
8838 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8839 argument is a pointer to or into the <tt>object</tt>. The second argument
8840 is a boolean and determines whether <tt>llvm.objectsize</tt> returns 0 (if
8841 true) or -1 (if false) when the object size is unknown.
8842 The second argument only accepts constants.</p>
8845 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to a constant representing
8846 the size of the object concerned. If the size cannot be determined at compile
8847 time, <tt>llvm.objectsize</tt> returns <tt>i32/i64 -1 or 0</tt>
8848 (depending on the <tt>min</tt> argument).</p>
8851 <!-- _______________________________________________________________________ -->
8853 <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a>
8860 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8861 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8865 <p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the
8866 most probable) value of <tt>val</tt>, which can be used by optimizers.</p>
8869 <p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first
8870 argument is a value. The second argument is an expected value, this needs to
8871 be a constant value, variables are not allowed.</p>
8874 <p>This intrinsic is lowered to the <tt>val</tt>.</p>
8877 <!-- _______________________________________________________________________ -->
8879 <a name="int_donothing">'<tt>llvm.donothing</tt>' Intrinsic</a>
8886 declare void @llvm.donothing() nounwind readnone
8890 <p>The <tt>llvm.donothing</tt> intrinsic doesn't perform any operation. It's the
8891 only intrinsic that can be called with an invoke instruction.</p>
8897 <p>This intrinsic does nothing, and it's removed by optimizers and ignored by
8904 <!-- *********************************************************************** -->
8907 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8908 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8909 <a href="http://validator.w3.org/check/referer"><img
8910 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8912 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8913 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8914 Last modified: $Date$