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_exp2">'<tt>llvm.exp2.*</tt>' Intrinsic</a></li>
260 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
261 <li><a href="#int_log10">'<tt>llvm.log10.*</tt>' Intrinsic</a></li>
262 <li><a href="#int_log2">'<tt>llvm.log2.*</tt>' Intrinsic</a></li>
263 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
264 <li><a href="#int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a></li>
265 <li><a href="#int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a></li>
266 <li><a href="#int_ceil">'<tt>llvm.ceil.*</tt>' Intrinsic</a></li>
267 <li><a href="#int_trunc">'<tt>llvm.trunc.*</tt>' Intrinsic</a></li>
268 <li><a href="#int_rint">'<tt>llvm.rint.*</tt>' Intrinsic</a></li>
269 <li><a href="#int_nearbyint">'<tt>llvm.nearbyint.*</tt>' Intrinsic</a></li>
272 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
274 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
275 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
276 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
277 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
280 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
282 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
283 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
284 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
285 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
286 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
287 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
290 <li><a href="#spec_arithmetic">Specialised Arithmetic Intrinsics</a>
292 <li><a href="#fmuladd">'<tt>llvm.fmuladd</tt> Intrinsic</a></li>
295 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
297 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
298 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
301 <li><a href="#int_debugger">Debugger intrinsics</a></li>
302 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
303 <li><a href="#int_trampoline">Trampoline Intrinsics</a>
305 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
306 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
309 <li><a href="#int_memorymarkers">Memory Use Markers</a>
311 <li><a href="#int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a></li>
312 <li><a href="#int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a></li>
313 <li><a href="#int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a></li>
314 <li><a href="#int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a></li>
317 <li><a href="#int_general">General intrinsics</a>
319 <li><a href="#int_var_annotation">
320 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
321 <li><a href="#int_annotation">
322 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
323 <li><a href="#int_trap">
324 '<tt>llvm.trap</tt>' Intrinsic</a></li>
325 <li><a href="#int_debugtrap">
326 '<tt>llvm.debugtrap</tt>' Intrinsic</a></li>
327 <li><a href="#int_stackprotector">
328 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
329 <li><a href="#int_objectsize">
330 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
331 <li><a href="#int_expect">
332 '<tt>llvm.expect</tt>' Intrinsic</a></li>
333 <li><a href="#int_donothing">
334 '<tt>llvm.donothing</tt>' Intrinsic</a></li>
341 <div class="doc_author">
342 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
343 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
346 <!-- *********************************************************************** -->
347 <h2><a name="abstract">Abstract</a></h2>
348 <!-- *********************************************************************** -->
352 <p>This document is a reference manual for the LLVM assembly language. LLVM is
353 a Static Single Assignment (SSA) based representation that provides type
354 safety, low-level operations, flexibility, and the capability of representing
355 'all' high-level languages cleanly. It is the common code representation
356 used throughout all phases of the LLVM compilation strategy.</p>
360 <!-- *********************************************************************** -->
361 <h2><a name="introduction">Introduction</a></h2>
362 <!-- *********************************************************************** -->
366 <p>The LLVM code representation is designed to be used in three different forms:
367 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
368 for fast loading by a Just-In-Time compiler), and as a human readable
369 assembly language representation. This allows LLVM to provide a powerful
370 intermediate representation for efficient compiler transformations and
371 analysis, while providing a natural means to debug and visualize the
372 transformations. The three different forms of LLVM are all equivalent. This
373 document describes the human readable representation and notation.</p>
375 <p>The LLVM representation aims to be light-weight and low-level while being
376 expressive, typed, and extensible at the same time. It aims to be a
377 "universal IR" of sorts, by being at a low enough level that high-level ideas
378 may be cleanly mapped to it (similar to how microprocessors are "universal
379 IR's", allowing many source languages to be mapped to them). By providing
380 type information, LLVM can be used as the target of optimizations: for
381 example, through pointer analysis, it can be proven that a C automatic
382 variable is never accessed outside of the current function, allowing it to
383 be promoted to a simple SSA value instead of a memory location.</p>
385 <!-- _______________________________________________________________________ -->
387 <a name="wellformed">Well-Formedness</a>
392 <p>It is important to note that this document describes 'well formed' LLVM
393 assembly language. There is a difference between what the parser accepts and
394 what is considered 'well formed'. For example, the following instruction is
395 syntactically okay, but not well formed:</p>
397 <pre class="doc_code">
398 %x = <a href="#i_add">add</a> i32 1, %x
401 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
402 LLVM infrastructure provides a verification pass that may be used to verify
403 that an LLVM module is well formed. This pass is automatically run by the
404 parser after parsing input assembly and by the optimizer before it outputs
405 bitcode. The violations pointed out by the verifier pass indicate bugs in
406 transformation passes or input to the parser.</p>
412 <!-- Describe the typesetting conventions here. -->
414 <!-- *********************************************************************** -->
415 <h2><a name="identifiers">Identifiers</a></h2>
416 <!-- *********************************************************************** -->
420 <p>LLVM identifiers come in two basic types: global and local. Global
421 identifiers (functions, global variables) begin with the <tt>'@'</tt>
422 character. Local identifiers (register names, types) begin with
423 the <tt>'%'</tt> character. Additionally, there are three different formats
424 for identifiers, for different purposes:</p>
427 <li>Named values are represented as a string of characters with their prefix.
428 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
429 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
430 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
431 other characters in their names can be surrounded with quotes. Special
432 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
433 ASCII code for the character in hexadecimal. In this way, any character
434 can be used in a name value, even quotes themselves.</li>
436 <li>Unnamed values are represented as an unsigned numeric value with their
437 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
439 <li>Constants, which are described in a <a href="#constants">section about
440 constants</a>, below.</li>
443 <p>LLVM requires that values start with a prefix for two reasons: Compilers
444 don't need to worry about name clashes with reserved words, and the set of
445 reserved words may be expanded in the future without penalty. Additionally,
446 unnamed identifiers allow a compiler to quickly come up with a temporary
447 variable without having to avoid symbol table conflicts.</p>
449 <p>Reserved words in LLVM are very similar to reserved words in other
450 languages. There are keywords for different opcodes
451 ('<tt><a href="#i_add">add</a></tt>',
452 '<tt><a href="#i_bitcast">bitcast</a></tt>',
453 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
454 ('<tt><a href="#t_void">void</a></tt>',
455 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
456 reserved words cannot conflict with variable names, because none of them
457 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
459 <p>Here is an example of LLVM code to multiply the integer variable
460 '<tt>%X</tt>' by 8:</p>
464 <pre class="doc_code">
465 %result = <a href="#i_mul">mul</a> i32 %X, 8
468 <p>After strength reduction:</p>
470 <pre class="doc_code">
471 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
474 <p>And the hard way:</p>
476 <pre class="doc_code">
477 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
478 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
479 %result = <a href="#i_add">add</a> i32 %1, %1
482 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
483 lexical features of LLVM:</p>
486 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
489 <li>Unnamed temporaries are created when the result of a computation is not
490 assigned to a named value.</li>
492 <li>Unnamed temporaries are numbered sequentially</li>
495 <p>It also shows a convention that we follow in this document. When
496 demonstrating instructions, we will follow an instruction with a comment that
497 defines the type and name of value produced. Comments are shown in italic
502 <!-- *********************************************************************** -->
503 <h2><a name="highlevel">High Level Structure</a></h2>
504 <!-- *********************************************************************** -->
506 <!-- ======================================================================= -->
508 <a name="modulestructure">Module Structure</a>
513 <p>LLVM programs are composed of <tt>Module</tt>s, each of which is a
514 translation unit of the input programs. Each module consists of functions,
515 global variables, and symbol table entries. Modules may be combined together
516 with the LLVM linker, which merges function (and global variable)
517 definitions, resolves forward declarations, and merges symbol table
518 entries. Here is an example of the "hello world" module:</p>
520 <pre class="doc_code">
521 <i>; Declare the string constant as a global constant.</i>
522 <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"
524 <i>; External declaration of the puts function</i>
525 <a href="#functionstructure">declare</a> i32 @puts(i8* <a href="#nocapture">nocapture</a>) <a href="#fnattrs">nounwind</a>
527 <i>; Definition of main function</i>
528 define i32 @main() { <i>; i32()* </i>
529 <i>; Convert [13 x i8]* to i8 *...</i>
530 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.str, i64 0, i64 0
532 <i>; Call puts function to write out the string to stdout.</i>
533 <a href="#i_call">call</a> i32 @puts(i8* %cast210)
534 <a href="#i_ret">ret</a> i32 0
537 <i>; Named metadata</i>
538 !1 = metadata !{i32 42}
542 <p>This example is made up of a <a href="#globalvars">global variable</a> named
543 "<tt>.str</tt>", an external declaration of the "<tt>puts</tt>" function,
544 a <a href="#functionstructure">function definition</a> for
545 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
548 <p>In general, a module is made up of a list of global values (where both
549 functions and global variables are global values). Global values are
550 represented by a pointer to a memory location (in this case, a pointer to an
551 array of char, and a pointer to a function), and have one of the
552 following <a href="#linkage">linkage types</a>.</p>
556 <!-- ======================================================================= -->
558 <a name="linkage">Linkage Types</a>
563 <p>All Global Variables and Functions have one of the following types of
567 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
568 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
569 by objects in the current module. In particular, linking code into a
570 module with an private global value may cause the private to be renamed as
571 necessary to avoid collisions. Because the symbol is private to the
572 module, all references can be updated. This doesn't show up in any symbol
573 table in the object file.</dd>
575 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
576 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
577 assembler and evaluated by the linker. Unlike normal strong symbols, they
578 are removed by the linker from the final linked image (executable or
579 dynamic library).</dd>
581 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
582 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
583 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
584 linker. The symbols are removed by the linker from the final linked image
585 (executable or dynamic library).</dd>
587 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
588 <dd>Similar to private, but the value shows as a local symbol
589 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
590 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
592 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
593 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
594 into the object file corresponding to the LLVM module. They exist to
595 allow inlining and other optimizations to take place given knowledge of
596 the definition of the global, which is known to be somewhere outside the
597 module. Globals with <tt>available_externally</tt> linkage are allowed to
598 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
599 This linkage type is only allowed on definitions, not declarations.</dd>
601 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
602 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
603 the same name when linkage occurs. This can be used to implement
604 some forms of inline functions, templates, or other code which must be
605 generated in each translation unit that uses it, but where the body may
606 be overridden with a more definitive definition later. Unreferenced
607 <tt>linkonce</tt> globals are allowed to be discarded. Note that
608 <tt>linkonce</tt> linkage does not actually allow the optimizer to
609 inline the body of this function into callers because it doesn't know if
610 this definition of the function is the definitive definition within the
611 program or whether it will be overridden by a stronger definition.
612 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
615 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
616 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
617 <tt>linkonce</tt> linkage, except that unreferenced globals with
618 <tt>weak</tt> linkage may not be discarded. This is used for globals that
619 are declared "weak" in C source code.</dd>
621 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
622 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
623 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
625 Symbols with "<tt>common</tt>" linkage are merged in the same way as
626 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
627 <tt>common</tt> symbols may not have an explicit section,
628 must have a zero initializer, and may not be marked '<a
629 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
630 have common linkage.</dd>
633 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
634 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
635 pointer to array type. When two global variables with appending linkage
636 are linked together, the two global arrays are appended together. This is
637 the LLVM, typesafe, equivalent of having the system linker append together
638 "sections" with identical names when .o files are linked.</dd>
640 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
641 <dd>The semantics of this linkage follow the ELF object file model: the symbol
642 is weak until linked, if not linked, the symbol becomes null instead of
643 being an undefined reference.</dd>
645 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
646 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
647 <dd>Some languages allow differing globals to be merged, such as two functions
648 with different semantics. Other languages, such as <tt>C++</tt>, ensure
649 that only equivalent globals are ever merged (the "one definition rule"
650 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
651 and <tt>weak_odr</tt> linkage types to indicate that the global will only
652 be merged with equivalent globals. These linkage types are otherwise the
653 same as their non-<tt>odr</tt> versions.</dd>
655 <dt><tt><b><a name="linkage_linkonce_odr_auto_hide">linkonce_odr_auto_hide</a></b></tt></dt>
656 <dd>Similar to "<tt>linkonce_odr</tt>", but nothing in the translation unit
657 takes the address of this definition. For instance, functions that had an
658 inline definition, but the compiler decided not to inline it.
659 <tt>linkonce_odr_auto_hide</tt> may have only <tt>default</tt> visibility.
660 The symbols are removed by the linker from the final linked image
661 (executable or dynamic library).</dd>
663 <dt><tt><b><a name="linkage_external">external</a></b></tt></dt>
664 <dd>If none of the above identifiers are used, the global is externally
665 visible, meaning that it participates in linkage and can be used to
666 resolve external symbol references.</dd>
669 <p>The next two types of linkage are targeted for Microsoft Windows platform
670 only. They are designed to support importing (exporting) symbols from (to)
671 DLLs (Dynamic Link Libraries).</p>
674 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
675 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
676 or variable via a global pointer to a pointer that is set up by the DLL
677 exporting the symbol. On Microsoft Windows targets, the pointer name is
678 formed by combining <code>__imp_</code> and the function or variable
681 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
682 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
683 pointer to a pointer in a DLL, so that it can be referenced with the
684 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
685 name is formed by combining <code>__imp_</code> and the function or
689 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
690 another module defined a "<tt>.LC0</tt>" variable and was linked with this
691 one, one of the two would be renamed, preventing a collision. Since
692 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
693 declarations), they are accessible outside of the current module.</p>
695 <p>It is illegal for a function <i>declaration</i> to have any linkage type
696 other than <tt>external</tt>, <tt>dllimport</tt>
697 or <tt>extern_weak</tt>.</p>
699 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
700 or <tt>weak_odr</tt> linkages.</p>
704 <!-- ======================================================================= -->
706 <a name="callingconv">Calling Conventions</a>
711 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
712 and <a href="#i_invoke">invokes</a> can all have an optional calling
713 convention specified for the call. The calling convention of any pair of
714 dynamic caller/callee must match, or the behavior of the program is
715 undefined. The following calling conventions are supported by LLVM, and more
716 may be added in the future:</p>
719 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
720 <dd>This calling convention (the default if no other calling convention is
721 specified) matches the target C calling conventions. This calling
722 convention supports varargs function calls and tolerates some mismatch in
723 the declared prototype and implemented declaration of the function (as
726 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
727 <dd>This calling convention attempts to make calls as fast as possible
728 (e.g. by passing things in registers). This calling convention allows the
729 target to use whatever tricks it wants to produce fast code for the
730 target, without having to conform to an externally specified ABI
731 (Application Binary Interface).
732 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
733 when this or the GHC convention is used.</a> This calling convention
734 does not support varargs and requires the prototype of all callees to
735 exactly match the prototype of the function definition.</dd>
737 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
738 <dd>This calling convention attempts to make code in the caller as efficient
739 as possible under the assumption that the call is not commonly executed.
740 As such, these calls often preserve all registers so that the call does
741 not break any live ranges in the caller side. This calling convention
742 does not support varargs and requires the prototype of all callees to
743 exactly match the prototype of the function definition.</dd>
745 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
746 <dd>This calling convention has been implemented specifically for use by the
747 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
748 It passes everything in registers, going to extremes to achieve this by
749 disabling callee save registers. This calling convention should not be
750 used lightly but only for specific situations such as an alternative to
751 the <em>register pinning</em> performance technique often used when
752 implementing functional programming languages.At the moment only X86
753 supports this convention and it has the following limitations:
755 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
756 floating point types are supported.</li>
757 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
758 6 floating point parameters.</li>
760 This calling convention supports
761 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
762 requires both the caller and callee are using it.
765 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
766 <dd>Any calling convention may be specified by number, allowing
767 target-specific calling conventions to be used. Target specific calling
768 conventions start at 64.</dd>
771 <p>More calling conventions can be added/defined on an as-needed basis, to
772 support Pascal conventions or any other well-known target-independent
777 <!-- ======================================================================= -->
779 <a name="visibility">Visibility Styles</a>
784 <p>All Global Variables and Functions have one of the following visibility
788 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
789 <dd>On targets that use the ELF object file format, default visibility means
790 that the declaration is visible to other modules and, in shared libraries,
791 means that the declared entity may be overridden. On Darwin, default
792 visibility means that the declaration is visible to other modules. Default
793 visibility corresponds to "external linkage" in the language.</dd>
795 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
796 <dd>Two declarations of an object with hidden visibility refer to the same
797 object if they are in the same shared object. Usually, hidden visibility
798 indicates that the symbol will not be placed into the dynamic symbol
799 table, so no other module (executable or shared library) can reference it
802 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
803 <dd>On ELF, protected visibility indicates that the symbol will be placed in
804 the dynamic symbol table, but that references within the defining module
805 will bind to the local symbol. That is, the symbol cannot be overridden by
811 <!-- ======================================================================= -->
813 <a name="namedtypes">Named Types</a>
818 <p>LLVM IR allows you to specify name aliases for certain types. This can make
819 it easier to read the IR and make the IR more condensed (particularly when
820 recursive types are involved). An example of a name specification is:</p>
822 <pre class="doc_code">
823 %mytype = type { %mytype*, i32 }
826 <p>You may give a name to any <a href="#typesystem">type</a> except
827 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
828 is expected with the syntax "%mytype".</p>
830 <p>Note that type names are aliases for the structural type that they indicate,
831 and that you can therefore specify multiple names for the same type. This
832 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
833 uses structural typing, the name is not part of the type. When printing out
834 LLVM IR, the printer will pick <em>one name</em> to render all types of a
835 particular shape. This means that if you have code where two different
836 source types end up having the same LLVM type, that the dumper will sometimes
837 print the "wrong" or unexpected type. This is an important design point and
838 isn't going to change.</p>
842 <!-- ======================================================================= -->
844 <a name="globalvars">Global Variables</a>
849 <p>Global variables define regions of memory allocated at compilation time
850 instead of run-time. Global variables may optionally be initialized, may
851 have an explicit section to be placed in, and may have an optional explicit
852 alignment specified.</p>
854 <p>A variable may be defined as <tt>thread_local</tt>, which
855 means that it will not be shared by threads (each thread will have a
856 separated copy of the variable). Not all targets support thread-local
857 variables. Optionally, a TLS model may be specified:</p>
860 <dt><b><tt>localdynamic</tt></b>:</dt>
861 <dd>For variables that are only used within the current shared library.</dd>
863 <dt><b><tt>initialexec</tt></b>:</dt>
864 <dd>For variables in modules that will not be loaded dynamically.</dd>
866 <dt><b><tt>localexec</tt></b>:</dt>
867 <dd>For variables defined in the executable and only used within it.</dd>
870 <p>The models correspond to the ELF TLS models; see
871 <a href="http://people.redhat.com/drepper/tls.pdf">ELF
872 Handling For Thread-Local Storage</a> for more information on under which
873 circumstances the different models may be used. The target may choose a
874 different TLS model if the specified model is not supported, or if a better
875 choice of model can be made.</p>
877 <p>A variable may be defined as a global
878 "constant," which indicates that the contents of the variable
879 will <b>never</b> be modified (enabling better optimization, allowing the
880 global data to be placed in the read-only section of an executable, etc).
881 Note that variables that need runtime initialization cannot be marked
882 "constant" as there is a store to the variable.</p>
884 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
885 constant, even if the final definition of the global is not. This capability
886 can be used to enable slightly better optimization of the program, but
887 requires the language definition to guarantee that optimizations based on the
888 'constantness' are valid for the translation units that do not include the
891 <p>As SSA values, global variables define pointer values that are in scope
892 (i.e. they dominate) all basic blocks in the program. Global variables
893 always define a pointer to their "content" type because they describe a
894 region of memory, and all memory objects in LLVM are accessed through
897 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
898 that the address is not significant, only the content. Constants marked
899 like this can be merged with other constants if they have the same
900 initializer. Note that a constant with significant address <em>can</em>
901 be merged with a <tt>unnamed_addr</tt> constant, the result being a
902 constant whose address is significant.</p>
904 <p>A global variable may be declared to reside in a target-specific numbered
905 address space. For targets that support them, address spaces may affect how
906 optimizations are performed and/or what target instructions are used to
907 access the variable. The default address space is zero. The address space
908 qualifier must precede any other attributes.</p>
910 <p>LLVM allows an explicit section to be specified for globals. If the target
911 supports it, it will emit globals to the section specified.</p>
913 <p>An explicit alignment may be specified for a global, which must be a power
914 of 2. If not present, or if the alignment is set to zero, the alignment of
915 the global is set by the target to whatever it feels convenient. If an
916 explicit alignment is specified, the global is forced to have exactly that
917 alignment. Targets and optimizers are not allowed to over-align the global
918 if the global has an assigned section. In this case, the extra alignment
919 could be observable: for example, code could assume that the globals are
920 densely packed in their section and try to iterate over them as an array,
921 alignment padding would break this iteration.</p>
923 <p>For example, the following defines a global in a numbered address space with
924 an initializer, section, and alignment:</p>
926 <pre class="doc_code">
927 @G = addrspace(5) constant float 1.0, section "foo", align 4
930 <p>The following example defines a thread-local global with
931 the <tt>initialexec</tt> TLS model:</p>
933 <pre class="doc_code">
934 @G = thread_local(initialexec) global i32 0, align 4
940 <!-- ======================================================================= -->
942 <a name="functionstructure">Functions</a>
947 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
948 optional <a href="#linkage">linkage type</a>, an optional
949 <a href="#visibility">visibility style</a>, an optional
950 <a href="#callingconv">calling convention</a>,
951 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
952 <a href="#paramattrs">parameter attribute</a> for the return type, a function
953 name, a (possibly empty) argument list (each with optional
954 <a href="#paramattrs">parameter attributes</a>), optional
955 <a href="#fnattrs">function attributes</a>, an optional section, an optional
956 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
957 curly brace, a list of basic blocks, and a closing curly brace.</p>
959 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
960 optional <a href="#linkage">linkage type</a>, an optional
961 <a href="#visibility">visibility style</a>, an optional
962 <a href="#callingconv">calling convention</a>,
963 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
964 <a href="#paramattrs">parameter attribute</a> for the return type, a function
965 name, a possibly empty list of arguments, an optional alignment, and an
966 optional <a href="#gc">garbage collector name</a>.</p>
968 <p>A function definition contains a list of basic blocks, forming the CFG
969 (Control Flow Graph) for the function. Each basic block may optionally start
970 with a label (giving the basic block a symbol table entry), contains a list
971 of instructions, and ends with a <a href="#terminators">terminator</a>
972 instruction (such as a branch or function return).</p>
974 <p>The first basic block in a function is special in two ways: it is immediately
975 executed on entrance to the function, and it is not allowed to have
976 predecessor basic blocks (i.e. there can not be any branches to the entry
977 block of a function). Because the block can have no predecessors, it also
978 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
980 <p>LLVM allows an explicit section to be specified for functions. If the target
981 supports it, it will emit functions to the section specified.</p>
983 <p>An explicit alignment may be specified for a function. If not present, or if
984 the alignment is set to zero, the alignment of the function is set by the
985 target to whatever it feels convenient. If an explicit alignment is
986 specified, the function is forced to have at least that much alignment. All
987 alignments must be a power of 2.</p>
989 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
990 be significant and two identical functions can be merged.</p>
993 <pre class="doc_code">
994 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
995 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
996 <ResultType> @<FunctionName> ([argument list])
997 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
998 [<a href="#gc">gc</a>] { ... }
1003 <!-- ======================================================================= -->
1005 <a name="aliasstructure">Aliases</a>
1010 <p>Aliases act as "second name" for the aliasee value (which can be either
1011 function, global variable, another alias or bitcast of global value). Aliases
1012 may have an optional <a href="#linkage">linkage type</a>, and an
1013 optional <a href="#visibility">visibility style</a>.</p>
1016 <pre class="doc_code">
1017 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
1022 <!-- ======================================================================= -->
1024 <a name="namedmetadatastructure">Named Metadata</a>
1029 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
1030 nodes</a> (but not metadata strings) are the only valid operands for
1031 a named metadata.</p>
1034 <pre class="doc_code">
1035 ; Some unnamed metadata nodes, which are referenced by the named metadata.
1036 !0 = metadata !{metadata !"zero"}
1037 !1 = metadata !{metadata !"one"}
1038 !2 = metadata !{metadata !"two"}
1040 !name = !{!0, !1, !2}
1045 <!-- ======================================================================= -->
1047 <a name="paramattrs">Parameter Attributes</a>
1052 <p>The return type and each parameter of a function type may have a set of
1053 <i>parameter attributes</i> associated with them. Parameter attributes are
1054 used to communicate additional information about the result or parameters of
1055 a function. Parameter attributes are considered to be part of the function,
1056 not of the function type, so functions with different parameter attributes
1057 can have the same function type.</p>
1059 <p>Parameter attributes are simple keywords that follow the type specified. If
1060 multiple parameter attributes are needed, they are space separated. For
1063 <pre class="doc_code">
1064 declare i32 @printf(i8* noalias nocapture, ...)
1065 declare i32 @atoi(i8 zeroext)
1066 declare signext i8 @returns_signed_char()
1069 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1070 <tt>readonly</tt>) come immediately after the argument list.</p>
1072 <p>Currently, only the following parameter attributes are defined:</p>
1075 <dt><tt><b>zeroext</b></tt></dt>
1076 <dd>This indicates to the code generator that the parameter or return value
1077 should be zero-extended to the extent required by the target's ABI (which
1078 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1079 parameter) or the callee (for a return value).</dd>
1081 <dt><tt><b>signext</b></tt></dt>
1082 <dd>This indicates to the code generator that the parameter or return value
1083 should be sign-extended to the extent required by the target's ABI (which
1084 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1087 <dt><tt><b>inreg</b></tt></dt>
1088 <dd>This indicates that this parameter or return value should be treated in a
1089 special target-dependent fashion during while emitting code for a function
1090 call or return (usually, by putting it in a register as opposed to memory,
1091 though some targets use it to distinguish between two different kinds of
1092 registers). Use of this attribute is target-specific.</dd>
1094 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1095 <dd><p>This indicates that the pointer parameter should really be passed by
1096 value to the function. The attribute implies that a hidden copy of the
1098 is made between the caller and the callee, so the callee is unable to
1099 modify the value in the caller. This attribute is only valid on LLVM
1100 pointer arguments. It is generally used to pass structs and arrays by
1101 value, but is also valid on pointers to scalars. The copy is considered
1102 to belong to the caller not the callee (for example,
1103 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1104 <tt>byval</tt> parameters). This is not a valid attribute for return
1107 <p>The byval attribute also supports specifying an alignment with
1108 the align attribute. It indicates the alignment of the stack slot to
1109 form and the known alignment of the pointer specified to the call site. If
1110 the alignment is not specified, then the code generator makes a
1111 target-specific assumption.</p></dd>
1113 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1114 <dd>This indicates that the pointer parameter specifies the address of a
1115 structure that is the return value of the function in the source program.
1116 This pointer must be guaranteed by the caller to be valid: loads and
1117 stores to the structure may be assumed by the callee to not to trap and
1118 to be properly aligned. This may only be applied to the first parameter.
1119 This is not a valid attribute for return values. </dd>
1121 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1122 <dd>This indicates that pointer values
1123 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1124 value do not alias pointer values which are not <i>based</i> on it,
1125 ignoring certain "irrelevant" dependencies.
1126 For a call to the parent function, dependencies between memory
1127 references from before or after the call and from those during the call
1128 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1129 return value used in that call.
1130 The caller shares the responsibility with the callee for ensuring that
1131 these requirements are met.
1132 For further details, please see the discussion of the NoAlias response in
1133 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1135 Note that this definition of <tt>noalias</tt> is intentionally
1136 similar to the definition of <tt>restrict</tt> in C99 for function
1137 arguments, though it is slightly weaker.
1139 For function return values, C99's <tt>restrict</tt> is not meaningful,
1140 while LLVM's <tt>noalias</tt> is.
1143 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1144 <dd>This indicates that the callee does not make any copies of the pointer
1145 that outlive the callee itself. This is not a valid attribute for return
1148 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1149 <dd>This indicates that the pointer parameter can be excised using the
1150 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1151 attribute for return values.</dd>
1156 <!-- ======================================================================= -->
1158 <a name="gc">Garbage Collector Names</a>
1163 <p>Each function may specify a garbage collector name, which is simply a
1166 <pre class="doc_code">
1167 define void @f() gc "name" { ... }
1170 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1171 collector which will cause the compiler to alter its output in order to
1172 support the named garbage collection algorithm.</p>
1176 <!-- ======================================================================= -->
1178 <a name="fnattrs">Function Attributes</a>
1183 <p>Function attributes are set to communicate additional information about a
1184 function. Function attributes are considered to be part of the function, not
1185 of the function type, so functions with different parameter attributes can
1186 have the same function type.</p>
1188 <p>Function attributes are simple keywords that follow the type specified. If
1189 multiple attributes are needed, they are space separated. For example:</p>
1191 <pre class="doc_code">
1192 define void @f() noinline { ... }
1193 define void @f() alwaysinline { ... }
1194 define void @f() alwaysinline optsize { ... }
1195 define void @f() optsize { ... }
1199 <dt><tt><b>address_safety</b></tt></dt>
1200 <dd>This attribute indicates that the address safety analysis
1201 is enabled for this function. </dd>
1203 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1204 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1205 the backend should forcibly align the stack pointer. Specify the
1206 desired alignment, which must be a power of two, in parentheses.
1208 <dt><tt><b>alwaysinline</b></tt></dt>
1209 <dd>This attribute indicates that the inliner should attempt to inline this
1210 function into callers whenever possible, ignoring any active inlining size
1211 threshold for this caller.</dd>
1213 <dt><tt><b>nonlazybind</b></tt></dt>
1214 <dd>This attribute suppresses lazy symbol binding for the function. This
1215 may make calls to the function faster, at the cost of extra program
1216 startup time if the function is not called during program startup.</dd>
1218 <dt><tt><b>inlinehint</b></tt></dt>
1219 <dd>This attribute indicates that the source code contained a hint that inlining
1220 this function is desirable (such as the "inline" keyword in C/C++). It
1221 is just a hint; it imposes no requirements on the inliner.</dd>
1223 <dt><tt><b>naked</b></tt></dt>
1224 <dd>This attribute disables prologue / epilogue emission for the function.
1225 This can have very system-specific consequences.</dd>
1227 <dt><tt><b>noimplicitfloat</b></tt></dt>
1228 <dd>This attributes disables implicit floating point instructions.</dd>
1230 <dt><tt><b>noinline</b></tt></dt>
1231 <dd>This attribute indicates that the inliner should never inline this
1232 function in any situation. This attribute may not be used together with
1233 the <tt>alwaysinline</tt> attribute.</dd>
1235 <dt><tt><b>noredzone</b></tt></dt>
1236 <dd>This attribute indicates that the code generator should not use a red
1237 zone, even if the target-specific ABI normally permits it.</dd>
1239 <dt><tt><b>noreturn</b></tt></dt>
1240 <dd>This function attribute indicates that the function never returns
1241 normally. This produces undefined behavior at runtime if the function
1242 ever does dynamically return.</dd>
1244 <dt><tt><b>nounwind</b></tt></dt>
1245 <dd>This function attribute indicates that the function never returns with an
1246 unwind or exceptional control flow. If the function does unwind, its
1247 runtime behavior is undefined.</dd>
1249 <dt><tt><b>optsize</b></tt></dt>
1250 <dd>This attribute suggests that optimization passes and code generator passes
1251 make choices that keep the code size of this function low, and otherwise
1252 do optimizations specifically to reduce code size.</dd>
1254 <dt><tt><b>readnone</b></tt></dt>
1255 <dd>This attribute indicates that the function computes its result (or decides
1256 to unwind an exception) based strictly on its arguments, without
1257 dereferencing any pointer arguments or otherwise accessing any mutable
1258 state (e.g. memory, control registers, etc) visible to caller functions.
1259 It does not write through any pointer arguments
1260 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1261 changes any state visible to callers. This means that it cannot unwind
1262 exceptions by calling the <tt>C++</tt> exception throwing methods.</dd>
1264 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1265 <dd>This attribute indicates that the function does not write through any
1266 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1267 arguments) or otherwise modify any state (e.g. memory, control registers,
1268 etc) visible to caller functions. It may dereference pointer arguments
1269 and read state that may be set in the caller. A readonly function always
1270 returns the same value (or unwinds an exception identically) when called
1271 with the same set of arguments and global state. It cannot unwind an
1272 exception by calling the <tt>C++</tt> exception throwing methods.</dd>
1274 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1275 <dd>This attribute indicates that this function can return twice. The
1276 C <code>setjmp</code> is an example of such a function. The compiler
1277 disables some optimizations (like tail calls) in the caller of these
1280 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1281 <dd>This attribute indicates that the function should emit a stack smashing
1282 protector. It is in the form of a "canary"—a random value placed on
1283 the stack before the local variables that's checked upon return from the
1284 function to see if it has been overwritten. A heuristic is used to
1285 determine if a function needs stack protectors or not.<br>
1287 If a function that has an <tt>ssp</tt> attribute is inlined into a
1288 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1289 function will have an <tt>ssp</tt> attribute.</dd>
1291 <dt><tt><b>sspreq</b></tt></dt>
1292 <dd>This attribute indicates that the function should <em>always</em> emit a
1293 stack smashing protector. This overrides
1294 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1296 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1297 function that doesn't have an <tt>sspreq</tt> attribute or which has
1298 an <tt>ssp</tt> attribute, then the resulting function will have
1299 an <tt>sspreq</tt> attribute.</dd>
1301 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1302 <dd>This attribute indicates that the ABI being targeted requires that
1303 an unwind table entry be produce for this function even if we can
1304 show that no exceptions passes by it. This is normally the case for
1305 the ELF x86-64 abi, but it can be disabled for some compilation
1311 <!-- ======================================================================= -->
1313 <a name="moduleasm">Module-Level Inline Assembly</a>
1318 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1319 the GCC "file scope inline asm" blocks. These blocks are internally
1320 concatenated by LLVM and treated as a single unit, but may be separated in
1321 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1323 <pre class="doc_code">
1324 module asm "inline asm code goes here"
1325 module asm "more can go here"
1328 <p>The strings can contain any character by escaping non-printable characters.
1329 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1332 <p>The inline asm code is simply printed to the machine code .s file when
1333 assembly code is generated.</p>
1337 <!-- ======================================================================= -->
1339 <a name="datalayout">Data Layout</a>
1344 <p>A module may specify a target specific data layout string that specifies how
1345 data is to be laid out in memory. The syntax for the data layout is
1348 <pre class="doc_code">
1349 target datalayout = "<i>layout specification</i>"
1352 <p>The <i>layout specification</i> consists of a list of specifications
1353 separated by the minus sign character ('-'). Each specification starts with
1354 a letter and may include other information after the letter to define some
1355 aspect of the data layout. The specifications accepted are as follows:</p>
1359 <dd>Specifies that the target lays out data in big-endian form. That is, the
1360 bits with the most significance have the lowest address location.</dd>
1363 <dd>Specifies that the target lays out data in little-endian form. That is,
1364 the bits with the least significance have the lowest address
1367 <dt><tt>S<i>size</i></tt></dt>
1368 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1369 of stack variables is limited to the natural stack alignment to avoid
1370 dynamic stack realignment. The stack alignment must be a multiple of
1371 8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1372 which does not prevent any alignment promotions.</dd>
1374 <dt><tt>p[n]:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1375 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1376 <i>preferred</i> alignments for address space <i>n</i>. All sizes are in
1377 bits. Specifying the <i>pref</i> alignment is optional. If omitted, the
1378 preceding <tt>:</tt> should be omitted too. The address space,
1379 <i>n</i> is optional, and if not specified, denotes the default address
1380 space 0. The value of <i>n</i> must be in the range [1,2^23).</dd>
1382 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1383 <dd>This specifies the alignment for an integer type of a given bit
1384 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1386 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1387 <dd>This specifies the alignment for a vector type of a given bit
1390 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1391 <dd>This specifies the alignment for a floating point type of a given bit
1392 <i>size</i>. Only values of <i>size</i> that are supported by the target
1393 will work. 32 (float) and 64 (double) are supported on all targets;
1394 80 or 128 (different flavors of long double) are also supported on some
1397 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1398 <dd>This specifies the alignment for an aggregate type of a given bit
1401 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1402 <dd>This specifies the alignment for a stack object of a given bit
1405 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1406 <dd>This specifies a set of native integer widths for the target CPU
1407 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1408 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1409 this set are considered to support most general arithmetic
1410 operations efficiently.</dd>
1413 <p>When constructing the data layout for a given target, LLVM starts with a
1414 default set of specifications which are then (possibly) overridden by the
1415 specifications in the <tt>datalayout</tt> keyword. The default specifications
1416 are given in this list:</p>
1419 <li><tt>E</tt> - big endian</li>
1420 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1421 <li><tt>p1:32:32:32</tt> - 32-bit pointers with 32-bit alignment for
1422 address space 1</li>
1423 <li><tt>p2:16:32:32</tt> - 16-bit pointers with 32-bit alignment for
1424 address space 2</li>
1425 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1426 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1427 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1428 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1429 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1430 alignment of 64-bits</li>
1431 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1432 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1433 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1434 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1435 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1436 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1439 <p>When LLVM is determining the alignment for a given type, it uses the
1440 following rules:</p>
1443 <li>If the type sought is an exact match for one of the specifications, that
1444 specification is used.</li>
1446 <li>If no match is found, and the type sought is an integer type, then the
1447 smallest integer type that is larger than the bitwidth of the sought type
1448 is used. If none of the specifications are larger than the bitwidth then
1449 the largest integer type is used. For example, given the default
1450 specifications above, the i7 type will use the alignment of i8 (next
1451 largest) while both i65 and i256 will use the alignment of i64 (largest
1454 <li>If no match is found, and the type sought is a vector type, then the
1455 largest vector type that is smaller than the sought vector type will be
1456 used as a fall back. This happens because <128 x double> can be
1457 implemented in terms of 64 <2 x double>, for example.</li>
1460 <p>The function of the data layout string may not be what you expect. Notably,
1461 this is not a specification from the frontend of what alignment the code
1462 generator should use.</p>
1464 <p>Instead, if specified, the target data layout is required to match what the
1465 ultimate <em>code generator</em> expects. This string is used by the
1466 mid-level optimizers to
1467 improve code, and this only works if it matches what the ultimate code
1468 generator uses. If you would like to generate IR that does not embed this
1469 target-specific detail into the IR, then you don't have to specify the
1470 string. This will disable some optimizations that require precise layout
1471 information, but this also prevents those optimizations from introducing
1472 target specificity into the IR.</p>
1478 <!-- ======================================================================= -->
1480 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1485 <p>Any memory access must be done through a pointer value associated
1486 with an address range of the memory access, otherwise the behavior
1487 is undefined. Pointer values are associated with address ranges
1488 according to the following rules:</p>
1491 <li>A pointer value is associated with the addresses associated with
1492 any value it is <i>based</i> on.
1493 <li>An address of a global variable is associated with the address
1494 range of the variable's storage.</li>
1495 <li>The result value of an allocation instruction is associated with
1496 the address range of the allocated storage.</li>
1497 <li>A null pointer in the default address-space is associated with
1499 <li>An integer constant other than zero or a pointer value returned
1500 from a function not defined within LLVM may be associated with address
1501 ranges allocated through mechanisms other than those provided by
1502 LLVM. Such ranges shall not overlap with any ranges of addresses
1503 allocated by mechanisms provided by LLVM.</li>
1506 <p>A pointer value is <i>based</i> on another pointer value according
1507 to the following rules:</p>
1510 <li>A pointer value formed from a
1511 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1512 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1513 <li>The result value of a
1514 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1515 of the <tt>bitcast</tt>.</li>
1516 <li>A pointer value formed by an
1517 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1518 pointer values that contribute (directly or indirectly) to the
1519 computation of the pointer's value.</li>
1520 <li>The "<i>based</i> on" relationship is transitive.</li>
1523 <p>Note that this definition of <i>"based"</i> is intentionally
1524 similar to the definition of <i>"based"</i> in C99, though it is
1525 slightly weaker.</p>
1527 <p>LLVM IR does not associate types with memory. The result type of a
1528 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1529 alignment of the memory from which to load, as well as the
1530 interpretation of the value. The first operand type of a
1531 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1532 and alignment of the store.</p>
1534 <p>Consequently, type-based alias analysis, aka TBAA, aka
1535 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1536 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1537 additional information which specialized optimization passes may use
1538 to implement type-based alias analysis.</p>
1542 <!-- ======================================================================= -->
1544 <a name="volatile">Volatile Memory Accesses</a>
1549 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1550 href="#i_store"><tt>store</tt></a>s, and <a
1551 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1552 The optimizers must not change the number of volatile operations or change their
1553 order of execution relative to other volatile operations. The optimizers
1554 <i>may</i> change the order of volatile operations relative to non-volatile
1555 operations. This is not Java's "volatile" and has no cross-thread
1556 synchronization behavior.</p>
1560 <!-- ======================================================================= -->
1562 <a name="memmodel">Memory Model for Concurrent Operations</a>
1567 <p>The LLVM IR does not define any way to start parallel threads of execution
1568 or to register signal handlers. Nonetheless, there are platform-specific
1569 ways to create them, and we define LLVM IR's behavior in their presence. This
1570 model is inspired by the C++0x memory model.</p>
1572 <p>For a more informal introduction to this model, see the
1573 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1575 <p>We define a <i>happens-before</i> partial order as the least partial order
1578 <li>Is a superset of single-thread program order, and</li>
1579 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1580 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1581 by platform-specific techniques, like pthread locks, thread
1582 creation, thread joining, etc., and by atomic instructions.
1583 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1587 <p>Note that program order does not introduce <i>happens-before</i> edges
1588 between a thread and signals executing inside that thread.</p>
1590 <p>Every (defined) read operation (load instructions, memcpy, atomic
1591 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1592 (defined) write operations (store instructions, atomic
1593 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1594 initialized globals are considered to have a write of the initializer which is
1595 atomic and happens before any other read or write of the memory in question.
1596 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1597 any write to the same byte, except:</p>
1600 <li>If <var>write<sub>1</sub></var> happens before
1601 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1602 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1603 does not see <var>write<sub>1</sub></var>.
1604 <li>If <var>R<sub>byte</sub></var> happens before
1605 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1606 see <var>write<sub>3</sub></var>.
1609 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1611 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1612 is supposed to give guarantees which can support
1613 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1614 addresses which do not behave like normal memory. It does not generally
1615 provide cross-thread synchronization.)
1616 <li>Otherwise, if there is no write to the same byte that happens before
1617 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1618 <tt>undef</tt> for that byte.
1619 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1620 <var>R<sub>byte</sub></var> returns the value written by that
1622 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1623 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1624 values written. See the <a href="#ordering">Atomic Memory Ordering
1625 Constraints</a> section for additional constraints on how the choice
1627 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1630 <p><var>R</var> returns the value composed of the series of bytes it read.
1631 This implies that some bytes within the value may be <tt>undef</tt>
1632 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1633 defines the semantics of the operation; it doesn't mean that targets will
1634 emit more than one instruction to read the series of bytes.</p>
1636 <p>Note that in cases where none of the atomic intrinsics are used, this model
1637 places only one restriction on IR transformations on top of what is required
1638 for single-threaded execution: introducing a store to a byte which might not
1639 otherwise be stored is not allowed in general. (Specifically, in the case
1640 where another thread might write to and read from an address, introducing a
1641 store can change a load that may see exactly one write into a load that may
1642 see multiple writes.)</p>
1644 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1645 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1646 none of the backends currently in the tree fall into this category; however,
1647 there might be targets which care. If there are, we want a paragraph
1650 Targets may specify that stores narrower than a certain width are not
1651 available; on such a target, for the purposes of this model, treat any
1652 non-atomic write with an alignment or width less than the minimum width
1653 as if it writes to the relevant surrounding bytes.
1658 <!-- ======================================================================= -->
1660 <a name="ordering">Atomic Memory Ordering Constraints</a>
1665 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1666 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1667 <a href="#i_fence"><code>fence</code></a>,
1668 <a href="#i_load"><code>atomic load</code></a>, and
1669 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1670 that determines which other atomic instructions on the same address they
1671 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1672 but are somewhat more colloquial. If these descriptions aren't precise enough,
1673 check those specs (see spec references in the
1674 <a href="Atomics.html#introduction">atomics guide</a>).
1675 <a href="#i_fence"><code>fence</code></a> instructions
1676 treat these orderings somewhat differently since they don't take an address.
1677 See that instruction's documentation for details.</p>
1679 <p>For a simpler introduction to the ordering constraints, see the
1680 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1683 <dt><code>unordered</code></dt>
1684 <dd>The set of values that can be read is governed by the happens-before
1685 partial order. A value cannot be read unless some operation wrote it.
1686 This is intended to provide a guarantee strong enough to model Java's
1687 non-volatile shared variables. This ordering cannot be specified for
1688 read-modify-write operations; it is not strong enough to make them atomic
1689 in any interesting way.</dd>
1690 <dt><code>monotonic</code></dt>
1691 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1692 total order for modifications by <code>monotonic</code> operations on each
1693 address. All modification orders must be compatible with the happens-before
1694 order. There is no guarantee that the modification orders can be combined to
1695 a global total order for the whole program (and this often will not be
1696 possible). The read in an atomic read-modify-write operation
1697 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1698 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1699 reads the value in the modification order immediately before the value it
1700 writes. If one atomic read happens before another atomic read of the same
1701 address, the later read must see the same value or a later value in the
1702 address's modification order. This disallows reordering of
1703 <code>monotonic</code> (or stronger) operations on the same address. If an
1704 address is written <code>monotonic</code>ally by one thread, and other threads
1705 <code>monotonic</code>ally read that address repeatedly, the other threads must
1706 eventually see the write. This corresponds to the C++0x/C1x
1707 <code>memory_order_relaxed</code>.</dd>
1708 <dt><code>acquire</code></dt>
1709 <dd>In addition to the guarantees of <code>monotonic</code>,
1710 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1711 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1712 <dt><code>release</code></dt>
1713 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1714 writes a value which is subsequently read by an <code>acquire</code> operation,
1715 it <i>synchronizes-with</i> that operation. (This isn't a complete
1716 description; see the C++0x definition of a release sequence.) This corresponds
1717 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1718 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1719 <code>acquire</code> and <code>release</code> operation on its address.
1720 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1721 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1722 <dd>In addition to the guarantees of <code>acq_rel</code>
1723 (<code>acquire</code> for an operation which only reads, <code>release</code>
1724 for an operation which only writes), there is a global total order on all
1725 sequentially-consistent operations on all addresses, which is consistent with
1726 the <i>happens-before</i> partial order and with the modification orders of
1727 all the affected addresses. Each sequentially-consistent read sees the last
1728 preceding write to the same address in this global order. This corresponds
1729 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1732 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1733 it only <i>synchronizes with</i> or participates in modification and seq_cst
1734 total orderings with other operations running in the same thread (for example,
1735 in signal handlers).</p>
1741 <!-- *********************************************************************** -->
1742 <h2><a name="typesystem">Type System</a></h2>
1743 <!-- *********************************************************************** -->
1747 <p>The LLVM type system is one of the most important features of the
1748 intermediate representation. Being typed enables a number of optimizations
1749 to be performed on the intermediate representation directly, without having
1750 to do extra analyses on the side before the transformation. A strong type
1751 system makes it easier to read the generated code and enables novel analyses
1752 and transformations that are not feasible to perform on normal three address
1753 code representations.</p>
1755 <!-- ======================================================================= -->
1757 <a name="t_classifications">Type Classifications</a>
1762 <p>The types fall into a few useful classifications:</p>
1764 <table border="1" cellspacing="0" cellpadding="4">
1766 <tr><th>Classification</th><th>Types</th></tr>
1768 <td><a href="#t_integer">integer</a></td>
1769 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1772 <td><a href="#t_floating">floating point</a></td>
1773 <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1776 <td><a name="t_firstclass">first class</a></td>
1777 <td><a href="#t_integer">integer</a>,
1778 <a href="#t_floating">floating point</a>,
1779 <a href="#t_pointer">pointer</a>,
1780 <a href="#t_vector">vector</a>,
1781 <a href="#t_struct">structure</a>,
1782 <a href="#t_array">array</a>,
1783 <a href="#t_label">label</a>,
1784 <a href="#t_metadata">metadata</a>.
1788 <td><a href="#t_primitive">primitive</a></td>
1789 <td><a href="#t_label">label</a>,
1790 <a href="#t_void">void</a>,
1791 <a href="#t_integer">integer</a>,
1792 <a href="#t_floating">floating point</a>,
1793 <a href="#t_x86mmx">x86mmx</a>,
1794 <a href="#t_metadata">metadata</a>.</td>
1797 <td><a href="#t_derived">derived</a></td>
1798 <td><a href="#t_array">array</a>,
1799 <a href="#t_function">function</a>,
1800 <a href="#t_pointer">pointer</a>,
1801 <a href="#t_struct">structure</a>,
1802 <a href="#t_vector">vector</a>,
1803 <a href="#t_opaque">opaque</a>.
1809 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1810 important. Values of these types are the only ones which can be produced by
1815 <!-- ======================================================================= -->
1817 <a name="t_primitive">Primitive Types</a>
1822 <p>The primitive types are the fundamental building blocks of the LLVM
1825 <!-- _______________________________________________________________________ -->
1827 <a name="t_integer">Integer Type</a>
1833 <p>The integer type is a very simple type that simply specifies an arbitrary
1834 bit width for the integer type desired. Any bit width from 1 bit to
1835 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1842 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1846 <table class="layout">
1848 <td class="left"><tt>i1</tt></td>
1849 <td class="left">a single-bit integer.</td>
1852 <td class="left"><tt>i32</tt></td>
1853 <td class="left">a 32-bit integer.</td>
1856 <td class="left"><tt>i1942652</tt></td>
1857 <td class="left">a really big integer of over 1 million bits.</td>
1863 <!-- _______________________________________________________________________ -->
1865 <a name="t_floating">Floating Point Types</a>
1872 <tr><th>Type</th><th>Description</th></tr>
1873 <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr>
1874 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1875 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1876 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1877 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1878 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1884 <!-- _______________________________________________________________________ -->
1886 <a name="t_x86mmx">X86mmx Type</a>
1892 <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>
1901 <!-- _______________________________________________________________________ -->
1903 <a name="t_void">Void Type</a>
1909 <p>The void type does not represent any value and has no size.</p>
1918 <!-- _______________________________________________________________________ -->
1920 <a name="t_label">Label Type</a>
1926 <p>The label type represents code labels.</p>
1935 <!-- _______________________________________________________________________ -->
1937 <a name="t_metadata">Metadata Type</a>
1943 <p>The metadata type represents embedded metadata. No derived types may be
1944 created from metadata except for <a href="#t_function">function</a>
1956 <!-- ======================================================================= -->
1958 <a name="t_derived">Derived Types</a>
1963 <p>The real power in LLVM comes from the derived types in the system. This is
1964 what allows a programmer to represent arrays, functions, pointers, and other
1965 useful types. Each of these types contain one or more element types which
1966 may be a primitive type, or another derived type. For example, it is
1967 possible to have a two dimensional array, using an array as the element type
1968 of another array.</p>
1970 <!-- _______________________________________________________________________ -->
1972 <a name="t_aggregate">Aggregate Types</a>
1977 <p>Aggregate Types are a subset of derived types that can contain multiple
1978 member types. <a href="#t_array">Arrays</a> and
1979 <a href="#t_struct">structs</a> are aggregate types.
1980 <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p>
1984 <!-- _______________________________________________________________________ -->
1986 <a name="t_array">Array Type</a>
1992 <p>The array type is a very simple derived type that arranges elements
1993 sequentially in memory. The array type requires a size (number of elements)
1994 and an underlying data type.</p>
1998 [<# elements> x <elementtype>]
2001 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
2002 be any type with a size.</p>
2005 <table class="layout">
2007 <td class="left"><tt>[40 x i32]</tt></td>
2008 <td class="left">Array of 40 32-bit integer values.</td>
2011 <td class="left"><tt>[41 x i32]</tt></td>
2012 <td class="left">Array of 41 32-bit integer values.</td>
2015 <td class="left"><tt>[4 x i8]</tt></td>
2016 <td class="left">Array of 4 8-bit integer values.</td>
2019 <p>Here are some examples of multidimensional arrays:</p>
2020 <table class="layout">
2022 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
2023 <td class="left">3x4 array of 32-bit integer values.</td>
2026 <td class="left"><tt>[12 x [10 x float]]</tt></td>
2027 <td class="left">12x10 array of single precision floating point values.</td>
2030 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
2031 <td class="left">2x3x4 array of 16-bit integer values.</td>
2035 <p>There is no restriction on indexing beyond the end of the array implied by
2036 a static type (though there are restrictions on indexing beyond the bounds
2037 of an allocated object in some cases). This means that single-dimension
2038 'variable sized array' addressing can be implemented in LLVM with a zero
2039 length array type. An implementation of 'pascal style arrays' in LLVM could
2040 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
2044 <!-- _______________________________________________________________________ -->
2046 <a name="t_function">Function Type</a>
2052 <p>The function type can be thought of as a function signature. It consists of
2053 a return type and a list of formal parameter types. The return type of a
2054 function type is a first class type or a void type.</p>
2058 <returntype> (<parameter list>)
2061 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
2062 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
2063 which indicates that the function takes a variable number of arguments.
2064 Variable argument functions can access their arguments with
2065 the <a href="#int_varargs">variable argument handling intrinsic</a>
2066 functions. '<tt><returntype></tt>' is any type except
2067 <a href="#t_label">label</a>.</p>
2070 <table class="layout">
2072 <td class="left"><tt>i32 (i32)</tt></td>
2073 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2075 </tr><tr class="layout">
2076 <td class="left"><tt>float (i16, i32 *) *
2078 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2079 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2080 returning <tt>float</tt>.
2082 </tr><tr class="layout">
2083 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2084 <td class="left">A vararg function that takes at least one
2085 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2086 which returns an integer. This is the signature for <tt>printf</tt> in
2089 </tr><tr class="layout">
2090 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2091 <td class="left">A function taking an <tt>i32</tt>, returning a
2092 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2099 <!-- _______________________________________________________________________ -->
2101 <a name="t_struct">Structure Type</a>
2107 <p>The structure type is used to represent a collection of data members together
2108 in memory. The elements of a structure may be any type that has a size.</p>
2110 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2111 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2112 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2113 Structures in registers are accessed using the
2114 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2115 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2117 <p>Structures may optionally be "packed" structures, which indicate that the
2118 alignment of the struct is one byte, and that there is no padding between
2119 the elements. In non-packed structs, padding between field types is inserted
2120 as defined by the DataLayout string in the module, which is required to match
2121 what the underlying code generator expects.</p>
2123 <p>Structures can either be "literal" or "identified". A literal structure is
2124 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2125 types are always defined at the top level with a name. Literal types are
2126 uniqued by their contents and can never be recursive or opaque since there is
2127 no way to write one. Identified types can be recursive, can be opaqued, and are
2133 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2134 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2138 <table class="layout">
2140 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2141 <td class="left">A triple of three <tt>i32</tt> values</td>
2144 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2145 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2146 second element is a <a href="#t_pointer">pointer</a> to a
2147 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2148 an <tt>i32</tt>.</td>
2151 <td class="left"><tt><{ i8, i32 }></tt></td>
2152 <td class="left">A packed struct known to be 5 bytes in size.</td>
2158 <!-- _______________________________________________________________________ -->
2160 <a name="t_opaque">Opaque Structure Types</a>
2166 <p>Opaque structure types are used to represent named structure types that do
2167 not have a body specified. This corresponds (for example) to the C notion of
2168 a forward declared structure.</p>
2177 <table class="layout">
2179 <td class="left"><tt>opaque</tt></td>
2180 <td class="left">An opaque type.</td>
2188 <!-- _______________________________________________________________________ -->
2190 <a name="t_pointer">Pointer Type</a>
2196 <p>The pointer type is used to specify memory locations.
2197 Pointers are commonly used to reference objects in memory.</p>
2199 <p>Pointer types may have an optional address space attribute defining the
2200 numbered address space where the pointed-to object resides. The default
2201 address space is number zero. The semantics of non-zero address
2202 spaces are target-specific.</p>
2204 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2205 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2213 <table class="layout">
2215 <td class="left"><tt>[4 x i32]*</tt></td>
2216 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2217 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2220 <td class="left"><tt>i32 (i32*) *</tt></td>
2221 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2222 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2226 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2227 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2228 that resides in address space #5.</td>
2234 <!-- _______________________________________________________________________ -->
2236 <a name="t_vector">Vector Type</a>
2242 <p>A vector type is a simple derived type that represents a vector of elements.
2243 Vector types are used when multiple primitive data are operated in parallel
2244 using a single instruction (SIMD). A vector type requires a size (number of
2245 elements) and an underlying primitive data type. Vector types are considered
2246 <a href="#t_firstclass">first class</a>.</p>
2250 < <# elements> x <elementtype> >
2253 <p>The number of elements is a constant integer value larger than 0; elementtype
2254 may be any integer or floating point type, or a pointer to these types.
2255 Vectors of size zero are not allowed. </p>
2258 <table class="layout">
2260 <td class="left"><tt><4 x i32></tt></td>
2261 <td class="left">Vector of 4 32-bit integer values.</td>
2264 <td class="left"><tt><8 x float></tt></td>
2265 <td class="left">Vector of 8 32-bit floating-point values.</td>
2268 <td class="left"><tt><2 x i64></tt></td>
2269 <td class="left">Vector of 2 64-bit integer values.</td>
2272 <td class="left"><tt><4 x i64*></tt></td>
2273 <td class="left">Vector of 4 pointers to 64-bit integer values.</td>
2283 <!-- *********************************************************************** -->
2284 <h2><a name="constants">Constants</a></h2>
2285 <!-- *********************************************************************** -->
2289 <p>LLVM has several different basic types of constants. This section describes
2290 them all and their syntax.</p>
2292 <!-- ======================================================================= -->
2294 <a name="simpleconstants">Simple Constants</a>
2300 <dt><b>Boolean constants</b></dt>
2301 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2302 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2304 <dt><b>Integer constants</b></dt>
2305 <dd>Standard integers (such as '4') are constants of
2306 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2307 with integer types.</dd>
2309 <dt><b>Floating point constants</b></dt>
2310 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2311 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2312 notation (see below). The assembler requires the exact decimal value of a
2313 floating-point constant. For example, the assembler accepts 1.25 but
2314 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2315 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2317 <dt><b>Null pointer constants</b></dt>
2318 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2319 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2322 <p>The one non-intuitive notation for constants is the hexadecimal form of
2323 floating point constants. For example, the form '<tt>double
2324 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2325 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2326 constants are required (and the only time that they are generated by the
2327 disassembler) is when a floating point constant must be emitted but it cannot
2328 be represented as a decimal floating point number in a reasonable number of
2329 digits. For example, NaN's, infinities, and other special values are
2330 represented in their IEEE hexadecimal format so that assembly and disassembly
2331 do not cause any bits to change in the constants.</p>
2333 <p>When using the hexadecimal form, constants of types half, float, and double are
2334 represented using the 16-digit form shown above (which matches the IEEE754
2335 representation for double); half and float values must, however, be exactly
2336 representable as IEE754 half and single precision, respectively.
2337 Hexadecimal format is always used
2338 for long double, and there are three forms of long double. The 80-bit format
2339 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2340 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2341 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2342 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2343 currently supported target uses this format. Long doubles will only work if
2344 they match the long double format on your target. The IEEE 16-bit format
2345 (half precision) is represented by <tt>0xH</tt> followed by 4 hexadecimal
2346 digits. All hexadecimal formats are big-endian (sign bit at the left).</p>
2348 <p>There are no constants of type x86mmx.</p>
2351 <!-- ======================================================================= -->
2353 <a name="aggregateconstants"></a> <!-- old anchor -->
2354 <a name="complexconstants">Complex Constants</a>
2359 <p>Complex constants are a (potentially recursive) combination of simple
2360 constants and smaller complex constants.</p>
2363 <dt><b>Structure constants</b></dt>
2364 <dd>Structure constants are represented with notation similar to structure
2365 type definitions (a comma separated list of elements, surrounded by braces
2366 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2367 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2368 Structure constants must have <a href="#t_struct">structure type</a>, and
2369 the number and types of elements must match those specified by the
2372 <dt><b>Array constants</b></dt>
2373 <dd>Array constants are represented with notation similar to array type
2374 definitions (a comma separated list of elements, surrounded by square
2375 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2376 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2377 the number and types of elements must match those specified by the
2380 <dt><b>Vector constants</b></dt>
2381 <dd>Vector constants are represented with notation similar to vector type
2382 definitions (a comma separated list of elements, surrounded by
2383 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2384 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2385 have <a href="#t_vector">vector type</a>, and the number and types of
2386 elements must match those specified by the type.</dd>
2388 <dt><b>Zero initialization</b></dt>
2389 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2390 value to zero of <em>any</em> type, including scalar and
2391 <a href="#t_aggregate">aggregate</a> types.
2392 This is often used to avoid having to print large zero initializers
2393 (e.g. for large arrays) and is always exactly equivalent to using explicit
2394 zero initializers.</dd>
2396 <dt><b>Metadata node</b></dt>
2397 <dd>A metadata node is a structure-like constant with
2398 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2399 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2400 be interpreted as part of the instruction stream, metadata is a place to
2401 attach additional information such as debug info.</dd>
2406 <!-- ======================================================================= -->
2408 <a name="globalconstants">Global Variable and Function Addresses</a>
2413 <p>The addresses of <a href="#globalvars">global variables</a>
2414 and <a href="#functionstructure">functions</a> are always implicitly valid
2415 (link-time) constants. These constants are explicitly referenced when
2416 the <a href="#identifiers">identifier for the global</a> is used and always
2417 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2418 legal LLVM file:</p>
2420 <pre class="doc_code">
2423 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2428 <!-- ======================================================================= -->
2430 <a name="undefvalues">Undefined Values</a>
2435 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2436 indicates that the user of the value may receive an unspecified bit-pattern.
2437 Undefined values may be of any type (other than '<tt>label</tt>'
2438 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2440 <p>Undefined values are useful because they indicate to the compiler that the
2441 program is well defined no matter what value is used. This gives the
2442 compiler more freedom to optimize. Here are some examples of (potentially
2443 surprising) transformations that are valid (in pseudo IR):</p>
2446 <pre class="doc_code">
2456 <p>This is safe because all of the output bits are affected by the undef bits.
2457 Any output bit can have a zero or one depending on the input bits.</p>
2459 <pre class="doc_code">
2470 <p>These logical operations have bits that are not always affected by the input.
2471 For example, if <tt>%X</tt> has a zero bit, then the output of the
2472 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2473 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2474 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2475 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2476 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2477 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2478 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2480 <pre class="doc_code">
2481 %A = select undef, %X, %Y
2482 %B = select undef, 42, %Y
2483 %C = select %X, %Y, undef
2494 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2495 branch) conditions can go <em>either way</em>, but they have to come from one
2496 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2497 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2498 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2499 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2500 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2503 <pre class="doc_code">
2504 %A = xor undef, undef
2522 <p>This example points out that two '<tt>undef</tt>' operands are not
2523 necessarily the same. This can be surprising to people (and also matches C
2524 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2525 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2526 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2527 its value over its "live range". This is true because the variable doesn't
2528 actually <em>have a live range</em>. Instead, the value is logically read
2529 from arbitrary registers that happen to be around when needed, so the value
2530 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2531 need to have the same semantics or the core LLVM "replace all uses with"
2532 concept would not hold.</p>
2534 <pre class="doc_code">
2542 <p>These examples show the crucial difference between an <em>undefined
2543 value</em> and <em>undefined behavior</em>. An undefined value (like
2544 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2545 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2546 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2547 defined on SNaN's. However, in the second example, we can make a more
2548 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2549 arbitrary value, we are allowed to assume that it could be zero. Since a
2550 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2551 the operation does not execute at all. This allows us to delete the divide and
2552 all code after it. Because the undefined operation "can't happen", the
2553 optimizer can assume that it occurs in dead code.</p>
2555 <pre class="doc_code">
2556 a: store undef -> %X
2557 b: store %X -> undef
2563 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2564 undefined value can be assumed to not have any effect; we can assume that the
2565 value is overwritten with bits that happen to match what was already there.
2566 However, a store <em>to</em> an undefined location could clobber arbitrary
2567 memory, therefore, it has undefined behavior.</p>
2571 <!-- ======================================================================= -->
2573 <a name="poisonvalues">Poison Values</a>
2578 <p>Poison values are similar to <a href="#undefvalues">undef values</a>, however
2579 they also represent the fact that an instruction or constant expression which
2580 cannot evoke side effects has nevertheless detected a condition which results
2581 in undefined behavior.</p>
2583 <p>There is currently no way of representing a poison value in the IR; they
2584 only exist when produced by operations such as
2585 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2587 <p>Poison value behavior is defined in terms of value <i>dependence</i>:</p>
2590 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2591 their operands.</li>
2593 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2594 to their dynamic predecessor basic block.</li>
2596 <li>Function arguments depend on the corresponding actual argument values in
2597 the dynamic callers of their functions.</li>
2599 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2600 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2601 control back to them.</li>
2603 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2604 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>,
2605 or exception-throwing call instructions that dynamically transfer control
2608 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2609 referenced memory addresses, following the order in the IR
2610 (including loads and stores implied by intrinsics such as
2611 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2613 <!-- TODO: In the case of multiple threads, this only applies if the store
2614 "happens-before" the load or store. -->
2616 <!-- TODO: floating-point exception state -->
2618 <li>An instruction with externally visible side effects depends on the most
2619 recent preceding instruction with externally visible side effects, following
2620 the order in the IR. (This includes
2621 <a href="#volatile">volatile operations</a>.)</li>
2623 <li>An instruction <i>control-depends</i> on a
2624 <a href="#terminators">terminator instruction</a>
2625 if the terminator instruction has multiple successors and the instruction
2626 is always executed when control transfers to one of the successors, and
2627 may not be executed when control is transferred to another.</li>
2629 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2630 instruction if the set of instructions it otherwise depends on would be
2631 different if the terminator had transferred control to a different
2634 <li>Dependence is transitive.</li>
2638 <p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>,
2639 with the additional affect that any instruction which has a <i>dependence</i>
2640 on a poison value has undefined behavior.</p>
2642 <p>Here are some examples:</p>
2644 <pre class="doc_code">
2646 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2647 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2648 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2649 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2651 store i32 %poison, i32* @g ; Poison value stored to memory.
2652 %poison2 = load i32* @g ; Poison value loaded back from memory.
2654 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2656 %narrowaddr = bitcast i32* @g to i16*
2657 %wideaddr = bitcast i32* @g to i64*
2658 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2659 %poison4 = load i64* %wideaddr ; Returns a poison value.
2661 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2662 br i1 %cmp, label %true, label %end ; Branch to either destination.
2665 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2666 ; it has undefined behavior.
2670 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2671 ; Both edges into this PHI are
2672 ; control-dependent on %cmp, so this
2673 ; always results in a poison value.
2675 store volatile i32 0, i32* @g ; This would depend on the store in %true
2676 ; if %cmp is true, or the store in %entry
2677 ; otherwise, so this is undefined behavior.
2679 br i1 %cmp, label %second_true, label %second_end
2680 ; The same branch again, but this time the
2681 ; true block doesn't have side effects.
2688 store volatile i32 0, i32* @g ; This time, the instruction always depends
2689 ; on the store in %end. Also, it is
2690 ; control-equivalent to %end, so this is
2691 ; well-defined (ignoring earlier undefined
2692 ; behavior in this example).
2697 <!-- ======================================================================= -->
2699 <a name="blockaddress">Addresses of Basic Blocks</a>
2704 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2706 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2707 basic block in the specified function, and always has an i8* type. Taking
2708 the address of the entry block is illegal.</p>
2710 <p>This value only has defined behavior when used as an operand to the
2711 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2712 comparisons against null. Pointer equality tests between labels addresses
2713 results in undefined behavior — though, again, comparison against null
2714 is ok, and no label is equal to the null pointer. This may be passed around
2715 as an opaque pointer sized value as long as the bits are not inspected. This
2716 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2717 long as the original value is reconstituted before the <tt>indirectbr</tt>
2720 <p>Finally, some targets may provide defined semantics when using the value as
2721 the operand to an inline assembly, but that is target specific.</p>
2726 <!-- ======================================================================= -->
2728 <a name="constantexprs">Constant Expressions</a>
2733 <p>Constant expressions are used to allow expressions involving other constants
2734 to be used as constants. Constant expressions may be of
2735 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2736 operation that does not have side effects (e.g. load and call are not
2737 supported). The following is the syntax for constant expressions:</p>
2740 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2741 <dd>Truncate a constant to another type. The bit size of CST must be larger
2742 than the bit size of TYPE. Both types must be integers.</dd>
2744 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2745 <dd>Zero extend a constant to another type. The bit size of CST must be
2746 smaller than the bit size of TYPE. Both types must be integers.</dd>
2748 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2749 <dd>Sign extend a constant to another type. The bit size of CST must be
2750 smaller than the bit size of TYPE. Both types must be integers.</dd>
2752 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2753 <dd>Truncate a floating point constant to another floating point type. The
2754 size of CST must be larger than the size of TYPE. Both types must be
2755 floating point.</dd>
2757 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2758 <dd>Floating point extend a constant to another type. The size of CST must be
2759 smaller or equal to the size of TYPE. Both types must be floating
2762 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2763 <dd>Convert a floating point constant to the corresponding unsigned integer
2764 constant. TYPE must be a scalar or vector integer type. CST must be of
2765 scalar or vector floating point type. Both CST and TYPE must be scalars,
2766 or vectors of the same number of elements. If the value won't fit in the
2767 integer type, the results are undefined.</dd>
2769 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2770 <dd>Convert a floating point constant to the corresponding signed integer
2771 constant. TYPE must be a scalar or vector integer type. CST must be of
2772 scalar or vector floating point type. Both CST and TYPE must be scalars,
2773 or vectors of the same number of elements. If the value won't fit in the
2774 integer type, the results are undefined.</dd>
2776 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2777 <dd>Convert an unsigned integer constant to the corresponding floating point
2778 constant. TYPE must be a scalar or vector floating point type. CST must be
2779 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2780 vectors of the same number of elements. If the value won't fit in the
2781 floating point type, the results are undefined.</dd>
2783 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2784 <dd>Convert a signed integer constant to the corresponding floating point
2785 constant. TYPE must be a scalar or vector floating point type. CST must be
2786 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2787 vectors of the same number of elements. If the value won't fit in the
2788 floating point type, the results are undefined.</dd>
2790 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2791 <dd>Convert a pointer typed constant to the corresponding integer constant
2792 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2793 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2794 make it fit in <tt>TYPE</tt>.</dd>
2796 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2797 <dd>Convert an integer constant to a pointer constant. TYPE must be a pointer
2798 type. CST must be of integer type. The CST value is zero extended,
2799 truncated, or unchanged to make it fit in a pointer size. This one is
2800 <i>really</i> dangerous!</dd>
2802 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2803 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2804 are the same as those for the <a href="#i_bitcast">bitcast
2805 instruction</a>.</dd>
2807 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2808 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2809 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2810 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2811 instruction, the index list may have zero or more indexes, which are
2812 required to make sense for the type of "CSTPTR".</dd>
2814 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2815 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2817 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2818 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2820 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2821 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2823 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2824 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2827 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2828 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2831 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2832 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2835 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2836 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2837 constants. The index list is interpreted in a similar manner as indices in
2838 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2839 index value must be specified.</dd>
2841 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2842 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2843 constants. The index list is interpreted in a similar manner as indices in
2844 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2845 index value must be specified.</dd>
2847 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2848 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2849 be any of the <a href="#binaryops">binary</a>
2850 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2851 on operands are the same as those for the corresponding instruction
2852 (e.g. no bitwise operations on floating point values are allowed).</dd>
2859 <!-- *********************************************************************** -->
2860 <h2><a name="othervalues">Other Values</a></h2>
2861 <!-- *********************************************************************** -->
2863 <!-- ======================================================================= -->
2865 <a name="inlineasm">Inline Assembler Expressions</a>
2870 <p>LLVM supports inline assembler expressions (as opposed
2871 to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of
2872 a special value. This value represents the inline assembler as a string
2873 (containing the instructions to emit), a list of operand constraints (stored
2874 as a string), a flag that indicates whether or not the inline asm
2875 expression has side effects, and a flag indicating whether the function
2876 containing the asm needs to align its stack conservatively. An example
2877 inline assembler expression is:</p>
2879 <pre class="doc_code">
2880 i32 (i32) asm "bswap $0", "=r,r"
2883 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2884 a <a href="#i_call"><tt>call</tt></a> or an
2885 <a href="#i_invoke"><tt>invoke</tt></a> instruction.
2886 Thus, typically we have:</p>
2888 <pre class="doc_code">
2889 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2892 <p>Inline asms with side effects not visible in the constraint list must be
2893 marked as having side effects. This is done through the use of the
2894 '<tt>sideeffect</tt>' keyword, like so:</p>
2896 <pre class="doc_code">
2897 call void asm sideeffect "eieio", ""()
2900 <p>In some cases inline asms will contain code that will not work unless the
2901 stack is aligned in some way, such as calls or SSE instructions on x86,
2902 yet will not contain code that does that alignment within the asm.
2903 The compiler should make conservative assumptions about what the asm might
2904 contain and should generate its usual stack alignment code in the prologue
2905 if the '<tt>alignstack</tt>' keyword is present:</p>
2907 <pre class="doc_code">
2908 call void asm alignstack "eieio", ""()
2911 <p>Inline asms also support using non-standard assembly dialects. The assumed
2912 dialect is ATT. When the '<tt>inteldialect</tt>' keyword is present, the
2913 inline asm is using the Intel dialect. Currently, ATT and Intel are the
2914 only supported dialects. An example is:</p>
2916 <pre class="doc_code">
2917 call void asm inteldialect "eieio", ""()
2920 <p>If multiple keywords appear the '<tt>sideeffect</tt>' keyword must come
2921 first, the '<tt>alignstack</tt>' keyword second and the
2922 '<tt>inteldialect</tt>' keyword last.</p>
2925 <p>TODO: The format of the asm and constraints string still need to be
2926 documented here. Constraints on what can be done (e.g. duplication, moving,
2927 etc need to be documented). This is probably best done by reference to
2928 another document that covers inline asm from a holistic perspective.</p>
2931 <!-- _______________________________________________________________________ -->
2933 <a name="inlineasm_md">Inline Asm Metadata</a>
2938 <p>The call instructions that wrap inline asm nodes may have a
2939 "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant
2940 integers. If present, the code generator will use the integer as the
2941 location cookie value when report errors through the <tt>LLVMContext</tt>
2942 error reporting mechanisms. This allows a front-end to correlate backend
2943 errors that occur with inline asm back to the source code that produced it.
2946 <pre class="doc_code">
2947 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2949 !42 = !{ i32 1234567 }
2952 <p>It is up to the front-end to make sense of the magic numbers it places in the
2953 IR. If the MDNode contains multiple constants, the code generator will use
2954 the one that corresponds to the line of the asm that the error occurs on.</p>
2960 <!-- ======================================================================= -->
2962 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2967 <p>LLVM IR allows metadata to be attached to instructions in the program that
2968 can convey extra information about the code to the optimizers and code
2969 generator. One example application of metadata is source-level debug
2970 information. There are two metadata primitives: strings and nodes. All
2971 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2972 preceding exclamation point ('<tt>!</tt>').</p>
2974 <p>A metadata string is a string surrounded by double quotes. It can contain
2975 any character by escaping non-printable characters with "<tt>\xx</tt>" where
2976 "<tt>xx</tt>" is the two digit hex code. For example:
2977 "<tt>!"test\00"</tt>".</p>
2979 <p>Metadata nodes are represented with notation similar to structure constants
2980 (a comma separated list of elements, surrounded by braces and preceded by an
2981 exclamation point). Metadata nodes can have any values as their operand. For
2984 <div class="doc_code">
2986 !{ metadata !"test\00", i32 10}
2990 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2991 metadata nodes, which can be looked up in the module symbol table. For
2994 <div class="doc_code">
2996 !foo = metadata !{!4, !3}
3000 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
3001 function is using two metadata arguments:</p>
3003 <div class="doc_code">
3005 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3009 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
3010 attached to the <tt>add</tt> instruction using the <tt>!dbg</tt>
3013 <div class="doc_code">
3015 %indvar.next = add i64 %indvar, 1, !dbg !21
3019 <p>More information about specific metadata nodes recognized by the optimizers
3020 and code generator is found below.</p>
3022 <!-- _______________________________________________________________________ -->
3024 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
3029 <p>In LLVM IR, memory does not have types, so LLVM's own type system is not
3030 suitable for doing TBAA. Instead, metadata is added to the IR to describe
3031 a type system of a higher level language. This can be used to implement
3032 typical C/C++ TBAA, but it can also be used to implement custom alias
3033 analysis behavior for other languages.</p>
3035 <p>The current metadata format is very simple. TBAA metadata nodes have up to
3036 three fields, e.g.:</p>
3038 <div class="doc_code">
3040 !0 = metadata !{ metadata !"an example type tree" }
3041 !1 = metadata !{ metadata !"int", metadata !0 }
3042 !2 = metadata !{ metadata !"float", metadata !0 }
3043 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
3047 <p>The first field is an identity field. It can be any value, usually
3048 a metadata string, which uniquely identifies the type. The most important
3049 name in the tree is the name of the root node. Two trees with
3050 different root node names are entirely disjoint, even if they
3051 have leaves with common names.</p>
3053 <p>The second field identifies the type's parent node in the tree, or
3054 is null or omitted for a root node. A type is considered to alias
3055 all of its descendants and all of its ancestors in the tree. Also,
3056 a type is considered to alias all types in other trees, so that
3057 bitcode produced from multiple front-ends is handled conservatively.</p>
3059 <p>If the third field is present, it's an integer which if equal to 1
3060 indicates that the type is "constant" (meaning
3061 <tt>pointsToConstantMemory</tt> should return true; see
3062 <a href="AliasAnalysis.html#OtherItfs">other useful
3063 <tt>AliasAnalysis</tt> methods</a>).</p>
3067 <!-- _______________________________________________________________________ -->
3069 <a name="tbaa.struct">'<tt>tbaa.struct</tt>' Metadata</a>
3074 <p>The <a href="#int_memcpy"><tt>llvm.memcpy</tt></a> is often used to implement
3075 aggregate assignment operations in C and similar languages, however it is
3076 defined to copy a contiguous region of memory, which is more than strictly
3077 necessary for aggregate types which contain holes due to padding. Also, it
3078 doesn't contain any TBAA information about the fields of the aggregate.</p>
3080 <p><tt>!tbaa.struct</tt> metadata can describe which memory subregions in a memcpy
3081 are padding and what the TBAA tags of the struct are.</p>
3083 <p>The current metadata format is very simple. <tt>!tbaa.struct</tt> metadata nodes
3084 are a list of operands which are in conceptual groups of three. For each
3085 group of three, the first operand gives the byte offset of a field in bytes,
3086 the second gives its size in bytes, and the third gives its
3089 <div class="doc_code">
3091 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
3095 <p>This describes a struct with two fields. The first is at offset 0 bytes
3096 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
3097 and has size 4 bytes and has tbaa tag !2.</p>
3099 <p>Note that the fields need not be contiguous. In this example, there is a
3100 4 byte gap between the two fields. This gap represents padding which
3101 does not carry useful data and need not be preserved.</p>
3105 <!-- _______________________________________________________________________ -->
3107 <a name="fpmath">'<tt>fpmath</tt>' Metadata</a>
3112 <p><tt>fpmath</tt> metadata may be attached to any instruction of floating point
3113 type. It can be used to express the maximum acceptable error in the result of
3114 that instruction, in ULPs, thus potentially allowing the compiler to use a
3115 more efficient but less accurate method of computing it. ULP is defined as
3120 <p>If <tt>x</tt> is a real number that lies between two finite consecutive
3121 floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one
3122 of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the
3123 distance between the two non-equal finite floating-point numbers nearest
3124 <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p>
3128 <p>The metadata node shall consist of a single positive floating point number
3129 representing the maximum relative error, for example:</p>
3131 <div class="doc_code">
3133 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3139 <!-- _______________________________________________________________________ -->
3141 <a name="range">'<tt>range</tt>' Metadata</a>
3145 <p><tt>range</tt> metadata may be attached only to loads of integer types. It
3146 expresses the possible ranges the loaded value is in. The ranges are
3147 represented with a flattened list of integers. The loaded value is known to
3148 be in the union of the ranges defined by each consecutive pair. Each pair
3149 has the following properties:</p>
3151 <li>The type must match the type loaded by the instruction.</li>
3152 <li>The pair <tt>a,b</tt> represents the range <tt>[a,b)</tt>.</li>
3153 <li>Both <tt>a</tt> and <tt>b</tt> are constants.</li>
3154 <li>The range is allowed to wrap.</li>
3155 <li>The range should not represent the full or empty set. That is,
3156 <tt>a!=b</tt>. </li>
3158 <p> In addition, the pairs must be in signed order of the lower bound and
3159 they must be non-contiguous.</p>
3162 <div class="doc_code">
3164 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
3165 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3166 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
3167 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
3169 !0 = metadata !{ i8 0, i8 2 }
3170 !1 = metadata !{ i8 255, i8 2 }
3171 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
3172 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
3180 <!-- *********************************************************************** -->
3182 <a name="module_flags">Module Flags Metadata</a>
3184 <!-- *********************************************************************** -->
3188 <p>Information about the module as a whole is difficult to convey to LLVM's
3189 subsystems. The LLVM IR isn't sufficient to transmit this
3190 information. The <tt>llvm.module.flags</tt> named metadata exists in order to
3191 facilitate this. These flags are in the form of key / value pairs —
3192 much like a dictionary — making it easy for any subsystem who cares
3193 about a flag to look it up.</p>
3195 <p>The <tt>llvm.module.flags</tt> metadata contains a list of metadata
3196 triplets. Each triplet has the following form:</p>
3199 <li>The first element is a <i>behavior</i> flag, which specifies the behavior
3200 when two (or more) modules are merged together, and it encounters two (or
3201 more) metadata with the same ID. The supported behaviors are described
3204 <li>The second element is a metadata string that is a unique ID for the
3205 metadata. How each ID is interpreted is documented below.</li>
3207 <li>The third element is the value of the flag.</li>
3210 <p>When two (or more) modules are merged together, the resulting
3211 <tt>llvm.module.flags</tt> metadata is the union of the
3212 modules' <tt>llvm.module.flags</tt> metadata. The only exception being a flag
3213 with the <i>Override</i> behavior, which may override another flag's value
3216 <p>The following behaviors are supported:</p>
3218 <table border="1" cellspacing="0" cellpadding="4">
3228 <dt><b>Error</b></dt>
3229 <dd>Emits an error if two values disagree. It is an error to have an ID
3230 with both an Error and a Warning behavior.</dd>
3238 <dt><b>Warning</b></dt>
3239 <dd>Emits a warning if two values disagree.</dd>
3247 <dt><b>Require</b></dt>
3248 <dd>Emits an error when the specified value is not present or doesn't
3249 have the specified value. It is an error for two (or more)
3250 <tt>llvm.module.flags</tt> with the same ID to have the Require
3251 behavior but different values. There may be multiple Require flags
3260 <dt><b>Override</b></dt>
3261 <dd>Uses the specified value if the two values disagree. It is an
3262 error for two (or more) <tt>llvm.module.flags</tt> with the same
3263 ID to have the Override behavior but different values.</dd>
3270 <p>An example of module flags:</p>
3272 <pre class="doc_code">
3273 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3274 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3275 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3276 !3 = metadata !{ i32 3, metadata !"qux",
3278 metadata !"foo", i32 1
3281 !llvm.module.flags = !{ !0, !1, !2, !3 }
3285 <li><p>Metadata <tt>!0</tt> has the ID <tt>!"foo"</tt> and the value '1'. The
3286 behavior if two or more <tt>!"foo"</tt> flags are seen is to emit an
3287 error if their values are not equal.</p></li>
3289 <li><p>Metadata <tt>!1</tt> has the ID <tt>!"bar"</tt> and the value '37'. The
3290 behavior if two or more <tt>!"bar"</tt> flags are seen is to use the
3291 value '37' if their values are not equal.</p></li>
3293 <li><p>Metadata <tt>!2</tt> has the ID <tt>!"qux"</tt> and the value '42'. The
3294 behavior if two or more <tt>!"qux"</tt> flags are seen is to emit a
3295 warning if their values are not equal.</p></li>
3297 <li><p>Metadata <tt>!3</tt> has the ID <tt>!"qux"</tt> and the value:</p>
3299 <pre class="doc_code">
3300 metadata !{ metadata !"foo", i32 1 }
3303 <p>The behavior is to emit an error if the <tt>llvm.module.flags</tt> does
3304 not contain a flag with the ID <tt>!"foo"</tt> that has the value
3305 '1'. If two or more <tt>!"qux"</tt> flags exist, then they must have
3306 the same value or an error will be issued.</p></li>
3310 <!-- ======================================================================= -->
3312 <a name="objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a>
3317 <p>On the Mach-O platform, Objective-C stores metadata about garbage collection
3318 in a special section called "image info". The metadata consists of a version
3319 number and a bitmask specifying what types of garbage collection are
3320 supported (if any) by the file. If two or more modules are linked together
3321 their garbage collection metadata needs to be merged rather than appended
3324 <p>The Objective-C garbage collection module flags metadata consists of the
3325 following key-value pairs:</p>
3327 <table border="1" cellspacing="0" cellpadding="4">
3335 <td><tt>Objective-C Version</tt></td>
3336 <td align="left"><b>[Required]</b> — The Objective-C ABI
3337 version. Valid values are 1 and 2.</td>
3340 <td><tt>Objective-C Image Info Version</tt></td>
3341 <td align="left"><b>[Required]</b> — The version of the image info
3342 section. Currently always 0.</td>
3345 <td><tt>Objective-C Image Info Section</tt></td>
3346 <td align="left"><b>[Required]</b> — The section to place the
3347 metadata. Valid values are <tt>"__OBJC, __image_info, regular"</tt> for
3348 Objective-C ABI version 1, and <tt>"__DATA,__objc_imageinfo, regular,
3349 no_dead_strip"</tt> for Objective-C ABI version 2.</td>
3352 <td><tt>Objective-C Garbage Collection</tt></td>
3353 <td align="left"><b>[Required]</b> — Specifies whether garbage
3354 collection is supported or not. Valid values are 0, for no garbage
3355 collection, and 2, for garbage collection supported.</td>
3358 <td><tt>Objective-C GC Only</tt></td>
3359 <td align="left"><b>[Optional]</b> — Specifies that only garbage
3360 collection is supported. If present, its value must be 6. This flag
3361 requires that the <tt>Objective-C Garbage Collection</tt> flag have the
3367 <p>Some important flag interactions:</p>
3370 <li>If a module with <tt>Objective-C Garbage Collection</tt> set to 0 is
3371 merged with a module with <tt>Objective-C Garbage Collection</tt> set to
3372 2, then the resulting module has the <tt>Objective-C Garbage
3373 Collection</tt> flag set to 0.</li>
3375 <li>A module with <tt>Objective-C Garbage Collection</tt> set to 0 cannot be
3376 merged with a module with <tt>Objective-C GC Only</tt> set to 6.</li>
3383 <!-- *********************************************************************** -->
3385 <a name="intrinsic_globals">Intrinsic Global Variables</a>
3387 <!-- *********************************************************************** -->
3389 <p>LLVM has a number of "magic" global variables that contain data that affect
3390 code generation or other IR semantics. These are documented here. All globals
3391 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
3392 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
3395 <!-- ======================================================================= -->
3397 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
3402 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
3403 href="#linkage_appending">appending linkage</a>. This array contains a list of
3404 pointers to global variables and functions which may optionally have a pointer
3405 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
3407 <div class="doc_code">
3412 @llvm.used = appending global [2 x i8*] [
3414 i8* bitcast (i32* @Y to i8*)
3415 ], section "llvm.metadata"
3419 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
3420 compiler, assembler, and linker are required to treat the symbol as if there
3421 is a reference to the global that it cannot see. For example, if a variable
3422 has internal linkage and no references other than that from
3423 the <tt>@llvm.used</tt> list, it cannot be deleted. This is commonly used to
3424 represent references from inline asms and other things the compiler cannot
3425 "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p>
3427 <p>On some targets, the code generator must emit a directive to the assembler or
3428 object file to prevent the assembler and linker from molesting the
3433 <!-- ======================================================================= -->
3435 <a name="intg_compiler_used">
3436 The '<tt>llvm.compiler.used</tt>' Global Variable
3442 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
3443 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
3444 touching the symbol. On targets that support it, this allows an intelligent
3445 linker to optimize references to the symbol without being impeded as it would
3446 be by <tt>@llvm.used</tt>.</p>
3448 <p>This is a rare construct that should only be used in rare circumstances, and
3449 should not be exposed to source languages.</p>
3453 <!-- ======================================================================= -->
3455 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
3460 <div class="doc_code">
3462 %0 = type { i32, void ()* }
3463 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3467 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor
3468 functions and associated priorities. The functions referenced by this array
3469 will be called in ascending order of priority (i.e. lowest first) when the
3470 module is loaded. The order of functions with the same priority is not
3475 <!-- ======================================================================= -->
3477 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3482 <div class="doc_code">
3484 %0 = type { i32, void ()* }
3485 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3489 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions
3490 and associated priorities. The functions referenced by this array will be
3491 called in descending order of priority (i.e. highest first) when the module
3492 is loaded. The order of functions with the same priority is not defined.</p>
3498 <!-- *********************************************************************** -->
3499 <h2><a name="instref">Instruction Reference</a></h2>
3500 <!-- *********************************************************************** -->
3504 <p>The LLVM instruction set consists of several different classifications of
3505 instructions: <a href="#terminators">terminator
3506 instructions</a>, <a href="#binaryops">binary instructions</a>,
3507 <a href="#bitwiseops">bitwise binary instructions</a>,
3508 <a href="#memoryops">memory instructions</a>, and
3509 <a href="#otherops">other instructions</a>.</p>
3511 <!-- ======================================================================= -->
3513 <a name="terminators">Terminator Instructions</a>
3518 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3519 in a program ends with a "Terminator" instruction, which indicates which
3520 block should be executed after the current block is finished. These
3521 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3522 control flow, not values (the one exception being the
3523 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3525 <p>The terminator instructions are:
3526 '<a href="#i_ret"><tt>ret</tt></a>',
3527 '<a href="#i_br"><tt>br</tt></a>',
3528 '<a href="#i_switch"><tt>switch</tt></a>',
3529 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3530 '<a href="#i_invoke"><tt>invoke</tt></a>',
3531 '<a href="#i_resume"><tt>resume</tt></a>', and
3532 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3534 <!-- _______________________________________________________________________ -->
3536 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3543 ret <type> <value> <i>; Return a value from a non-void function</i>
3544 ret void <i>; Return from void function</i>
3548 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3549 a value) from a function back to the caller.</p>
3551 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3552 value and then causes control flow, and one that just causes control flow to
3556 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3557 return value. The type of the return value must be a
3558 '<a href="#t_firstclass">first class</a>' type.</p>
3560 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3561 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3562 value or a return value with a type that does not match its type, or if it
3563 has a void return type and contains a '<tt>ret</tt>' instruction with a
3567 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3568 the calling function's context. If the caller is a
3569 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3570 instruction after the call. If the caller was an
3571 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3572 the beginning of the "normal" destination block. If the instruction returns
3573 a value, that value shall set the call or invoke instruction's return
3578 ret i32 5 <i>; Return an integer value of 5</i>
3579 ret void <i>; Return from a void function</i>
3580 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3584 <!-- _______________________________________________________________________ -->
3586 <a name="i_br">'<tt>br</tt>' Instruction</a>
3593 br i1 <cond>, label <iftrue>, label <iffalse>
3594 br label <dest> <i>; Unconditional branch</i>
3598 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3599 different basic block in the current function. There are two forms of this
3600 instruction, corresponding to a conditional branch and an unconditional
3604 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3605 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3606 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3610 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3611 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3612 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3613 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3618 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3619 br i1 %cond, label %IfEqual, label %IfUnequal
3621 <a href="#i_ret">ret</a> i32 1
3623 <a href="#i_ret">ret</a> i32 0
3628 <!-- _______________________________________________________________________ -->
3630 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3637 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3641 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3642 several different places. It is a generalization of the '<tt>br</tt>'
3643 instruction, allowing a branch to occur to one of many possible
3647 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3648 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3649 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3650 The table is not allowed to contain duplicate constant entries.</p>
3653 <p>The <tt>switch</tt> instruction specifies a table of values and
3654 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3655 is searched for the given value. If the value is found, control flow is
3656 transferred to the corresponding destination; otherwise, control flow is
3657 transferred to the default destination.</p>
3659 <h5>Implementation:</h5>
3660 <p>Depending on properties of the target machine and the particular
3661 <tt>switch</tt> instruction, this instruction may be code generated in
3662 different ways. For example, it could be generated as a series of chained
3663 conditional branches or with a lookup table.</p>
3667 <i>; Emulate a conditional br instruction</i>
3668 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3669 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3671 <i>; Emulate an unconditional br instruction</i>
3672 switch i32 0, label %dest [ ]
3674 <i>; Implement a jump table:</i>
3675 switch i32 %val, label %otherwise [ i32 0, label %onzero
3677 i32 2, label %ontwo ]
3683 <!-- _______________________________________________________________________ -->
3685 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3692 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3697 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3698 within the current function, whose address is specified by
3699 "<tt>address</tt>". Address must be derived from a <a
3700 href="#blockaddress">blockaddress</a> constant.</p>
3704 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3705 rest of the arguments indicate the full set of possible destinations that the
3706 address may point to. Blocks are allowed to occur multiple times in the
3707 destination list, though this isn't particularly useful.</p>
3709 <p>This destination list is required so that dataflow analysis has an accurate
3710 understanding of the CFG.</p>
3714 <p>Control transfers to the block specified in the address argument. All
3715 possible destination blocks must be listed in the label list, otherwise this
3716 instruction has undefined behavior. This implies that jumps to labels
3717 defined in other functions have undefined behavior as well.</p>
3719 <h5>Implementation:</h5>
3721 <p>This is typically implemented with a jump through a register.</p>
3725 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3731 <!-- _______________________________________________________________________ -->
3733 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3740 <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>]
3741 to label <normal label> unwind label <exception label>
3745 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3746 function, with the possibility of control flow transfer to either the
3747 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3748 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3749 control flow will return to the "normal" label. If the callee (or any
3750 indirect callees) returns via the "<a href="#i_resume"><tt>resume</tt></a>"
3751 instruction or other exception handling mechanism, control is interrupted and
3752 continued at the dynamically nearest "exception" label.</p>
3754 <p>The '<tt>exception</tt>' label is a
3755 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3756 exception. As such, '<tt>exception</tt>' label is required to have the
3757 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3758 the information about the behavior of the program after unwinding
3759 happens, as its first non-PHI instruction. The restrictions on the
3760 "<tt>landingpad</tt>" instruction's tightly couples it to the
3761 "<tt>invoke</tt>" instruction, so that the important information contained
3762 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3766 <p>This instruction requires several arguments:</p>
3769 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3770 convention</a> the call should use. If none is specified, the call
3771 defaults to using C calling conventions.</li>
3773 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3774 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3775 '<tt>inreg</tt>' attributes are valid here.</li>
3777 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3778 function value being invoked. In most cases, this is a direct function
3779 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3780 off an arbitrary pointer to function value.</li>
3782 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3783 function to be invoked. </li>
3785 <li>'<tt>function args</tt>': argument list whose types match the function
3786 signature argument types and parameter attributes. All arguments must be
3787 of <a href="#t_firstclass">first class</a> type. If the function
3788 signature indicates the function accepts a variable number of arguments,
3789 the extra arguments can be specified.</li>
3791 <li>'<tt>normal label</tt>': the label reached when the called function
3792 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3794 <li>'<tt>exception label</tt>': the label reached when a callee returns via
3795 the <a href="#i_resume"><tt>resume</tt></a> instruction or other exception
3796 handling mechanism.</li>
3798 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3799 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3800 '<tt>readnone</tt>' attributes are valid here.</li>
3804 <p>This instruction is designed to operate as a standard
3805 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3806 primary difference is that it establishes an association with a label, which
3807 is used by the runtime library to unwind the stack.</p>
3809 <p>This instruction is used in languages with destructors to ensure that proper
3810 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3811 exception. Additionally, this is important for implementation of
3812 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3814 <p>For the purposes of the SSA form, the definition of the value returned by the
3815 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3816 block to the "normal" label. If the callee unwinds then no return value is
3821 %retval = invoke i32 @Test(i32 15) to label %Continue
3822 unwind label %TestCleanup <i>; {i32}:retval set</i>
3823 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3824 unwind label %TestCleanup <i>; {i32}:retval set</i>
3829 <!-- _______________________________________________________________________ -->
3832 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3839 resume <type> <value>
3843 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3847 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3848 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3852 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3853 (in-flight) exception whose unwinding was interrupted with
3854 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3858 resume { i8*, i32 } %exn
3863 <!-- _______________________________________________________________________ -->
3866 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3877 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3878 instruction is used to inform the optimizer that a particular portion of the
3879 code is not reachable. This can be used to indicate that the code after a
3880 no-return function cannot be reached, and other facts.</p>
3883 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3889 <!-- ======================================================================= -->
3891 <a name="binaryops">Binary Operations</a>
3896 <p>Binary operators are used to do most of the computation in a program. They
3897 require two operands of the same type, execute an operation on them, and
3898 produce a single value. The operands might represent multiple data, as is
3899 the case with the <a href="#t_vector">vector</a> data type. The result value
3900 has the same type as its operands.</p>
3902 <p>There are several different binary operators:</p>
3904 <!-- _______________________________________________________________________ -->
3906 <a name="i_add">'<tt>add</tt>' Instruction</a>
3913 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3914 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3915 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3916 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3920 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3923 <p>The two arguments to the '<tt>add</tt>' instruction must
3924 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3925 integer values. Both arguments must have identical types.</p>
3928 <p>The value produced is the integer sum of the two operands.</p>
3930 <p>If the sum has unsigned overflow, the result returned is the mathematical
3931 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3933 <p>Because LLVM integers use a two's complement representation, this instruction
3934 is appropriate for both signed and unsigned integers.</p>
3936 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3937 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3938 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3939 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3940 respectively, occurs.</p>
3944 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3949 <!-- _______________________________________________________________________ -->
3951 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3958 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3962 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3965 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3966 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3967 floating point values. Both arguments must have identical types.</p>
3970 <p>The value produced is the floating point sum of the two operands.</p>
3974 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3979 <!-- _______________________________________________________________________ -->
3981 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3988 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3989 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3990 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3991 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3995 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3998 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3999 '<tt>neg</tt>' instruction present in most other intermediate
4000 representations.</p>
4003 <p>The two arguments to the '<tt>sub</tt>' instruction must
4004 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4005 integer values. Both arguments must have identical types.</p>
4008 <p>The value produced is the integer difference of the two operands.</p>
4010 <p>If the difference has unsigned overflow, the result returned is the
4011 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
4014 <p>Because LLVM integers use a two's complement representation, this instruction
4015 is appropriate for both signed and unsigned integers.</p>
4017 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
4018 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
4019 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
4020 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4021 respectively, occurs.</p>
4025 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
4026 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
4031 <!-- _______________________________________________________________________ -->
4033 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
4040 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4044 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
4047 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
4048 '<tt>fneg</tt>' instruction present in most other intermediate
4049 representations.</p>
4052 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
4053 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4054 floating point values. Both arguments must have identical types.</p>
4057 <p>The value produced is the floating point difference of the two operands.</p>
4061 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
4062 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
4067 <!-- _______________________________________________________________________ -->
4069 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
4076 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4077 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4078 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4079 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4083 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
4086 <p>The two arguments to the '<tt>mul</tt>' instruction must
4087 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4088 integer values. Both arguments must have identical types.</p>
4091 <p>The value produced is the integer product of the two operands.</p>
4093 <p>If the result of the multiplication has unsigned overflow, the result
4094 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
4095 width of the result.</p>
4097 <p>Because LLVM integers use a two's complement representation, and the result
4098 is the same width as the operands, this instruction returns the correct
4099 result for both signed and unsigned integers. If a full product
4100 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
4101 be sign-extended or zero-extended as appropriate to the width of the full
4104 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
4105 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
4106 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
4107 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4108 respectively, occurs.</p>
4112 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
4117 <!-- _______________________________________________________________________ -->
4119 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
4126 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4130 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
4133 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
4134 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4135 floating point values. Both arguments must have identical types.</p>
4138 <p>The value produced is the floating point product of the two operands.</p>
4142 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
4147 <!-- _______________________________________________________________________ -->
4149 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
4156 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4157 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4161 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
4164 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
4165 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4166 values. Both arguments must have identical types.</p>
4169 <p>The value produced is the unsigned integer quotient of the two operands.</p>
4171 <p>Note that unsigned integer division and signed integer division are distinct
4172 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
4174 <p>Division by zero leads to undefined behavior.</p>
4176 <p>If the <tt>exact</tt> keyword is present, the result value of the
4177 <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a
4178 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
4183 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4188 <!-- _______________________________________________________________________ -->
4190 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
4197 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4198 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4202 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
4205 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
4206 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4207 values. Both arguments must have identical types.</p>
4210 <p>The value produced is the signed integer quotient of the two operands rounded
4213 <p>Note that signed integer division and unsigned integer division are distinct
4214 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
4216 <p>Division by zero leads to undefined behavior. Overflow also leads to
4217 undefined behavior; this is a rare case, but can occur, for example, by doing
4218 a 32-bit division of -2147483648 by -1.</p>
4220 <p>If the <tt>exact</tt> keyword is present, the result value of the
4221 <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would
4226 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4231 <!-- _______________________________________________________________________ -->
4233 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
4240 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4244 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
4247 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
4248 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4249 floating point values. Both arguments must have identical types.</p>
4252 <p>The value produced is the floating point quotient of the two operands.</p>
4256 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
4261 <!-- _______________________________________________________________________ -->
4263 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
4270 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4274 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
4275 division of its two arguments.</p>
4278 <p>The two arguments to the '<tt>urem</tt>' instruction must be
4279 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4280 values. Both arguments must have identical types.</p>
4283 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
4284 This instruction always performs an unsigned division to get the
4287 <p>Note that unsigned integer remainder and signed integer remainder are
4288 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
4290 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
4294 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4299 <!-- _______________________________________________________________________ -->
4301 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
4308 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4312 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
4313 division of its two operands. This instruction can also take
4314 <a href="#t_vector">vector</a> versions of the values in which case the
4315 elements must be integers.</p>
4318 <p>The two arguments to the '<tt>srem</tt>' instruction must be
4319 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4320 values. Both arguments must have identical types.</p>
4323 <p>This instruction returns the <i>remainder</i> of a division (where the result
4324 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
4325 <i>modulo</i> operator (where the result is either zero or has the same sign
4326 as the divisor, <tt>op2</tt>) of a value.
4327 For more information about the difference,
4328 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
4329 Math Forum</a>. For a table of how this is implemented in various languages,
4330 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
4331 Wikipedia: modulo operation</a>.</p>
4333 <p>Note that signed integer remainder and unsigned integer remainder are
4334 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
4336 <p>Taking the remainder of a division by zero leads to undefined behavior.
4337 Overflow also leads to undefined behavior; this is a rare case, but can
4338 occur, for example, by taking the remainder of a 32-bit division of
4339 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
4340 lets srem be implemented using instructions that return both the result of
4341 the division and the remainder.)</p>
4345 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4350 <!-- _______________________________________________________________________ -->
4352 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
4359 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4363 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
4364 its two operands.</p>
4367 <p>The two arguments to the '<tt>frem</tt>' instruction must be
4368 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4369 floating point values. Both arguments must have identical types.</p>
4372 <p>This instruction returns the <i>remainder</i> of a division. The remainder
4373 has the same sign as the dividend.</p>
4377 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
4384 <!-- ======================================================================= -->
4386 <a name="bitwiseops">Bitwise Binary Operations</a>
4391 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
4392 program. They are generally very efficient instructions and can commonly be
4393 strength reduced from other instructions. They require two operands of the
4394 same type, execute an operation on them, and produce a single value. The
4395 resulting value is the same type as its operands.</p>
4397 <!-- _______________________________________________________________________ -->
4399 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
4406 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4407 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4408 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4409 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4413 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
4414 a specified number of bits.</p>
4417 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
4418 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4419 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4422 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
4423 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
4424 is (statically or dynamically) negative or equal to or larger than the number
4425 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4426 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4427 shift amount in <tt>op2</tt>.</p>
4429 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
4430 <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits. If
4431 the <tt>nsw</tt> keyword is present, then the shift produces a
4432 <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree
4433 with the resultant sign bit. As such, NUW/NSW have the same semantics as
4434 they would if the shift were expressed as a mul instruction with the same
4435 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
4439 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
4440 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
4441 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
4442 <result> = shl i32 1, 32 <i>; undefined</i>
4443 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4448 <!-- _______________________________________________________________________ -->
4450 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4457 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4458 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4462 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4463 operand shifted to the right a specified number of bits with zero fill.</p>
4466 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4467 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4468 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4471 <p>This instruction always performs a logical shift right operation. The most
4472 significant bits of the result will be filled with zero bits after the shift.
4473 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4474 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4475 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4476 shift amount in <tt>op2</tt>.</p>
4478 <p>If the <tt>exact</tt> keyword is present, the result value of the
4479 <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4480 shifted out are non-zero.</p>
4485 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4486 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4487 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4488 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4489 <result> = lshr i32 1, 32 <i>; undefined</i>
4490 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4495 <!-- _______________________________________________________________________ -->
4497 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4504 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4505 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4509 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4510 operand shifted to the right a specified number of bits with sign
4514 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4515 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4516 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4519 <p>This instruction always performs an arithmetic shift right operation, The
4520 most significant bits of the result will be filled with the sign bit
4521 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4522 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4523 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4524 the corresponding shift amount in <tt>op2</tt>.</p>
4526 <p>If the <tt>exact</tt> keyword is present, the result value of the
4527 <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4528 shifted out are non-zero.</p>
4532 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4533 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4534 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4535 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4536 <result> = ashr i32 1, 32 <i>; undefined</i>
4537 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4542 <!-- _______________________________________________________________________ -->
4544 <a name="i_and">'<tt>and</tt>' Instruction</a>
4551 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4555 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4559 <p>The two arguments to the '<tt>and</tt>' instruction must be
4560 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4561 values. Both arguments must have identical types.</p>
4564 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4566 <table border="1" cellspacing="0" cellpadding="4">
4598 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4599 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4600 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4603 <!-- _______________________________________________________________________ -->
4605 <a name="i_or">'<tt>or</tt>' Instruction</a>
4612 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4616 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4620 <p>The two arguments to the '<tt>or</tt>' instruction must be
4621 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4622 values. Both arguments must have identical types.</p>
4625 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4627 <table border="1" cellspacing="0" cellpadding="4">
4659 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4660 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4661 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4666 <!-- _______________________________________________________________________ -->
4668 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4675 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4679 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4680 its two operands. The <tt>xor</tt> is used to implement the "one's
4681 complement" operation, which is the "~" operator in C.</p>
4684 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4685 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4686 values. Both arguments must have identical types.</p>
4689 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4691 <table border="1" cellspacing="0" cellpadding="4">
4723 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4724 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4725 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4726 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4733 <!-- ======================================================================= -->
4735 <a name="vectorops">Vector Operations</a>
4740 <p>LLVM supports several instructions to represent vector operations in a
4741 target-independent manner. These instructions cover the element-access and
4742 vector-specific operations needed to process vectors effectively. While LLVM
4743 does directly support these vector operations, many sophisticated algorithms
4744 will want to use target-specific intrinsics to take full advantage of a
4745 specific target.</p>
4747 <!-- _______________________________________________________________________ -->
4749 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4756 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4760 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4761 from a vector at a specified index.</p>
4765 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4766 of <a href="#t_vector">vector</a> type. The second operand is an index
4767 indicating the position from which to extract the element. The index may be
4771 <p>The result is a scalar of the same type as the element type of
4772 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4773 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4774 results are undefined.</p>
4778 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4783 <!-- _______________________________________________________________________ -->
4785 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4792 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4796 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4797 vector at a specified index.</p>
4800 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4801 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4802 whose type must equal the element type of the first operand. The third
4803 operand is an index indicating the position at which to insert the value.
4804 The index may be a variable.</p>
4807 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4808 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4809 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4810 results are undefined.</p>
4814 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4819 <!-- _______________________________________________________________________ -->
4821 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4828 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4832 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4833 from two input vectors, returning a vector with the same element type as the
4834 input and length that is the same as the shuffle mask.</p>
4837 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4838 with the same type. The third argument is a shuffle mask whose
4839 element type is always 'i32'. The result of the instruction is a vector
4840 whose length is the same as the shuffle mask and whose element type is the
4841 same as the element type of the first two operands.</p>
4843 <p>The shuffle mask operand is required to be a constant vector with either
4844 constant integer or undef values.</p>
4847 <p>The elements of the two input vectors are numbered from left to right across
4848 both of the vectors. The shuffle mask operand specifies, for each element of
4849 the result vector, which element of the two input vectors the result element
4850 gets. The element selector may be undef (meaning "don't care") and the
4851 second operand may be undef if performing a shuffle from only one vector.</p>
4855 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4856 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4857 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4858 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4859 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4860 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4861 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4862 <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>
4869 <!-- ======================================================================= -->
4871 <a name="aggregateops">Aggregate Operations</a>
4876 <p>LLVM supports several instructions for working with
4877 <a href="#t_aggregate">aggregate</a> values.</p>
4879 <!-- _______________________________________________________________________ -->
4881 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4888 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4892 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4893 from an <a href="#t_aggregate">aggregate</a> value.</p>
4896 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4897 of <a href="#t_struct">struct</a> or
4898 <a href="#t_array">array</a> type. The operands are constant indices to
4899 specify which value to extract in a similar manner as indices in a
4900 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4901 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4903 <li>Since the value being indexed is not a pointer, the first index is
4904 omitted and assumed to be zero.</li>
4905 <li>At least one index must be specified.</li>
4906 <li>Not only struct indices but also array indices must be in
4911 <p>The result is the value at the position in the aggregate specified by the
4916 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4921 <!-- _______________________________________________________________________ -->
4923 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4930 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4934 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4935 in an <a href="#t_aggregate">aggregate</a> value.</p>
4938 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4939 of <a href="#t_struct">struct</a> or
4940 <a href="#t_array">array</a> type. The second operand is a first-class
4941 value to insert. The following operands are constant indices indicating
4942 the position at which to insert the value in a similar manner as indices in a
4943 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4944 value to insert must have the same type as the value identified by the
4948 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4949 that of <tt>val</tt> except that the value at the position specified by the
4950 indices is that of <tt>elt</tt>.</p>
4954 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4955 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4956 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4963 <!-- ======================================================================= -->
4965 <a name="memoryops">Memory Access and Addressing Operations</a>
4970 <p>A key design point of an SSA-based representation is how it represents
4971 memory. In LLVM, no memory locations are in SSA form, which makes things
4972 very simple. This section describes how to read, write, and allocate
4975 <!-- _______________________________________________________________________ -->
4977 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4984 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4988 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4989 currently executing function, to be automatically released when this function
4990 returns to its caller. The object is always allocated in the generic address
4991 space (address space zero).</p>
4994 <p>The '<tt>alloca</tt>' instruction
4995 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4996 runtime stack, returning a pointer of the appropriate type to the program.
4997 If "NumElements" is specified, it is the number of elements allocated,
4998 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4999 specified, the value result of the allocation is guaranteed to be aligned to
5000 at least that boundary. If not specified, or if zero, the target can choose
5001 to align the allocation on any convenient boundary compatible with the
5004 <p>'<tt>type</tt>' may be any sized type.</p>
5007 <p>Memory is allocated; a pointer is returned. The operation is undefined if
5008 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
5009 memory is automatically released when the function returns. The
5010 '<tt>alloca</tt>' instruction is commonly used to represent automatic
5011 variables that must have an address available. When the function returns
5012 (either with the <tt><a href="#i_ret">ret</a></tt>
5013 or <tt><a href="#i_resume">resume</a></tt> instructions), the memory is
5014 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5015 The order in which memory is allocated (ie., which way the stack grows) is
5022 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
5023 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
5024 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
5025 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
5030 <!-- _______________________________________________________________________ -->
5032 <a name="i_load">'<tt>load</tt>' Instruction</a>
5039 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
5040 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5041 !<index> = !{ i32 1 }
5045 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
5048 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
5049 from which to load. The pointer must point to
5050 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
5051 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
5052 number or order of execution of this <tt>load</tt> with other <a
5053 href="#volatile">volatile operations</a>.</p>
5055 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
5056 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5057 argument. The <code>release</code> and <code>acq_rel</code> orderings are
5058 not valid on <code>load</code> instructions. Atomic loads produce <a
5059 href="#memorymodel">defined</a> results when they may see multiple atomic
5060 stores. The type of the pointee must be an integer type whose bit width
5061 is a power of two greater than or equal to eight and less than or equal
5062 to a target-specific size limit. <code>align</code> must be explicitly
5063 specified on atomic loads, and the load has undefined behavior if the
5064 alignment is not set to a value which is at least the size in bytes of
5065 the pointee. <code>!nontemporal</code> does not have any defined semantics
5066 for atomic loads.</p>
5068 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
5069 operation (that is, the alignment of the memory address). A value of 0 or an
5070 omitted <tt>align</tt> argument means that the operation has the abi
5071 alignment for the target. It is the responsibility of the code emitter to
5072 ensure that the alignment information is correct. Overestimating the
5073 alignment results in undefined behavior. Underestimating the alignment may
5074 produce less efficient code. An alignment of 1 is always safe.</p>
5076 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
5077 metatadata name <index> corresponding to a metadata node with
5078 one <tt>i32</tt> entry of value 1. The existence of
5079 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
5080 and code generator that this load is not expected to be reused in the cache.
5081 The code generator may select special instructions to save cache bandwidth,
5082 such as the <tt>MOVNT</tt> instruction on x86.</p>
5084 <p>The optional <tt>!invariant.load</tt> metadata must reference a single
5085 metatadata name <index> corresponding to a metadata node with no
5086 entries. The existence of the <tt>!invariant.load</tt> metatadata on the
5087 instruction tells the optimizer and code generator that this load address
5088 points to memory which does not change value during program execution.
5089 The optimizer may then move this load around, for example, by hoisting it
5090 out of loops using loop invariant code motion.</p>
5093 <p>The location of memory pointed to is loaded. If the value being loaded is of
5094 scalar type then the number of bytes read does not exceed the minimum number
5095 of bytes needed to hold all bits of the type. For example, loading an
5096 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
5097 <tt>i20</tt> with a size that is not an integral number of bytes, the result
5098 is undefined if the value was not originally written using a store of the
5103 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5104 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
5105 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
5110 <!-- _______________________________________________________________________ -->
5112 <a name="i_store">'<tt>store</tt>' Instruction</a>
5119 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
5120 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
5124 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
5127 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
5128 and an address at which to store it. The type of the
5129 '<tt><pointer></tt>' operand must be a pointer to
5130 the <a href="#t_firstclass">first class</a> type of the
5131 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
5132 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
5133 order of execution of this <tt>store</tt> with other <a
5134 href="#volatile">volatile operations</a>.</p>
5136 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
5137 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5138 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
5139 valid on <code>store</code> instructions. Atomic loads produce <a
5140 href="#memorymodel">defined</a> results when they may see multiple atomic
5141 stores. The type of the pointee must be an integer type whose bit width
5142 is a power of two greater than or equal to eight and less than or equal
5143 to a target-specific size limit. <code>align</code> must be explicitly
5144 specified on atomic stores, and the store has undefined behavior if the
5145 alignment is not set to a value which is at least the size in bytes of
5146 the pointee. <code>!nontemporal</code> does not have any defined semantics
5147 for atomic stores.</p>
5149 <p>The optional constant "align" argument specifies the alignment of the
5150 operation (that is, the alignment of the memory address). A value of 0 or an
5151 omitted "align" argument means that the operation has the abi
5152 alignment for the target. It is the responsibility of the code emitter to
5153 ensure that the alignment information is correct. Overestimating the
5154 alignment results in an undefined behavior. Underestimating the alignment may
5155 produce less efficient code. An alignment of 1 is always safe.</p>
5157 <p>The optional !nontemporal metadata must reference a single metatadata
5158 name <index> corresponding to a metadata node with one i32 entry of
5159 value 1. The existence of the !nontemporal metatadata on the
5160 instruction tells the optimizer and code generator that this load is
5161 not expected to be reused in the cache. The code generator may
5162 select special instructions to save cache bandwidth, such as the
5163 MOVNT instruction on x86.</p>
5167 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
5168 location specified by the '<tt><pointer></tt>' operand. If
5169 '<tt><value></tt>' is of scalar type then the number of bytes written
5170 does not exceed the minimum number of bytes needed to hold all bits of the
5171 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
5172 writing a value of a type like <tt>i20</tt> with a size that is not an
5173 integral number of bytes, it is unspecified what happens to the extra bits
5174 that do not belong to the type, but they will typically be overwritten.</p>
5178 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5179 store i32 3, i32* %ptr <i>; yields {void}</i>
5180 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
5185 <!-- _______________________________________________________________________ -->
5187 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
5194 fence [singlethread] <ordering> <i>; yields {void}</i>
5198 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
5199 between operations.</p>
5201 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
5202 href="#ordering">ordering</a> argument which defines what
5203 <i>synchronizes-with</i> edges they add. They can only be given
5204 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
5205 <code>seq_cst</code> orderings.</p>
5208 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
5209 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
5210 <code>acquire</code> ordering semantics if and only if there exist atomic
5211 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
5212 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
5213 <var>X</var> modifies <var>M</var> (either directly or through some side effect
5214 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
5215 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
5216 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
5217 than an explicit <code>fence</code>, one (but not both) of the atomic operations
5218 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
5219 <code>acquire</code> (resp.) ordering constraint and still
5220 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
5221 <i>happens-before</i> edge.</p>
5223 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
5224 having both <code>acquire</code> and <code>release</code> semantics specified
5225 above, participates in the global program order of other <code>seq_cst</code>
5226 operations and/or fences.</p>
5228 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
5229 specifies that the fence only synchronizes with other fences in the same
5230 thread. (This is useful for interacting with signal handlers.)</p>
5234 fence acquire <i>; yields {void}</i>
5235 fence singlethread seq_cst <i>; yields {void}</i>
5240 <!-- _______________________________________________________________________ -->
5242 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
5249 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
5253 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
5254 It loads a value in memory and compares it to a given value. If they are
5255 equal, it stores a new value into the memory.</p>
5258 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
5259 address to operate on, a value to compare to the value currently be at that
5260 address, and a new value to place at that address if the compared values are
5261 equal. The type of '<var><cmp></var>' must be an integer type whose
5262 bit width is a power of two greater than or equal to eight and less than
5263 or equal to a target-specific size limit. '<var><cmp></var>' and
5264 '<var><new></var>' must have the same type, and the type of
5265 '<var><pointer></var>' must be a pointer to that type. If the
5266 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
5267 optimizer is not allowed to modify the number or order of execution
5268 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
5271 <!-- FIXME: Extend allowed types. -->
5273 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
5274 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
5276 <p>The optional "<code>singlethread</code>" argument declares that the
5277 <code>cmpxchg</code> is only atomic with respect to code (usually signal
5278 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
5279 cmpxchg is atomic with respect to all other code in the system.</p>
5281 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
5282 the size in memory of the operand.
5285 <p>The contents of memory at the location specified by the
5286 '<tt><pointer></tt>' operand is read and compared to
5287 '<tt><cmp></tt>'; if the read value is the equal,
5288 '<tt><new></tt>' is written. The original value at the location
5291 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
5292 purpose of identifying <a href="#release_sequence">release sequences</a>. A
5293 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
5294 parameter determined by dropping any <code>release</code> part of the
5295 <code>cmpxchg</code>'s ordering.</p>
5298 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
5299 optimization work on ARM.)
5301 FIXME: Is a weaker ordering constraint on failure helpful in practice?
5307 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
5308 <a href="#i_br">br</a> label %loop
5311 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
5312 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
5313 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
5314 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
5315 <a href="#i_br">br</a> i1 %success, label %done, label %loop
5323 <!-- _______________________________________________________________________ -->
5325 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
5332 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
5336 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
5339 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
5340 operation to apply, an address whose value to modify, an argument to the
5341 operation. The operation must be one of the following keywords:</p>
5356 <p>The type of '<var><value></var>' must be an integer type whose
5357 bit width is a power of two greater than or equal to eight and less than
5358 or equal to a target-specific size limit. The type of the
5359 '<code><pointer></code>' operand must be a pointer to that type.
5360 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
5361 optimizer is not allowed to modify the number or order of execution of this
5362 <code>atomicrmw</code> with other <a href="#volatile">volatile
5365 <!-- FIXME: Extend allowed types. -->
5368 <p>The contents of memory at the location specified by the
5369 '<tt><pointer></tt>' operand are atomically read, modified, and written
5370 back. The original value at the location is returned. The modification is
5371 specified by the <var>operation</var> argument:</p>
5374 <li>xchg: <code>*ptr = val</code></li>
5375 <li>add: <code>*ptr = *ptr + val</code></li>
5376 <li>sub: <code>*ptr = *ptr - val</code></li>
5377 <li>and: <code>*ptr = *ptr & val</code></li>
5378 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
5379 <li>or: <code>*ptr = *ptr | val</code></li>
5380 <li>xor: <code>*ptr = *ptr ^ val</code></li>
5381 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
5382 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
5383 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
5384 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
5389 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
5394 <!-- _______________________________________________________________________ -->
5396 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
5403 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5404 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5405 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5409 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
5410 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
5411 It performs address calculation only and does not access memory.</p>
5414 <p>The first argument is always a pointer or a vector of pointers,
5415 and forms the basis of the
5416 calculation. The remaining arguments are indices that indicate which of the
5417 elements of the aggregate object are indexed. The interpretation of each
5418 index is dependent on the type being indexed into. The first index always
5419 indexes the pointer value given as the first argument, the second index
5420 indexes a value of the type pointed to (not necessarily the value directly
5421 pointed to, since the first index can be non-zero), etc. The first type
5422 indexed into must be a pointer value, subsequent types can be arrays,
5423 vectors, and structs. Note that subsequent types being indexed into
5424 can never be pointers, since that would require loading the pointer before
5425 continuing calculation.</p>
5427 <p>The type of each index argument depends on the type it is indexing into.
5428 When indexing into a (optionally packed) structure, only <tt>i32</tt>
5429 integer <b>constants</b> are allowed (when using a vector of indices they
5430 must all be the <b>same</b> <tt>i32</tt> integer constant). When indexing
5431 into an array, pointer or vector, integers of any width are allowed, and
5432 they are not required to be constant. These integers are treated as signed
5433 values where relevant.</p>
5435 <p>For example, let's consider a C code fragment and how it gets compiled to
5438 <pre class="doc_code">
5450 int *foo(struct ST *s) {
5451 return &s[1].Z.B[5][13];
5455 <p>The LLVM code generated by Clang is:</p>
5457 <pre class="doc_code">
5458 %struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 }
5459 %struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT }
5461 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5463 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5469 <p>In the example above, the first index is indexing into the
5470 '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a
5471 '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a
5472 structure. The second index indexes into the third element of the structure,
5473 yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>'
5474 type, another structure. The third index indexes into the second element of
5475 the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The
5476 two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>'
5477 type. The '<tt>getelementptr</tt>' instruction returns a pointer to this
5478 element, thus computing a value of '<tt>i32*</tt>' type.</p>
5480 <p>Note that it is perfectly legal to index partially through a structure,
5481 returning a pointer to an inner element. Because of this, the LLVM code for
5482 the given testcase is equivalent to:</p>
5484 <pre class="doc_code">
5485 define i32* @foo(%struct.ST* %s) {
5486 %t1 = getelementptr %struct.ST* %s, i32 1 <i>; yields %struct.ST*:%t1</i>
5487 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 <i>; yields %struct.RT*:%t2</i>
5488 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5489 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5490 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5495 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5496 <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the
5497 base pointer is not an <i>in bounds</i> address of an allocated object,
5498 or if any of the addresses that would be formed by successive addition of
5499 the offsets implied by the indices to the base address with infinitely
5500 precise signed arithmetic are not an <i>in bounds</i> address of that
5501 allocated object. The <i>in bounds</i> addresses for an allocated object
5502 are all the addresses that point into the object, plus the address one
5504 In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword
5505 applies to each of the computations element-wise. </p>
5507 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5508 the base address with silently-wrapping two's complement arithmetic. If the
5509 offsets have a different width from the pointer, they are sign-extended or
5510 truncated to the width of the pointer. The result value of the
5511 <tt>getelementptr</tt> may be outside the object pointed to by the base
5512 pointer. The result value may not necessarily be used to access memory
5513 though, even if it happens to point into allocated storage. See the
5514 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5517 <p>The getelementptr instruction is often confusing. For some more insight into
5518 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5522 <i>; yields [12 x i8]*:aptr</i>
5523 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5524 <i>; yields i8*:vptr</i>
5525 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5526 <i>; yields i8*:eptr</i>
5527 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5528 <i>; yields i32*:iptr</i>
5529 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5532 <p>In cases where the pointer argument is a vector of pointers, each index must
5533 be a vector with the same number of elements. For example: </p>
5534 <pre class="doc_code">
5535 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5542 <!-- ======================================================================= -->
5544 <a name="convertops">Conversion Operations</a>
5549 <p>The instructions in this category are the conversion instructions (casting)
5550 which all take a single operand and a type. They perform various bit
5551 conversions on the operand.</p>
5553 <!-- _______________________________________________________________________ -->
5555 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5562 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5566 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5567 type <tt>ty2</tt>.</p>
5570 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5571 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5572 of the same number of integers.
5573 The bit size of the <tt>value</tt> must be larger than
5574 the bit size of the destination type, <tt>ty2</tt>.
5575 Equal sized types are not allowed.</p>
5578 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5579 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5580 source size must be larger than the destination size, <tt>trunc</tt> cannot
5581 be a <i>no-op cast</i>. It will always truncate bits.</p>
5585 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5586 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5587 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5588 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5593 <!-- _______________________________________________________________________ -->
5595 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5602 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5606 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5611 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5612 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5613 of the same number of integers.
5614 The bit size of the <tt>value</tt> must be smaller than
5615 the bit size of the destination type,
5619 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5620 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5622 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5626 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5627 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5628 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5633 <!-- _______________________________________________________________________ -->
5635 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5642 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5646 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5649 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5650 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5651 of the same number of integers.
5652 The bit size of the <tt>value</tt> must be smaller than
5653 the bit size of the destination type,
5657 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5658 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5659 of the type <tt>ty2</tt>.</p>
5661 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5665 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5666 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5667 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5672 <!-- _______________________________________________________________________ -->
5674 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5681 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5685 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5689 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5690 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5691 to cast it to. The size of <tt>value</tt> must be larger than the size of
5692 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5693 <i>no-op cast</i>.</p>
5696 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5697 <a href="#t_floating">floating point</a> type to a smaller
5698 <a href="#t_floating">floating point</a> type. If the value cannot fit
5699 within the destination type, <tt>ty2</tt>, then the results are
5704 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5705 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5710 <!-- _______________________________________________________________________ -->
5712 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5719 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5723 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5724 floating point value.</p>
5727 <p>The '<tt>fpext</tt>' instruction takes a
5728 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5729 a <a href="#t_floating">floating point</a> type to cast it to. The source
5730 type must be smaller than the destination type.</p>
5733 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5734 <a href="#t_floating">floating point</a> type to a larger
5735 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5736 used to make a <i>no-op cast</i> because it always changes bits. Use
5737 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5741 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5742 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5747 <!-- _______________________________________________________________________ -->
5749 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5756 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5760 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5761 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5764 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5765 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5766 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5767 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5768 vector integer type with the same number of elements as <tt>ty</tt></p>
5771 <p>The '<tt>fptoui</tt>' instruction converts its
5772 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5773 towards zero) unsigned integer value. If the value cannot fit
5774 in <tt>ty2</tt>, the results are undefined.</p>
5778 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5779 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5780 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5785 <!-- _______________________________________________________________________ -->
5787 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5794 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5798 <p>The '<tt>fptosi</tt>' instruction converts
5799 <a href="#t_floating">floating point</a> <tt>value</tt> to
5800 type <tt>ty2</tt>.</p>
5803 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5804 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5805 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5806 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5807 vector integer type with the same number of elements as <tt>ty</tt></p>
5810 <p>The '<tt>fptosi</tt>' instruction converts its
5811 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5812 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5813 the results are undefined.</p>
5817 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5818 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5819 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5824 <!-- _______________________________________________________________________ -->
5826 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5833 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5837 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5838 integer and converts that value to the <tt>ty2</tt> type.</p>
5841 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5842 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5843 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5844 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5845 floating point type with the same number of elements as <tt>ty</tt></p>
5848 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5849 integer quantity and converts it to the corresponding floating point
5850 value. If the value cannot fit in the floating point value, the results are
5855 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5856 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5861 <!-- _______________________________________________________________________ -->
5863 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5870 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5874 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5875 and converts that value to the <tt>ty2</tt> type.</p>
5878 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5879 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5880 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5881 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5882 floating point type with the same number of elements as <tt>ty</tt></p>
5885 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5886 quantity and converts it to the corresponding floating point value. If the
5887 value cannot fit in the floating point value, the results are undefined.</p>
5891 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5892 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5897 <!-- _______________________________________________________________________ -->
5899 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5906 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5910 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of
5911 pointers <tt>value</tt> to
5912 the integer (or vector of integers) type <tt>ty2</tt>.</p>
5915 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5916 must be a a value of type <a href="#t_pointer">pointer</a> or a vector of
5917 pointers, and a type to cast it to
5918 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector
5919 of integers type.</p>
5922 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5923 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5924 truncating or zero extending that value to the size of the integer type. If
5925 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5926 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5927 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5932 %X = ptrtoint i32* %P to i8 <i>; yields truncation on 32-bit architecture</i>
5933 %Y = ptrtoint i32* %P to i64 <i>; yields zero extension on 32-bit architecture</i>
5934 %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>
5939 <!-- _______________________________________________________________________ -->
5941 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5948 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5952 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5953 pointer type, <tt>ty2</tt>.</p>
5956 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5957 value to cast, and a type to cast it to, which must be a
5958 <a href="#t_pointer">pointer</a> type.</p>
5961 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5962 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5963 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5964 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5965 than the size of a pointer then a zero extension is done. If they are the
5966 same size, nothing is done (<i>no-op cast</i>).</p>
5970 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5971 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5972 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5973 %Z = inttoptr <4 x i32> %G to <4 x i8*><i>; yields truncation of vector G to four pointers</i>
5978 <!-- _______________________________________________________________________ -->
5980 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5987 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5991 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5992 <tt>ty2</tt> without changing any bits.</p>
5995 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5996 non-aggregate first class value, and a type to cast it to, which must also be
5997 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5998 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5999 identical. If the source type is a pointer, the destination type must also be
6000 a pointer. This instruction supports bitwise conversion of vectors to
6001 integers and to vectors of other types (as long as they have the same
6005 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
6006 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
6007 this conversion. The conversion is done as if the <tt>value</tt> had been
6008 stored to memory and read back as type <tt>ty2</tt>.
6009 Pointer (or vector of pointers) types may only be converted to other pointer
6010 (or vector of pointers) types with this instruction. To convert
6011 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
6012 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
6016 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
6017 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
6018 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
6019 %Z = bitcast <2 x i32*> %V to <2 x i64*> <i>; yields <2 x i64*></i>
6026 <!-- ======================================================================= -->
6028 <a name="otherops">Other Operations</a>
6033 <p>The instructions in this category are the "miscellaneous" instructions, which
6034 defy better classification.</p>
6036 <!-- _______________________________________________________________________ -->
6038 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
6045 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
6049 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
6050 boolean values based on comparison of its two integer, integer vector,
6051 pointer, or pointer vector operands.</p>
6054 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
6055 the condition code indicating the kind of comparison to perform. It is not a
6056 value, just a keyword. The possible condition code are:</p>
6059 <li><tt>eq</tt>: equal</li>
6060 <li><tt>ne</tt>: not equal </li>
6061 <li><tt>ugt</tt>: unsigned greater than</li>
6062 <li><tt>uge</tt>: unsigned greater or equal</li>
6063 <li><tt>ult</tt>: unsigned less than</li>
6064 <li><tt>ule</tt>: unsigned less or equal</li>
6065 <li><tt>sgt</tt>: signed greater than</li>
6066 <li><tt>sge</tt>: signed greater or equal</li>
6067 <li><tt>slt</tt>: signed less than</li>
6068 <li><tt>sle</tt>: signed less or equal</li>
6071 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
6072 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
6073 typed. They must also be identical types.</p>
6076 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
6077 condition code given as <tt>cond</tt>. The comparison performed always yields
6078 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
6079 result, as follows:</p>
6082 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
6083 <tt>false</tt> otherwise. No sign interpretation is necessary or
6086 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
6087 <tt>false</tt> otherwise. No sign interpretation is necessary or
6090 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
6091 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6093 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
6094 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6095 to <tt>op2</tt>.</li>
6097 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
6098 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6100 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
6101 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6103 <li><tt>sgt</tt>: interprets the operands as signed values and yields
6104 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6106 <li><tt>sge</tt>: interprets the operands as signed values and yields
6107 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6108 to <tt>op2</tt>.</li>
6110 <li><tt>slt</tt>: interprets the operands as signed values and yields
6111 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6113 <li><tt>sle</tt>: interprets the operands as signed values and yields
6114 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6117 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
6118 values are compared as if they were integers.</p>
6120 <p>If the operands are integer vectors, then they are compared element by
6121 element. The result is an <tt>i1</tt> vector with the same number of elements
6122 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
6126 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
6127 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
6128 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
6129 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
6130 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
6131 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
6134 <p>Note that the code generator does not yet support vector types with
6135 the <tt>icmp</tt> instruction.</p>
6139 <!-- _______________________________________________________________________ -->
6141 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
6148 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
6152 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
6153 values based on comparison of its operands.</p>
6155 <p>If the operands are floating point scalars, then the result type is a boolean
6156 (<a href="#t_integer"><tt>i1</tt></a>).</p>
6158 <p>If the operands are floating point vectors, then the result type is a vector
6159 of boolean with the same number of elements as the operands being
6163 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
6164 the condition code indicating the kind of comparison to perform. It is not a
6165 value, just a keyword. The possible condition code are:</p>
6168 <li><tt>false</tt>: no comparison, always returns false</li>
6169 <li><tt>oeq</tt>: ordered and equal</li>
6170 <li><tt>ogt</tt>: ordered and greater than </li>
6171 <li><tt>oge</tt>: ordered and greater than or equal</li>
6172 <li><tt>olt</tt>: ordered and less than </li>
6173 <li><tt>ole</tt>: ordered and less than or equal</li>
6174 <li><tt>one</tt>: ordered and not equal</li>
6175 <li><tt>ord</tt>: ordered (no nans)</li>
6176 <li><tt>ueq</tt>: unordered or equal</li>
6177 <li><tt>ugt</tt>: unordered or greater than </li>
6178 <li><tt>uge</tt>: unordered or greater than or equal</li>
6179 <li><tt>ult</tt>: unordered or less than </li>
6180 <li><tt>ule</tt>: unordered or less than or equal</li>
6181 <li><tt>une</tt>: unordered or not equal</li>
6182 <li><tt>uno</tt>: unordered (either nans)</li>
6183 <li><tt>true</tt>: no comparison, always returns true</li>
6186 <p><i>Ordered</i> means that neither operand is a QNAN while
6187 <i>unordered</i> means that either operand may be a QNAN.</p>
6189 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
6190 a <a href="#t_floating">floating point</a> type or
6191 a <a href="#t_vector">vector</a> of floating point type. They must have
6192 identical types.</p>
6195 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
6196 according to the condition code given as <tt>cond</tt>. If the operands are
6197 vectors, then the vectors are compared element by element. Each comparison
6198 performed always yields an <a href="#t_integer">i1</a> result, as
6202 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
6204 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6205 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6207 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6208 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6210 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6211 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6213 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6214 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6216 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6217 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6219 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6220 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6222 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
6224 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
6225 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6227 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
6228 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6230 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
6231 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6233 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
6234 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6236 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
6237 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6239 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
6240 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6242 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
6244 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
6249 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
6250 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
6251 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
6252 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
6255 <p>Note that the code generator does not yet support vector types with
6256 the <tt>fcmp</tt> instruction.</p>
6260 <!-- _______________________________________________________________________ -->
6262 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
6269 <result> = phi <ty> [ <val0>, <label0>], ...
6273 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
6274 SSA graph representing the function.</p>
6277 <p>The type of the incoming values is specified with the first type field. After
6278 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
6279 one pair for each predecessor basic block of the current block. Only values
6280 of <a href="#t_firstclass">first class</a> type may be used as the value
6281 arguments to the PHI node. Only labels may be used as the label
6284 <p>There must be no non-phi instructions between the start of a basic block and
6285 the PHI instructions: i.e. PHI instructions must be first in a basic
6288 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
6289 occur on the edge from the corresponding predecessor block to the current
6290 block (but after any definition of an '<tt>invoke</tt>' instruction's return
6291 value on the same edge).</p>
6294 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
6295 specified by the pair corresponding to the predecessor basic block that
6296 executed just prior to the current block.</p>
6300 Loop: ; Infinite loop that counts from 0 on up...
6301 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6302 %nextindvar = add i32 %indvar, 1
6308 <!-- _______________________________________________________________________ -->
6310 <a name="i_select">'<tt>select</tt>' Instruction</a>
6317 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
6319 <i>selty</i> is either i1 or {<N x i1>}
6323 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
6324 condition, without branching.</p>
6328 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
6329 values indicating the condition, and two values of the
6330 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
6331 vectors and the condition is a scalar, then entire vectors are selected, not
6332 individual elements.</p>
6335 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
6336 first value argument; otherwise, it returns the second value argument.</p>
6338 <p>If the condition is a vector of i1, then the value arguments must be vectors
6339 of the same size, and the selection is done element by element.</p>
6343 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
6348 <!-- _______________________________________________________________________ -->
6350 <a name="i_call">'<tt>call</tt>' Instruction</a>
6357 <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>]
6361 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
6364 <p>This instruction requires several arguments:</p>
6367 <li>The optional "tail" marker indicates that the callee function does not
6368 access any allocas or varargs in the caller. Note that calls may be
6369 marked "tail" even if they do not occur before
6370 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
6371 present, the function call is eligible for tail call optimization,
6372 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
6373 optimized into a jump</a>. The code generator may optimize calls marked
6374 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
6375 sibling call optimization</a> when the caller and callee have
6376 matching signatures, or 2) forced tail call optimization when the
6377 following extra requirements are met:
6379 <li>Caller and callee both have the calling
6380 convention <tt>fastcc</tt>.</li>
6381 <li>The call is in tail position (ret immediately follows call and ret
6382 uses value of call or is void).</li>
6383 <li>Option <tt>-tailcallopt</tt> is enabled,
6384 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
6385 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
6386 constraints are met.</a></li>
6390 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
6391 convention</a> the call should use. If none is specified, the call
6392 defaults to using C calling conventions. The calling convention of the
6393 call must match the calling convention of the target function, or else the
6394 behavior is undefined.</li>
6396 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
6397 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
6398 '<tt>inreg</tt>' attributes are valid here.</li>
6400 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
6401 type of the return value. Functions that return no value are marked
6402 <tt><a href="#t_void">void</a></tt>.</li>
6404 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
6405 being invoked. The argument types must match the types implied by this
6406 signature. This type can be omitted if the function is not varargs and if
6407 the function type does not return a pointer to a function.</li>
6409 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
6410 be invoked. In most cases, this is a direct function invocation, but
6411 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
6412 to function value.</li>
6414 <li>'<tt>function args</tt>': argument list whose types match the function
6415 signature argument types and parameter attributes. All arguments must be
6416 of <a href="#t_firstclass">first class</a> type. If the function
6417 signature indicates the function accepts a variable number of arguments,
6418 the extra arguments can be specified.</li>
6420 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
6421 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
6422 '<tt>readnone</tt>' attributes are valid here.</li>
6426 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
6427 a specified function, with its incoming arguments bound to the specified
6428 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
6429 function, control flow continues with the instruction after the function
6430 call, and the return value of the function is bound to the result
6435 %retval = call i32 @test(i32 %argc)
6436 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
6437 %X = tail call i32 @foo() <i>; yields i32</i>
6438 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
6439 call void %foo(i8 97 signext)
6441 %struct.A = type { i32, i8 }
6442 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
6443 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
6444 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
6445 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
6446 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
6449 <p>llvm treats calls to some functions with names and arguments that match the
6450 standard C99 library as being the C99 library functions, and may perform
6451 optimizations or generate code for them under that assumption. This is
6452 something we'd like to change in the future to provide better support for
6453 freestanding environments and non-C-based languages.</p>
6457 <!-- _______________________________________________________________________ -->
6459 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
6466 <resultval> = va_arg <va_list*> <arglist>, <argty>
6470 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6471 the "variable argument" area of a function call. It is used to implement the
6472 <tt>va_arg</tt> macro in C.</p>
6475 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6476 argument. It returns a value of the specified argument type and increments
6477 the <tt>va_list</tt> to point to the next argument. The actual type
6478 of <tt>va_list</tt> is target specific.</p>
6481 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6482 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6483 to the next argument. For more information, see the variable argument
6484 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6486 <p>It is legal for this instruction to be called in a function which does not
6487 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6490 <p><tt>va_arg</tt> is an LLVM instruction instead of
6491 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6495 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6497 <p>Note that the code generator does not yet fully support va_arg on many
6498 targets. Also, it does not currently support va_arg with aggregate types on
6503 <!-- _______________________________________________________________________ -->
6505 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6512 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6513 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6515 <clause> := catch <type> <value>
6516 <clause> := filter <array constant type> <array constant>
6520 <p>The '<tt>landingpad</tt>' instruction is used by
6521 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6522 system</a> to specify that a basic block is a landing pad — one where
6523 the exception lands, and corresponds to the code found in the
6524 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6525 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6526 re-entry to the function. The <tt>resultval</tt> has the
6527 type <tt>resultty</tt>.</p>
6530 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6531 function associated with the unwinding mechanism. The optional
6532 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6534 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6535 or <tt>filter</tt> — and contains the global variable representing the
6536 "type" that may be caught or filtered respectively. Unlike the
6537 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6538 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6539 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6540 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6543 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6544 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6545 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6546 calling conventions, how the personality function results are represented in
6547 LLVM IR is target specific.</p>
6549 <p>The clauses are applied in order from top to bottom. If two
6550 <tt>landingpad</tt> instructions are merged together through inlining, the
6551 clauses from the calling function are appended to the list of clauses.
6552 When the call stack is being unwound due to an exception being thrown, the
6553 exception is compared against each <tt>clause</tt> in turn. If it doesn't
6554 match any of the clauses, and the <tt>cleanup</tt> flag is not set, then
6555 unwinding continues further up the call stack.</p>
6557 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6560 <li>A landing pad block is a basic block which is the unwind destination of an
6561 '<tt>invoke</tt>' instruction.</li>
6562 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6563 first non-PHI instruction.</li>
6564 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6566 <li>A basic block that is not a landing pad block may not include a
6567 '<tt>landingpad</tt>' instruction.</li>
6568 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6569 personality function.</li>
6574 ;; A landing pad which can catch an integer.
6575 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6577 ;; A landing pad that is a cleanup.
6578 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6580 ;; A landing pad which can catch an integer and can only throw a double.
6581 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6583 filter [1 x i8**] [@_ZTId]
6592 <!-- *********************************************************************** -->
6593 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6594 <!-- *********************************************************************** -->
6598 <p>LLVM supports the notion of an "intrinsic function". These functions have
6599 well known names and semantics and are required to follow certain
6600 restrictions. Overall, these intrinsics represent an extension mechanism for
6601 the LLVM language that does not require changing all of the transformations
6602 in LLVM when adding to the language (or the bitcode reader/writer, the
6603 parser, etc...).</p>
6605 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6606 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6607 begin with this prefix. Intrinsic functions must always be external
6608 functions: you cannot define the body of intrinsic functions. Intrinsic
6609 functions may only be used in call or invoke instructions: it is illegal to
6610 take the address of an intrinsic function. Additionally, because intrinsic
6611 functions are part of the LLVM language, it is required if any are added that
6612 they be documented here.</p>
6614 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6615 family of functions that perform the same operation but on different data
6616 types. Because LLVM can represent over 8 million different integer types,
6617 overloading is used commonly to allow an intrinsic function to operate on any
6618 integer type. One or more of the argument types or the result type can be
6619 overloaded to accept any integer type. Argument types may also be defined as
6620 exactly matching a previous argument's type or the result type. This allows
6621 an intrinsic function which accepts multiple arguments, but needs all of them
6622 to be of the same type, to only be overloaded with respect to a single
6623 argument or the result.</p>
6625 <p>Overloaded intrinsics will have the names of its overloaded argument types
6626 encoded into its function name, each preceded by a period. Only those types
6627 which are overloaded result in a name suffix. Arguments whose type is matched
6628 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6629 can take an integer of any width and returns an integer of exactly the same
6630 integer width. This leads to a family of functions such as
6631 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6632 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6633 suffix is required. Because the argument's type is matched against the return
6634 type, it does not require its own name suffix.</p>
6636 <p>To learn how to add an intrinsic function, please see the
6637 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6639 <!-- ======================================================================= -->
6641 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6646 <p>Variable argument support is defined in LLVM with
6647 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6648 intrinsic functions. These functions are related to the similarly named
6649 macros defined in the <tt><stdarg.h></tt> header file.</p>
6651 <p>All of these functions operate on arguments that use a target-specific value
6652 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6653 not define what this type is, so all transformations should be prepared to
6654 handle these functions regardless of the type used.</p>
6656 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6657 instruction and the variable argument handling intrinsic functions are
6660 <pre class="doc_code">
6661 define i32 @test(i32 %X, ...) {
6662 ; Initialize variable argument processing
6664 %ap2 = bitcast i8** %ap to i8*
6665 call void @llvm.va_start(i8* %ap2)
6667 ; Read a single integer argument
6668 %tmp = va_arg i8** %ap, i32
6670 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6672 %aq2 = bitcast i8** %aq to i8*
6673 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6674 call void @llvm.va_end(i8* %aq2)
6676 ; Stop processing of arguments.
6677 call void @llvm.va_end(i8* %ap2)
6681 declare void @llvm.va_start(i8*)
6682 declare void @llvm.va_copy(i8*, i8*)
6683 declare void @llvm.va_end(i8*)
6686 <!-- _______________________________________________________________________ -->
6688 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6696 declare void %llvm.va_start(i8* <arglist>)
6700 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6701 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6704 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6707 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6708 macro available in C. In a target-dependent way, it initializes
6709 the <tt>va_list</tt> element to which the argument points, so that the next
6710 call to <tt>va_arg</tt> will produce the first variable argument passed to
6711 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6712 need to know the last argument of the function as the compiler can figure
6717 <!-- _______________________________________________________________________ -->
6719 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6726 declare void @llvm.va_end(i8* <arglist>)
6730 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6731 which has been initialized previously
6732 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6733 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6736 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6739 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6740 macro available in C. In a target-dependent way, it destroys
6741 the <tt>va_list</tt> element to which the argument points. Calls
6742 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6743 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6744 with calls to <tt>llvm.va_end</tt>.</p>
6748 <!-- _______________________________________________________________________ -->
6750 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6757 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6761 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6762 from the source argument list to the destination argument list.</p>
6765 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6766 The second argument is a pointer to a <tt>va_list</tt> element to copy
6770 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6771 macro available in C. In a target-dependent way, it copies the
6772 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6773 element. This intrinsic is necessary because
6774 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6775 arbitrarily complex and require, for example, memory allocation.</p>
6781 <!-- ======================================================================= -->
6783 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6788 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6789 Collection</a> (GC) requires the implementation and generation of these
6790 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6791 roots on the stack</a>, as well as garbage collector implementations that
6792 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6793 barriers. Front-ends for type-safe garbage collected languages should generate
6794 these intrinsics to make use of the LLVM garbage collectors. For more details,
6795 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6798 <p>The garbage collection intrinsics only operate on objects in the generic
6799 address space (address space zero).</p>
6801 <!-- _______________________________________________________________________ -->
6803 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6810 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6814 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6815 the code generator, and allows some metadata to be associated with it.</p>
6818 <p>The first argument specifies the address of a stack object that contains the
6819 root pointer. The second pointer (which must be either a constant or a
6820 global value address) contains the meta-data to be associated with the
6824 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6825 location. At compile-time, the code generator generates information to allow
6826 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6827 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6832 <!-- _______________________________________________________________________ -->
6834 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6841 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6845 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6846 locations, allowing garbage collector implementations that require read
6850 <p>The second argument is the address to read from, which should be an address
6851 allocated from the garbage collector. The first object is a pointer to the
6852 start of the referenced object, if needed by the language runtime (otherwise
6856 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6857 instruction, but may be replaced with substantially more complex code by the
6858 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6859 may only be used in a function which <a href="#gc">specifies a GC
6864 <!-- _______________________________________________________________________ -->
6866 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6873 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6877 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6878 locations, allowing garbage collector implementations that require write
6879 barriers (such as generational or reference counting collectors).</p>
6882 <p>The first argument is the reference to store, the second is the start of the
6883 object to store it to, and the third is the address of the field of Obj to
6884 store to. If the runtime does not require a pointer to the object, Obj may
6888 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6889 instruction, but may be replaced with substantially more complex code by the
6890 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6891 may only be used in a function which <a href="#gc">specifies a GC
6898 <!-- ======================================================================= -->
6900 <a name="int_codegen">Code Generator Intrinsics</a>
6905 <p>These intrinsics are provided by LLVM to expose special features that may
6906 only be implemented with code generator support.</p>
6908 <!-- _______________________________________________________________________ -->
6910 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6917 declare i8 *@llvm.returnaddress(i32 <level>)
6921 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6922 target-specific value indicating the return address of the current function
6923 or one of its callers.</p>
6926 <p>The argument to this intrinsic indicates which function to return the address
6927 for. Zero indicates the calling function, one indicates its caller, etc.
6928 The argument is <b>required</b> to be a constant integer value.</p>
6931 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6932 indicating the return address of the specified call frame, or zero if it
6933 cannot be identified. The value returned by this intrinsic is likely to be
6934 incorrect or 0 for arguments other than zero, so it should only be used for
6935 debugging purposes.</p>
6937 <p>Note that calling this intrinsic does not prevent function inlining or other
6938 aggressive transformations, so the value returned may not be that of the
6939 obvious source-language caller.</p>
6943 <!-- _______________________________________________________________________ -->
6945 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6952 declare i8* @llvm.frameaddress(i32 <level>)
6956 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6957 target-specific frame pointer value for the specified stack frame.</p>
6960 <p>The argument to this intrinsic indicates which function to return the frame
6961 pointer for. Zero indicates the calling function, one indicates its caller,
6962 etc. The argument is <b>required</b> to be a constant integer value.</p>
6965 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6966 indicating the frame address of the specified call frame, or zero if it
6967 cannot be identified. The value returned by this intrinsic is likely to be
6968 incorrect or 0 for arguments other than zero, so it should only be used for
6969 debugging purposes.</p>
6971 <p>Note that calling this intrinsic does not prevent function inlining or other
6972 aggressive transformations, so the value returned may not be that of the
6973 obvious source-language caller.</p>
6977 <!-- _______________________________________________________________________ -->
6979 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6986 declare i8* @llvm.stacksave()
6990 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6991 of the function stack, for use
6992 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6993 useful for implementing language features like scoped automatic variable
6994 sized arrays in C99.</p>
6997 <p>This intrinsic returns a opaque pointer value that can be passed
6998 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6999 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
7000 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
7001 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
7002 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
7003 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
7007 <!-- _______________________________________________________________________ -->
7009 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
7016 declare void @llvm.stackrestore(i8* %ptr)
7020 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
7021 the function stack to the state it was in when the
7022 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
7023 executed. This is useful for implementing language features like scoped
7024 automatic variable sized arrays in C99.</p>
7027 <p>See the description
7028 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
7032 <!-- _______________________________________________________________________ -->
7034 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
7041 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7045 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
7046 insert a prefetch instruction if supported; otherwise, it is a noop.
7047 Prefetches have no effect on the behavior of the program but can change its
7048 performance characteristics.</p>
7051 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
7052 specifier determining if the fetch should be for a read (0) or write (1),
7053 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
7054 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
7055 specifies whether the prefetch is performed on the data (1) or instruction (0)
7056 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
7057 must be constant integers.</p>
7060 <p>This intrinsic does not modify the behavior of the program. In particular,
7061 prefetches cannot trap and do not produce a value. On targets that support
7062 this intrinsic, the prefetch can provide hints to the processor cache for
7063 better performance.</p>
7067 <!-- _______________________________________________________________________ -->
7069 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
7076 declare void @llvm.pcmarker(i32 <id>)
7080 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
7081 Counter (PC) in a region of code to simulators and other tools. The method
7082 is target specific, but it is expected that the marker will use exported
7083 symbols to transmit the PC of the marker. The marker makes no guarantees
7084 that it will remain with any specific instruction after optimizations. It is
7085 possible that the presence of a marker will inhibit optimizations. The
7086 intended use is to be inserted after optimizations to allow correlations of
7087 simulation runs.</p>
7090 <p><tt>id</tt> is a numerical id identifying the marker.</p>
7093 <p>This intrinsic does not modify the behavior of the program. Backends that do
7094 not support this intrinsic may ignore it.</p>
7098 <!-- _______________________________________________________________________ -->
7100 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
7107 declare i64 @llvm.readcyclecounter()
7111 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
7112 counter register (or similar low latency, high accuracy clocks) on those
7113 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7114 should map to RPCC. As the backing counters overflow quickly (on the order
7115 of 9 seconds on alpha), this should only be used for small timings.</p>
7118 <p>When directly supported, reading the cycle counter should not modify any
7119 memory. Implementations are allowed to either return a application specific
7120 value or a system wide value. On backends without support, this is lowered
7121 to a constant 0.</p>
7127 <!-- ======================================================================= -->
7129 <a name="int_libc">Standard C Library Intrinsics</a>
7134 <p>LLVM provides intrinsics for a few important standard C library functions.
7135 These intrinsics allow source-language front-ends to pass information about
7136 the alignment of the pointer arguments to the code generator, providing
7137 opportunity for more efficient code generation.</p>
7139 <!-- _______________________________________________________________________ -->
7141 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
7147 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
7148 integer bit width and for different address spaces. Not all targets support
7149 all bit widths however.</p>
7152 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7153 i32 <len>, i32 <align>, i1 <isvolatile>)
7154 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7155 i64 <len>, i32 <align>, i1 <isvolatile>)
7159 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7160 source location to the destination location.</p>
7162 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
7163 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7164 and the pointers can be in specified address spaces.</p>
7168 <p>The first argument is a pointer to the destination, the second is a pointer
7169 to the source. The third argument is an integer argument specifying the
7170 number of bytes to copy, the fourth argument is the alignment of the
7171 source and destination locations, and the fifth is a boolean indicating a
7172 volatile access.</p>
7174 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7175 then the caller guarantees that both the source and destination pointers are
7176 aligned to that boundary.</p>
7178 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7179 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
7180 The detailed access behavior is not very cleanly specified and it is unwise
7181 to depend on it.</p>
7185 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7186 source location to the destination location, which are not allowed to
7187 overlap. It copies "len" bytes of memory over. If the argument is known to
7188 be aligned to some boundary, this can be specified as the fourth argument,
7189 otherwise it should be set to 0 or 1.</p>
7193 <!-- _______________________________________________________________________ -->
7195 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
7201 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
7202 width and for different address space. Not all targets support all bit
7206 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7207 i32 <len>, i32 <align>, i1 <isvolatile>)
7208 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7209 i64 <len>, i32 <align>, i1 <isvolatile>)
7213 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
7214 source location to the destination location. It is similar to the
7215 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
7218 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
7219 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7220 and the pointers can be in specified address spaces.</p>
7224 <p>The first argument is a pointer to the destination, the second is a pointer
7225 to the source. The third argument is an integer argument specifying the
7226 number of bytes to copy, the fourth argument is the alignment of the
7227 source and destination locations, and the fifth is a boolean indicating a
7228 volatile access.</p>
7230 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7231 then the caller guarantees that the source and destination pointers are
7232 aligned to that boundary.</p>
7234 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7235 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
7236 The detailed access behavior is not very cleanly specified and it is unwise
7237 to depend on it.</p>
7241 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
7242 source location to the destination location, which may overlap. It copies
7243 "len" bytes of memory over. If the argument is known to be aligned to some
7244 boundary, this can be specified as the fourth argument, otherwise it should
7245 be set to 0 or 1.</p>
7249 <!-- _______________________________________________________________________ -->
7251 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
7257 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
7258 width and for different address spaces. However, not all targets support all
7262 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7263 i32 <len>, i32 <align>, i1 <isvolatile>)
7264 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7265 i64 <len>, i32 <align>, i1 <isvolatile>)
7269 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
7270 particular byte value.</p>
7272 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
7273 intrinsic does not return a value and takes extra alignment/volatile
7274 arguments. Also, the destination can be in an arbitrary address space.</p>
7277 <p>The first argument is a pointer to the destination to fill, the second is the
7278 byte value with which to fill it, the third argument is an integer argument
7279 specifying the number of bytes to fill, and the fourth argument is the known
7280 alignment of the destination location.</p>
7282 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7283 then the caller guarantees that the destination pointer is aligned to that
7286 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7287 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
7288 The detailed access behavior is not very cleanly specified and it is unwise
7289 to depend on it.</p>
7292 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
7293 at the destination location. If the argument is known to be aligned to some
7294 boundary, this can be specified as the fourth argument, otherwise it should
7295 be set to 0 or 1.</p>
7299 <!-- _______________________________________________________________________ -->
7301 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
7307 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
7308 floating point or vector of floating point type. Not all targets support all
7312 declare float @llvm.sqrt.f32(float %Val)
7313 declare double @llvm.sqrt.f64(double %Val)
7314 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7315 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7316 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7320 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
7321 returning the same value as the libm '<tt>sqrt</tt>' functions would.
7322 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
7323 behavior for negative numbers other than -0.0 (which allows for better
7324 optimization, because there is no need to worry about errno being
7325 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
7328 <p>The argument and return value are floating point numbers of the same
7332 <p>This function returns the sqrt of the specified operand if it is a
7333 nonnegative floating point number.</p>
7337 <!-- _______________________________________________________________________ -->
7339 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
7345 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
7346 floating point or vector of floating point type. Not all targets support all
7350 declare float @llvm.powi.f32(float %Val, i32 %power)
7351 declare double @llvm.powi.f64(double %Val, i32 %power)
7352 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7353 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7354 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7358 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
7359 specified (positive or negative) power. The order of evaluation of
7360 multiplications is not defined. When a vector of floating point type is
7361 used, the second argument remains a scalar integer value.</p>
7364 <p>The second argument is an integer power, and the first is a value to raise to
7368 <p>This function returns the first value raised to the second power with an
7369 unspecified sequence of rounding operations.</p>
7373 <!-- _______________________________________________________________________ -->
7375 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
7381 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
7382 floating point or vector of floating point type. Not all targets support all
7386 declare float @llvm.sin.f32(float %Val)
7387 declare double @llvm.sin.f64(double %Val)
7388 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7389 declare fp128 @llvm.sin.f128(fp128 %Val)
7390 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7394 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
7397 <p>The argument and return value are floating point numbers of the same
7401 <p>This function returns the sine of the specified operand, returning the same
7402 values as the libm <tt>sin</tt> functions would, and handles error conditions
7403 in the same way.</p>
7407 <!-- _______________________________________________________________________ -->
7409 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
7415 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
7416 floating point or vector of floating point type. Not all targets support all
7420 declare float @llvm.cos.f32(float %Val)
7421 declare double @llvm.cos.f64(double %Val)
7422 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7423 declare fp128 @llvm.cos.f128(fp128 %Val)
7424 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7428 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
7431 <p>The argument and return value are floating point numbers of the same
7435 <p>This function returns the cosine of the specified operand, returning the same
7436 values as the libm <tt>cos</tt> functions would, and handles error conditions
7437 in the same way.</p>
7441 <!-- _______________________________________________________________________ -->
7443 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
7449 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
7450 floating point or vector of floating point type. Not all targets support all
7454 declare float @llvm.pow.f32(float %Val, float %Power)
7455 declare double @llvm.pow.f64(double %Val, double %Power)
7456 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7457 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7458 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7462 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
7463 specified (positive or negative) power.</p>
7466 <p>The second argument is a floating point power, and the first is a value to
7467 raise to that power.</p>
7470 <p>This function returns the first value raised to the second power, returning
7471 the same values as the libm <tt>pow</tt> functions would, and handles error
7472 conditions in the same way.</p>
7476 <!-- _______________________________________________________________________ -->
7478 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7484 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7485 floating point or vector of floating point type. Not all targets support all
7489 declare float @llvm.exp.f32(float %Val)
7490 declare double @llvm.exp.f64(double %Val)
7491 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7492 declare fp128 @llvm.exp.f128(fp128 %Val)
7493 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7497 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7500 <p>The argument and return value are floating point numbers of the same
7504 <p>This function returns the same values as the libm <tt>exp</tt> functions
7505 would, and handles error conditions in the same way.</p>
7509 <!-- _______________________________________________________________________ -->
7511 <a name="int_exp2">'<tt>llvm.exp2.*</tt>' Intrinsic</a>
7517 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp2</tt> on any
7518 floating point or vector of floating point type. Not all targets support all
7522 declare float @llvm.exp2.f32(float %Val)
7523 declare double @llvm.exp2.f64(double %Val)
7524 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7525 declare fp128 @llvm.exp2.f128(fp128 %Val)
7526 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7530 <p>The '<tt>llvm.exp2.*</tt>' intrinsics perform the exp2 function.</p>
7533 <p>The argument and return value are floating point numbers of the same
7537 <p>This function returns the same values as the libm <tt>exp2</tt> functions
7538 would, and handles error conditions in the same way.</p>
7542 <!-- _______________________________________________________________________ -->
7544 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7550 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7551 floating point or vector of floating point type. Not all targets support all
7555 declare float @llvm.log.f32(float %Val)
7556 declare double @llvm.log.f64(double %Val)
7557 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7558 declare fp128 @llvm.log.f128(fp128 %Val)
7559 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7563 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7566 <p>The argument and return value are floating point numbers of the same
7570 <p>This function returns the same values as the libm <tt>log</tt> functions
7571 would, and handles error conditions in the same way.</p>
7575 <!-- _______________________________________________________________________ -->
7577 <a name="int_log10">'<tt>llvm.log10.*</tt>' Intrinsic</a>
7583 <p>This is an overloaded intrinsic. You can use <tt>llvm.log10</tt> on any
7584 floating point or vector of floating point type. Not all targets support all
7588 declare float @llvm.log10.f32(float %Val)
7589 declare double @llvm.log10.f64(double %Val)
7590 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7591 declare fp128 @llvm.log10.f128(fp128 %Val)
7592 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7596 <p>The '<tt>llvm.log10.*</tt>' intrinsics perform the log10 function.</p>
7599 <p>The argument and return value are floating point numbers of the same
7603 <p>This function returns the same values as the libm <tt>log10</tt> functions
7604 would, and handles error conditions in the same way.</p>
7608 <!-- _______________________________________________________________________ -->
7610 <a name="int_log2">'<tt>llvm.log2.*</tt>' Intrinsic</a>
7616 <p>This is an overloaded intrinsic. You can use <tt>llvm.log2</tt> on any
7617 floating point or vector of floating point type. Not all targets support all
7621 declare float @llvm.log2.f32(float %Val)
7622 declare double @llvm.log2.f64(double %Val)
7623 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7624 declare fp128 @llvm.log2.f128(fp128 %Val)
7625 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7629 <p>The '<tt>llvm.log2.*</tt>' intrinsics perform the log2 function.</p>
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>log2</tt> functions
7637 would, and handles error conditions in the same way.</p>
7641 <!-- _______________________________________________________________________ -->
7643 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7649 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7650 floating point or vector of floating point type. Not all targets support all
7654 declare float @llvm.fma.f32(float %a, float %b, float %c)
7655 declare double @llvm.fma.f64(double %a, double %b, double %c)
7656 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7657 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7658 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7662 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
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>fma</tt> functions
7675 <!-- _______________________________________________________________________ -->
7677 <a name="int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a>
7683 <p>This is an overloaded intrinsic. You can use <tt>llvm.fabs</tt> on any
7684 floating point or vector of floating point type. Not all targets support all
7688 declare float @llvm.fabs.f32(float %Val)
7689 declare double @llvm.fabs.f64(double %Val)
7690 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7691 declare fp128 @llvm.fabs.f128(fp128 %Val)
7692 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7696 <p>The '<tt>llvm.fabs.*</tt>' intrinsics return the absolute value of
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>fabs</tt> functions
7705 would, and handles error conditions in the same way.</p>
7709 <!-- _______________________________________________________________________ -->
7711 <a name="int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a>
7717 <p>This is an overloaded intrinsic. You can use <tt>llvm.floor</tt> on any
7718 floating point or vector of floating point type. Not all targets support all
7722 declare float @llvm.floor.f32(float %Val)
7723 declare double @llvm.floor.f64(double %Val)
7724 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7725 declare fp128 @llvm.floor.f128(fp128 %Val)
7726 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7730 <p>The '<tt>llvm.floor.*</tt>' intrinsics return the floor of
7734 <p>The argument and return value are floating point numbers of the same
7738 <p>This function returns the same values as the libm <tt>floor</tt> functions
7739 would, and handles error conditions in the same way.</p>
7743 <!-- _______________________________________________________________________ -->
7745 <a name="int_ceil">'<tt>llvm.ceil.*</tt>' Intrinsic</a>
7751 <p>This is an overloaded intrinsic. You can use <tt>llvm.ceil</tt> on any
7752 floating point or vector of floating point type. Not all targets support all
7756 declare float @llvm.ceil.f32(float %Val)
7757 declare double @llvm.ceil.f64(double %Val)
7758 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7759 declare fp128 @llvm.ceil.f128(fp128 %Val)
7760 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7764 <p>The '<tt>llvm.ceil.*</tt>' intrinsics return the ceiling of
7768 <p>The argument and return value are floating point numbers of the same
7772 <p>This function returns the same values as the libm <tt>ceil</tt> functions
7773 would, and handles error conditions in the same way.</p>
7777 <!-- _______________________________________________________________________ -->
7779 <a name="int_trunc">'<tt>llvm.trunc.*</tt>' Intrinsic</a>
7785 <p>This is an overloaded intrinsic. You can use <tt>llvm.trunc</tt> on any
7786 floating point or vector of floating point type. Not all targets support all
7790 declare float @llvm.trunc.f32(float %Val)
7791 declare double @llvm.trunc.f64(double %Val)
7792 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7793 declare fp128 @llvm.trunc.f128(fp128 %Val)
7794 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7798 <p>The '<tt>llvm.trunc.*</tt>' intrinsics returns the operand rounded to the
7799 nearest integer not larger in magnitude than the operand.</p>
7802 <p>The argument and return value are floating point numbers of the same
7806 <p>This function returns the same values as the libm <tt>trunc</tt> functions
7807 would, and handles error conditions in the same way.</p>
7811 <!-- _______________________________________________________________________ -->
7813 <a name="int_rint">'<tt>llvm.rint.*</tt>' Intrinsic</a>
7819 <p>This is an overloaded intrinsic. You can use <tt>llvm.rint</tt> on any
7820 floating point or vector of floating point type. Not all targets support all
7824 declare float @llvm.rint.f32(float %Val)
7825 declare double @llvm.rint.f64(double %Val)
7826 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7827 declare fp128 @llvm.rint.f128(fp128 %Val)
7828 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7832 <p>The '<tt>llvm.rint.*</tt>' intrinsics returns the operand rounded to the
7833 nearest integer. It may raise an inexact floating-point exception if the
7834 operand isn't an integer.</p>
7837 <p>The argument and return value are floating point numbers of the same
7841 <p>This function returns the same values as the libm <tt>rint</tt> functions
7842 would, and handles error conditions in the same way.</p>
7846 <!-- _______________________________________________________________________ -->
7848 <a name="int_nearbyint">'<tt>llvm.nearbyint.*</tt>' Intrinsic</a>
7854 <p>This is an overloaded intrinsic. You can use <tt>llvm.nearbyint</tt> on any
7855 floating point or vector of floating point type. Not all targets support all
7859 declare float @llvm.nearbyint.f32(float %Val)
7860 declare double @llvm.nearbyint.f64(double %Val)
7861 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7862 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7863 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7867 <p>The '<tt>llvm.nearbyint.*</tt>' intrinsics returns the operand rounded to the
7868 nearest integer.</p>
7871 <p>The argument and return value are floating point numbers of the same
7875 <p>This function returns the same values as the libm <tt>nearbyint</tt>
7876 functions would, and handles error conditions in the same way.</p>
7882 <!-- ======================================================================= -->
7884 <a name="int_manip">Bit Manipulation Intrinsics</a>
7889 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7890 These allow efficient code generation for some algorithms.</p>
7892 <!-- _______________________________________________________________________ -->
7894 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7900 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7901 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7904 declare i16 @llvm.bswap.i16(i16 <id>)
7905 declare i32 @llvm.bswap.i32(i32 <id>)
7906 declare i64 @llvm.bswap.i64(i64 <id>)
7910 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7911 values with an even number of bytes (positive multiple of 16 bits). These
7912 are useful for performing operations on data that is not in the target's
7913 native byte order.</p>
7916 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7917 and low byte of the input i16 swapped. Similarly,
7918 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7919 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7920 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7921 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7922 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7923 more, respectively).</p>
7927 <!-- _______________________________________________________________________ -->
7929 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7935 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7936 width, or on any vector with integer elements. Not all targets support all
7937 bit widths or vector types, however.</p>
7940 declare i8 @llvm.ctpop.i8(i8 <src>)
7941 declare i16 @llvm.ctpop.i16(i16 <src>)
7942 declare i32 @llvm.ctpop.i32(i32 <src>)
7943 declare i64 @llvm.ctpop.i64(i64 <src>)
7944 declare i256 @llvm.ctpop.i256(i256 <src>)
7945 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7949 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7953 <p>The only argument is the value to be counted. The argument may be of any
7954 integer type, or a vector with integer elements.
7955 The return type must match the argument type.</p>
7958 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7959 element of a vector.</p>
7963 <!-- _______________________________________________________________________ -->
7965 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7971 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7972 integer bit width, or any vector whose elements are integers. Not all
7973 targets support all bit widths or vector types, however.</p>
7976 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7977 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7978 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7979 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7980 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7981 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7985 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7986 leading zeros in a variable.</p>
7989 <p>The first argument is the value to be counted. This argument may be of any
7990 integer type, or a vectory with integer element type. The return type
7991 must match the first argument type.</p>
7993 <p>The second argument must be a constant and is a flag to indicate whether the
7994 intrinsic should ensure that a zero as the first argument produces a defined
7995 result. Historically some architectures did not provide a defined result for
7996 zero values as efficiently, and many algorithms are now predicated on
7997 avoiding zero-value inputs.</p>
8000 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
8001 zeros in a variable, or within each element of the vector.
8002 If <tt>src == 0</tt> then the result is the size in bits of the type of
8003 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
8004 For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
8008 <!-- _______________________________________________________________________ -->
8010 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
8016 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
8017 integer bit width, or any vector of integer elements. Not all targets
8018 support all bit widths or vector types, however.</p>
8021 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
8022 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
8023 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
8024 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
8025 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
8026 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8030 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
8034 <p>The first argument is the value to be counted. This argument may be of any
8035 integer type, or a vectory with integer element type. The return type
8036 must match the first argument type.</p>
8038 <p>The second argument must be a constant and is a flag to indicate whether the
8039 intrinsic should ensure that a zero as the first argument produces a defined
8040 result. Historically some architectures did not provide a defined result for
8041 zero values as efficiently, and many algorithms are now predicated on
8042 avoiding zero-value inputs.</p>
8045 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
8046 zeros in a variable, or within each element of a vector.
8047 If <tt>src == 0</tt> then the result is the size in bits of the type of
8048 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
8049 For example, <tt>llvm.cttz(2) = 1</tt>.</p>
8055 <!-- ======================================================================= -->
8057 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
8062 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
8064 <!-- _______________________________________________________________________ -->
8066 <a name="int_sadd_overflow">
8067 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
8074 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
8075 on any integer bit width.</p>
8078 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8079 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8080 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8084 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
8085 a signed addition of the two arguments, and indicate whether an overflow
8086 occurred during the signed summation.</p>
8089 <p>The arguments (%a and %b) and the first element of the result structure may
8090 be of integer types of any bit width, but they must have the same bit
8091 width. The second element of the result structure must be of
8092 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8093 undergo signed addition.</p>
8096 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
8097 a signed addition of the two variables. They return a structure — the
8098 first element of which is the signed summation, and the second element of
8099 which is a bit specifying if the signed summation resulted in an
8104 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8105 %sum = extractvalue {i32, i1} %res, 0
8106 %obit = extractvalue {i32, i1} %res, 1
8107 br i1 %obit, label %overflow, label %normal
8112 <!-- _______________________________________________________________________ -->
8114 <a name="int_uadd_overflow">
8115 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
8122 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
8123 on any integer bit width.</p>
8126 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8127 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8128 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8132 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
8133 an unsigned addition of the two arguments, and indicate whether a carry
8134 occurred during the unsigned summation.</p>
8137 <p>The arguments (%a and %b) and the first element of the result structure may
8138 be of integer types of any bit width, but they must have the same bit
8139 width. The second element of the result structure must be of
8140 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8141 undergo unsigned addition.</p>
8144 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
8145 an unsigned addition of the two arguments. They return a structure —
8146 the first element of which is the sum, and the second element of which is a
8147 bit specifying if the unsigned summation resulted in a carry.</p>
8151 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8152 %sum = extractvalue {i32, i1} %res, 0
8153 %obit = extractvalue {i32, i1} %res, 1
8154 br i1 %obit, label %carry, label %normal
8159 <!-- _______________________________________________________________________ -->
8161 <a name="int_ssub_overflow">
8162 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
8169 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
8170 on any integer bit width.</p>
8173 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8174 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8175 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8179 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
8180 a signed subtraction of the two arguments, and indicate whether an overflow
8181 occurred during the signed subtraction.</p>
8184 <p>The arguments (%a and %b) and the first element of the result structure may
8185 be of integer types of any bit width, but they must have the same bit
8186 width. The second element of the result structure must be of
8187 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8188 undergo signed subtraction.</p>
8191 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
8192 a signed subtraction of the two arguments. They return a structure —
8193 the first element of which is the subtraction, and the second element of
8194 which is a bit specifying if the signed subtraction resulted in an
8199 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8200 %sum = extractvalue {i32, i1} %res, 0
8201 %obit = extractvalue {i32, i1} %res, 1
8202 br i1 %obit, label %overflow, label %normal
8207 <!-- _______________________________________________________________________ -->
8209 <a name="int_usub_overflow">
8210 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
8217 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
8218 on any integer bit width.</p>
8221 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8222 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8223 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8227 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
8228 an unsigned subtraction of the two arguments, and indicate whether an
8229 overflow occurred during the unsigned subtraction.</p>
8232 <p>The arguments (%a and %b) and the first element of the result structure may
8233 be of integer types of any bit width, but they must have the same bit
8234 width. The second element of the result structure must be of
8235 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8236 undergo unsigned subtraction.</p>
8239 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
8240 an unsigned subtraction of the two arguments. They return a structure —
8241 the first element of which is the subtraction, and the second element of
8242 which is a bit specifying if the unsigned subtraction resulted in an
8247 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8248 %sum = extractvalue {i32, i1} %res, 0
8249 %obit = extractvalue {i32, i1} %res, 1
8250 br i1 %obit, label %overflow, label %normal
8255 <!-- _______________________________________________________________________ -->
8257 <a name="int_smul_overflow">
8258 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
8265 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
8266 on any integer bit width.</p>
8269 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8270 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8271 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8276 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
8277 a signed multiplication of the two arguments, and indicate whether an
8278 overflow occurred during the signed multiplication.</p>
8281 <p>The arguments (%a and %b) and the first element of the result structure may
8282 be of integer types of any bit width, but they must have the same bit
8283 width. The second element of the result structure must be of
8284 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8285 undergo signed multiplication.</p>
8288 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
8289 a signed multiplication of the two arguments. They return a structure —
8290 the first element of which is the multiplication, and the second element of
8291 which is a bit specifying if the signed multiplication resulted in an
8296 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8297 %sum = extractvalue {i32, i1} %res, 0
8298 %obit = extractvalue {i32, i1} %res, 1
8299 br i1 %obit, label %overflow, label %normal
8304 <!-- _______________________________________________________________________ -->
8306 <a name="int_umul_overflow">
8307 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
8314 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
8315 on any integer bit width.</p>
8318 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8319 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8320 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8324 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
8325 a unsigned multiplication of the two arguments, and indicate whether an
8326 overflow occurred during the unsigned multiplication.</p>
8329 <p>The arguments (%a and %b) and the first element of the result structure may
8330 be of integer types of any bit width, but they must have the same bit
8331 width. The second element of the result structure must be of
8332 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8333 undergo unsigned multiplication.</p>
8336 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
8337 an unsigned multiplication of the two arguments. They return a structure
8338 — the first element of which is the multiplication, and the second
8339 element of which is a bit specifying if the unsigned multiplication resulted
8344 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8345 %sum = extractvalue {i32, i1} %res, 0
8346 %obit = extractvalue {i32, i1} %res, 1
8347 br i1 %obit, label %overflow, label %normal
8354 <!-- ======================================================================= -->
8356 <a name="spec_arithmetic">Specialised Arithmetic Intrinsics</a>
8359 <!-- _______________________________________________________________________ -->
8362 <a name="fmuladd">'<tt>llvm.fmuladd.*</tt>' Intrinsic</a>
8369 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8370 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8374 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsic functions represent multiply-add
8375 expressions that can be fused if the code generator determines that the fused
8376 expression would be legal and efficient.</p>
8379 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsics each take three arguments: two
8380 multiplicands, a and b, and an addend c.</p>
8383 <p>The expression:</p>
8385 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8387 <p>is equivalent to the expression a * b + c, except that rounding will not be
8388 performed between the multiplication and addition steps if the code generator
8389 fuses the operations. Fusion is not guaranteed, even if the target platform
8390 supports it. If a fused multiply-add is required the corresponding llvm.fma.*
8391 intrinsic function should be used instead.</p>
8395 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8400 <!-- ======================================================================= -->
8402 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
8407 <p>For most target platforms, half precision floating point is a storage-only
8408 format. This means that it is
8409 a dense encoding (in memory) but does not support computation in the
8412 <p>This means that code must first load the half-precision floating point
8413 value as an i16, then convert it to float with <a
8414 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
8415 Computation can then be performed on the float value (including extending to
8416 double etc). To store the value back to memory, it is first converted to
8417 float if needed, then converted to i16 with
8418 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
8419 storing as an i16 value.</p>
8421 <!-- _______________________________________________________________________ -->
8423 <a name="int_convert_to_fp16">
8424 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
8432 declare i16 @llvm.convert.to.fp16(f32 %a)
8436 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8437 a conversion from single precision floating point format to half precision
8438 floating point format.</p>
8441 <p>The intrinsic function contains single argument - the value to be
8445 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8446 a conversion from single precision floating point format to half precision
8447 floating point format. The return value is an <tt>i16</tt> which
8448 contains the converted number.</p>
8452 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8453 store i16 %res, i16* @x, align 2
8458 <!-- _______________________________________________________________________ -->
8460 <a name="int_convert_from_fp16">
8461 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
8469 declare f32 @llvm.convert.from.fp16(i16 %a)
8473 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
8474 a conversion from half precision floating point format to single precision
8475 floating point format.</p>
8478 <p>The intrinsic function contains single argument - the value to be
8482 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
8483 conversion from half single precision floating point format to single
8484 precision floating point format. The input half-float value is represented by
8485 an <tt>i16</tt> value.</p>
8489 %a = load i16* @x, align 2
8490 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8497 <!-- ======================================================================= -->
8499 <a name="int_debugger">Debugger Intrinsics</a>
8504 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
8505 prefix), are described in
8506 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
8507 Level Debugging</a> document.</p>
8511 <!-- ======================================================================= -->
8513 <a name="int_eh">Exception Handling Intrinsics</a>
8518 <p>The LLVM exception handling intrinsics (which all start with
8519 <tt>llvm.eh.</tt> prefix), are described in
8520 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
8521 Handling</a> document.</p>
8525 <!-- ======================================================================= -->
8527 <a name="int_trampoline">Trampoline Intrinsics</a>
8532 <p>These intrinsics make it possible to excise one parameter, marked with
8533 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
8534 The result is a callable
8535 function pointer lacking the nest parameter - the caller does not need to
8536 provide a value for it. Instead, the value to use is stored in advance in a
8537 "trampoline", a block of memory usually allocated on the stack, which also
8538 contains code to splice the nest value into the argument list. This is used
8539 to implement the GCC nested function address extension.</p>
8541 <p>For example, if the function is
8542 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
8543 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
8546 <pre class="doc_code">
8547 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8548 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8549 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8550 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8551 %fp = bitcast i8* %p to i32 (i32, i32)*
8554 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
8555 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
8557 <!-- _______________________________________________________________________ -->
8560 '<tt>llvm.init.trampoline</tt>' Intrinsic
8568 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8572 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
8573 turning it into a trampoline.</p>
8576 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
8577 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
8578 sufficiently aligned block of memory; this memory is written to by the
8579 intrinsic. Note that the size and the alignment are target-specific - LLVM
8580 currently provides no portable way of determining them, so a front-end that
8581 generates this intrinsic needs to have some target-specific knowledge.
8582 The <tt>func</tt> argument must hold a function bitcast to
8583 an <tt>i8*</tt>.</p>
8586 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
8587 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
8588 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
8589 which can be <a href="#int_trampoline">bitcast (to a new function) and
8590 called</a>. The new function's signature is the same as that of
8591 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
8592 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
8593 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
8594 with the same argument list, but with <tt>nval</tt> used for the missing
8595 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
8596 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
8597 to the returned function pointer is undefined.</p>
8600 <!-- _______________________________________________________________________ -->
8603 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
8611 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8615 <p>This performs any required machine-specific adjustment to the address of a
8616 trampoline (passed as <tt>tramp</tt>).</p>
8619 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
8620 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
8624 <p>On some architectures the address of the code to be executed needs to be
8625 different to the address where the trampoline is actually stored. This
8626 intrinsic returns the executable address corresponding to <tt>tramp</tt>
8627 after performing the required machine specific adjustments.
8628 The pointer returned can then be <a href="#int_trampoline"> bitcast and
8636 <!-- ======================================================================= -->
8638 <a name="int_memorymarkers">Memory Use Markers</a>
8643 <p>This class of intrinsics exists to information about the lifetime of memory
8644 objects and ranges where variables are immutable.</p>
8646 <!-- _______________________________________________________________________ -->
8648 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8655 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8659 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8660 object's lifetime.</p>
8663 <p>The first argument is a constant integer representing the size of the
8664 object, or -1 if it is variable sized. The second argument is a pointer to
8668 <p>This intrinsic indicates that before this point in the code, the value of the
8669 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8670 never be used and has an undefined value. A load from the pointer that
8671 precedes this intrinsic can be replaced with
8672 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8676 <!-- _______________________________________________________________________ -->
8678 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8685 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8689 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8690 object's lifetime.</p>
8693 <p>The first argument is a constant integer representing the size of the
8694 object, or -1 if it is variable sized. The second argument is a pointer to
8698 <p>This intrinsic indicates that after this point in the code, the value of the
8699 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8700 never be used and has an undefined value. Any stores into the memory object
8701 following this intrinsic may be removed as dead.
8705 <!-- _______________________________________________________________________ -->
8707 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8714 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8718 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8719 a memory object will not change.</p>
8722 <p>The first argument is a constant integer representing the size of the
8723 object, or -1 if it is variable sized. The second argument is a pointer to
8727 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8728 the return value, the referenced memory location is constant and
8733 <!-- _______________________________________________________________________ -->
8735 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8742 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8746 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8747 a memory object are mutable.</p>
8750 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8751 The second argument is a constant integer representing the size of the
8752 object, or -1 if it is variable sized and the third argument is a pointer
8756 <p>This intrinsic indicates that the memory is mutable again.</p>
8762 <!-- ======================================================================= -->
8764 <a name="int_general">General Intrinsics</a>
8769 <p>This class of intrinsics is designed to be generic and has no specific
8772 <!-- _______________________________________________________________________ -->
8774 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8781 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8785 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8788 <p>The first argument is a pointer to a value, the second is a pointer to a
8789 global string, the third is a pointer to a global string which is the source
8790 file name, and the last argument is the line number.</p>
8793 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8794 This can be useful for special purpose optimizations that want to look for
8795 these annotations. These have no other defined use; they are ignored by code
8796 generation and optimization.</p>
8800 <!-- _______________________________________________________________________ -->
8802 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8808 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8809 any integer bit width.</p>
8812 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8813 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8814 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8815 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8816 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8820 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8823 <p>The first argument is an integer value (result of some expression), the
8824 second is a pointer to a global string, the third is a pointer to a global
8825 string which is the source file name, and the last argument is the line
8826 number. It returns the value of the first argument.</p>
8829 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8830 arbitrary strings. This can be useful for special purpose optimizations that
8831 want to look for these annotations. These have no other defined use; they
8832 are ignored by code generation and optimization.</p>
8836 <!-- _______________________________________________________________________ -->
8838 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8845 declare void @llvm.trap() noreturn nounwind
8849 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8855 <p>This intrinsic is lowered to the target dependent trap instruction. If the
8856 target does not have a trap instruction, this intrinsic will be lowered to
8857 a call of the <tt>abort()</tt> function.</p>
8861 <!-- _______________________________________________________________________ -->
8863 <a name="int_debugtrap">'<tt>llvm.debugtrap</tt>' Intrinsic</a>
8870 declare void @llvm.debugtrap() nounwind
8874 <p>The '<tt>llvm.debugtrap</tt>' intrinsic.</p>
8880 <p>This intrinsic is lowered to code which is intended to cause an execution
8881 trap with the intention of requesting the attention of a debugger.</p>
8885 <!-- _______________________________________________________________________ -->
8887 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8894 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8898 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8899 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8900 ensure that it is placed on the stack before local variables.</p>
8903 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8904 arguments. The first argument is the value loaded from the stack
8905 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8906 that has enough space to hold the value of the guard.</p>
8909 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8910 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8911 stack. This is to ensure that if a local variable on the stack is
8912 overwritten, it will destroy the value of the guard. When the function exits,
8913 the guard on the stack is checked against the original guard. If they are
8914 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8919 <!-- _______________________________________________________________________ -->
8921 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8928 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8929 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8933 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8934 the optimizers to determine at compile time whether a) an operation (like
8935 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8936 runtime check for overflow isn't necessary. An object in this context means
8937 an allocation of a specific class, structure, array, or other object.</p>
8940 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8941 argument is a pointer to or into the <tt>object</tt>. The second argument
8942 is a boolean and determines whether <tt>llvm.objectsize</tt> returns 0 (if
8943 true) or -1 (if false) when the object size is unknown.
8944 The second argument only accepts constants.</p>
8947 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to a constant representing
8948 the size of the object concerned. If the size cannot be determined at compile
8949 time, <tt>llvm.objectsize</tt> returns <tt>i32/i64 -1 or 0</tt>
8950 (depending on the <tt>min</tt> argument).</p>
8953 <!-- _______________________________________________________________________ -->
8955 <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a>
8962 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8963 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8967 <p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the
8968 most probable) value of <tt>val</tt>, which can be used by optimizers.</p>
8971 <p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first
8972 argument is a value. The second argument is an expected value, this needs to
8973 be a constant value, variables are not allowed.</p>
8976 <p>This intrinsic is lowered to the <tt>val</tt>.</p>
8979 <!-- _______________________________________________________________________ -->
8981 <a name="int_donothing">'<tt>llvm.donothing</tt>' Intrinsic</a>
8988 declare void @llvm.donothing() nounwind readnone
8992 <p>The <tt>llvm.donothing</tt> intrinsic doesn't perform any operation. It's the
8993 only intrinsic that can be called with an invoke instruction.</p>
8999 <p>This intrinsic does nothing, and it's removed by optimizers and ignored by
9006 <!-- *********************************************************************** -->
9009 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
9010 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
9011 <a href="http://validator.w3.org/check/referer"><img
9012 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
9014 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
9015 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
9016 Last modified: $Date$