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#id80">Tail calls can only be optimized
733 when this, the GHC or the HiPE convention is used.</a> This calling
734 convention does not support varargs and requires the prototype of all
735 callees to 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#id80">tail call optimization</a> but
762 requires both the caller and callee are using it.
765 <dt><b>"<tt>cc <em>11</em></tt>" - The HiPE calling convention</b>:</dt>
766 <dd>This calling convention has been implemented specifically for use by the
767 <a href="http://www.it.uu.se/research/group/hipe/">High-Performance Erlang
768 (HiPE)</a> compiler, <em>the</em> native code compiler of the
769 <a href="http://www.erlang.org/download.shtml">Ericsson's Open Source
770 Erlang/OTP system</a>. It uses more registers for argument passing than
771 the ordinary C calling convention and defines no callee-saved registers.
772 The calling convention properly supports
773 <a href="CodeGenerator.html#id80">tail call optimization</a> but requires
774 that both the caller and the callee use it. It uses a <em>register
775 pinning</em> mechanism, similar to GHC's convention, for keeping
776 frequently accessed runtime components pinned to specific hardware
777 registers. At the moment only X86 supports this convention (both 32 and 64
780 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
781 <dd>Any calling convention may be specified by number, allowing
782 target-specific calling conventions to be used. Target specific calling
783 conventions start at 64.</dd>
786 <p>More calling conventions can be added/defined on an as-needed basis, to
787 support Pascal conventions or any other well-known target-independent
792 <!-- ======================================================================= -->
794 <a name="visibility">Visibility Styles</a>
799 <p>All Global Variables and Functions have one of the following visibility
803 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
804 <dd>On targets that use the ELF object file format, default visibility means
805 that the declaration is visible to other modules and, in shared libraries,
806 means that the declared entity may be overridden. On Darwin, default
807 visibility means that the declaration is visible to other modules. Default
808 visibility corresponds to "external linkage" in the language.</dd>
810 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
811 <dd>Two declarations of an object with hidden visibility refer to the same
812 object if they are in the same shared object. Usually, hidden visibility
813 indicates that the symbol will not be placed into the dynamic symbol
814 table, so no other module (executable or shared library) can reference it
817 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
818 <dd>On ELF, protected visibility indicates that the symbol will be placed in
819 the dynamic symbol table, but that references within the defining module
820 will bind to the local symbol. That is, the symbol cannot be overridden by
826 <!-- ======================================================================= -->
828 <a name="namedtypes">Named Types</a>
833 <p>LLVM IR allows you to specify name aliases for certain types. This can make
834 it easier to read the IR and make the IR more condensed (particularly when
835 recursive types are involved). An example of a name specification is:</p>
837 <pre class="doc_code">
838 %mytype = type { %mytype*, i32 }
841 <p>You may give a name to any <a href="#typesystem">type</a> except
842 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
843 is expected with the syntax "%mytype".</p>
845 <p>Note that type names are aliases for the structural type that they indicate,
846 and that you can therefore specify multiple names for the same type. This
847 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
848 uses structural typing, the name is not part of the type. When printing out
849 LLVM IR, the printer will pick <em>one name</em> to render all types of a
850 particular shape. This means that if you have code where two different
851 source types end up having the same LLVM type, that the dumper will sometimes
852 print the "wrong" or unexpected type. This is an important design point and
853 isn't going to change.</p>
857 <!-- ======================================================================= -->
859 <a name="globalvars">Global Variables</a>
864 <p>Global variables define regions of memory allocated at compilation time
865 instead of run-time. Global variables may optionally be initialized, may
866 have an explicit section to be placed in, and may have an optional explicit
867 alignment specified.</p>
869 <p>A variable may be defined as <tt>thread_local</tt>, which
870 means that it will not be shared by threads (each thread will have a
871 separated copy of the variable). Not all targets support thread-local
872 variables. Optionally, a TLS model may be specified:</p>
875 <dt><b><tt>localdynamic</tt></b>:</dt>
876 <dd>For variables that are only used within the current shared library.</dd>
878 <dt><b><tt>initialexec</tt></b>:</dt>
879 <dd>For variables in modules that will not be loaded dynamically.</dd>
881 <dt><b><tt>localexec</tt></b>:</dt>
882 <dd>For variables defined in the executable and only used within it.</dd>
885 <p>The models correspond to the ELF TLS models; see
886 <a href="http://people.redhat.com/drepper/tls.pdf">ELF
887 Handling For Thread-Local Storage</a> for more information on under which
888 circumstances the different models may be used. The target may choose a
889 different TLS model if the specified model is not supported, or if a better
890 choice of model can be made.</p>
892 <p>A variable may be defined as a global
893 "constant," which indicates that the contents of the variable
894 will <b>never</b> be modified (enabling better optimization, allowing the
895 global data to be placed in the read-only section of an executable, etc).
896 Note that variables that need runtime initialization cannot be marked
897 "constant" as there is a store to the variable.</p>
899 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
900 constant, even if the final definition of the global is not. This capability
901 can be used to enable slightly better optimization of the program, but
902 requires the language definition to guarantee that optimizations based on the
903 'constantness' are valid for the translation units that do not include the
906 <p>As SSA values, global variables define pointer values that are in scope
907 (i.e. they dominate) all basic blocks in the program. Global variables
908 always define a pointer to their "content" type because they describe a
909 region of memory, and all memory objects in LLVM are accessed through
912 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
913 that the address is not significant, only the content. Constants marked
914 like this can be merged with other constants if they have the same
915 initializer. Note that a constant with significant address <em>can</em>
916 be merged with a <tt>unnamed_addr</tt> constant, the result being a
917 constant whose address is significant.</p>
919 <p>A global variable may be declared to reside in a target-specific numbered
920 address space. For targets that support them, address spaces may affect how
921 optimizations are performed and/or what target instructions are used to
922 access the variable. The default address space is zero. The address space
923 qualifier must precede any other attributes.</p>
925 <p>LLVM allows an explicit section to be specified for globals. If the target
926 supports it, it will emit globals to the section specified.</p>
928 <p>An explicit alignment may be specified for a global, which must be a power
929 of 2. If not present, or if the alignment is set to zero, the alignment of
930 the global is set by the target to whatever it feels convenient. If an
931 explicit alignment is specified, the global is forced to have exactly that
932 alignment. Targets and optimizers are not allowed to over-align the global
933 if the global has an assigned section. In this case, the extra alignment
934 could be observable: for example, code could assume that the globals are
935 densely packed in their section and try to iterate over them as an array,
936 alignment padding would break this iteration.</p>
938 <p>For example, the following defines a global in a numbered address space with
939 an initializer, section, and alignment:</p>
941 <pre class="doc_code">
942 @G = addrspace(5) constant float 1.0, section "foo", align 4
945 <p>The following example defines a thread-local global with
946 the <tt>initialexec</tt> TLS model:</p>
948 <pre class="doc_code">
949 @G = thread_local(initialexec) global i32 0, align 4
955 <!-- ======================================================================= -->
957 <a name="functionstructure">Functions</a>
962 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
963 optional <a href="#linkage">linkage type</a>, an optional
964 <a href="#visibility">visibility style</a>, an optional
965 <a href="#callingconv">calling convention</a>,
966 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
967 <a href="#paramattrs">parameter attribute</a> for the return type, a function
968 name, a (possibly empty) argument list (each with optional
969 <a href="#paramattrs">parameter attributes</a>), optional
970 <a href="#fnattrs">function attributes</a>, an optional section, an optional
971 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
972 curly brace, a list of basic blocks, and a closing curly brace.</p>
974 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
975 optional <a href="#linkage">linkage type</a>, an optional
976 <a href="#visibility">visibility style</a>, an optional
977 <a href="#callingconv">calling convention</a>,
978 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
979 <a href="#paramattrs">parameter attribute</a> for the return type, a function
980 name, a possibly empty list of arguments, an optional alignment, and an
981 optional <a href="#gc">garbage collector name</a>.</p>
983 <p>A function definition contains a list of basic blocks, forming the CFG
984 (Control Flow Graph) for the function. Each basic block may optionally start
985 with a label (giving the basic block a symbol table entry), contains a list
986 of instructions, and ends with a <a href="#terminators">terminator</a>
987 instruction (such as a branch or function return).</p>
989 <p>The first basic block in a function is special in two ways: it is immediately
990 executed on entrance to the function, and it is not allowed to have
991 predecessor basic blocks (i.e. there can not be any branches to the entry
992 block of a function). Because the block can have no predecessors, it also
993 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
995 <p>LLVM allows an explicit section to be specified for functions. If the target
996 supports it, it will emit functions to the section specified.</p>
998 <p>An explicit alignment may be specified for a function. If not present, or if
999 the alignment is set to zero, the alignment of the function is set by the
1000 target to whatever it feels convenient. If an explicit alignment is
1001 specified, the function is forced to have at least that much alignment. All
1002 alignments must be a power of 2.</p>
1004 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
1005 be significant and two identical functions can be merged.</p>
1008 <pre class="doc_code">
1009 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
1010 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
1011 <ResultType> @<FunctionName> ([argument list])
1012 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
1013 [<a href="#gc">gc</a>] { ... }
1018 <!-- ======================================================================= -->
1020 <a name="aliasstructure">Aliases</a>
1025 <p>Aliases act as "second name" for the aliasee value (which can be either
1026 function, global variable, another alias or bitcast of global value). Aliases
1027 may have an optional <a href="#linkage">linkage type</a>, and an
1028 optional <a href="#visibility">visibility style</a>.</p>
1031 <pre class="doc_code">
1032 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
1037 <!-- ======================================================================= -->
1039 <a name="namedmetadatastructure">Named Metadata</a>
1044 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
1045 nodes</a> (but not metadata strings) are the only valid operands for
1046 a named metadata.</p>
1049 <pre class="doc_code">
1050 ; Some unnamed metadata nodes, which are referenced by the named metadata.
1051 !0 = metadata !{metadata !"zero"}
1052 !1 = metadata !{metadata !"one"}
1053 !2 = metadata !{metadata !"two"}
1055 !name = !{!0, !1, !2}
1060 <!-- ======================================================================= -->
1062 <a name="paramattrs">Parameter Attributes</a>
1067 <p>The return type and each parameter of a function type may have a set of
1068 <i>parameter attributes</i> associated with them. Parameter attributes are
1069 used to communicate additional information about the result or parameters of
1070 a function. Parameter attributes are considered to be part of the function,
1071 not of the function type, so functions with different parameter attributes
1072 can have the same function type.</p>
1074 <p>Parameter attributes are simple keywords that follow the type specified. If
1075 multiple parameter attributes are needed, they are space separated. For
1078 <pre class="doc_code">
1079 declare i32 @printf(i8* noalias nocapture, ...)
1080 declare i32 @atoi(i8 zeroext)
1081 declare signext i8 @returns_signed_char()
1084 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1085 <tt>readonly</tt>) come immediately after the argument list.</p>
1087 <p>Currently, only the following parameter attributes are defined:</p>
1090 <dt><tt><b>zeroext</b></tt></dt>
1091 <dd>This indicates to the code generator that the parameter or return value
1092 should be zero-extended to the extent required by the target's ABI (which
1093 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1094 parameter) or the callee (for a return value).</dd>
1096 <dt><tt><b>signext</b></tt></dt>
1097 <dd>This indicates to the code generator that the parameter or return value
1098 should be sign-extended to the extent required by the target's ABI (which
1099 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1102 <dt><tt><b>inreg</b></tt></dt>
1103 <dd>This indicates that this parameter or return value should be treated in a
1104 special target-dependent fashion during while emitting code for a function
1105 call or return (usually, by putting it in a register as opposed to memory,
1106 though some targets use it to distinguish between two different kinds of
1107 registers). Use of this attribute is target-specific.</dd>
1109 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1110 <dd><p>This indicates that the pointer parameter should really be passed by
1111 value to the function. The attribute implies that a hidden copy of the
1113 is made between the caller and the callee, so the callee is unable to
1114 modify the value in the caller. This attribute is only valid on LLVM
1115 pointer arguments. It is generally used to pass structs and arrays by
1116 value, but is also valid on pointers to scalars. The copy is considered
1117 to belong to the caller not the callee (for example,
1118 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1119 <tt>byval</tt> parameters). This is not a valid attribute for return
1122 <p>The byval attribute also supports specifying an alignment with
1123 the align attribute. It indicates the alignment of the stack slot to
1124 form and the known alignment of the pointer specified to the call site. If
1125 the alignment is not specified, then the code generator makes a
1126 target-specific assumption.</p></dd>
1128 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1129 <dd>This indicates that the pointer parameter specifies the address of a
1130 structure that is the return value of the function in the source program.
1131 This pointer must be guaranteed by the caller to be valid: loads and
1132 stores to the structure may be assumed by the callee to not to trap and
1133 to be properly aligned. This may only be applied to the first parameter.
1134 This is not a valid attribute for return values. </dd>
1136 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1137 <dd>This indicates that pointer values
1138 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1139 value do not alias pointer values which are not <i>based</i> on it,
1140 ignoring certain "irrelevant" dependencies.
1141 For a call to the parent function, dependencies between memory
1142 references from before or after the call and from those during the call
1143 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1144 return value used in that call.
1145 The caller shares the responsibility with the callee for ensuring that
1146 these requirements are met.
1147 For further details, please see the discussion of the NoAlias response in
1148 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1150 Note that this definition of <tt>noalias</tt> is intentionally
1151 similar to the definition of <tt>restrict</tt> in C99 for function
1152 arguments, though it is slightly weaker.
1154 For function return values, C99's <tt>restrict</tt> is not meaningful,
1155 while LLVM's <tt>noalias</tt> is.
1158 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1159 <dd>This indicates that the callee does not make any copies of the pointer
1160 that outlive the callee itself. This is not a valid attribute for return
1163 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1164 <dd>This indicates that the pointer parameter can be excised using the
1165 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1166 attribute for return values.</dd>
1171 <!-- ======================================================================= -->
1173 <a name="gc">Garbage Collector Names</a>
1178 <p>Each function may specify a garbage collector name, which is simply a
1181 <pre class="doc_code">
1182 define void @f() gc "name" { ... }
1185 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1186 collector which will cause the compiler to alter its output in order to
1187 support the named garbage collection algorithm.</p>
1191 <!-- ======================================================================= -->
1193 <a name="fnattrs">Function Attributes</a>
1198 <p>Function attributes are set to communicate additional information about a
1199 function. Function attributes are considered to be part of the function, not
1200 of the function type, so functions with different function attributes can
1201 have the same function type.</p>
1203 <p>Function attributes are simple keywords that follow the type specified. If
1204 multiple attributes are needed, they are space separated. For example:</p>
1206 <pre class="doc_code">
1207 define void @f() noinline { ... }
1208 define void @f() alwaysinline { ... }
1209 define void @f() alwaysinline optsize { ... }
1210 define void @f() optsize { ... }
1214 <dt><tt><b>address_safety</b></tt></dt>
1215 <dd>This attribute indicates that the address safety analysis
1216 is enabled for this function. </dd>
1218 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1219 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1220 the backend should forcibly align the stack pointer. Specify the
1221 desired alignment, which must be a power of two, in parentheses.
1223 <dt><tt><b>alwaysinline</b></tt></dt>
1224 <dd>This attribute indicates that the inliner should attempt to inline this
1225 function into callers whenever possible, ignoring any active inlining size
1226 threshold for this caller.</dd>
1228 <dt><tt><b>nonlazybind</b></tt></dt>
1229 <dd>This attribute suppresses lazy symbol binding for the function. This
1230 may make calls to the function faster, at the cost of extra program
1231 startup time if the function is not called during program startup.</dd>
1233 <dt><tt><b>inlinehint</b></tt></dt>
1234 <dd>This attribute indicates that the source code contained a hint that inlining
1235 this function is desirable (such as the "inline" keyword in C/C++). It
1236 is just a hint; it imposes no requirements on the inliner.</dd>
1238 <dt><tt><b>naked</b></tt></dt>
1239 <dd>This attribute disables prologue / epilogue emission for the function.
1240 This can have very system-specific consequences.</dd>
1242 <dt><tt><b>noimplicitfloat</b></tt></dt>
1243 <dd>This attributes disables implicit floating point instructions.</dd>
1245 <dt><tt><b>noinline</b></tt></dt>
1246 <dd>This attribute indicates that the inliner should never inline this
1247 function in any situation. This attribute may not be used together with
1248 the <tt>alwaysinline</tt> attribute.</dd>
1250 <dt><tt><b>noredzone</b></tt></dt>
1251 <dd>This attribute indicates that the code generator should not use a red
1252 zone, even if the target-specific ABI normally permits it.</dd>
1254 <dt><tt><b>noreturn</b></tt></dt>
1255 <dd>This function attribute indicates that the function never returns
1256 normally. This produces undefined behavior at runtime if the function
1257 ever does dynamically return.</dd>
1259 <dt><tt><b>nounwind</b></tt></dt>
1260 <dd>This function attribute indicates that the function never returns with an
1261 unwind or exceptional control flow. If the function does unwind, its
1262 runtime behavior is undefined.</dd>
1264 <dt><tt><b>optsize</b></tt></dt>
1265 <dd>This attribute suggests that optimization passes and code generator passes
1266 make choices that keep the code size of this function low, and otherwise
1267 do optimizations specifically to reduce code size.</dd>
1269 <dt><tt><b>readnone</b></tt></dt>
1270 <dd>This attribute indicates that the function computes its result (or decides
1271 to unwind an exception) based strictly on its arguments, without
1272 dereferencing any pointer arguments or otherwise accessing any mutable
1273 state (e.g. memory, control registers, etc) visible to caller functions.
1274 It does not write through any pointer arguments
1275 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1276 changes any state visible to callers. This means that it cannot unwind
1277 exceptions by calling the <tt>C++</tt> exception throwing methods.</dd>
1279 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1280 <dd>This attribute indicates that the function does not write through any
1281 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1282 arguments) or otherwise modify any state (e.g. memory, control registers,
1283 etc) visible to caller functions. It may dereference pointer arguments
1284 and read state that may be set in the caller. A readonly function always
1285 returns the same value (or unwinds an exception identically) when called
1286 with the same set of arguments and global state. It cannot unwind an
1287 exception by calling the <tt>C++</tt> exception throwing methods.</dd>
1289 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1290 <dd>This attribute indicates that this function can return twice. The
1291 C <code>setjmp</code> is an example of such a function. The compiler
1292 disables some optimizations (like tail calls) in the caller of these
1295 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1296 <dd>This attribute indicates that the function should emit a stack smashing
1297 protector. It is in the form of a "canary"—a random value placed on
1298 the stack before the local variables that's checked upon return from the
1299 function to see if it has been overwritten. A heuristic is used to
1300 determine if a function needs stack protectors or not.<br>
1302 If a function that has an <tt>ssp</tt> attribute is inlined into a
1303 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1304 function will have an <tt>ssp</tt> attribute.</dd>
1306 <dt><tt><b>sspreq</b></tt></dt>
1307 <dd>This attribute indicates that the function should <em>always</em> emit a
1308 stack smashing protector. This overrides
1309 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1311 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1312 function that doesn't have an <tt>sspreq</tt> attribute or which has
1313 an <tt>ssp</tt> attribute, then the resulting function will have
1314 an <tt>sspreq</tt> attribute.</dd>
1316 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1317 <dd>This attribute indicates that the ABI being targeted requires that
1318 an unwind table entry be produce for this function even if we can
1319 show that no exceptions passes by it. This is normally the case for
1320 the ELF x86-64 abi, but it can be disabled for some compilation
1326 <!-- ======================================================================= -->
1328 <a name="moduleasm">Module-Level Inline Assembly</a>
1333 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1334 the GCC "file scope inline asm" blocks. These blocks are internally
1335 concatenated by LLVM and treated as a single unit, but may be separated in
1336 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1338 <pre class="doc_code">
1339 module asm "inline asm code goes here"
1340 module asm "more can go here"
1343 <p>The strings can contain any character by escaping non-printable characters.
1344 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1347 <p>The inline asm code is simply printed to the machine code .s file when
1348 assembly code is generated.</p>
1352 <!-- ======================================================================= -->
1354 <a name="datalayout">Data Layout</a>
1359 <p>A module may specify a target specific data layout string that specifies how
1360 data is to be laid out in memory. The syntax for the data layout is
1363 <pre class="doc_code">
1364 target datalayout = "<i>layout specification</i>"
1367 <p>The <i>layout specification</i> consists of a list of specifications
1368 separated by the minus sign character ('-'). Each specification starts with
1369 a letter and may include other information after the letter to define some
1370 aspect of the data layout. The specifications accepted are as follows:</p>
1374 <dd>Specifies that the target lays out data in big-endian form. That is, the
1375 bits with the most significance have the lowest address location.</dd>
1378 <dd>Specifies that the target lays out data in little-endian form. That is,
1379 the bits with the least significance have the lowest address
1382 <dt><tt>S<i>size</i></tt></dt>
1383 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1384 of stack variables is limited to the natural stack alignment to avoid
1385 dynamic stack realignment. The stack alignment must be a multiple of
1386 8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1387 which does not prevent any alignment promotions.</dd>
1389 <dt><tt>p[n]:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1390 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1391 <i>preferred</i> alignments for address space <i>n</i>. All sizes are in
1392 bits. Specifying the <i>pref</i> alignment is optional. If omitted, the
1393 preceding <tt>:</tt> should be omitted too. The address space,
1394 <i>n</i> is optional, and if not specified, denotes the default address
1395 space 0. The value of <i>n</i> must be in the range [1,2^23).</dd>
1397 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1398 <dd>This specifies the alignment for an integer type of a given bit
1399 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1401 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1402 <dd>This specifies the alignment for a vector type of a given bit
1405 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1406 <dd>This specifies the alignment for a floating point type of a given bit
1407 <i>size</i>. Only values of <i>size</i> that are supported by the target
1408 will work. 32 (float) and 64 (double) are supported on all targets;
1409 80 or 128 (different flavors of long double) are also supported on some
1412 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1413 <dd>This specifies the alignment for an aggregate type of a given bit
1416 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1417 <dd>This specifies the alignment for a stack object of a given bit
1420 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1421 <dd>This specifies a set of native integer widths for the target CPU
1422 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1423 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1424 this set are considered to support most general arithmetic
1425 operations efficiently.</dd>
1428 <p>When constructing the data layout for a given target, LLVM starts with a
1429 default set of specifications which are then (possibly) overridden by the
1430 specifications in the <tt>datalayout</tt> keyword. The default specifications
1431 are given in this list:</p>
1434 <li><tt>E</tt> - big endian</li>
1435 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1436 <li><tt>p1:32:32:32</tt> - 32-bit pointers with 32-bit alignment for
1437 address space 1</li>
1438 <li><tt>p2:16:32:32</tt> - 16-bit pointers with 32-bit alignment for
1439 address space 2</li>
1440 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1441 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1442 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1443 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1444 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1445 alignment of 64-bits</li>
1446 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1447 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1448 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1449 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1450 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1451 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1454 <p>When LLVM is determining the alignment for a given type, it uses the
1455 following rules:</p>
1458 <li>If the type sought is an exact match for one of the specifications, that
1459 specification is used.</li>
1461 <li>If no match is found, and the type sought is an integer type, then the
1462 smallest integer type that is larger than the bitwidth of the sought type
1463 is used. If none of the specifications are larger than the bitwidth then
1464 the largest integer type is used. For example, given the default
1465 specifications above, the i7 type will use the alignment of i8 (next
1466 largest) while both i65 and i256 will use the alignment of i64 (largest
1469 <li>If no match is found, and the type sought is a vector type, then the
1470 largest vector type that is smaller than the sought vector type will be
1471 used as a fall back. This happens because <128 x double> can be
1472 implemented in terms of 64 <2 x double>, for example.</li>
1475 <p>The function of the data layout string may not be what you expect. Notably,
1476 this is not a specification from the frontend of what alignment the code
1477 generator should use.</p>
1479 <p>Instead, if specified, the target data layout is required to match what the
1480 ultimate <em>code generator</em> expects. This string is used by the
1481 mid-level optimizers to
1482 improve code, and this only works if it matches what the ultimate code
1483 generator uses. If you would like to generate IR that does not embed this
1484 target-specific detail into the IR, then you don't have to specify the
1485 string. This will disable some optimizations that require precise layout
1486 information, but this also prevents those optimizations from introducing
1487 target specificity into the IR.</p>
1493 <!-- ======================================================================= -->
1495 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1500 <p>Any memory access must be done through a pointer value associated
1501 with an address range of the memory access, otherwise the behavior
1502 is undefined. Pointer values are associated with address ranges
1503 according to the following rules:</p>
1506 <li>A pointer value is associated with the addresses associated with
1507 any value it is <i>based</i> on.
1508 <li>An address of a global variable is associated with the address
1509 range of the variable's storage.</li>
1510 <li>The result value of an allocation instruction is associated with
1511 the address range of the allocated storage.</li>
1512 <li>A null pointer in the default address-space is associated with
1514 <li>An integer constant other than zero or a pointer value returned
1515 from a function not defined within LLVM may be associated with address
1516 ranges allocated through mechanisms other than those provided by
1517 LLVM. Such ranges shall not overlap with any ranges of addresses
1518 allocated by mechanisms provided by LLVM.</li>
1521 <p>A pointer value is <i>based</i> on another pointer value according
1522 to the following rules:</p>
1525 <li>A pointer value formed from a
1526 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1527 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1528 <li>The result value of a
1529 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1530 of the <tt>bitcast</tt>.</li>
1531 <li>A pointer value formed by an
1532 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1533 pointer values that contribute (directly or indirectly) to the
1534 computation of the pointer's value.</li>
1535 <li>The "<i>based</i> on" relationship is transitive.</li>
1538 <p>Note that this definition of <i>"based"</i> is intentionally
1539 similar to the definition of <i>"based"</i> in C99, though it is
1540 slightly weaker.</p>
1542 <p>LLVM IR does not associate types with memory. The result type of a
1543 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1544 alignment of the memory from which to load, as well as the
1545 interpretation of the value. The first operand type of a
1546 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1547 and alignment of the store.</p>
1549 <p>Consequently, type-based alias analysis, aka TBAA, aka
1550 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1551 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1552 additional information which specialized optimization passes may use
1553 to implement type-based alias analysis.</p>
1557 <!-- ======================================================================= -->
1559 <a name="volatile">Volatile Memory Accesses</a>
1564 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1565 href="#i_store"><tt>store</tt></a>s, and <a
1566 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1567 The optimizers must not change the number of volatile operations or change their
1568 order of execution relative to other volatile operations. The optimizers
1569 <i>may</i> change the order of volatile operations relative to non-volatile
1570 operations. This is not Java's "volatile" and has no cross-thread
1571 synchronization behavior.</p>
1575 <!-- ======================================================================= -->
1577 <a name="memmodel">Memory Model for Concurrent Operations</a>
1582 <p>The LLVM IR does not define any way to start parallel threads of execution
1583 or to register signal handlers. Nonetheless, there are platform-specific
1584 ways to create them, and we define LLVM IR's behavior in their presence. This
1585 model is inspired by the C++0x memory model.</p>
1587 <p>For a more informal introduction to this model, see the
1588 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1590 <p>We define a <i>happens-before</i> partial order as the least partial order
1593 <li>Is a superset of single-thread program order, and</li>
1594 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1595 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1596 by platform-specific techniques, like pthread locks, thread
1597 creation, thread joining, etc., and by atomic instructions.
1598 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1602 <p>Note that program order does not introduce <i>happens-before</i> edges
1603 between a thread and signals executing inside that thread.</p>
1605 <p>Every (defined) read operation (load instructions, memcpy, atomic
1606 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1607 (defined) write operations (store instructions, atomic
1608 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1609 initialized globals are considered to have a write of the initializer which is
1610 atomic and happens before any other read or write of the memory in question.
1611 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1612 any write to the same byte, except:</p>
1615 <li>If <var>write<sub>1</sub></var> happens before
1616 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1617 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1618 does not see <var>write<sub>1</sub></var>.
1619 <li>If <var>R<sub>byte</sub></var> happens before
1620 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1621 see <var>write<sub>3</sub></var>.
1624 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1626 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1627 is supposed to give guarantees which can support
1628 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1629 addresses which do not behave like normal memory. It does not generally
1630 provide cross-thread synchronization.)
1631 <li>Otherwise, if there is no write to the same byte that happens before
1632 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1633 <tt>undef</tt> for that byte.
1634 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1635 <var>R<sub>byte</sub></var> returns the value written by that
1637 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1638 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1639 values written. See the <a href="#ordering">Atomic Memory Ordering
1640 Constraints</a> section for additional constraints on how the choice
1642 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1645 <p><var>R</var> returns the value composed of the series of bytes it read.
1646 This implies that some bytes within the value may be <tt>undef</tt>
1647 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1648 defines the semantics of the operation; it doesn't mean that targets will
1649 emit more than one instruction to read the series of bytes.</p>
1651 <p>Note that in cases where none of the atomic intrinsics are used, this model
1652 places only one restriction on IR transformations on top of what is required
1653 for single-threaded execution: introducing a store to a byte which might not
1654 otherwise be stored is not allowed in general. (Specifically, in the case
1655 where another thread might write to and read from an address, introducing a
1656 store can change a load that may see exactly one write into a load that may
1657 see multiple writes.)</p>
1659 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1660 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1661 none of the backends currently in the tree fall into this category; however,
1662 there might be targets which care. If there are, we want a paragraph
1665 Targets may specify that stores narrower than a certain width are not
1666 available; on such a target, for the purposes of this model, treat any
1667 non-atomic write with an alignment or width less than the minimum width
1668 as if it writes to the relevant surrounding bytes.
1673 <!-- ======================================================================= -->
1675 <a name="ordering">Atomic Memory Ordering Constraints</a>
1680 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1681 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1682 <a href="#i_fence"><code>fence</code></a>,
1683 <a href="#i_load"><code>atomic load</code></a>, and
1684 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1685 that determines which other atomic instructions on the same address they
1686 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1687 but are somewhat more colloquial. If these descriptions aren't precise enough,
1688 check those specs (see spec references in the
1689 <a href="Atomics.html#introduction">atomics guide</a>).
1690 <a href="#i_fence"><code>fence</code></a> instructions
1691 treat these orderings somewhat differently since they don't take an address.
1692 See that instruction's documentation for details.</p>
1694 <p>For a simpler introduction to the ordering constraints, see the
1695 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1698 <dt><code>unordered</code></dt>
1699 <dd>The set of values that can be read is governed by the happens-before
1700 partial order. A value cannot be read unless some operation wrote it.
1701 This is intended to provide a guarantee strong enough to model Java's
1702 non-volatile shared variables. This ordering cannot be specified for
1703 read-modify-write operations; it is not strong enough to make them atomic
1704 in any interesting way.</dd>
1705 <dt><code>monotonic</code></dt>
1706 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1707 total order for modifications by <code>monotonic</code> operations on each
1708 address. All modification orders must be compatible with the happens-before
1709 order. There is no guarantee that the modification orders can be combined to
1710 a global total order for the whole program (and this often will not be
1711 possible). The read in an atomic read-modify-write operation
1712 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1713 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1714 reads the value in the modification order immediately before the value it
1715 writes. If one atomic read happens before another atomic read of the same
1716 address, the later read must see the same value or a later value in the
1717 address's modification order. This disallows reordering of
1718 <code>monotonic</code> (or stronger) operations on the same address. If an
1719 address is written <code>monotonic</code>ally by one thread, and other threads
1720 <code>monotonic</code>ally read that address repeatedly, the other threads must
1721 eventually see the write. This corresponds to the C++0x/C1x
1722 <code>memory_order_relaxed</code>.</dd>
1723 <dt><code>acquire</code></dt>
1724 <dd>In addition to the guarantees of <code>monotonic</code>,
1725 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1726 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1727 <dt><code>release</code></dt>
1728 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1729 writes a value which is subsequently read by an <code>acquire</code> operation,
1730 it <i>synchronizes-with</i> that operation. (This isn't a complete
1731 description; see the C++0x definition of a release sequence.) This corresponds
1732 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1733 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1734 <code>acquire</code> and <code>release</code> operation on its address.
1735 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1736 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1737 <dd>In addition to the guarantees of <code>acq_rel</code>
1738 (<code>acquire</code> for an operation which only reads, <code>release</code>
1739 for an operation which only writes), there is a global total order on all
1740 sequentially-consistent operations on all addresses, which is consistent with
1741 the <i>happens-before</i> partial order and with the modification orders of
1742 all the affected addresses. Each sequentially-consistent read sees the last
1743 preceding write to the same address in this global order. This corresponds
1744 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1747 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1748 it only <i>synchronizes with</i> or participates in modification and seq_cst
1749 total orderings with other operations running in the same thread (for example,
1750 in signal handlers).</p>
1756 <!-- *********************************************************************** -->
1757 <h2><a name="typesystem">Type System</a></h2>
1758 <!-- *********************************************************************** -->
1762 <p>The LLVM type system is one of the most important features of the
1763 intermediate representation. Being typed enables a number of optimizations
1764 to be performed on the intermediate representation directly, without having
1765 to do extra analyses on the side before the transformation. A strong type
1766 system makes it easier to read the generated code and enables novel analyses
1767 and transformations that are not feasible to perform on normal three address
1768 code representations.</p>
1770 <!-- ======================================================================= -->
1772 <a name="t_classifications">Type Classifications</a>
1777 <p>The types fall into a few useful classifications:</p>
1779 <table border="1" cellspacing="0" cellpadding="4">
1781 <tr><th>Classification</th><th>Types</th></tr>
1783 <td><a href="#t_integer">integer</a></td>
1784 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1787 <td><a href="#t_floating">floating point</a></td>
1788 <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1791 <td><a name="t_firstclass">first class</a></td>
1792 <td><a href="#t_integer">integer</a>,
1793 <a href="#t_floating">floating point</a>,
1794 <a href="#t_pointer">pointer</a>,
1795 <a href="#t_vector">vector</a>,
1796 <a href="#t_struct">structure</a>,
1797 <a href="#t_array">array</a>,
1798 <a href="#t_label">label</a>,
1799 <a href="#t_metadata">metadata</a>.
1803 <td><a href="#t_primitive">primitive</a></td>
1804 <td><a href="#t_label">label</a>,
1805 <a href="#t_void">void</a>,
1806 <a href="#t_integer">integer</a>,
1807 <a href="#t_floating">floating point</a>,
1808 <a href="#t_x86mmx">x86mmx</a>,
1809 <a href="#t_metadata">metadata</a>.</td>
1812 <td><a href="#t_derived">derived</a></td>
1813 <td><a href="#t_array">array</a>,
1814 <a href="#t_function">function</a>,
1815 <a href="#t_pointer">pointer</a>,
1816 <a href="#t_struct">structure</a>,
1817 <a href="#t_vector">vector</a>,
1818 <a href="#t_opaque">opaque</a>.
1824 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1825 important. Values of these types are the only ones which can be produced by
1830 <!-- ======================================================================= -->
1832 <a name="t_primitive">Primitive Types</a>
1837 <p>The primitive types are the fundamental building blocks of the LLVM
1840 <!-- _______________________________________________________________________ -->
1842 <a name="t_integer">Integer Type</a>
1848 <p>The integer type is a very simple type that simply specifies an arbitrary
1849 bit width for the integer type desired. Any bit width from 1 bit to
1850 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1857 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1861 <table class="layout">
1863 <td class="left"><tt>i1</tt></td>
1864 <td class="left">a single-bit integer.</td>
1867 <td class="left"><tt>i32</tt></td>
1868 <td class="left">a 32-bit integer.</td>
1871 <td class="left"><tt>i1942652</tt></td>
1872 <td class="left">a really big integer of over 1 million bits.</td>
1878 <!-- _______________________________________________________________________ -->
1880 <a name="t_floating">Floating Point Types</a>
1887 <tr><th>Type</th><th>Description</th></tr>
1888 <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr>
1889 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1890 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1891 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1892 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1893 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1899 <!-- _______________________________________________________________________ -->
1901 <a name="t_x86mmx">X86mmx Type</a>
1907 <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>
1916 <!-- _______________________________________________________________________ -->
1918 <a name="t_void">Void Type</a>
1924 <p>The void type does not represent any value and has no size.</p>
1933 <!-- _______________________________________________________________________ -->
1935 <a name="t_label">Label Type</a>
1941 <p>The label type represents code labels.</p>
1950 <!-- _______________________________________________________________________ -->
1952 <a name="t_metadata">Metadata Type</a>
1958 <p>The metadata type represents embedded metadata. No derived types may be
1959 created from metadata except for <a href="#t_function">function</a>
1971 <!-- ======================================================================= -->
1973 <a name="t_derived">Derived Types</a>
1978 <p>The real power in LLVM comes from the derived types in the system. This is
1979 what allows a programmer to represent arrays, functions, pointers, and other
1980 useful types. Each of these types contain one or more element types which
1981 may be a primitive type, or another derived type. For example, it is
1982 possible to have a two dimensional array, using an array as the element type
1983 of another array.</p>
1985 <!-- _______________________________________________________________________ -->
1987 <a name="t_aggregate">Aggregate Types</a>
1992 <p>Aggregate Types are a subset of derived types that can contain multiple
1993 member types. <a href="#t_array">Arrays</a> and
1994 <a href="#t_struct">structs</a> are aggregate types.
1995 <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p>
1999 <!-- _______________________________________________________________________ -->
2001 <a name="t_array">Array Type</a>
2007 <p>The array type is a very simple derived type that arranges elements
2008 sequentially in memory. The array type requires a size (number of elements)
2009 and an underlying data type.</p>
2013 [<# elements> x <elementtype>]
2016 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
2017 be any type with a size.</p>
2020 <table class="layout">
2022 <td class="left"><tt>[40 x i32]</tt></td>
2023 <td class="left">Array of 40 32-bit integer values.</td>
2026 <td class="left"><tt>[41 x i32]</tt></td>
2027 <td class="left">Array of 41 32-bit integer values.</td>
2030 <td class="left"><tt>[4 x i8]</tt></td>
2031 <td class="left">Array of 4 8-bit integer values.</td>
2034 <p>Here are some examples of multidimensional arrays:</p>
2035 <table class="layout">
2037 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
2038 <td class="left">3x4 array of 32-bit integer values.</td>
2041 <td class="left"><tt>[12 x [10 x float]]</tt></td>
2042 <td class="left">12x10 array of single precision floating point values.</td>
2045 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
2046 <td class="left">2x3x4 array of 16-bit integer values.</td>
2050 <p>There is no restriction on indexing beyond the end of the array implied by
2051 a static type (though there are restrictions on indexing beyond the bounds
2052 of an allocated object in some cases). This means that single-dimension
2053 'variable sized array' addressing can be implemented in LLVM with a zero
2054 length array type. An implementation of 'pascal style arrays' in LLVM could
2055 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
2059 <!-- _______________________________________________________________________ -->
2061 <a name="t_function">Function Type</a>
2067 <p>The function type can be thought of as a function signature. It consists of
2068 a return type and a list of formal parameter types. The return type of a
2069 function type is a first class type or a void type.</p>
2073 <returntype> (<parameter list>)
2076 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
2077 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
2078 which indicates that the function takes a variable number of arguments.
2079 Variable argument functions can access their arguments with
2080 the <a href="#int_varargs">variable argument handling intrinsic</a>
2081 functions. '<tt><returntype></tt>' is any type except
2082 <a href="#t_label">label</a>.</p>
2085 <table class="layout">
2087 <td class="left"><tt>i32 (i32)</tt></td>
2088 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2090 </tr><tr class="layout">
2091 <td class="left"><tt>float (i16, i32 *) *
2093 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2094 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2095 returning <tt>float</tt>.
2097 </tr><tr class="layout">
2098 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2099 <td class="left">A vararg function that takes at least one
2100 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2101 which returns an integer. This is the signature for <tt>printf</tt> in
2104 </tr><tr class="layout">
2105 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2106 <td class="left">A function taking an <tt>i32</tt>, returning a
2107 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2114 <!-- _______________________________________________________________________ -->
2116 <a name="t_struct">Structure Type</a>
2122 <p>The structure type is used to represent a collection of data members together
2123 in memory. The elements of a structure may be any type that has a size.</p>
2125 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2126 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2127 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2128 Structures in registers are accessed using the
2129 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2130 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2132 <p>Structures may optionally be "packed" structures, which indicate that the
2133 alignment of the struct is one byte, and that there is no padding between
2134 the elements. In non-packed structs, padding between field types is inserted
2135 as defined by the DataLayout string in the module, which is required to match
2136 what the underlying code generator expects.</p>
2138 <p>Structures can either be "literal" or "identified". A literal structure is
2139 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2140 types are always defined at the top level with a name. Literal types are
2141 uniqued by their contents and can never be recursive or opaque since there is
2142 no way to write one. Identified types can be recursive, can be opaqued, and are
2148 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2149 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2153 <table class="layout">
2155 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2156 <td class="left">A triple of three <tt>i32</tt> values</td>
2159 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2160 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2161 second element is a <a href="#t_pointer">pointer</a> to a
2162 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2163 an <tt>i32</tt>.</td>
2166 <td class="left"><tt><{ i8, i32 }></tt></td>
2167 <td class="left">A packed struct known to be 5 bytes in size.</td>
2173 <!-- _______________________________________________________________________ -->
2175 <a name="t_opaque">Opaque Structure Types</a>
2181 <p>Opaque structure types are used to represent named structure types that do
2182 not have a body specified. This corresponds (for example) to the C notion of
2183 a forward declared structure.</p>
2192 <table class="layout">
2194 <td class="left"><tt>opaque</tt></td>
2195 <td class="left">An opaque type.</td>
2203 <!-- _______________________________________________________________________ -->
2205 <a name="t_pointer">Pointer Type</a>
2211 <p>The pointer type is used to specify memory locations.
2212 Pointers are commonly used to reference objects in memory.</p>
2214 <p>Pointer types may have an optional address space attribute defining the
2215 numbered address space where the pointed-to object resides. The default
2216 address space is number zero. The semantics of non-zero address
2217 spaces are target-specific.</p>
2219 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2220 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2228 <table class="layout">
2230 <td class="left"><tt>[4 x i32]*</tt></td>
2231 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2232 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2235 <td class="left"><tt>i32 (i32*) *</tt></td>
2236 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2237 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2241 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2242 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2243 that resides in address space #5.</td>
2249 <!-- _______________________________________________________________________ -->
2251 <a name="t_vector">Vector Type</a>
2257 <p>A vector type is a simple derived type that represents a vector of elements.
2258 Vector types are used when multiple primitive data are operated in parallel
2259 using a single instruction (SIMD). A vector type requires a size (number of
2260 elements) and an underlying primitive data type. Vector types are considered
2261 <a href="#t_firstclass">first class</a>.</p>
2265 < <# elements> x <elementtype> >
2268 <p>The number of elements is a constant integer value larger than 0; elementtype
2269 may be any integer or floating point type, or a pointer to these types.
2270 Vectors of size zero are not allowed. </p>
2273 <table class="layout">
2275 <td class="left"><tt><4 x i32></tt></td>
2276 <td class="left">Vector of 4 32-bit integer values.</td>
2279 <td class="left"><tt><8 x float></tt></td>
2280 <td class="left">Vector of 8 32-bit floating-point values.</td>
2283 <td class="left"><tt><2 x i64></tt></td>
2284 <td class="left">Vector of 2 64-bit integer values.</td>
2287 <td class="left"><tt><4 x i64*></tt></td>
2288 <td class="left">Vector of 4 pointers to 64-bit integer values.</td>
2298 <!-- *********************************************************************** -->
2299 <h2><a name="constants">Constants</a></h2>
2300 <!-- *********************************************************************** -->
2304 <p>LLVM has several different basic types of constants. This section describes
2305 them all and their syntax.</p>
2307 <!-- ======================================================================= -->
2309 <a name="simpleconstants">Simple Constants</a>
2315 <dt><b>Boolean constants</b></dt>
2316 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2317 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2319 <dt><b>Integer constants</b></dt>
2320 <dd>Standard integers (such as '4') are constants of
2321 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2322 with integer types.</dd>
2324 <dt><b>Floating point constants</b></dt>
2325 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2326 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2327 notation (see below). The assembler requires the exact decimal value of a
2328 floating-point constant. For example, the assembler accepts 1.25 but
2329 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2330 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2332 <dt><b>Null pointer constants</b></dt>
2333 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2334 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2337 <p>The one non-intuitive notation for constants is the hexadecimal form of
2338 floating point constants. For example, the form '<tt>double
2339 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2340 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2341 constants are required (and the only time that they are generated by the
2342 disassembler) is when a floating point constant must be emitted but it cannot
2343 be represented as a decimal floating point number in a reasonable number of
2344 digits. For example, NaN's, infinities, and other special values are
2345 represented in their IEEE hexadecimal format so that assembly and disassembly
2346 do not cause any bits to change in the constants.</p>
2348 <p>When using the hexadecimal form, constants of types half, float, and double are
2349 represented using the 16-digit form shown above (which matches the IEEE754
2350 representation for double); half and float values must, however, be exactly
2351 representable as IEE754 half and single precision, respectively.
2352 Hexadecimal format is always used
2353 for long double, and there are three forms of long double. The 80-bit format
2354 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2355 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2356 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2357 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2358 currently supported target uses this format. Long doubles will only work if
2359 they match the long double format on your target. The IEEE 16-bit format
2360 (half precision) is represented by <tt>0xH</tt> followed by 4 hexadecimal
2361 digits. All hexadecimal formats are big-endian (sign bit at the left).</p>
2363 <p>There are no constants of type x86mmx.</p>
2366 <!-- ======================================================================= -->
2368 <a name="aggregateconstants"></a> <!-- old anchor -->
2369 <a name="complexconstants">Complex Constants</a>
2374 <p>Complex constants are a (potentially recursive) combination of simple
2375 constants and smaller complex constants.</p>
2378 <dt><b>Structure constants</b></dt>
2379 <dd>Structure constants are represented with notation similar to structure
2380 type definitions (a comma separated list of elements, surrounded by braces
2381 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2382 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2383 Structure constants must have <a href="#t_struct">structure type</a>, and
2384 the number and types of elements must match those specified by the
2387 <dt><b>Array constants</b></dt>
2388 <dd>Array constants are represented with notation similar to array type
2389 definitions (a comma separated list of elements, surrounded by square
2390 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2391 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2392 the number and types of elements must match those specified by the
2395 <dt><b>Vector constants</b></dt>
2396 <dd>Vector constants are represented with notation similar to vector type
2397 definitions (a comma separated list of elements, surrounded by
2398 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2399 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2400 have <a href="#t_vector">vector type</a>, and the number and types of
2401 elements must match those specified by the type.</dd>
2403 <dt><b>Zero initialization</b></dt>
2404 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2405 value to zero of <em>any</em> type, including scalar and
2406 <a href="#t_aggregate">aggregate</a> types.
2407 This is often used to avoid having to print large zero initializers
2408 (e.g. for large arrays) and is always exactly equivalent to using explicit
2409 zero initializers.</dd>
2411 <dt><b>Metadata node</b></dt>
2412 <dd>A metadata node is a structure-like constant with
2413 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2414 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2415 be interpreted as part of the instruction stream, metadata is a place to
2416 attach additional information such as debug info.</dd>
2421 <!-- ======================================================================= -->
2423 <a name="globalconstants">Global Variable and Function Addresses</a>
2428 <p>The addresses of <a href="#globalvars">global variables</a>
2429 and <a href="#functionstructure">functions</a> are always implicitly valid
2430 (link-time) constants. These constants are explicitly referenced when
2431 the <a href="#identifiers">identifier for the global</a> is used and always
2432 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2433 legal LLVM file:</p>
2435 <pre class="doc_code">
2438 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2443 <!-- ======================================================================= -->
2445 <a name="undefvalues">Undefined Values</a>
2450 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2451 indicates that the user of the value may receive an unspecified bit-pattern.
2452 Undefined values may be of any type (other than '<tt>label</tt>'
2453 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2455 <p>Undefined values are useful because they indicate to the compiler that the
2456 program is well defined no matter what value is used. This gives the
2457 compiler more freedom to optimize. Here are some examples of (potentially
2458 surprising) transformations that are valid (in pseudo IR):</p>
2461 <pre class="doc_code">
2471 <p>This is safe because all of the output bits are affected by the undef bits.
2472 Any output bit can have a zero or one depending on the input bits.</p>
2474 <pre class="doc_code">
2485 <p>These logical operations have bits that are not always affected by the input.
2486 For example, if <tt>%X</tt> has a zero bit, then the output of the
2487 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2488 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2489 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2490 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2491 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2492 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2493 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2495 <pre class="doc_code">
2496 %A = select undef, %X, %Y
2497 %B = select undef, 42, %Y
2498 %C = select %X, %Y, undef
2509 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2510 branch) conditions can go <em>either way</em>, but they have to come from one
2511 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2512 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2513 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2514 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2515 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2518 <pre class="doc_code">
2519 %A = xor undef, undef
2537 <p>This example points out that two '<tt>undef</tt>' operands are not
2538 necessarily the same. This can be surprising to people (and also matches C
2539 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2540 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2541 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2542 its value over its "live range". This is true because the variable doesn't
2543 actually <em>have a live range</em>. Instead, the value is logically read
2544 from arbitrary registers that happen to be around when needed, so the value
2545 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2546 need to have the same semantics or the core LLVM "replace all uses with"
2547 concept would not hold.</p>
2549 <pre class="doc_code">
2557 <p>These examples show the crucial difference between an <em>undefined
2558 value</em> and <em>undefined behavior</em>. An undefined value (like
2559 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2560 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2561 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2562 defined on SNaN's. However, in the second example, we can make a more
2563 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2564 arbitrary value, we are allowed to assume that it could be zero. Since a
2565 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2566 the operation does not execute at all. This allows us to delete the divide and
2567 all code after it. Because the undefined operation "can't happen", the
2568 optimizer can assume that it occurs in dead code.</p>
2570 <pre class="doc_code">
2571 a: store undef -> %X
2572 b: store %X -> undef
2578 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2579 undefined value can be assumed to not have any effect; we can assume that the
2580 value is overwritten with bits that happen to match what was already there.
2581 However, a store <em>to</em> an undefined location could clobber arbitrary
2582 memory, therefore, it has undefined behavior.</p>
2586 <!-- ======================================================================= -->
2588 <a name="poisonvalues">Poison Values</a>
2593 <p>Poison values are similar to <a href="#undefvalues">undef values</a>, however
2594 they also represent the fact that an instruction or constant expression which
2595 cannot evoke side effects has nevertheless detected a condition which results
2596 in undefined behavior.</p>
2598 <p>There is currently no way of representing a poison value in the IR; they
2599 only exist when produced by operations such as
2600 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2602 <p>Poison value behavior is defined in terms of value <i>dependence</i>:</p>
2605 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2606 their operands.</li>
2608 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2609 to their dynamic predecessor basic block.</li>
2611 <li>Function arguments depend on the corresponding actual argument values in
2612 the dynamic callers of their functions.</li>
2614 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2615 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2616 control back to them.</li>
2618 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2619 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>,
2620 or exception-throwing call instructions that dynamically transfer control
2623 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2624 referenced memory addresses, following the order in the IR
2625 (including loads and stores implied by intrinsics such as
2626 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2628 <!-- TODO: In the case of multiple threads, this only applies if the store
2629 "happens-before" the load or store. -->
2631 <!-- TODO: floating-point exception state -->
2633 <li>An instruction with externally visible side effects depends on the most
2634 recent preceding instruction with externally visible side effects, following
2635 the order in the IR. (This includes
2636 <a href="#volatile">volatile operations</a>.)</li>
2638 <li>An instruction <i>control-depends</i> on a
2639 <a href="#terminators">terminator instruction</a>
2640 if the terminator instruction has multiple successors and the instruction
2641 is always executed when control transfers to one of the successors, and
2642 may not be executed when control is transferred to another.</li>
2644 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2645 instruction if the set of instructions it otherwise depends on would be
2646 different if the terminator had transferred control to a different
2649 <li>Dependence is transitive.</li>
2653 <p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>,
2654 with the additional affect that any instruction which has a <i>dependence</i>
2655 on a poison value has undefined behavior.</p>
2657 <p>Here are some examples:</p>
2659 <pre class="doc_code">
2661 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2662 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2663 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2664 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2666 store i32 %poison, i32* @g ; Poison value stored to memory.
2667 %poison2 = load i32* @g ; Poison value loaded back from memory.
2669 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2671 %narrowaddr = bitcast i32* @g to i16*
2672 %wideaddr = bitcast i32* @g to i64*
2673 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2674 %poison4 = load i64* %wideaddr ; Returns a poison value.
2676 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2677 br i1 %cmp, label %true, label %end ; Branch to either destination.
2680 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2681 ; it has undefined behavior.
2685 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2686 ; Both edges into this PHI are
2687 ; control-dependent on %cmp, so this
2688 ; always results in a poison value.
2690 store volatile i32 0, i32* @g ; This would depend on the store in %true
2691 ; if %cmp is true, or the store in %entry
2692 ; otherwise, so this is undefined behavior.
2694 br i1 %cmp, label %second_true, label %second_end
2695 ; The same branch again, but this time the
2696 ; true block doesn't have side effects.
2703 store volatile i32 0, i32* @g ; This time, the instruction always depends
2704 ; on the store in %end. Also, it is
2705 ; control-equivalent to %end, so this is
2706 ; well-defined (ignoring earlier undefined
2707 ; behavior in this example).
2712 <!-- ======================================================================= -->
2714 <a name="blockaddress">Addresses of Basic Blocks</a>
2719 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2721 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2722 basic block in the specified function, and always has an i8* type. Taking
2723 the address of the entry block is illegal.</p>
2725 <p>This value only has defined behavior when used as an operand to the
2726 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2727 comparisons against null. Pointer equality tests between labels addresses
2728 results in undefined behavior — though, again, comparison against null
2729 is ok, and no label is equal to the null pointer. This may be passed around
2730 as an opaque pointer sized value as long as the bits are not inspected. This
2731 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2732 long as the original value is reconstituted before the <tt>indirectbr</tt>
2735 <p>Finally, some targets may provide defined semantics when using the value as
2736 the operand to an inline assembly, but that is target specific.</p>
2741 <!-- ======================================================================= -->
2743 <a name="constantexprs">Constant Expressions</a>
2748 <p>Constant expressions are used to allow expressions involving other constants
2749 to be used as constants. Constant expressions may be of
2750 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2751 operation that does not have side effects (e.g. load and call are not
2752 supported). The following is the syntax for constant expressions:</p>
2755 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2756 <dd>Truncate a constant to another type. The bit size of CST must be larger
2757 than the bit size of TYPE. Both types must be integers.</dd>
2759 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2760 <dd>Zero extend a constant to another type. The bit size of CST must be
2761 smaller than the bit size of TYPE. Both types must be integers.</dd>
2763 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2764 <dd>Sign extend a constant to another type. The bit size of CST must be
2765 smaller than the bit size of TYPE. Both types must be integers.</dd>
2767 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2768 <dd>Truncate a floating point constant to another floating point type. The
2769 size of CST must be larger than the size of TYPE. Both types must be
2770 floating point.</dd>
2772 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2773 <dd>Floating point extend a constant to another type. The size of CST must be
2774 smaller or equal to the size of TYPE. Both types must be floating
2777 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2778 <dd>Convert a floating point constant to the corresponding unsigned integer
2779 constant. TYPE must be a scalar or vector integer type. CST must be of
2780 scalar or vector floating point type. Both CST and TYPE must be scalars,
2781 or vectors of the same number of elements. If the value won't fit in the
2782 integer type, the results are undefined.</dd>
2784 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2785 <dd>Convert a floating point constant to the corresponding signed integer
2786 constant. TYPE must be a scalar or vector integer type. CST must be of
2787 scalar or vector floating point type. Both CST and TYPE must be scalars,
2788 or vectors of the same number of elements. If the value won't fit in the
2789 integer type, the results are undefined.</dd>
2791 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2792 <dd>Convert an unsigned integer constant to the corresponding floating point
2793 constant. TYPE must be a scalar or vector floating point type. CST must be
2794 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2795 vectors of the same number of elements. If the value won't fit in the
2796 floating point type, the results are undefined.</dd>
2798 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2799 <dd>Convert a signed integer constant to the corresponding floating point
2800 constant. TYPE must be a scalar or vector floating point type. CST must be
2801 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2802 vectors of the same number of elements. If the value won't fit in the
2803 floating point type, the results are undefined.</dd>
2805 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2806 <dd>Convert a pointer typed constant to the corresponding integer constant
2807 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2808 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2809 make it fit in <tt>TYPE</tt>.</dd>
2811 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2812 <dd>Convert an integer constant to a pointer constant. TYPE must be a pointer
2813 type. CST must be of integer type. The CST value is zero extended,
2814 truncated, or unchanged to make it fit in a pointer size. This one is
2815 <i>really</i> dangerous!</dd>
2817 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2818 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2819 are the same as those for the <a href="#i_bitcast">bitcast
2820 instruction</a>.</dd>
2822 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2823 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2824 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2825 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2826 instruction, the index list may have zero or more indexes, which are
2827 required to make sense for the type of "CSTPTR".</dd>
2829 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2830 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2832 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2833 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2835 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2836 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2838 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2839 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2842 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2843 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2846 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2847 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2850 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2851 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2852 constants. The index list is interpreted in a similar manner as indices in
2853 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2854 index value must be specified.</dd>
2856 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2857 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2858 constants. The index list is interpreted in a similar manner as indices in
2859 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2860 index value must be specified.</dd>
2862 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2863 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2864 be any of the <a href="#binaryops">binary</a>
2865 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2866 on operands are the same as those for the corresponding instruction
2867 (e.g. no bitwise operations on floating point values are allowed).</dd>
2874 <!-- *********************************************************************** -->
2875 <h2><a name="othervalues">Other Values</a></h2>
2876 <!-- *********************************************************************** -->
2878 <!-- ======================================================================= -->
2880 <a name="inlineasm">Inline Assembler Expressions</a>
2885 <p>LLVM supports inline assembler expressions (as opposed
2886 to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of
2887 a special value. This value represents the inline assembler as a string
2888 (containing the instructions to emit), a list of operand constraints (stored
2889 as a string), a flag that indicates whether or not the inline asm
2890 expression has side effects, and a flag indicating whether the function
2891 containing the asm needs to align its stack conservatively. An example
2892 inline assembler expression is:</p>
2894 <pre class="doc_code">
2895 i32 (i32) asm "bswap $0", "=r,r"
2898 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2899 a <a href="#i_call"><tt>call</tt></a> or an
2900 <a href="#i_invoke"><tt>invoke</tt></a> instruction.
2901 Thus, typically we have:</p>
2903 <pre class="doc_code">
2904 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2907 <p>Inline asms with side effects not visible in the constraint list must be
2908 marked as having side effects. This is done through the use of the
2909 '<tt>sideeffect</tt>' keyword, like so:</p>
2911 <pre class="doc_code">
2912 call void asm sideeffect "eieio", ""()
2915 <p>In some cases inline asms will contain code that will not work unless the
2916 stack is aligned in some way, such as calls or SSE instructions on x86,
2917 yet will not contain code that does that alignment within the asm.
2918 The compiler should make conservative assumptions about what the asm might
2919 contain and should generate its usual stack alignment code in the prologue
2920 if the '<tt>alignstack</tt>' keyword is present:</p>
2922 <pre class="doc_code">
2923 call void asm alignstack "eieio", ""()
2926 <p>Inline asms also support using non-standard assembly dialects. The assumed
2927 dialect is ATT. When the '<tt>inteldialect</tt>' keyword is present, the
2928 inline asm is using the Intel dialect. Currently, ATT and Intel are the
2929 only supported dialects. An example is:</p>
2931 <pre class="doc_code">
2932 call void asm inteldialect "eieio", ""()
2935 <p>If multiple keywords appear the '<tt>sideeffect</tt>' keyword must come
2936 first, the '<tt>alignstack</tt>' keyword second and the
2937 '<tt>inteldialect</tt>' keyword last.</p>
2940 <p>TODO: The format of the asm and constraints string still need to be
2941 documented here. Constraints on what can be done (e.g. duplication, moving,
2942 etc need to be documented). This is probably best done by reference to
2943 another document that covers inline asm from a holistic perspective.</p>
2946 <!-- _______________________________________________________________________ -->
2948 <a name="inlineasm_md">Inline Asm Metadata</a>
2953 <p>The call instructions that wrap inline asm nodes may have a
2954 "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant
2955 integers. If present, the code generator will use the integer as the
2956 location cookie value when report errors through the <tt>LLVMContext</tt>
2957 error reporting mechanisms. This allows a front-end to correlate backend
2958 errors that occur with inline asm back to the source code that produced it.
2961 <pre class="doc_code">
2962 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2964 !42 = !{ i32 1234567 }
2967 <p>It is up to the front-end to make sense of the magic numbers it places in the
2968 IR. If the MDNode contains multiple constants, the code generator will use
2969 the one that corresponds to the line of the asm that the error occurs on.</p>
2975 <!-- ======================================================================= -->
2977 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2982 <p>LLVM IR allows metadata to be attached to instructions in the program that
2983 can convey extra information about the code to the optimizers and code
2984 generator. One example application of metadata is source-level debug
2985 information. There are two metadata primitives: strings and nodes. All
2986 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2987 preceding exclamation point ('<tt>!</tt>').</p>
2989 <p>A metadata string is a string surrounded by double quotes. It can contain
2990 any character by escaping non-printable characters with "<tt>\xx</tt>" where
2991 "<tt>xx</tt>" is the two digit hex code. For example:
2992 "<tt>!"test\00"</tt>".</p>
2994 <p>Metadata nodes are represented with notation similar to structure constants
2995 (a comma separated list of elements, surrounded by braces and preceded by an
2996 exclamation point). Metadata nodes can have any values as their operand. For
2999 <div class="doc_code">
3001 !{ metadata !"test\00", i32 10}
3005 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
3006 metadata nodes, which can be looked up in the module symbol table. For
3009 <div class="doc_code">
3011 !foo = metadata !{!4, !3}
3015 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
3016 function is using two metadata arguments:</p>
3018 <div class="doc_code">
3020 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3024 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
3025 attached to the <tt>add</tt> instruction using the <tt>!dbg</tt>
3028 <div class="doc_code">
3030 %indvar.next = add i64 %indvar, 1, !dbg !21
3034 <p>More information about specific metadata nodes recognized by the optimizers
3035 and code generator is found below.</p>
3037 <!-- _______________________________________________________________________ -->
3039 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
3044 <p>In LLVM IR, memory does not have types, so LLVM's own type system is not
3045 suitable for doing TBAA. Instead, metadata is added to the IR to describe
3046 a type system of a higher level language. This can be used to implement
3047 typical C/C++ TBAA, but it can also be used to implement custom alias
3048 analysis behavior for other languages.</p>
3050 <p>The current metadata format is very simple. TBAA metadata nodes have up to
3051 three fields, e.g.:</p>
3053 <div class="doc_code">
3055 !0 = metadata !{ metadata !"an example type tree" }
3056 !1 = metadata !{ metadata !"int", metadata !0 }
3057 !2 = metadata !{ metadata !"float", metadata !0 }
3058 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
3062 <p>The first field is an identity field. It can be any value, usually
3063 a metadata string, which uniquely identifies the type. The most important
3064 name in the tree is the name of the root node. Two trees with
3065 different root node names are entirely disjoint, even if they
3066 have leaves with common names.</p>
3068 <p>The second field identifies the type's parent node in the tree, or
3069 is null or omitted for a root node. A type is considered to alias
3070 all of its descendants and all of its ancestors in the tree. Also,
3071 a type is considered to alias all types in other trees, so that
3072 bitcode produced from multiple front-ends is handled conservatively.</p>
3074 <p>If the third field is present, it's an integer which if equal to 1
3075 indicates that the type is "constant" (meaning
3076 <tt>pointsToConstantMemory</tt> should return true; see
3077 <a href="AliasAnalysis.html#OtherItfs">other useful
3078 <tt>AliasAnalysis</tt> methods</a>).</p>
3082 <!-- _______________________________________________________________________ -->
3084 <a name="tbaa.struct">'<tt>tbaa.struct</tt>' Metadata</a>
3089 <p>The <a href="#int_memcpy"><tt>llvm.memcpy</tt></a> is often used to implement
3090 aggregate assignment operations in C and similar languages, however it is
3091 defined to copy a contiguous region of memory, which is more than strictly
3092 necessary for aggregate types which contain holes due to padding. Also, it
3093 doesn't contain any TBAA information about the fields of the aggregate.</p>
3095 <p><tt>!tbaa.struct</tt> metadata can describe which memory subregions in a memcpy
3096 are padding and what the TBAA tags of the struct are.</p>
3098 <p>The current metadata format is very simple. <tt>!tbaa.struct</tt> metadata nodes
3099 are a list of operands which are in conceptual groups of three. For each
3100 group of three, the first operand gives the byte offset of a field in bytes,
3101 the second gives its size in bytes, and the third gives its
3104 <div class="doc_code">
3106 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
3110 <p>This describes a struct with two fields. The first is at offset 0 bytes
3111 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
3112 and has size 4 bytes and has tbaa tag !2.</p>
3114 <p>Note that the fields need not be contiguous. In this example, there is a
3115 4 byte gap between the two fields. This gap represents padding which
3116 does not carry useful data and need not be preserved.</p>
3120 <!-- _______________________________________________________________________ -->
3122 <a name="fpmath">'<tt>fpmath</tt>' Metadata</a>
3127 <p><tt>fpmath</tt> metadata may be attached to any instruction of floating point
3128 type. It can be used to express the maximum acceptable error in the result of
3129 that instruction, in ULPs, thus potentially allowing the compiler to use a
3130 more efficient but less accurate method of computing it. ULP is defined as
3135 <p>If <tt>x</tt> is a real number that lies between two finite consecutive
3136 floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one
3137 of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the
3138 distance between the two non-equal finite floating-point numbers nearest
3139 <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p>
3143 <p>The metadata node shall consist of a single positive floating point number
3144 representing the maximum relative error, for example:</p>
3146 <div class="doc_code">
3148 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3154 <!-- _______________________________________________________________________ -->
3156 <a name="range">'<tt>range</tt>' Metadata</a>
3160 <p><tt>range</tt> metadata may be attached only to loads of integer types. It
3161 expresses the possible ranges the loaded value is in. The ranges are
3162 represented with a flattened list of integers. The loaded value is known to
3163 be in the union of the ranges defined by each consecutive pair. Each pair
3164 has the following properties:</p>
3166 <li>The type must match the type loaded by the instruction.</li>
3167 <li>The pair <tt>a,b</tt> represents the range <tt>[a,b)</tt>.</li>
3168 <li>Both <tt>a</tt> and <tt>b</tt> are constants.</li>
3169 <li>The range is allowed to wrap.</li>
3170 <li>The range should not represent the full or empty set. That is,
3171 <tt>a!=b</tt>. </li>
3173 <p> In addition, the pairs must be in signed order of the lower bound and
3174 they must be non-contiguous.</p>
3177 <div class="doc_code">
3179 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
3180 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3181 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
3182 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
3184 !0 = metadata !{ i8 0, i8 2 }
3185 !1 = metadata !{ i8 255, i8 2 }
3186 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
3187 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
3195 <!-- *********************************************************************** -->
3197 <a name="module_flags">Module Flags Metadata</a>
3199 <!-- *********************************************************************** -->
3203 <p>Information about the module as a whole is difficult to convey to LLVM's
3204 subsystems. The LLVM IR isn't sufficient to transmit this
3205 information. The <tt>llvm.module.flags</tt> named metadata exists in order to
3206 facilitate this. These flags are in the form of key / value pairs —
3207 much like a dictionary — making it easy for any subsystem who cares
3208 about a flag to look it up.</p>
3210 <p>The <tt>llvm.module.flags</tt> metadata contains a list of metadata
3211 triplets. Each triplet has the following form:</p>
3214 <li>The first element is a <i>behavior</i> flag, which specifies the behavior
3215 when two (or more) modules are merged together, and it encounters two (or
3216 more) metadata with the same ID. The supported behaviors are described
3219 <li>The second element is a metadata string that is a unique ID for the
3220 metadata. How each ID is interpreted is documented below.</li>
3222 <li>The third element is the value of the flag.</li>
3225 <p>When two (or more) modules are merged together, the resulting
3226 <tt>llvm.module.flags</tt> metadata is the union of the
3227 modules' <tt>llvm.module.flags</tt> metadata. The only exception being a flag
3228 with the <i>Override</i> behavior, which may override another flag's value
3231 <p>The following behaviors are supported:</p>
3233 <table border="1" cellspacing="0" cellpadding="4">
3243 <dt><b>Error</b></dt>
3244 <dd>Emits an error if two values disagree. It is an error to have an ID
3245 with both an Error and a Warning behavior.</dd>
3253 <dt><b>Warning</b></dt>
3254 <dd>Emits a warning if two values disagree.</dd>
3262 <dt><b>Require</b></dt>
3263 <dd>Emits an error when the specified value is not present or doesn't
3264 have the specified value. It is an error for two (or more)
3265 <tt>llvm.module.flags</tt> with the same ID to have the Require
3266 behavior but different values. There may be multiple Require flags
3275 <dt><b>Override</b></dt>
3276 <dd>Uses the specified value if the two values disagree. It is an
3277 error for two (or more) <tt>llvm.module.flags</tt> with the same
3278 ID to have the Override behavior but different values.</dd>
3285 <p>An example of module flags:</p>
3287 <pre class="doc_code">
3288 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3289 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3290 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3291 !3 = metadata !{ i32 3, metadata !"qux",
3293 metadata !"foo", i32 1
3296 !llvm.module.flags = !{ !0, !1, !2, !3 }
3300 <li><p>Metadata <tt>!0</tt> has the ID <tt>!"foo"</tt> and the value '1'. The
3301 behavior if two or more <tt>!"foo"</tt> flags are seen is to emit an
3302 error if their values are not equal.</p></li>
3304 <li><p>Metadata <tt>!1</tt> has the ID <tt>!"bar"</tt> and the value '37'. The
3305 behavior if two or more <tt>!"bar"</tt> flags are seen is to use the
3306 value '37' if their values are not equal.</p></li>
3308 <li><p>Metadata <tt>!2</tt> has the ID <tt>!"qux"</tt> and the value '42'. The
3309 behavior if two or more <tt>!"qux"</tt> flags are seen is to emit a
3310 warning if their values are not equal.</p></li>
3312 <li><p>Metadata <tt>!3</tt> has the ID <tt>!"qux"</tt> and the value:</p>
3314 <pre class="doc_code">
3315 metadata !{ metadata !"foo", i32 1 }
3318 <p>The behavior is to emit an error if the <tt>llvm.module.flags</tt> does
3319 not contain a flag with the ID <tt>!"foo"</tt> that has the value
3320 '1'. If two or more <tt>!"qux"</tt> flags exist, then they must have
3321 the same value or an error will be issued.</p></li>
3325 <!-- ======================================================================= -->
3327 <a name="objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a>
3332 <p>On the Mach-O platform, Objective-C stores metadata about garbage collection
3333 in a special section called "image info". The metadata consists of a version
3334 number and a bitmask specifying what types of garbage collection are
3335 supported (if any) by the file. If two or more modules are linked together
3336 their garbage collection metadata needs to be merged rather than appended
3339 <p>The Objective-C garbage collection module flags metadata consists of the
3340 following key-value pairs:</p>
3342 <table border="1" cellspacing="0" cellpadding="4">
3350 <td><tt>Objective-C Version</tt></td>
3351 <td align="left"><b>[Required]</b> — The Objective-C ABI
3352 version. Valid values are 1 and 2.</td>
3355 <td><tt>Objective-C Image Info Version</tt></td>
3356 <td align="left"><b>[Required]</b> — The version of the image info
3357 section. Currently always 0.</td>
3360 <td><tt>Objective-C Image Info Section</tt></td>
3361 <td align="left"><b>[Required]</b> — The section to place the
3362 metadata. Valid values are <tt>"__OBJC, __image_info, regular"</tt> for
3363 Objective-C ABI version 1, and <tt>"__DATA,__objc_imageinfo, regular,
3364 no_dead_strip"</tt> for Objective-C ABI version 2.</td>
3367 <td><tt>Objective-C Garbage Collection</tt></td>
3368 <td align="left"><b>[Required]</b> — Specifies whether garbage
3369 collection is supported or not. Valid values are 0, for no garbage
3370 collection, and 2, for garbage collection supported.</td>
3373 <td><tt>Objective-C GC Only</tt></td>
3374 <td align="left"><b>[Optional]</b> — Specifies that only garbage
3375 collection is supported. If present, its value must be 6. This flag
3376 requires that the <tt>Objective-C Garbage Collection</tt> flag have the
3382 <p>Some important flag interactions:</p>
3385 <li>If a module with <tt>Objective-C Garbage Collection</tt> set to 0 is
3386 merged with a module with <tt>Objective-C Garbage Collection</tt> set to
3387 2, then the resulting module has the <tt>Objective-C Garbage
3388 Collection</tt> flag set to 0.</li>
3390 <li>A module with <tt>Objective-C Garbage Collection</tt> set to 0 cannot be
3391 merged with a module with <tt>Objective-C GC Only</tt> set to 6.</li>
3398 <!-- *********************************************************************** -->
3400 <a name="intrinsic_globals">Intrinsic Global Variables</a>
3402 <!-- *********************************************************************** -->
3404 <p>LLVM has a number of "magic" global variables that contain data that affect
3405 code generation or other IR semantics. These are documented here. All globals
3406 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
3407 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
3410 <!-- ======================================================================= -->
3412 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
3417 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
3418 href="#linkage_appending">appending linkage</a>. This array contains a list of
3419 pointers to global variables and functions which may optionally have a pointer
3420 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
3422 <div class="doc_code">
3427 @llvm.used = appending global [2 x i8*] [
3429 i8* bitcast (i32* @Y to i8*)
3430 ], section "llvm.metadata"
3434 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
3435 compiler, assembler, and linker are required to treat the symbol as if there
3436 is a reference to the global that it cannot see. For example, if a variable
3437 has internal linkage and no references other than that from
3438 the <tt>@llvm.used</tt> list, it cannot be deleted. This is commonly used to
3439 represent references from inline asms and other things the compiler cannot
3440 "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p>
3442 <p>On some targets, the code generator must emit a directive to the assembler or
3443 object file to prevent the assembler and linker from molesting the
3448 <!-- ======================================================================= -->
3450 <a name="intg_compiler_used">
3451 The '<tt>llvm.compiler.used</tt>' Global Variable
3457 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
3458 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
3459 touching the symbol. On targets that support it, this allows an intelligent
3460 linker to optimize references to the symbol without being impeded as it would
3461 be by <tt>@llvm.used</tt>.</p>
3463 <p>This is a rare construct that should only be used in rare circumstances, and
3464 should not be exposed to source languages.</p>
3468 <!-- ======================================================================= -->
3470 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
3475 <div class="doc_code">
3477 %0 = type { i32, void ()* }
3478 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3482 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor
3483 functions and associated priorities. The functions referenced by this array
3484 will be called in ascending order of priority (i.e. lowest first) when the
3485 module is loaded. The order of functions with the same priority is not
3490 <!-- ======================================================================= -->
3492 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3497 <div class="doc_code">
3499 %0 = type { i32, void ()* }
3500 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3504 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions
3505 and associated priorities. The functions referenced by this array will be
3506 called in descending order of priority (i.e. highest first) when the module
3507 is loaded. The order of functions with the same priority is not defined.</p>
3513 <!-- *********************************************************************** -->
3514 <h2><a name="instref">Instruction Reference</a></h2>
3515 <!-- *********************************************************************** -->
3519 <p>The LLVM instruction set consists of several different classifications of
3520 instructions: <a href="#terminators">terminator
3521 instructions</a>, <a href="#binaryops">binary instructions</a>,
3522 <a href="#bitwiseops">bitwise binary instructions</a>,
3523 <a href="#memoryops">memory instructions</a>, and
3524 <a href="#otherops">other instructions</a>.</p>
3526 <!-- ======================================================================= -->
3528 <a name="terminators">Terminator Instructions</a>
3533 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3534 in a program ends with a "Terminator" instruction, which indicates which
3535 block should be executed after the current block is finished. These
3536 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3537 control flow, not values (the one exception being the
3538 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3540 <p>The terminator instructions are:
3541 '<a href="#i_ret"><tt>ret</tt></a>',
3542 '<a href="#i_br"><tt>br</tt></a>',
3543 '<a href="#i_switch"><tt>switch</tt></a>',
3544 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3545 '<a href="#i_invoke"><tt>invoke</tt></a>',
3546 '<a href="#i_resume"><tt>resume</tt></a>', and
3547 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3549 <!-- _______________________________________________________________________ -->
3551 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3558 ret <type> <value> <i>; Return a value from a non-void function</i>
3559 ret void <i>; Return from void function</i>
3563 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3564 a value) from a function back to the caller.</p>
3566 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3567 value and then causes control flow, and one that just causes control flow to
3571 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3572 return value. The type of the return value must be a
3573 '<a href="#t_firstclass">first class</a>' type.</p>
3575 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3576 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3577 value or a return value with a type that does not match its type, or if it
3578 has a void return type and contains a '<tt>ret</tt>' instruction with a
3582 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3583 the calling function's context. If the caller is a
3584 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3585 instruction after the call. If the caller was an
3586 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3587 the beginning of the "normal" destination block. If the instruction returns
3588 a value, that value shall set the call or invoke instruction's return
3593 ret i32 5 <i>; Return an integer value of 5</i>
3594 ret void <i>; Return from a void function</i>
3595 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3599 <!-- _______________________________________________________________________ -->
3601 <a name="i_br">'<tt>br</tt>' Instruction</a>
3608 br i1 <cond>, label <iftrue>, label <iffalse>
3609 br label <dest> <i>; Unconditional branch</i>
3613 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3614 different basic block in the current function. There are two forms of this
3615 instruction, corresponding to a conditional branch and an unconditional
3619 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3620 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3621 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3625 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3626 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3627 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3628 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3633 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3634 br i1 %cond, label %IfEqual, label %IfUnequal
3636 <a href="#i_ret">ret</a> i32 1
3638 <a href="#i_ret">ret</a> i32 0
3643 <!-- _______________________________________________________________________ -->
3645 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3652 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3656 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3657 several different places. It is a generalization of the '<tt>br</tt>'
3658 instruction, allowing a branch to occur to one of many possible
3662 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3663 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3664 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3665 The table is not allowed to contain duplicate constant entries.</p>
3668 <p>The <tt>switch</tt> instruction specifies a table of values and
3669 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3670 is searched for the given value. If the value is found, control flow is
3671 transferred to the corresponding destination; otherwise, control flow is
3672 transferred to the default destination.</p>
3674 <h5>Implementation:</h5>
3675 <p>Depending on properties of the target machine and the particular
3676 <tt>switch</tt> instruction, this instruction may be code generated in
3677 different ways. For example, it could be generated as a series of chained
3678 conditional branches or with a lookup table.</p>
3682 <i>; Emulate a conditional br instruction</i>
3683 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3684 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3686 <i>; Emulate an unconditional br instruction</i>
3687 switch i32 0, label %dest [ ]
3689 <i>; Implement a jump table:</i>
3690 switch i32 %val, label %otherwise [ i32 0, label %onzero
3692 i32 2, label %ontwo ]
3698 <!-- _______________________________________________________________________ -->
3700 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3707 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3712 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3713 within the current function, whose address is specified by
3714 "<tt>address</tt>". Address must be derived from a <a
3715 href="#blockaddress">blockaddress</a> constant.</p>
3719 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3720 rest of the arguments indicate the full set of possible destinations that the
3721 address may point to. Blocks are allowed to occur multiple times in the
3722 destination list, though this isn't particularly useful.</p>
3724 <p>This destination list is required so that dataflow analysis has an accurate
3725 understanding of the CFG.</p>
3729 <p>Control transfers to the block specified in the address argument. All
3730 possible destination blocks must be listed in the label list, otherwise this
3731 instruction has undefined behavior. This implies that jumps to labels
3732 defined in other functions have undefined behavior as well.</p>
3734 <h5>Implementation:</h5>
3736 <p>This is typically implemented with a jump through a register.</p>
3740 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3746 <!-- _______________________________________________________________________ -->
3748 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3755 <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>]
3756 to label <normal label> unwind label <exception label>
3760 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3761 function, with the possibility of control flow transfer to either the
3762 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3763 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3764 control flow will return to the "normal" label. If the callee (or any
3765 indirect callees) returns via the "<a href="#i_resume"><tt>resume</tt></a>"
3766 instruction or other exception handling mechanism, control is interrupted and
3767 continued at the dynamically nearest "exception" label.</p>
3769 <p>The '<tt>exception</tt>' label is a
3770 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3771 exception. As such, '<tt>exception</tt>' label is required to have the
3772 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3773 the information about the behavior of the program after unwinding
3774 happens, as its first non-PHI instruction. The restrictions on the
3775 "<tt>landingpad</tt>" instruction's tightly couples it to the
3776 "<tt>invoke</tt>" instruction, so that the important information contained
3777 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3781 <p>This instruction requires several arguments:</p>
3784 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3785 convention</a> the call should use. If none is specified, the call
3786 defaults to using C calling conventions.</li>
3788 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3789 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3790 '<tt>inreg</tt>' attributes are valid here.</li>
3792 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3793 function value being invoked. In most cases, this is a direct function
3794 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3795 off an arbitrary pointer to function value.</li>
3797 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3798 function to be invoked. </li>
3800 <li>'<tt>function args</tt>': argument list whose types match the function
3801 signature argument types and parameter attributes. All arguments must be
3802 of <a href="#t_firstclass">first class</a> type. If the function
3803 signature indicates the function accepts a variable number of arguments,
3804 the extra arguments can be specified.</li>
3806 <li>'<tt>normal label</tt>': the label reached when the called function
3807 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3809 <li>'<tt>exception label</tt>': the label reached when a callee returns via
3810 the <a href="#i_resume"><tt>resume</tt></a> instruction or other exception
3811 handling mechanism.</li>
3813 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3814 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3815 '<tt>readnone</tt>' attributes are valid here.</li>
3819 <p>This instruction is designed to operate as a standard
3820 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3821 primary difference is that it establishes an association with a label, which
3822 is used by the runtime library to unwind the stack.</p>
3824 <p>This instruction is used in languages with destructors to ensure that proper
3825 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3826 exception. Additionally, this is important for implementation of
3827 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3829 <p>For the purposes of the SSA form, the definition of the value returned by the
3830 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3831 block to the "normal" label. If the callee unwinds then no return value is
3836 %retval = invoke i32 @Test(i32 15) to label %Continue
3837 unwind label %TestCleanup <i>; {i32}:retval set</i>
3838 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3839 unwind label %TestCleanup <i>; {i32}:retval set</i>
3844 <!-- _______________________________________________________________________ -->
3847 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3854 resume <type> <value>
3858 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3862 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3863 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3867 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3868 (in-flight) exception whose unwinding was interrupted with
3869 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3873 resume { i8*, i32 } %exn
3878 <!-- _______________________________________________________________________ -->
3881 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3892 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3893 instruction is used to inform the optimizer that a particular portion of the
3894 code is not reachable. This can be used to indicate that the code after a
3895 no-return function cannot be reached, and other facts.</p>
3898 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3904 <!-- ======================================================================= -->
3906 <a name="binaryops">Binary Operations</a>
3911 <p>Binary operators are used to do most of the computation in a program. They
3912 require two operands of the same type, execute an operation on them, and
3913 produce a single value. The operands might represent multiple data, as is
3914 the case with the <a href="#t_vector">vector</a> data type. The result value
3915 has the same type as its operands.</p>
3917 <p>There are several different binary operators:</p>
3919 <!-- _______________________________________________________________________ -->
3921 <a name="i_add">'<tt>add</tt>' Instruction</a>
3928 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3929 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3930 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3931 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3935 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3938 <p>The two arguments to the '<tt>add</tt>' instruction must
3939 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3940 integer values. Both arguments must have identical types.</p>
3943 <p>The value produced is the integer sum of the two operands.</p>
3945 <p>If the sum has unsigned overflow, the result returned is the mathematical
3946 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3948 <p>Because LLVM integers use a two's complement representation, this instruction
3949 is appropriate for both signed and unsigned integers.</p>
3951 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3952 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3953 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3954 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3955 respectively, occurs.</p>
3959 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3964 <!-- _______________________________________________________________________ -->
3966 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3973 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3977 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3980 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3981 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3982 floating point values. Both arguments must have identical types.</p>
3985 <p>The value produced is the floating point sum of the two operands. This
3986 instruction can also take any number of fast-math flags, which are
3987 optimization hints to enable otherwise unsafe floating point
3991 <li><tt>nnan</tt>: No NaNs - Allow optimizations to assume the arguments and
3992 result are not NaN. Such optimizations are required to retain defined behavior
3993 over NaNs, but the value of the result is undefined.</li>
3995 <li><tt>ninf</tt>: No Inf - Allow optimizations to assume the arguments and
3996 result are not +/-Inf. Such optimizations are required to retain defined
3997 behavior over +/-Inf, but the value of the result is undefined.</li>
3999 <li><tt>nsz</tt>: No Signed Zeros: Allow optimizations to treat the
4000 sign of a zero argument or result as insignificant. </li>
4002 <li><tt>fast</tt>: Allow algebraically equivalent transformations that may
4003 dramatically change results in floating point (e.g. reassociate). This flag
4004 implies all the others.</li>
4010 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
4015 <!-- _______________________________________________________________________ -->
4017 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
4024 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4025 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4026 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4027 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4031 <p>The '<tt>sub</tt>' instruction returns the difference of its two
4034 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
4035 '<tt>neg</tt>' instruction present in most other intermediate
4036 representations.</p>
4039 <p>The two arguments to the '<tt>sub</tt>' instruction must
4040 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4041 integer values. Both arguments must have identical types.</p>
4044 <p>The value produced is the integer difference of the two operands.</p>
4046 <p>If the difference has unsigned overflow, the result returned is the
4047 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
4050 <p>Because LLVM integers use a two's complement representation, this instruction
4051 is appropriate for both signed and unsigned integers.</p>
4053 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
4054 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
4055 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
4056 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4057 respectively, occurs.</p>
4061 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
4062 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
4067 <!-- _______________________________________________________________________ -->
4069 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
4076 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4080 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
4083 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
4084 '<tt>fneg</tt>' instruction present in most other intermediate
4085 representations.</p>
4088 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
4089 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4090 floating point values. Both arguments must have identical types.</p>
4093 <p>The value produced is the floating point difference of the two operands. This
4094 instruction can also take any number of fast-math flags, which are
4095 optimization hints to enable otherwise unsafe floating point
4099 <li><tt>nnan</tt>: No NaNs - Allow optimizations to assume the arguments and
4100 result are not NaN. Such optimizations are required to retain defined behavior
4101 over NaNs, but the value of the result is undefined.</li>
4103 <li><tt>ninf</tt>: No Inf - Allow optimizations to assume the arguments and
4104 result are not +/-Inf. Such optimizations are required to retain defined
4105 behavior over +/-Inf, but the value of the result is undefined.</li>
4107 <li><tt>nsz</tt>: No Signed Zeros: Allow optimizations to treat the
4108 sign of a zero argument or result as insignificant. </li>
4110 <li><tt>fast</tt>: Allow algebraically equivalent transformations that may
4111 dramatically change results in floating point (e.g. reassociate). This flag
4112 implies all the others.</li>
4118 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
4119 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
4124 <!-- _______________________________________________________________________ -->
4126 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
4133 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4134 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4135 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4136 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4140 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
4143 <p>The two arguments to the '<tt>mul</tt>' instruction must
4144 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4145 integer values. Both arguments must have identical types.</p>
4148 <p>The value produced is the integer product of the two operands.</p>
4150 <p>If the result of the multiplication has unsigned overflow, the result
4151 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
4152 width of the result.</p>
4154 <p>Because LLVM integers use a two's complement representation, and the result
4155 is the same width as the operands, this instruction returns the correct
4156 result for both signed and unsigned integers. If a full product
4157 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
4158 be sign-extended or zero-extended as appropriate to the width of the full
4161 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
4162 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
4163 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
4164 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4165 respectively, occurs.</p>
4169 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
4174 <!-- _______________________________________________________________________ -->
4176 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
4183 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4187 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
4190 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
4191 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4192 floating point values. Both arguments must have identical types.</p>
4195 <p>The value produced is the floating point product of the two operands. This
4196 instruction can also take any number of fast-math flags, which are
4197 optimization hints to enable otherwise unsafe floating point
4201 <li><tt>nnan</tt>: No NaNs - Allow optimizations to assume the arguments and
4202 result are not NaN. Such optimizations are required to retain defined behavior
4203 over NaNs, but the value of the result is undefined.</li>
4205 <li><tt>ninf</tt>: No Inf - Allow optimizations to assume the arguments and
4206 result are not +/-Inf. Such optimizations are required to retain defined
4207 behavior over +/-Inf, but the value of the result is undefined.</li>
4209 <li><tt>nsz</tt>: No Signed Zeros: Allow optimizations to treat the
4210 sign of a zero argument or result as insignificant. </li>
4212 <li><tt>fast</tt>: Allow algebraically equivalent transformations that may
4213 dramatically change results in floating point (e.g. reassociate). This flag
4214 implies all the others.</li>
4220 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
4225 <!-- _______________________________________________________________________ -->
4227 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
4234 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4235 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4239 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
4242 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
4243 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4244 values. Both arguments must have identical types.</p>
4247 <p>The value produced is the unsigned integer quotient of the two operands.</p>
4249 <p>Note that unsigned integer division and signed integer division are distinct
4250 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
4252 <p>Division by zero leads to undefined behavior.</p>
4254 <p>If the <tt>exact</tt> keyword is present, the result value of the
4255 <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a
4256 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
4261 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4266 <!-- _______________________________________________________________________ -->
4268 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
4275 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4276 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4280 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
4283 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
4284 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4285 values. Both arguments must have identical types.</p>
4288 <p>The value produced is the signed integer quotient of the two operands rounded
4291 <p>Note that signed integer division and unsigned integer division are distinct
4292 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
4294 <p>Division by zero leads to undefined behavior. Overflow also leads to
4295 undefined behavior; this is a rare case, but can occur, for example, by doing
4296 a 32-bit division of -2147483648 by -1.</p>
4298 <p>If the <tt>exact</tt> keyword is present, the result value of the
4299 <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would
4304 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4309 <!-- _______________________________________________________________________ -->
4311 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
4318 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4322 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
4325 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
4326 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4327 floating point values. Both arguments must have identical types.</p>
4330 <p>The value produced is the floating point quotient of the two operands. This
4331 instruction can also take any number of fast-math flags, which are
4332 optimization hints to enable otherwise unsafe floating point
4336 <li><tt>nnan</tt>: No NaNs - Allow optimizations to assume the arguments and
4337 result are not NaN. Such optimizations are required to retain defined behavior
4338 over NaNs, but the value of the result is undefined.</li>
4340 <li><tt>ninf</tt>: No Inf - Allow optimizations to assume the arguments and
4341 result are not +/-Inf. Such optimizations are required to retain defined
4342 behavior over +/-Inf, but the value of the result is undefined.</li>
4344 <li><tt>nsz</tt>: No Signed Zeros: Allow optimizations to treat the
4345 sign of a zero argument or result as insignificant. </li>
4347 <li><tt>arcp</tt>: Allow Reciprocal: Allow optimizations to use the reciprocal
4348 of an argument rather than perform division. </li>
4350 <li><tt>fast</tt>: Allow algebraically equivalent transformations that may
4351 dramatically change results in floating point (e.g. reassociate). This flag
4352 implies all the others.</li>
4359 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
4364 <!-- _______________________________________________________________________ -->
4366 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
4373 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4377 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
4378 division of its two arguments.</p>
4381 <p>The two arguments to the '<tt>urem</tt>' instruction must be
4382 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4383 values. Both arguments must have identical types.</p>
4386 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
4387 This instruction always performs an unsigned division to get the
4390 <p>Note that unsigned integer remainder and signed integer remainder are
4391 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
4393 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
4397 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4402 <!-- _______________________________________________________________________ -->
4404 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
4411 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4415 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
4416 division of its two operands. This instruction can also take
4417 <a href="#t_vector">vector</a> versions of the values in which case the
4418 elements must be integers.</p>
4421 <p>The two arguments to the '<tt>srem</tt>' instruction must be
4422 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4423 values. Both arguments must have identical types.</p>
4426 <p>This instruction returns the <i>remainder</i> of a division (where the result
4427 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
4428 <i>modulo</i> operator (where the result is either zero or has the same sign
4429 as the divisor, <tt>op2</tt>) of a value.
4430 For more information about the difference,
4431 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
4432 Math Forum</a>. For a table of how this is implemented in various languages,
4433 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
4434 Wikipedia: modulo operation</a>.</p>
4436 <p>Note that signed integer remainder and unsigned integer remainder are
4437 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
4439 <p>Taking the remainder of a division by zero leads to undefined behavior.
4440 Overflow also leads to undefined behavior; this is a rare case, but can
4441 occur, for example, by taking the remainder of a 32-bit division of
4442 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
4443 lets srem be implemented using instructions that return both the result of
4444 the division and the remainder.)</p>
4448 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4453 <!-- _______________________________________________________________________ -->
4455 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
4462 <result> = frem [fast-math flags]* <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4466 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
4467 its two operands.</p>
4470 <p>The two arguments to the '<tt>frem</tt>' instruction must be
4471 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4472 floating point values. Both arguments must have identical types.</p>
4475 <p>This instruction returns the <i>remainder</i> of a division. The remainder
4476 has the same sign as the dividend. This instruction can also take any number
4477 of fast-math flags, which are optimization hints to enable otherwise unsafe
4478 floating point optimizations:</p>
4482 <li><tt>nnan</tt>: No NaNs - Allow optimizations to assume the arguments and
4483 result are not NaN. Such optimizations are required to retain defined behavior
4484 over NaNs, but the value of the result is undefined.</li>
4486 <li><tt>ninf</tt>: No Inf - Allow optimizations to assume the arguments and
4487 result are not +/-Inf. Such optimizations are required to retain defined
4488 behavior over +/-Inf, but the value of the result is undefined.</li>
4490 <li><tt>nsz</tt>: No Signed Zeros: Allow optimizations to treat the
4491 sign of a zero argument or result as insignificant. </li>
4493 <li><tt>arcp</tt>: Allow Reciprocal: Allow optimizations to use the reciprocal
4494 of an argument rather than perform division. </li>
4496 <li><tt>fast</tt>: Allow algebraically equivalent transformations that may
4497 dramatically change results in floating point (e.g. reassociate). This flag
4498 implies all the others.</li>
4504 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
4511 <!-- ======================================================================= -->
4513 <a name="bitwiseops">Bitwise Binary Operations</a>
4518 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
4519 program. They are generally very efficient instructions and can commonly be
4520 strength reduced from other instructions. They require two operands of the
4521 same type, execute an operation on them, and produce a single value. The
4522 resulting value is the same type as its operands.</p>
4524 <!-- _______________________________________________________________________ -->
4526 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
4533 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4534 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4535 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4536 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4540 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
4541 a specified number of bits.</p>
4544 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
4545 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4546 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4549 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
4550 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
4551 is (statically or dynamically) negative or equal to or larger than the number
4552 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4553 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4554 shift amount in <tt>op2</tt>.</p>
4556 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
4557 <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits. If
4558 the <tt>nsw</tt> keyword is present, then the shift produces a
4559 <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree
4560 with the resultant sign bit. As such, NUW/NSW have the same semantics as
4561 they would if the shift were expressed as a mul instruction with the same
4562 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
4566 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
4567 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
4568 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
4569 <result> = shl i32 1, 32 <i>; undefined</i>
4570 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4575 <!-- _______________________________________________________________________ -->
4577 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4584 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4585 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4589 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4590 operand shifted to the right a specified number of bits with zero fill.</p>
4593 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4594 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4595 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4598 <p>This instruction always performs a logical shift right operation. The most
4599 significant bits of the result will be filled with zero bits after the shift.
4600 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4601 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4602 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4603 shift amount in <tt>op2</tt>.</p>
4605 <p>If the <tt>exact</tt> keyword is present, the result value of the
4606 <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4607 shifted out are non-zero.</p>
4612 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4613 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4614 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4615 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4616 <result> = lshr i32 1, 32 <i>; undefined</i>
4617 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4622 <!-- _______________________________________________________________________ -->
4624 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4631 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4632 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4636 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4637 operand shifted to the right a specified number of bits with sign
4641 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4642 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4643 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4646 <p>This instruction always performs an arithmetic shift right operation, The
4647 most significant bits of the result will be filled with the sign bit
4648 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4649 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4650 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4651 the corresponding shift amount in <tt>op2</tt>.</p>
4653 <p>If the <tt>exact</tt> keyword is present, the result value of the
4654 <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4655 shifted out are non-zero.</p>
4659 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4660 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4661 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4662 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4663 <result> = ashr i32 1, 32 <i>; undefined</i>
4664 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4669 <!-- _______________________________________________________________________ -->
4671 <a name="i_and">'<tt>and</tt>' Instruction</a>
4678 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4682 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4686 <p>The two arguments to the '<tt>and</tt>' instruction must be
4687 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4688 values. Both arguments must have identical types.</p>
4691 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4693 <table border="1" cellspacing="0" cellpadding="4">
4725 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4726 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4727 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4730 <!-- _______________________________________________________________________ -->
4732 <a name="i_or">'<tt>or</tt>' Instruction</a>
4739 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4743 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4747 <p>The two arguments to the '<tt>or</tt>' instruction must be
4748 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4749 values. Both arguments must have identical types.</p>
4752 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4754 <table border="1" cellspacing="0" cellpadding="4">
4786 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4787 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4788 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4793 <!-- _______________________________________________________________________ -->
4795 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4802 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4806 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4807 its two operands. The <tt>xor</tt> is used to implement the "one's
4808 complement" operation, which is the "~" operator in C.</p>
4811 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4812 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4813 values. Both arguments must have identical types.</p>
4816 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4818 <table border="1" cellspacing="0" cellpadding="4">
4850 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4851 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4852 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4853 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4860 <!-- ======================================================================= -->
4862 <a name="vectorops">Vector Operations</a>
4867 <p>LLVM supports several instructions to represent vector operations in a
4868 target-independent manner. These instructions cover the element-access and
4869 vector-specific operations needed to process vectors effectively. While LLVM
4870 does directly support these vector operations, many sophisticated algorithms
4871 will want to use target-specific intrinsics to take full advantage of a
4872 specific target.</p>
4874 <!-- _______________________________________________________________________ -->
4876 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4883 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4887 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4888 from a vector at a specified index.</p>
4892 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4893 of <a href="#t_vector">vector</a> type. The second operand is an index
4894 indicating the position from which to extract the element. The index may be
4898 <p>The result is a scalar of the same type as the element type of
4899 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4900 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4901 results are undefined.</p>
4905 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4910 <!-- _______________________________________________________________________ -->
4912 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4919 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4923 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4924 vector at a specified index.</p>
4927 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4928 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4929 whose type must equal the element type of the first operand. The third
4930 operand is an index indicating the position at which to insert the value.
4931 The index may be a variable.</p>
4934 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4935 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4936 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4937 results are undefined.</p>
4941 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4946 <!-- _______________________________________________________________________ -->
4948 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4955 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4959 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4960 from two input vectors, returning a vector with the same element type as the
4961 input and length that is the same as the shuffle mask.</p>
4964 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4965 with the same type. The third argument is a shuffle mask whose
4966 element type is always 'i32'. The result of the instruction is a vector
4967 whose length is the same as the shuffle mask and whose element type is the
4968 same as the element type of the first two operands.</p>
4970 <p>The shuffle mask operand is required to be a constant vector with either
4971 constant integer or undef values.</p>
4974 <p>The elements of the two input vectors are numbered from left to right across
4975 both of the vectors. The shuffle mask operand specifies, for each element of
4976 the result vector, which element of the two input vectors the result element
4977 gets. The element selector may be undef (meaning "don't care") and the
4978 second operand may be undef if performing a shuffle from only one vector.</p>
4982 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4983 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4984 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4985 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4986 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4987 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4988 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4989 <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>
4996 <!-- ======================================================================= -->
4998 <a name="aggregateops">Aggregate Operations</a>
5003 <p>LLVM supports several instructions for working with
5004 <a href="#t_aggregate">aggregate</a> values.</p>
5006 <!-- _______________________________________________________________________ -->
5008 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
5015 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
5019 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
5020 from an <a href="#t_aggregate">aggregate</a> value.</p>
5023 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
5024 of <a href="#t_struct">struct</a> or
5025 <a href="#t_array">array</a> type. The operands are constant indices to
5026 specify which value to extract in a similar manner as indices in a
5027 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
5028 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
5030 <li>Since the value being indexed is not a pointer, the first index is
5031 omitted and assumed to be zero.</li>
5032 <li>At least one index must be specified.</li>
5033 <li>Not only struct indices but also array indices must be in
5038 <p>The result is the value at the position in the aggregate specified by the
5043 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
5048 <!-- _______________________________________________________________________ -->
5050 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
5057 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
5061 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
5062 in an <a href="#t_aggregate">aggregate</a> value.</p>
5065 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
5066 of <a href="#t_struct">struct</a> or
5067 <a href="#t_array">array</a> type. The second operand is a first-class
5068 value to insert. The following operands are constant indices indicating
5069 the position at which to insert the value in a similar manner as indices in a
5070 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
5071 value to insert must have the same type as the value identified by the
5075 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
5076 that of <tt>val</tt> except that the value at the position specified by the
5077 indices is that of <tt>elt</tt>.</p>
5081 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
5082 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
5083 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
5090 <!-- ======================================================================= -->
5092 <a name="memoryops">Memory Access and Addressing Operations</a>
5097 <p>A key design point of an SSA-based representation is how it represents
5098 memory. In LLVM, no memory locations are in SSA form, which makes things
5099 very simple. This section describes how to read, write, and allocate
5102 <!-- _______________________________________________________________________ -->
5104 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
5111 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
5115 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
5116 currently executing function, to be automatically released when this function
5117 returns to its caller. The object is always allocated in the generic address
5118 space (address space zero).</p>
5121 <p>The '<tt>alloca</tt>' instruction
5122 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
5123 runtime stack, returning a pointer of the appropriate type to the program.
5124 If "NumElements" is specified, it is the number of elements allocated,
5125 otherwise "NumElements" is defaulted to be one. If a constant alignment is
5126 specified, the value result of the allocation is guaranteed to be aligned to
5127 at least that boundary. If not specified, or if zero, the target can choose
5128 to align the allocation on any convenient boundary compatible with the
5131 <p>'<tt>type</tt>' may be any sized type.</p>
5134 <p>Memory is allocated; a pointer is returned. The operation is undefined if
5135 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
5136 memory is automatically released when the function returns. The
5137 '<tt>alloca</tt>' instruction is commonly used to represent automatic
5138 variables that must have an address available. When the function returns
5139 (either with the <tt><a href="#i_ret">ret</a></tt>
5140 or <tt><a href="#i_resume">resume</a></tt> instructions), the memory is
5141 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5142 The order in which memory is allocated (ie., which way the stack grows) is
5149 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
5150 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
5151 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
5152 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
5157 <!-- _______________________________________________________________________ -->
5159 <a name="i_load">'<tt>load</tt>' Instruction</a>
5166 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
5167 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5168 !<index> = !{ i32 1 }
5172 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
5175 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
5176 from which to load. The pointer must point to
5177 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
5178 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
5179 number or order of execution of this <tt>load</tt> with other <a
5180 href="#volatile">volatile operations</a>.</p>
5182 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
5183 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5184 argument. The <code>release</code> and <code>acq_rel</code> orderings are
5185 not valid on <code>load</code> instructions. Atomic loads produce <a
5186 href="#memorymodel">defined</a> results when they may see multiple atomic
5187 stores. The type of the pointee must be an integer type whose bit width
5188 is a power of two greater than or equal to eight and less than or equal
5189 to a target-specific size limit. <code>align</code> must be explicitly
5190 specified on atomic loads, and the load has undefined behavior if the
5191 alignment is not set to a value which is at least the size in bytes of
5192 the pointee. <code>!nontemporal</code> does not have any defined semantics
5193 for atomic loads.</p>
5195 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
5196 operation (that is, the alignment of the memory address). A value of 0 or an
5197 omitted <tt>align</tt> argument means that the operation has the abi
5198 alignment for the target. It is the responsibility of the code emitter to
5199 ensure that the alignment information is correct. Overestimating the
5200 alignment results in undefined behavior. Underestimating the alignment may
5201 produce less efficient code. An alignment of 1 is always safe.</p>
5203 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
5204 metatadata name <index> corresponding to a metadata node with
5205 one <tt>i32</tt> entry of value 1. The existence of
5206 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
5207 and code generator that this load is not expected to be reused in the cache.
5208 The code generator may select special instructions to save cache bandwidth,
5209 such as the <tt>MOVNT</tt> instruction on x86.</p>
5211 <p>The optional <tt>!invariant.load</tt> metadata must reference a single
5212 metatadata name <index> corresponding to a metadata node with no
5213 entries. The existence of the <tt>!invariant.load</tt> metatadata on the
5214 instruction tells the optimizer and code generator that this load address
5215 points to memory which does not change value during program execution.
5216 The optimizer may then move this load around, for example, by hoisting it
5217 out of loops using loop invariant code motion.</p>
5220 <p>The location of memory pointed to is loaded. If the value being loaded is of
5221 scalar type then the number of bytes read does not exceed the minimum number
5222 of bytes needed to hold all bits of the type. For example, loading an
5223 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
5224 <tt>i20</tt> with a size that is not an integral number of bytes, the result
5225 is undefined if the value was not originally written using a store of the
5230 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5231 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
5232 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
5237 <!-- _______________________________________________________________________ -->
5239 <a name="i_store">'<tt>store</tt>' Instruction</a>
5246 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
5247 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
5251 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
5254 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
5255 and an address at which to store it. The type of the
5256 '<tt><pointer></tt>' operand must be a pointer to
5257 the <a href="#t_firstclass">first class</a> type of the
5258 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
5259 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
5260 order of execution of this <tt>store</tt> with other <a
5261 href="#volatile">volatile operations</a>.</p>
5263 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
5264 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5265 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
5266 valid on <code>store</code> instructions. Atomic loads produce <a
5267 href="#memorymodel">defined</a> results when they may see multiple atomic
5268 stores. The type of the pointee must be an integer type whose bit width
5269 is a power of two greater than or equal to eight and less than or equal
5270 to a target-specific size limit. <code>align</code> must be explicitly
5271 specified on atomic stores, and the store has undefined behavior if the
5272 alignment is not set to a value which is at least the size in bytes of
5273 the pointee. <code>!nontemporal</code> does not have any defined semantics
5274 for atomic stores.</p>
5276 <p>The optional constant "align" argument specifies the alignment of the
5277 operation (that is, the alignment of the memory address). A value of 0 or an
5278 omitted "align" argument means that the operation has the abi
5279 alignment for the target. It is the responsibility of the code emitter to
5280 ensure that the alignment information is correct. Overestimating the
5281 alignment results in an undefined behavior. Underestimating the alignment may
5282 produce less efficient code. An alignment of 1 is always safe.</p>
5284 <p>The optional !nontemporal metadata must reference a single metatadata
5285 name <index> corresponding to a metadata node with one i32 entry of
5286 value 1. The existence of the !nontemporal metatadata on the
5287 instruction tells the optimizer and code generator that this load is
5288 not expected to be reused in the cache. The code generator may
5289 select special instructions to save cache bandwidth, such as the
5290 MOVNT instruction on x86.</p>
5294 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
5295 location specified by the '<tt><pointer></tt>' operand. If
5296 '<tt><value></tt>' is of scalar type then the number of bytes written
5297 does not exceed the minimum number of bytes needed to hold all bits of the
5298 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
5299 writing a value of a type like <tt>i20</tt> with a size that is not an
5300 integral number of bytes, it is unspecified what happens to the extra bits
5301 that do not belong to the type, but they will typically be overwritten.</p>
5305 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5306 store i32 3, i32* %ptr <i>; yields {void}</i>
5307 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
5312 <!-- _______________________________________________________________________ -->
5314 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
5321 fence [singlethread] <ordering> <i>; yields {void}</i>
5325 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
5326 between operations.</p>
5328 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
5329 href="#ordering">ordering</a> argument which defines what
5330 <i>synchronizes-with</i> edges they add. They can only be given
5331 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
5332 <code>seq_cst</code> orderings.</p>
5335 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
5336 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
5337 <code>acquire</code> ordering semantics if and only if there exist atomic
5338 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
5339 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
5340 <var>X</var> modifies <var>M</var> (either directly or through some side effect
5341 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
5342 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
5343 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
5344 than an explicit <code>fence</code>, one (but not both) of the atomic operations
5345 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
5346 <code>acquire</code> (resp.) ordering constraint and still
5347 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
5348 <i>happens-before</i> edge.</p>
5350 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
5351 having both <code>acquire</code> and <code>release</code> semantics specified
5352 above, participates in the global program order of other <code>seq_cst</code>
5353 operations and/or fences.</p>
5355 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
5356 specifies that the fence only synchronizes with other fences in the same
5357 thread. (This is useful for interacting with signal handlers.)</p>
5361 fence acquire <i>; yields {void}</i>
5362 fence singlethread seq_cst <i>; yields {void}</i>
5367 <!-- _______________________________________________________________________ -->
5369 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
5376 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
5380 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
5381 It loads a value in memory and compares it to a given value. If they are
5382 equal, it stores a new value into the memory.</p>
5385 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
5386 address to operate on, a value to compare to the value currently be at that
5387 address, and a new value to place at that address if the compared values are
5388 equal. The type of '<var><cmp></var>' must be an integer type whose
5389 bit width is a power of two greater than or equal to eight and less than
5390 or equal to a target-specific size limit. '<var><cmp></var>' and
5391 '<var><new></var>' must have the same type, and the type of
5392 '<var><pointer></var>' must be a pointer to that type. If the
5393 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
5394 optimizer is not allowed to modify the number or order of execution
5395 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
5398 <!-- FIXME: Extend allowed types. -->
5400 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
5401 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
5403 <p>The optional "<code>singlethread</code>" argument declares that the
5404 <code>cmpxchg</code> is only atomic with respect to code (usually signal
5405 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
5406 cmpxchg is atomic with respect to all other code in the system.</p>
5408 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
5409 the size in memory of the operand.
5412 <p>The contents of memory at the location specified by the
5413 '<tt><pointer></tt>' operand is read and compared to
5414 '<tt><cmp></tt>'; if the read value is the equal,
5415 '<tt><new></tt>' is written. The original value at the location
5418 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
5419 purpose of identifying <a href="#release_sequence">release sequences</a>. A
5420 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
5421 parameter determined by dropping any <code>release</code> part of the
5422 <code>cmpxchg</code>'s ordering.</p>
5425 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
5426 optimization work on ARM.)
5428 FIXME: Is a weaker ordering constraint on failure helpful in practice?
5434 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
5435 <a href="#i_br">br</a> label %loop
5438 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
5439 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
5440 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
5441 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
5442 <a href="#i_br">br</a> i1 %success, label %done, label %loop
5450 <!-- _______________________________________________________________________ -->
5452 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
5459 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
5463 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
5466 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
5467 operation to apply, an address whose value to modify, an argument to the
5468 operation. The operation must be one of the following keywords:</p>
5483 <p>The type of '<var><value></var>' must be an integer type whose
5484 bit width is a power of two greater than or equal to eight and less than
5485 or equal to a target-specific size limit. The type of the
5486 '<code><pointer></code>' operand must be a pointer to that type.
5487 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
5488 optimizer is not allowed to modify the number or order of execution of this
5489 <code>atomicrmw</code> with other <a href="#volatile">volatile
5492 <!-- FIXME: Extend allowed types. -->
5495 <p>The contents of memory at the location specified by the
5496 '<tt><pointer></tt>' operand are atomically read, modified, and written
5497 back. The original value at the location is returned. The modification is
5498 specified by the <var>operation</var> argument:</p>
5501 <li>xchg: <code>*ptr = val</code></li>
5502 <li>add: <code>*ptr = *ptr + val</code></li>
5503 <li>sub: <code>*ptr = *ptr - val</code></li>
5504 <li>and: <code>*ptr = *ptr & val</code></li>
5505 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
5506 <li>or: <code>*ptr = *ptr | val</code></li>
5507 <li>xor: <code>*ptr = *ptr ^ val</code></li>
5508 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
5509 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
5510 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
5511 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
5516 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
5521 <!-- _______________________________________________________________________ -->
5523 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
5530 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5531 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5532 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5536 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
5537 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
5538 It performs address calculation only and does not access memory.</p>
5541 <p>The first argument is always a pointer or a vector of pointers,
5542 and forms the basis of the
5543 calculation. The remaining arguments are indices that indicate which of the
5544 elements of the aggregate object are indexed. The interpretation of each
5545 index is dependent on the type being indexed into. The first index always
5546 indexes the pointer value given as the first argument, the second index
5547 indexes a value of the type pointed to (not necessarily the value directly
5548 pointed to, since the first index can be non-zero), etc. The first type
5549 indexed into must be a pointer value, subsequent types can be arrays,
5550 vectors, and structs. Note that subsequent types being indexed into
5551 can never be pointers, since that would require loading the pointer before
5552 continuing calculation.</p>
5554 <p>The type of each index argument depends on the type it is indexing into.
5555 When indexing into a (optionally packed) structure, only <tt>i32</tt>
5556 integer <b>constants</b> are allowed (when using a vector of indices they
5557 must all be the <b>same</b> <tt>i32</tt> integer constant). When indexing
5558 into an array, pointer or vector, integers of any width are allowed, and
5559 they are not required to be constant. These integers are treated as signed
5560 values where relevant.</p>
5562 <p>For example, let's consider a C code fragment and how it gets compiled to
5565 <pre class="doc_code">
5577 int *foo(struct ST *s) {
5578 return &s[1].Z.B[5][13];
5582 <p>The LLVM code generated by Clang is:</p>
5584 <pre class="doc_code">
5585 %struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 }
5586 %struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT }
5588 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5590 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5596 <p>In the example above, the first index is indexing into the
5597 '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a
5598 '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a
5599 structure. The second index indexes into the third element of the structure,
5600 yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>'
5601 type, another structure. The third index indexes into the second element of
5602 the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The
5603 two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>'
5604 type. The '<tt>getelementptr</tt>' instruction returns a pointer to this
5605 element, thus computing a value of '<tt>i32*</tt>' type.</p>
5607 <p>Note that it is perfectly legal to index partially through a structure,
5608 returning a pointer to an inner element. Because of this, the LLVM code for
5609 the given testcase is equivalent to:</p>
5611 <pre class="doc_code">
5612 define i32* @foo(%struct.ST* %s) {
5613 %t1 = getelementptr %struct.ST* %s, i32 1 <i>; yields %struct.ST*:%t1</i>
5614 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 <i>; yields %struct.RT*:%t2</i>
5615 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5616 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5617 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5622 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5623 <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the
5624 base pointer is not an <i>in bounds</i> address of an allocated object,
5625 or if any of the addresses that would be formed by successive addition of
5626 the offsets implied by the indices to the base address with infinitely
5627 precise signed arithmetic are not an <i>in bounds</i> address of that
5628 allocated object. The <i>in bounds</i> addresses for an allocated object
5629 are all the addresses that point into the object, plus the address one
5631 In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword
5632 applies to each of the computations element-wise. </p>
5634 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5635 the base address with silently-wrapping two's complement arithmetic. If the
5636 offsets have a different width from the pointer, they are sign-extended or
5637 truncated to the width of the pointer. The result value of the
5638 <tt>getelementptr</tt> may be outside the object pointed to by the base
5639 pointer. The result value may not necessarily be used to access memory
5640 though, even if it happens to point into allocated storage. See the
5641 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5644 <p>The getelementptr instruction is often confusing. For some more insight into
5645 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5649 <i>; yields [12 x i8]*:aptr</i>
5650 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5651 <i>; yields i8*:vptr</i>
5652 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5653 <i>; yields i8*:eptr</i>
5654 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5655 <i>; yields i32*:iptr</i>
5656 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5659 <p>In cases where the pointer argument is a vector of pointers, each index must
5660 be a vector with the same number of elements. For example: </p>
5661 <pre class="doc_code">
5662 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5669 <!-- ======================================================================= -->
5671 <a name="convertops">Conversion Operations</a>
5676 <p>The instructions in this category are the conversion instructions (casting)
5677 which all take a single operand and a type. They perform various bit
5678 conversions on the operand.</p>
5680 <!-- _______________________________________________________________________ -->
5682 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5689 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5693 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5694 type <tt>ty2</tt>.</p>
5697 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5698 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5699 of the same number of integers.
5700 The bit size of the <tt>value</tt> must be larger than
5701 the bit size of the destination type, <tt>ty2</tt>.
5702 Equal sized types are not allowed.</p>
5705 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5706 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5707 source size must be larger than the destination size, <tt>trunc</tt> cannot
5708 be a <i>no-op cast</i>. It will always truncate bits.</p>
5712 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5713 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5714 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5715 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5720 <!-- _______________________________________________________________________ -->
5722 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5729 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5733 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5738 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5739 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5740 of the same number of integers.
5741 The bit size of the <tt>value</tt> must be smaller than
5742 the bit size of the destination type,
5746 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5747 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5749 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5753 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5754 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5755 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5760 <!-- _______________________________________________________________________ -->
5762 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5769 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5773 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5776 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5777 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5778 of the same number of integers.
5779 The bit size of the <tt>value</tt> must be smaller than
5780 the bit size of the destination type,
5784 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5785 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5786 of the type <tt>ty2</tt>.</p>
5788 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5792 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5793 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5794 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5799 <!-- _______________________________________________________________________ -->
5801 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5808 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5812 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5816 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5817 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5818 to cast it to. The size of <tt>value</tt> must be larger than the size of
5819 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5820 <i>no-op cast</i>.</p>
5823 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5824 <a href="#t_floating">floating point</a> type to a smaller
5825 <a href="#t_floating">floating point</a> type. If the value cannot fit
5826 within the destination type, <tt>ty2</tt>, then the results are
5831 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5832 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5837 <!-- _______________________________________________________________________ -->
5839 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5846 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5850 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5851 floating point value.</p>
5854 <p>The '<tt>fpext</tt>' instruction takes a
5855 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5856 a <a href="#t_floating">floating point</a> type to cast it to. The source
5857 type must be smaller than the destination type.</p>
5860 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5861 <a href="#t_floating">floating point</a> type to a larger
5862 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5863 used to make a <i>no-op cast</i> because it always changes bits. Use
5864 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5868 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5869 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5874 <!-- _______________________________________________________________________ -->
5876 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5883 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5887 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5888 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5891 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5892 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5893 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5894 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5895 vector integer type with the same number of elements as <tt>ty</tt></p>
5898 <p>The '<tt>fptoui</tt>' instruction converts its
5899 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5900 towards zero) unsigned integer value. If the value cannot fit
5901 in <tt>ty2</tt>, the results are undefined.</p>
5905 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5906 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5907 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5912 <!-- _______________________________________________________________________ -->
5914 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5921 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5925 <p>The '<tt>fptosi</tt>' instruction converts
5926 <a href="#t_floating">floating point</a> <tt>value</tt> to
5927 type <tt>ty2</tt>.</p>
5930 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5931 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5932 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5933 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5934 vector integer type with the same number of elements as <tt>ty</tt></p>
5937 <p>The '<tt>fptosi</tt>' instruction converts its
5938 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5939 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5940 the results are undefined.</p>
5944 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5945 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5946 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5951 <!-- _______________________________________________________________________ -->
5953 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5960 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5964 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5965 integer and converts that value to the <tt>ty2</tt> type.</p>
5968 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5969 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5970 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5971 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5972 floating point type with the same number of elements as <tt>ty</tt></p>
5975 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5976 integer quantity and converts it to the corresponding floating point
5977 value. If the value cannot fit in the floating point value, the results are
5982 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5983 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5988 <!-- _______________________________________________________________________ -->
5990 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5997 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
6001 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
6002 and converts that value to the <tt>ty2</tt> type.</p>
6005 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
6006 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
6007 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
6008 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
6009 floating point type with the same number of elements as <tt>ty</tt></p>
6012 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
6013 quantity and converts it to the corresponding floating point value. If the
6014 value cannot fit in the floating point value, the results are undefined.</p>
6018 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
6019 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
6024 <!-- _______________________________________________________________________ -->
6026 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
6033 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
6037 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of
6038 pointers <tt>value</tt> to
6039 the integer (or vector of integers) type <tt>ty2</tt>.</p>
6042 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
6043 must be a a value of type <a href="#t_pointer">pointer</a> or a vector of
6044 pointers, and a type to cast it to
6045 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector
6046 of integers type.</p>
6049 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
6050 <tt>ty2</tt> by interpreting the pointer value as an integer and either
6051 truncating or zero extending that value to the size of the integer type. If
6052 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
6053 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
6054 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
6059 %X = ptrtoint i32* %P to i8 <i>; yields truncation on 32-bit architecture</i>
6060 %Y = ptrtoint i32* %P to i64 <i>; yields zero extension on 32-bit architecture</i>
6061 %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>
6066 <!-- _______________________________________________________________________ -->
6068 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
6075 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
6079 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
6080 pointer type, <tt>ty2</tt>.</p>
6083 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
6084 value to cast, and a type to cast it to, which must be a
6085 <a href="#t_pointer">pointer</a> type.</p>
6088 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
6089 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
6090 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
6091 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
6092 than the size of a pointer then a zero extension is done. If they are the
6093 same size, nothing is done (<i>no-op cast</i>).</p>
6097 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
6098 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
6099 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
6100 %Z = inttoptr <4 x i32> %G to <4 x i8*><i>; yields truncation of vector G to four pointers</i>
6105 <!-- _______________________________________________________________________ -->
6107 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
6114 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
6118 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
6119 <tt>ty2</tt> without changing any bits.</p>
6122 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
6123 non-aggregate first class value, and a type to cast it to, which must also be
6124 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
6125 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
6126 identical. If the source type is a pointer, the destination type must also be
6127 a pointer. This instruction supports bitwise conversion of vectors to
6128 integers and to vectors of other types (as long as they have the same
6132 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
6133 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
6134 this conversion. The conversion is done as if the <tt>value</tt> had been
6135 stored to memory and read back as type <tt>ty2</tt>.
6136 Pointer (or vector of pointers) types may only be converted to other pointer
6137 (or vector of pointers) types with this instruction. To convert
6138 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
6139 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
6143 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
6144 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
6145 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
6146 %Z = bitcast <2 x i32*> %V to <2 x i64*> <i>; yields <2 x i64*></i>
6153 <!-- ======================================================================= -->
6155 <a name="otherops">Other Operations</a>
6160 <p>The instructions in this category are the "miscellaneous" instructions, which
6161 defy better classification.</p>
6163 <!-- _______________________________________________________________________ -->
6165 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
6172 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
6176 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
6177 boolean values based on comparison of its two integer, integer vector,
6178 pointer, or pointer vector operands.</p>
6181 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
6182 the condition code indicating the kind of comparison to perform. It is not a
6183 value, just a keyword. The possible condition code are:</p>
6186 <li><tt>eq</tt>: equal</li>
6187 <li><tt>ne</tt>: not equal </li>
6188 <li><tt>ugt</tt>: unsigned greater than</li>
6189 <li><tt>uge</tt>: unsigned greater or equal</li>
6190 <li><tt>ult</tt>: unsigned less than</li>
6191 <li><tt>ule</tt>: unsigned less or equal</li>
6192 <li><tt>sgt</tt>: signed greater than</li>
6193 <li><tt>sge</tt>: signed greater or equal</li>
6194 <li><tt>slt</tt>: signed less than</li>
6195 <li><tt>sle</tt>: signed less or equal</li>
6198 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
6199 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
6200 typed. They must also be identical types.</p>
6203 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
6204 condition code given as <tt>cond</tt>. The comparison performed always yields
6205 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
6206 result, as follows:</p>
6209 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
6210 <tt>false</tt> otherwise. No sign interpretation is necessary or
6213 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
6214 <tt>false</tt> otherwise. No sign interpretation is necessary or
6217 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
6218 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6220 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
6221 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6222 to <tt>op2</tt>.</li>
6224 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
6225 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6227 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
6228 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6230 <li><tt>sgt</tt>: interprets the operands as signed values and yields
6231 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6233 <li><tt>sge</tt>: interprets the operands as signed values and yields
6234 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6235 to <tt>op2</tt>.</li>
6237 <li><tt>slt</tt>: interprets the operands as signed values and yields
6238 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6240 <li><tt>sle</tt>: interprets the operands as signed values and yields
6241 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6244 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
6245 values are compared as if they were integers.</p>
6247 <p>If the operands are integer vectors, then they are compared element by
6248 element. The result is an <tt>i1</tt> vector with the same number of elements
6249 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
6253 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
6254 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
6255 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
6256 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
6257 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
6258 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
6261 <p>Note that the code generator does not yet support vector types with
6262 the <tt>icmp</tt> instruction.</p>
6266 <!-- _______________________________________________________________________ -->
6268 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
6275 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
6279 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
6280 values based on comparison of its operands.</p>
6282 <p>If the operands are floating point scalars, then the result type is a boolean
6283 (<a href="#t_integer"><tt>i1</tt></a>).</p>
6285 <p>If the operands are floating point vectors, then the result type is a vector
6286 of boolean with the same number of elements as the operands being
6290 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
6291 the condition code indicating the kind of comparison to perform. It is not a
6292 value, just a keyword. The possible condition code are:</p>
6295 <li><tt>false</tt>: no comparison, always returns false</li>
6296 <li><tt>oeq</tt>: ordered and equal</li>
6297 <li><tt>ogt</tt>: ordered and greater than </li>
6298 <li><tt>oge</tt>: ordered and greater than or equal</li>
6299 <li><tt>olt</tt>: ordered and less than </li>
6300 <li><tt>ole</tt>: ordered and less than or equal</li>
6301 <li><tt>one</tt>: ordered and not equal</li>
6302 <li><tt>ord</tt>: ordered (no nans)</li>
6303 <li><tt>ueq</tt>: unordered or equal</li>
6304 <li><tt>ugt</tt>: unordered or greater than </li>
6305 <li><tt>uge</tt>: unordered or greater than or equal</li>
6306 <li><tt>ult</tt>: unordered or less than </li>
6307 <li><tt>ule</tt>: unordered or less than or equal</li>
6308 <li><tt>une</tt>: unordered or not equal</li>
6309 <li><tt>uno</tt>: unordered (either nans)</li>
6310 <li><tt>true</tt>: no comparison, always returns true</li>
6313 <p><i>Ordered</i> means that neither operand is a QNAN while
6314 <i>unordered</i> means that either operand may be a QNAN.</p>
6316 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
6317 a <a href="#t_floating">floating point</a> type or
6318 a <a href="#t_vector">vector</a> of floating point type. They must have
6319 identical types.</p>
6322 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
6323 according to the condition code given as <tt>cond</tt>. If the operands are
6324 vectors, then the vectors are compared element by element. Each comparison
6325 performed always yields an <a href="#t_integer">i1</a> result, as
6329 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
6331 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6332 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6334 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6335 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6337 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6338 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6340 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6341 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6343 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6344 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6346 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6347 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6349 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
6351 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
6352 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6354 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
6355 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6357 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
6358 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6360 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
6361 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6363 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
6364 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6366 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
6367 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6369 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
6371 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
6376 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
6377 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
6378 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
6379 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
6382 <p>Note that the code generator does not yet support vector types with
6383 the <tt>fcmp</tt> instruction.</p>
6387 <!-- _______________________________________________________________________ -->
6389 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
6396 <result> = phi <ty> [ <val0>, <label0>], ...
6400 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
6401 SSA graph representing the function.</p>
6404 <p>The type of the incoming values is specified with the first type field. After
6405 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
6406 one pair for each predecessor basic block of the current block. Only values
6407 of <a href="#t_firstclass">first class</a> type may be used as the value
6408 arguments to the PHI node. Only labels may be used as the label
6411 <p>There must be no non-phi instructions between the start of a basic block and
6412 the PHI instructions: i.e. PHI instructions must be first in a basic
6415 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
6416 occur on the edge from the corresponding predecessor block to the current
6417 block (but after any definition of an '<tt>invoke</tt>' instruction's return
6418 value on the same edge).</p>
6421 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
6422 specified by the pair corresponding to the predecessor basic block that
6423 executed just prior to the current block.</p>
6427 Loop: ; Infinite loop that counts from 0 on up...
6428 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6429 %nextindvar = add i32 %indvar, 1
6435 <!-- _______________________________________________________________________ -->
6437 <a name="i_select">'<tt>select</tt>' Instruction</a>
6444 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
6446 <i>selty</i> is either i1 or {<N x i1>}
6450 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
6451 condition, without branching.</p>
6455 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
6456 values indicating the condition, and two values of the
6457 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
6458 vectors and the condition is a scalar, then entire vectors are selected, not
6459 individual elements.</p>
6462 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
6463 first value argument; otherwise, it returns the second value argument.</p>
6465 <p>If the condition is a vector of i1, then the value arguments must be vectors
6466 of the same size, and the selection is done element by element.</p>
6470 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
6475 <!-- _______________________________________________________________________ -->
6477 <a name="i_call">'<tt>call</tt>' Instruction</a>
6484 <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>]
6488 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
6491 <p>This instruction requires several arguments:</p>
6494 <li>The optional "tail" marker indicates that the callee function does not
6495 access any allocas or varargs in the caller. Note that calls may be
6496 marked "tail" even if they do not occur before
6497 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
6498 present, the function call is eligible for tail call optimization,
6499 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
6500 optimized into a jump</a>. The code generator may optimize calls marked
6501 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
6502 sibling call optimization</a> when the caller and callee have
6503 matching signatures, or 2) forced tail call optimization when the
6504 following extra requirements are met:
6506 <li>Caller and callee both have the calling
6507 convention <tt>fastcc</tt>.</li>
6508 <li>The call is in tail position (ret immediately follows call and ret
6509 uses value of call or is void).</li>
6510 <li>Option <tt>-tailcallopt</tt> is enabled,
6511 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
6512 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
6513 constraints are met.</a></li>
6517 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
6518 convention</a> the call should use. If none is specified, the call
6519 defaults to using C calling conventions. The calling convention of the
6520 call must match the calling convention of the target function, or else the
6521 behavior is undefined.</li>
6523 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
6524 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
6525 '<tt>inreg</tt>' attributes are valid here.</li>
6527 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
6528 type of the return value. Functions that return no value are marked
6529 <tt><a href="#t_void">void</a></tt>.</li>
6531 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
6532 being invoked. The argument types must match the types implied by this
6533 signature. This type can be omitted if the function is not varargs and if
6534 the function type does not return a pointer to a function.</li>
6536 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
6537 be invoked. In most cases, this is a direct function invocation, but
6538 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
6539 to function value.</li>
6541 <li>'<tt>function args</tt>': argument list whose types match the function
6542 signature argument types and parameter attributes. All arguments must be
6543 of <a href="#t_firstclass">first class</a> type. If the function
6544 signature indicates the function accepts a variable number of arguments,
6545 the extra arguments can be specified.</li>
6547 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
6548 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
6549 '<tt>readnone</tt>' attributes are valid here.</li>
6553 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
6554 a specified function, with its incoming arguments bound to the specified
6555 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
6556 function, control flow continues with the instruction after the function
6557 call, and the return value of the function is bound to the result
6562 %retval = call i32 @test(i32 %argc)
6563 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
6564 %X = tail call i32 @foo() <i>; yields i32</i>
6565 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
6566 call void %foo(i8 97 signext)
6568 %struct.A = type { i32, i8 }
6569 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
6570 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
6571 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
6572 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
6573 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
6576 <p>llvm treats calls to some functions with names and arguments that match the
6577 standard C99 library as being the C99 library functions, and may perform
6578 optimizations or generate code for them under that assumption. This is
6579 something we'd like to change in the future to provide better support for
6580 freestanding environments and non-C-based languages.</p>
6584 <!-- _______________________________________________________________________ -->
6586 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
6593 <resultval> = va_arg <va_list*> <arglist>, <argty>
6597 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6598 the "variable argument" area of a function call. It is used to implement the
6599 <tt>va_arg</tt> macro in C.</p>
6602 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6603 argument. It returns a value of the specified argument type and increments
6604 the <tt>va_list</tt> to point to the next argument. The actual type
6605 of <tt>va_list</tt> is target specific.</p>
6608 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6609 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6610 to the next argument. For more information, see the variable argument
6611 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6613 <p>It is legal for this instruction to be called in a function which does not
6614 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6617 <p><tt>va_arg</tt> is an LLVM instruction instead of
6618 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6622 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6624 <p>Note that the code generator does not yet fully support va_arg on many
6625 targets. Also, it does not currently support va_arg with aggregate types on
6630 <!-- _______________________________________________________________________ -->
6632 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6639 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6640 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6642 <clause> := catch <type> <value>
6643 <clause> := filter <array constant type> <array constant>
6647 <p>The '<tt>landingpad</tt>' instruction is used by
6648 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6649 system</a> to specify that a basic block is a landing pad — one where
6650 the exception lands, and corresponds to the code found in the
6651 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6652 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6653 re-entry to the function. The <tt>resultval</tt> has the
6654 type <tt>resultty</tt>.</p>
6657 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6658 function associated with the unwinding mechanism. The optional
6659 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6661 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6662 or <tt>filter</tt> — and contains the global variable representing the
6663 "type" that may be caught or filtered respectively. Unlike the
6664 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6665 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6666 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6667 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6670 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6671 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6672 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6673 calling conventions, how the personality function results are represented in
6674 LLVM IR is target specific.</p>
6676 <p>The clauses are applied in order from top to bottom. If two
6677 <tt>landingpad</tt> instructions are merged together through inlining, the
6678 clauses from the calling function are appended to the list of clauses.
6679 When the call stack is being unwound due to an exception being thrown, the
6680 exception is compared against each <tt>clause</tt> in turn. If it doesn't
6681 match any of the clauses, and the <tt>cleanup</tt> flag is not set, then
6682 unwinding continues further up the call stack.</p>
6684 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6687 <li>A landing pad block is a basic block which is the unwind destination of an
6688 '<tt>invoke</tt>' instruction.</li>
6689 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6690 first non-PHI instruction.</li>
6691 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6693 <li>A basic block that is not a landing pad block may not include a
6694 '<tt>landingpad</tt>' instruction.</li>
6695 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6696 personality function.</li>
6701 ;; A landing pad which can catch an integer.
6702 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6704 ;; A landing pad that is a cleanup.
6705 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6707 ;; A landing pad which can catch an integer and can only throw a double.
6708 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6710 filter [1 x i8**] [@_ZTId]
6719 <!-- *********************************************************************** -->
6720 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6721 <!-- *********************************************************************** -->
6725 <p>LLVM supports the notion of an "intrinsic function". These functions have
6726 well known names and semantics and are required to follow certain
6727 restrictions. Overall, these intrinsics represent an extension mechanism for
6728 the LLVM language that does not require changing all of the transformations
6729 in LLVM when adding to the language (or the bitcode reader/writer, the
6730 parser, etc...).</p>
6732 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6733 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6734 begin with this prefix. Intrinsic functions must always be external
6735 functions: you cannot define the body of intrinsic functions. Intrinsic
6736 functions may only be used in call or invoke instructions: it is illegal to
6737 take the address of an intrinsic function. Additionally, because intrinsic
6738 functions are part of the LLVM language, it is required if any are added that
6739 they be documented here.</p>
6741 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6742 family of functions that perform the same operation but on different data
6743 types. Because LLVM can represent over 8 million different integer types,
6744 overloading is used commonly to allow an intrinsic function to operate on any
6745 integer type. One or more of the argument types or the result type can be
6746 overloaded to accept any integer type. Argument types may also be defined as
6747 exactly matching a previous argument's type or the result type. This allows
6748 an intrinsic function which accepts multiple arguments, but needs all of them
6749 to be of the same type, to only be overloaded with respect to a single
6750 argument or the result.</p>
6752 <p>Overloaded intrinsics will have the names of its overloaded argument types
6753 encoded into its function name, each preceded by a period. Only those types
6754 which are overloaded result in a name suffix. Arguments whose type is matched
6755 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6756 can take an integer of any width and returns an integer of exactly the same
6757 integer width. This leads to a family of functions such as
6758 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6759 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6760 suffix is required. Because the argument's type is matched against the return
6761 type, it does not require its own name suffix.</p>
6763 <p>To learn how to add an intrinsic function, please see the
6764 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6766 <!-- ======================================================================= -->
6768 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6773 <p>Variable argument support is defined in LLVM with
6774 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6775 intrinsic functions. These functions are related to the similarly named
6776 macros defined in the <tt><stdarg.h></tt> header file.</p>
6778 <p>All of these functions operate on arguments that use a target-specific value
6779 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6780 not define what this type is, so all transformations should be prepared to
6781 handle these functions regardless of the type used.</p>
6783 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6784 instruction and the variable argument handling intrinsic functions are
6787 <pre class="doc_code">
6788 define i32 @test(i32 %X, ...) {
6789 ; Initialize variable argument processing
6791 %ap2 = bitcast i8** %ap to i8*
6792 call void @llvm.va_start(i8* %ap2)
6794 ; Read a single integer argument
6795 %tmp = va_arg i8** %ap, i32
6797 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6799 %aq2 = bitcast i8** %aq to i8*
6800 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6801 call void @llvm.va_end(i8* %aq2)
6803 ; Stop processing of arguments.
6804 call void @llvm.va_end(i8* %ap2)
6808 declare void @llvm.va_start(i8*)
6809 declare void @llvm.va_copy(i8*, i8*)
6810 declare void @llvm.va_end(i8*)
6813 <!-- _______________________________________________________________________ -->
6815 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6823 declare void %llvm.va_start(i8* <arglist>)
6827 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6828 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6831 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6834 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6835 macro available in C. In a target-dependent way, it initializes
6836 the <tt>va_list</tt> element to which the argument points, so that the next
6837 call to <tt>va_arg</tt> will produce the first variable argument passed to
6838 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6839 need to know the last argument of the function as the compiler can figure
6844 <!-- _______________________________________________________________________ -->
6846 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6853 declare void @llvm.va_end(i8* <arglist>)
6857 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6858 which has been initialized previously
6859 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6860 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6863 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6866 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6867 macro available in C. In a target-dependent way, it destroys
6868 the <tt>va_list</tt> element to which the argument points. Calls
6869 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6870 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6871 with calls to <tt>llvm.va_end</tt>.</p>
6875 <!-- _______________________________________________________________________ -->
6877 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6884 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6888 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6889 from the source argument list to the destination argument list.</p>
6892 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6893 The second argument is a pointer to a <tt>va_list</tt> element to copy
6897 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6898 macro available in C. In a target-dependent way, it copies the
6899 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6900 element. This intrinsic is necessary because
6901 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6902 arbitrarily complex and require, for example, memory allocation.</p>
6908 <!-- ======================================================================= -->
6910 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6915 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6916 Collection</a> (GC) requires the implementation and generation of these
6917 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6918 roots on the stack</a>, as well as garbage collector implementations that
6919 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6920 barriers. Front-ends for type-safe garbage collected languages should generate
6921 these intrinsics to make use of the LLVM garbage collectors. For more details,
6922 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6925 <p>The garbage collection intrinsics only operate on objects in the generic
6926 address space (address space zero).</p>
6928 <!-- _______________________________________________________________________ -->
6930 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6937 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6941 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6942 the code generator, and allows some metadata to be associated with it.</p>
6945 <p>The first argument specifies the address of a stack object that contains the
6946 root pointer. The second pointer (which must be either a constant or a
6947 global value address) contains the meta-data to be associated with the
6951 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6952 location. At compile-time, the code generator generates information to allow
6953 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6954 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6959 <!-- _______________________________________________________________________ -->
6961 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6968 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6972 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6973 locations, allowing garbage collector implementations that require read
6977 <p>The second argument is the address to read from, which should be an address
6978 allocated from the garbage collector. The first object is a pointer to the
6979 start of the referenced object, if needed by the language runtime (otherwise
6983 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6984 instruction, but may be replaced with substantially more complex code by the
6985 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6986 may only be used in a function which <a href="#gc">specifies a GC
6991 <!-- _______________________________________________________________________ -->
6993 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
7000 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
7004 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
7005 locations, allowing garbage collector implementations that require write
7006 barriers (such as generational or reference counting collectors).</p>
7009 <p>The first argument is the reference to store, the second is the start of the
7010 object to store it to, and the third is the address of the field of Obj to
7011 store to. If the runtime does not require a pointer to the object, Obj may
7015 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
7016 instruction, but may be replaced with substantially more complex code by the
7017 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
7018 may only be used in a function which <a href="#gc">specifies a GC
7025 <!-- ======================================================================= -->
7027 <a name="int_codegen">Code Generator Intrinsics</a>
7032 <p>These intrinsics are provided by LLVM to expose special features that may
7033 only be implemented with code generator support.</p>
7035 <!-- _______________________________________________________________________ -->
7037 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
7044 declare i8 *@llvm.returnaddress(i32 <level>)
7048 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
7049 target-specific value indicating the return address of the current function
7050 or one of its callers.</p>
7053 <p>The argument to this intrinsic indicates which function to return the address
7054 for. Zero indicates the calling function, one indicates its caller, etc.
7055 The argument is <b>required</b> to be a constant integer value.</p>
7058 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
7059 indicating the return address of the specified call frame, or zero if it
7060 cannot be identified. The value returned by this intrinsic is likely to be
7061 incorrect or 0 for arguments other than zero, so it should only be used for
7062 debugging purposes.</p>
7064 <p>Note that calling this intrinsic does not prevent function inlining or other
7065 aggressive transformations, so the value returned may not be that of the
7066 obvious source-language caller.</p>
7070 <!-- _______________________________________________________________________ -->
7072 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
7079 declare i8* @llvm.frameaddress(i32 <level>)
7083 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
7084 target-specific frame pointer value for the specified stack frame.</p>
7087 <p>The argument to this intrinsic indicates which function to return the frame
7088 pointer for. Zero indicates the calling function, one indicates its caller,
7089 etc. The argument is <b>required</b> to be a constant integer value.</p>
7092 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
7093 indicating the frame address of the specified call frame, or zero if it
7094 cannot be identified. The value returned by this intrinsic is likely to be
7095 incorrect or 0 for arguments other than zero, so it should only be used for
7096 debugging purposes.</p>
7098 <p>Note that calling this intrinsic does not prevent function inlining or other
7099 aggressive transformations, so the value returned may not be that of the
7100 obvious source-language caller.</p>
7104 <!-- _______________________________________________________________________ -->
7106 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
7113 declare i8* @llvm.stacksave()
7117 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
7118 of the function stack, for use
7119 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
7120 useful for implementing language features like scoped automatic variable
7121 sized arrays in C99.</p>
7124 <p>This intrinsic returns a opaque pointer value that can be passed
7125 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
7126 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
7127 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
7128 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
7129 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
7130 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
7134 <!-- _______________________________________________________________________ -->
7136 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
7143 declare void @llvm.stackrestore(i8* %ptr)
7147 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
7148 the function stack to the state it was in when the
7149 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
7150 executed. This is useful for implementing language features like scoped
7151 automatic variable sized arrays in C99.</p>
7154 <p>See the description
7155 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
7159 <!-- _______________________________________________________________________ -->
7161 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
7168 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7172 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
7173 insert a prefetch instruction if supported; otherwise, it is a noop.
7174 Prefetches have no effect on the behavior of the program but can change its
7175 performance characteristics.</p>
7178 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
7179 specifier determining if the fetch should be for a read (0) or write (1),
7180 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
7181 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
7182 specifies whether the prefetch is performed on the data (1) or instruction (0)
7183 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
7184 must be constant integers.</p>
7187 <p>This intrinsic does not modify the behavior of the program. In particular,
7188 prefetches cannot trap and do not produce a value. On targets that support
7189 this intrinsic, the prefetch can provide hints to the processor cache for
7190 better performance.</p>
7194 <!-- _______________________________________________________________________ -->
7196 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
7203 declare void @llvm.pcmarker(i32 <id>)
7207 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
7208 Counter (PC) in a region of code to simulators and other tools. The method
7209 is target specific, but it is expected that the marker will use exported
7210 symbols to transmit the PC of the marker. The marker makes no guarantees
7211 that it will remain with any specific instruction after optimizations. It is
7212 possible that the presence of a marker will inhibit optimizations. The
7213 intended use is to be inserted after optimizations to allow correlations of
7214 simulation runs.</p>
7217 <p><tt>id</tt> is a numerical id identifying the marker.</p>
7220 <p>This intrinsic does not modify the behavior of the program. Backends that do
7221 not support this intrinsic may ignore it.</p>
7225 <!-- _______________________________________________________________________ -->
7227 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
7234 declare i64 @llvm.readcyclecounter()
7238 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
7239 counter register (or similar low latency, high accuracy clocks) on those
7240 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7241 should map to RPCC. As the backing counters overflow quickly (on the order
7242 of 9 seconds on alpha), this should only be used for small timings.</p>
7245 <p>When directly supported, reading the cycle counter should not modify any
7246 memory. Implementations are allowed to either return a application specific
7247 value or a system wide value. On backends without support, this is lowered
7248 to a constant 0.</p>
7254 <!-- ======================================================================= -->
7256 <a name="int_libc">Standard C Library Intrinsics</a>
7261 <p>LLVM provides intrinsics for a few important standard C library functions.
7262 These intrinsics allow source-language front-ends to pass information about
7263 the alignment of the pointer arguments to the code generator, providing
7264 opportunity for more efficient code generation.</p>
7266 <!-- _______________________________________________________________________ -->
7268 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
7274 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
7275 integer bit width and for different address spaces. Not all targets support
7276 all bit widths however.</p>
7279 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7280 i32 <len>, i32 <align>, i1 <isvolatile>)
7281 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7282 i64 <len>, i32 <align>, i1 <isvolatile>)
7286 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7287 source location to the destination location.</p>
7289 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
7290 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7291 and the pointers can be in specified address spaces.</p>
7295 <p>The first argument is a pointer to the destination, the second is a pointer
7296 to the source. The third argument is an integer argument specifying the
7297 number of bytes to copy, the fourth argument is the alignment of the
7298 source and destination locations, and the fifth is a boolean indicating a
7299 volatile access.</p>
7301 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7302 then the caller guarantees that both the source and destination pointers are
7303 aligned to that boundary.</p>
7305 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7306 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
7307 The detailed access behavior is not very cleanly specified and it is unwise
7308 to depend on it.</p>
7312 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7313 source location to the destination location, which are not allowed to
7314 overlap. It copies "len" bytes of memory over. If the argument is known to
7315 be aligned to some boundary, this can be specified as the fourth argument,
7316 otherwise it should be set to 0 or 1.</p>
7320 <!-- _______________________________________________________________________ -->
7322 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
7328 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
7329 width and for different address space. Not all targets support all bit
7333 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7334 i32 <len>, i32 <align>, i1 <isvolatile>)
7335 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7336 i64 <len>, i32 <align>, i1 <isvolatile>)
7340 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
7341 source location to the destination location. It is similar to the
7342 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
7345 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
7346 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7347 and the pointers can be in specified address spaces.</p>
7351 <p>The first argument is a pointer to the destination, the second is a pointer
7352 to the source. The third argument is an integer argument specifying the
7353 number of bytes to copy, the fourth argument is the alignment of the
7354 source and destination locations, and the fifth is a boolean indicating a
7355 volatile access.</p>
7357 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7358 then the caller guarantees that the source and destination pointers are
7359 aligned to that boundary.</p>
7361 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7362 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
7363 The detailed access behavior is not very cleanly specified and it is unwise
7364 to depend on it.</p>
7368 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
7369 source location to the destination location, which may overlap. It copies
7370 "len" bytes of memory over. If the argument is known to be aligned to some
7371 boundary, this can be specified as the fourth argument, otherwise it should
7372 be set to 0 or 1.</p>
7376 <!-- _______________________________________________________________________ -->
7378 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
7384 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
7385 width and for different address spaces. However, not all targets support all
7389 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7390 i32 <len>, i32 <align>, i1 <isvolatile>)
7391 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7392 i64 <len>, i32 <align>, i1 <isvolatile>)
7396 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
7397 particular byte value.</p>
7399 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
7400 intrinsic does not return a value and takes extra alignment/volatile
7401 arguments. Also, the destination can be in an arbitrary address space.</p>
7404 <p>The first argument is a pointer to the destination to fill, the second is the
7405 byte value with which to fill it, the third argument is an integer argument
7406 specifying the number of bytes to fill, and the fourth argument is the known
7407 alignment of the destination location.</p>
7409 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7410 then the caller guarantees that the destination pointer is aligned to that
7413 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7414 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
7415 The detailed access behavior is not very cleanly specified and it is unwise
7416 to depend on it.</p>
7419 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
7420 at the destination location. If the argument is known to be aligned to some
7421 boundary, this can be specified as the fourth argument, otherwise it should
7422 be set to 0 or 1.</p>
7426 <!-- _______________________________________________________________________ -->
7428 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
7434 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
7435 floating point or vector of floating point type. Not all targets support all
7439 declare float @llvm.sqrt.f32(float %Val)
7440 declare double @llvm.sqrt.f64(double %Val)
7441 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7442 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7443 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7447 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
7448 returning the same value as the libm '<tt>sqrt</tt>' functions would.
7449 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
7450 behavior for negative numbers other than -0.0 (which allows for better
7451 optimization, because there is no need to worry about errno being
7452 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
7455 <p>The argument and return value are floating point numbers of the same
7459 <p>This function returns the sqrt of the specified operand if it is a
7460 nonnegative floating point number.</p>
7464 <!-- _______________________________________________________________________ -->
7466 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
7472 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
7473 floating point or vector of floating point type. Not all targets support all
7477 declare float @llvm.powi.f32(float %Val, i32 %power)
7478 declare double @llvm.powi.f64(double %Val, i32 %power)
7479 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7480 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7481 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7485 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
7486 specified (positive or negative) power. The order of evaluation of
7487 multiplications is not defined. When a vector of floating point type is
7488 used, the second argument remains a scalar integer value.</p>
7491 <p>The second argument is an integer power, and the first is a value to raise to
7495 <p>This function returns the first value raised to the second power with an
7496 unspecified sequence of rounding operations.</p>
7500 <!-- _______________________________________________________________________ -->
7502 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
7508 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
7509 floating point or vector of floating point type. Not all targets support all
7513 declare float @llvm.sin.f32(float %Val)
7514 declare double @llvm.sin.f64(double %Val)
7515 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7516 declare fp128 @llvm.sin.f128(fp128 %Val)
7517 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7521 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
7524 <p>The argument and return value are floating point numbers of the same
7528 <p>This function returns the sine of the specified operand, returning the same
7529 values as the libm <tt>sin</tt> functions would, and handles error conditions
7530 in the same way.</p>
7534 <!-- _______________________________________________________________________ -->
7536 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
7542 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
7543 floating point or vector of floating point type. Not all targets support all
7547 declare float @llvm.cos.f32(float %Val)
7548 declare double @llvm.cos.f64(double %Val)
7549 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7550 declare fp128 @llvm.cos.f128(fp128 %Val)
7551 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7555 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
7558 <p>The argument and return value are floating point numbers of the same
7562 <p>This function returns the cosine of the specified operand, returning the same
7563 values as the libm <tt>cos</tt> functions would, and handles error conditions
7564 in the same way.</p>
7568 <!-- _______________________________________________________________________ -->
7570 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
7576 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
7577 floating point or vector of floating point type. Not all targets support all
7581 declare float @llvm.pow.f32(float %Val, float %Power)
7582 declare double @llvm.pow.f64(double %Val, double %Power)
7583 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7584 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7585 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7589 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
7590 specified (positive or negative) power.</p>
7593 <p>The second argument is a floating point power, and the first is a value to
7594 raise to that power.</p>
7597 <p>This function returns the first value raised to the second power, returning
7598 the same values as the libm <tt>pow</tt> functions would, and handles error
7599 conditions in the same way.</p>
7603 <!-- _______________________________________________________________________ -->
7605 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7611 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7612 floating point or vector of floating point type. Not all targets support all
7616 declare float @llvm.exp.f32(float %Val)
7617 declare double @llvm.exp.f64(double %Val)
7618 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7619 declare fp128 @llvm.exp.f128(fp128 %Val)
7620 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7624 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7627 <p>The argument and return value are floating point numbers of the same
7631 <p>This function returns the same values as the libm <tt>exp</tt> functions
7632 would, and handles error conditions in the same way.</p>
7636 <!-- _______________________________________________________________________ -->
7638 <a name="int_exp2">'<tt>llvm.exp2.*</tt>' Intrinsic</a>
7644 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp2</tt> on any
7645 floating point or vector of floating point type. Not all targets support all
7649 declare float @llvm.exp2.f32(float %Val)
7650 declare double @llvm.exp2.f64(double %Val)
7651 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7652 declare fp128 @llvm.exp2.f128(fp128 %Val)
7653 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7657 <p>The '<tt>llvm.exp2.*</tt>' intrinsics perform the exp2 function.</p>
7660 <p>The argument and return value are floating point numbers of the same
7664 <p>This function returns the same values as the libm <tt>exp2</tt> functions
7665 would, and handles error conditions in the same way.</p>
7669 <!-- _______________________________________________________________________ -->
7671 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7677 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7678 floating point or vector of floating point type. Not all targets support all
7682 declare float @llvm.log.f32(float %Val)
7683 declare double @llvm.log.f64(double %Val)
7684 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7685 declare fp128 @llvm.log.f128(fp128 %Val)
7686 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7690 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7693 <p>The argument and return value are floating point numbers of the same
7697 <p>This function returns the same values as the libm <tt>log</tt> functions
7698 would, and handles error conditions in the same way.</p>
7702 <!-- _______________________________________________________________________ -->
7704 <a name="int_log10">'<tt>llvm.log10.*</tt>' Intrinsic</a>
7710 <p>This is an overloaded intrinsic. You can use <tt>llvm.log10</tt> on any
7711 floating point or vector of floating point type. Not all targets support all
7715 declare float @llvm.log10.f32(float %Val)
7716 declare double @llvm.log10.f64(double %Val)
7717 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7718 declare fp128 @llvm.log10.f128(fp128 %Val)
7719 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7723 <p>The '<tt>llvm.log10.*</tt>' intrinsics perform the log10 function.</p>
7726 <p>The argument and return value are floating point numbers of the same
7730 <p>This function returns the same values as the libm <tt>log10</tt> functions
7731 would, and handles error conditions in the same way.</p>
7735 <!-- _______________________________________________________________________ -->
7737 <a name="int_log2">'<tt>llvm.log2.*</tt>' Intrinsic</a>
7743 <p>This is an overloaded intrinsic. You can use <tt>llvm.log2</tt> on any
7744 floating point or vector of floating point type. Not all targets support all
7748 declare float @llvm.log2.f32(float %Val)
7749 declare double @llvm.log2.f64(double %Val)
7750 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7751 declare fp128 @llvm.log2.f128(fp128 %Val)
7752 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7756 <p>The '<tt>llvm.log2.*</tt>' intrinsics perform the log2 function.</p>
7759 <p>The argument and return value are floating point numbers of the same
7763 <p>This function returns the same values as the libm <tt>log2</tt> functions
7764 would, and handles error conditions in the same way.</p>
7768 <!-- _______________________________________________________________________ -->
7770 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7776 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7777 floating point or vector of floating point type. Not all targets support all
7781 declare float @llvm.fma.f32(float %a, float %b, float %c)
7782 declare double @llvm.fma.f64(double %a, double %b, double %c)
7783 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7784 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7785 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7789 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7793 <p>The argument and return value are floating point numbers of the same
7797 <p>This function returns the same values as the libm <tt>fma</tt> functions
7802 <!-- _______________________________________________________________________ -->
7804 <a name="int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a>
7810 <p>This is an overloaded intrinsic. You can use <tt>llvm.fabs</tt> on any
7811 floating point or vector of floating point type. Not all targets support all
7815 declare float @llvm.fabs.f32(float %Val)
7816 declare double @llvm.fabs.f64(double %Val)
7817 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7818 declare fp128 @llvm.fabs.f128(fp128 %Val)
7819 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7823 <p>The '<tt>llvm.fabs.*</tt>' intrinsics return the absolute value of
7827 <p>The argument and return value are floating point numbers of the same
7831 <p>This function returns the same values as the libm <tt>fabs</tt> functions
7832 would, and handles error conditions in the same way.</p>
7836 <!-- _______________________________________________________________________ -->
7838 <a name="int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a>
7844 <p>This is an overloaded intrinsic. You can use <tt>llvm.floor</tt> on any
7845 floating point or vector of floating point type. Not all targets support all
7849 declare float @llvm.floor.f32(float %Val)
7850 declare double @llvm.floor.f64(double %Val)
7851 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7852 declare fp128 @llvm.floor.f128(fp128 %Val)
7853 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7857 <p>The '<tt>llvm.floor.*</tt>' intrinsics return the floor of
7861 <p>The argument and return value are floating point numbers of the same
7865 <p>This function returns the same values as the libm <tt>floor</tt> functions
7866 would, and handles error conditions in the same way.</p>
7870 <!-- _______________________________________________________________________ -->
7872 <a name="int_ceil">'<tt>llvm.ceil.*</tt>' Intrinsic</a>
7878 <p>This is an overloaded intrinsic. You can use <tt>llvm.ceil</tt> on any
7879 floating point or vector of floating point type. Not all targets support all
7883 declare float @llvm.ceil.f32(float %Val)
7884 declare double @llvm.ceil.f64(double %Val)
7885 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7886 declare fp128 @llvm.ceil.f128(fp128 %Val)
7887 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7891 <p>The '<tt>llvm.ceil.*</tt>' intrinsics return the ceiling of
7895 <p>The argument and return value are floating point numbers of the same
7899 <p>This function returns the same values as the libm <tt>ceil</tt> functions
7900 would, and handles error conditions in the same way.</p>
7904 <!-- _______________________________________________________________________ -->
7906 <a name="int_trunc">'<tt>llvm.trunc.*</tt>' Intrinsic</a>
7912 <p>This is an overloaded intrinsic. You can use <tt>llvm.trunc</tt> on any
7913 floating point or vector of floating point type. Not all targets support all
7917 declare float @llvm.trunc.f32(float %Val)
7918 declare double @llvm.trunc.f64(double %Val)
7919 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7920 declare fp128 @llvm.trunc.f128(fp128 %Val)
7921 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7925 <p>The '<tt>llvm.trunc.*</tt>' intrinsics returns the operand rounded to the
7926 nearest integer not larger in magnitude than the operand.</p>
7929 <p>The argument and return value are floating point numbers of the same
7933 <p>This function returns the same values as the libm <tt>trunc</tt> functions
7934 would, and handles error conditions in the same way.</p>
7938 <!-- _______________________________________________________________________ -->
7940 <a name="int_rint">'<tt>llvm.rint.*</tt>' Intrinsic</a>
7946 <p>This is an overloaded intrinsic. You can use <tt>llvm.rint</tt> on any
7947 floating point or vector of floating point type. Not all targets support all
7951 declare float @llvm.rint.f32(float %Val)
7952 declare double @llvm.rint.f64(double %Val)
7953 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7954 declare fp128 @llvm.rint.f128(fp128 %Val)
7955 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7959 <p>The '<tt>llvm.rint.*</tt>' intrinsics returns the operand rounded to the
7960 nearest integer. It may raise an inexact floating-point exception if the
7961 operand isn't an integer.</p>
7964 <p>The argument and return value are floating point numbers of the same
7968 <p>This function returns the same values as the libm <tt>rint</tt> functions
7969 would, and handles error conditions in the same way.</p>
7973 <!-- _______________________________________________________________________ -->
7975 <a name="int_nearbyint">'<tt>llvm.nearbyint.*</tt>' Intrinsic</a>
7981 <p>This is an overloaded intrinsic. You can use <tt>llvm.nearbyint</tt> on any
7982 floating point or vector of floating point type. Not all targets support all
7986 declare float @llvm.nearbyint.f32(float %Val)
7987 declare double @llvm.nearbyint.f64(double %Val)
7988 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7989 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7990 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7994 <p>The '<tt>llvm.nearbyint.*</tt>' intrinsics returns the operand rounded to the
7995 nearest integer.</p>
7998 <p>The argument and return value are floating point numbers of the same
8002 <p>This function returns the same values as the libm <tt>nearbyint</tt>
8003 functions would, and handles error conditions in the same way.</p>
8009 <!-- ======================================================================= -->
8011 <a name="int_manip">Bit Manipulation Intrinsics</a>
8016 <p>LLVM provides intrinsics for a few important bit manipulation operations.
8017 These allow efficient code generation for some algorithms.</p>
8019 <!-- _______________________________________________________________________ -->
8021 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
8027 <p>This is an overloaded intrinsic function. You can use bswap on any integer
8028 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
8031 declare i16 @llvm.bswap.i16(i16 <id>)
8032 declare i32 @llvm.bswap.i32(i32 <id>)
8033 declare i64 @llvm.bswap.i64(i64 <id>)
8037 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
8038 values with an even number of bytes (positive multiple of 16 bits). These
8039 are useful for performing operations on data that is not in the target's
8040 native byte order.</p>
8043 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
8044 and low byte of the input i16 swapped. Similarly,
8045 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
8046 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
8047 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
8048 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
8049 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
8050 more, respectively).</p>
8054 <!-- _______________________________________________________________________ -->
8056 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
8062 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
8063 width, or on any vector with integer elements. Not all targets support all
8064 bit widths or vector types, however.</p>
8067 declare i8 @llvm.ctpop.i8(i8 <src>)
8068 declare i16 @llvm.ctpop.i16(i16 <src>)
8069 declare i32 @llvm.ctpop.i32(i32 <src>)
8070 declare i64 @llvm.ctpop.i64(i64 <src>)
8071 declare i256 @llvm.ctpop.i256(i256 <src>)
8072 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
8076 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
8080 <p>The only argument is the value to be counted. The argument may be of any
8081 integer type, or a vector with integer elements.
8082 The return type must match the argument type.</p>
8085 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
8086 element of a vector.</p>
8090 <!-- _______________________________________________________________________ -->
8092 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
8098 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
8099 integer bit width, or any vector whose elements are integers. Not all
8100 targets support all bit widths or vector types, however.</p>
8103 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
8104 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
8105 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
8106 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
8107 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
8108 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8112 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
8113 leading zeros in a variable.</p>
8116 <p>The first argument is the value to be counted. This argument may be of any
8117 integer type, or a vectory with integer element type. The return type
8118 must match the first argument type.</p>
8120 <p>The second argument must be a constant and is a flag to indicate whether the
8121 intrinsic should ensure that a zero as the first argument produces a defined
8122 result. Historically some architectures did not provide a defined result for
8123 zero values as efficiently, and many algorithms are now predicated on
8124 avoiding zero-value inputs.</p>
8127 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
8128 zeros in a variable, or within each element of the vector.
8129 If <tt>src == 0</tt> then the result is the size in bits of the type of
8130 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
8131 For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
8135 <!-- _______________________________________________________________________ -->
8137 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
8143 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
8144 integer bit width, or any vector of integer elements. Not all targets
8145 support all bit widths or vector types, however.</p>
8148 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
8149 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
8150 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
8151 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
8152 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
8153 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8157 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
8161 <p>The first argument is the value to be counted. This argument may be of any
8162 integer type, or a vectory with integer element type. The return type
8163 must match the first argument type.</p>
8165 <p>The second argument must be a constant and is a flag to indicate whether the
8166 intrinsic should ensure that a zero as the first argument produces a defined
8167 result. Historically some architectures did not provide a defined result for
8168 zero values as efficiently, and many algorithms are now predicated on
8169 avoiding zero-value inputs.</p>
8172 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
8173 zeros in a variable, or within each element of a vector.
8174 If <tt>src == 0</tt> then the result is the size in bits of the type of
8175 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
8176 For example, <tt>llvm.cttz(2) = 1</tt>.</p>
8182 <!-- ======================================================================= -->
8184 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
8189 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
8191 <!-- _______________________________________________________________________ -->
8193 <a name="int_sadd_overflow">
8194 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
8201 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
8202 on any integer bit width.</p>
8205 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8206 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8207 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8211 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
8212 a signed addition of the two arguments, and indicate whether an overflow
8213 occurred during the signed summation.</p>
8216 <p>The arguments (%a and %b) and the first element of the result structure may
8217 be of integer types of any bit width, but they must have the same bit
8218 width. The second element of the result structure must be of
8219 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8220 undergo signed addition.</p>
8223 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
8224 a signed addition of the two variables. They return a structure — the
8225 first element of which is the signed summation, and the second element of
8226 which is a bit specifying if the signed summation resulted in an
8231 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8232 %sum = extractvalue {i32, i1} %res, 0
8233 %obit = extractvalue {i32, i1} %res, 1
8234 br i1 %obit, label %overflow, label %normal
8239 <!-- _______________________________________________________________________ -->
8241 <a name="int_uadd_overflow">
8242 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
8249 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
8250 on any integer bit width.</p>
8253 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8254 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8255 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8259 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
8260 an unsigned addition of the two arguments, and indicate whether a carry
8261 occurred during the unsigned summation.</p>
8264 <p>The arguments (%a and %b) and the first element of the result structure may
8265 be of integer types of any bit width, but they must have the same bit
8266 width. The second element of the result structure must be of
8267 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8268 undergo unsigned addition.</p>
8271 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
8272 an unsigned addition of the two arguments. They return a structure —
8273 the first element of which is the sum, and the second element of which is a
8274 bit specifying if the unsigned summation resulted in a carry.</p>
8278 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8279 %sum = extractvalue {i32, i1} %res, 0
8280 %obit = extractvalue {i32, i1} %res, 1
8281 br i1 %obit, label %carry, label %normal
8286 <!-- _______________________________________________________________________ -->
8288 <a name="int_ssub_overflow">
8289 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
8296 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
8297 on any integer bit width.</p>
8300 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8301 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8302 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8306 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
8307 a signed subtraction of the two arguments, and indicate whether an overflow
8308 occurred during the signed subtraction.</p>
8311 <p>The arguments (%a and %b) and the first element of the result structure may
8312 be of integer types of any bit width, but they must have the same bit
8313 width. The second element of the result structure must be of
8314 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8315 undergo signed subtraction.</p>
8318 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
8319 a signed subtraction of the two arguments. They return a structure —
8320 the first element of which is the subtraction, and the second element of
8321 which is a bit specifying if the signed subtraction resulted in an
8326 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8327 %sum = extractvalue {i32, i1} %res, 0
8328 %obit = extractvalue {i32, i1} %res, 1
8329 br i1 %obit, label %overflow, label %normal
8334 <!-- _______________________________________________________________________ -->
8336 <a name="int_usub_overflow">
8337 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
8344 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
8345 on any integer bit width.</p>
8348 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8349 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8350 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8354 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
8355 an unsigned subtraction of the two arguments, and indicate whether an
8356 overflow occurred during the unsigned subtraction.</p>
8359 <p>The arguments (%a and %b) and the first element of the result structure may
8360 be of integer types of any bit width, but they must have the same bit
8361 width. The second element of the result structure must be of
8362 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8363 undergo unsigned subtraction.</p>
8366 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
8367 an unsigned subtraction of the two arguments. They return a structure —
8368 the first element of which is the subtraction, and the second element of
8369 which is a bit specifying if the unsigned subtraction resulted in an
8374 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8375 %sum = extractvalue {i32, i1} %res, 0
8376 %obit = extractvalue {i32, i1} %res, 1
8377 br i1 %obit, label %overflow, label %normal
8382 <!-- _______________________________________________________________________ -->
8384 <a name="int_smul_overflow">
8385 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
8392 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
8393 on any integer bit width.</p>
8396 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8397 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8398 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8403 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
8404 a signed multiplication of the two arguments, and indicate whether an
8405 overflow occurred during the signed multiplication.</p>
8408 <p>The arguments (%a and %b) and the first element of the result structure may
8409 be of integer types of any bit width, but they must have the same bit
8410 width. The second element of the result structure must be of
8411 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8412 undergo signed multiplication.</p>
8415 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
8416 a signed multiplication of the two arguments. They return a structure —
8417 the first element of which is the multiplication, and the second element of
8418 which is a bit specifying if the signed multiplication resulted in an
8423 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8424 %sum = extractvalue {i32, i1} %res, 0
8425 %obit = extractvalue {i32, i1} %res, 1
8426 br i1 %obit, label %overflow, label %normal
8431 <!-- _______________________________________________________________________ -->
8433 <a name="int_umul_overflow">
8434 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
8441 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
8442 on any integer bit width.</p>
8445 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8446 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8447 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8451 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
8452 a unsigned multiplication of the two arguments, and indicate whether an
8453 overflow occurred during the unsigned multiplication.</p>
8456 <p>The arguments (%a and %b) and the first element of the result structure may
8457 be of integer types of any bit width, but they must have the same bit
8458 width. The second element of the result structure must be of
8459 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8460 undergo unsigned multiplication.</p>
8463 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
8464 an unsigned multiplication of the two arguments. They return a structure
8465 — the first element of which is the multiplication, and the second
8466 element of which is a bit specifying if the unsigned multiplication resulted
8471 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8472 %sum = extractvalue {i32, i1} %res, 0
8473 %obit = extractvalue {i32, i1} %res, 1
8474 br i1 %obit, label %overflow, label %normal
8481 <!-- ======================================================================= -->
8483 <a name="spec_arithmetic">Specialised Arithmetic Intrinsics</a>
8486 <!-- _______________________________________________________________________ -->
8489 <a name="fmuladd">'<tt>llvm.fmuladd.*</tt>' Intrinsic</a>
8496 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8497 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8501 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsic functions represent multiply-add
8502 expressions that can be fused if the code generator determines that the fused
8503 expression would be legal and efficient.</p>
8506 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsics each take three arguments: two
8507 multiplicands, a and b, and an addend c.</p>
8510 <p>The expression:</p>
8512 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8514 <p>is equivalent to the expression a * b + c, except that rounding will not be
8515 performed between the multiplication and addition steps if the code generator
8516 fuses the operations. Fusion is not guaranteed, even if the target platform
8517 supports it. If a fused multiply-add is required the corresponding llvm.fma.*
8518 intrinsic function should be used instead.</p>
8522 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8527 <!-- ======================================================================= -->
8529 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
8534 <p>For most target platforms, half precision floating point is a storage-only
8535 format. This means that it is
8536 a dense encoding (in memory) but does not support computation in the
8539 <p>This means that code must first load the half-precision floating point
8540 value as an i16, then convert it to float with <a
8541 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
8542 Computation can then be performed on the float value (including extending to
8543 double etc). To store the value back to memory, it is first converted to
8544 float if needed, then converted to i16 with
8545 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
8546 storing as an i16 value.</p>
8548 <!-- _______________________________________________________________________ -->
8550 <a name="int_convert_to_fp16">
8551 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
8559 declare i16 @llvm.convert.to.fp16(f32 %a)
8563 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8564 a conversion from single precision floating point format to half precision
8565 floating point format.</p>
8568 <p>The intrinsic function contains single argument - the value to be
8572 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8573 a conversion from single precision floating point format to half precision
8574 floating point format. The return value is an <tt>i16</tt> which
8575 contains the converted number.</p>
8579 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8580 store i16 %res, i16* @x, align 2
8585 <!-- _______________________________________________________________________ -->
8587 <a name="int_convert_from_fp16">
8588 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
8596 declare f32 @llvm.convert.from.fp16(i16 %a)
8600 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
8601 a conversion from half precision floating point format to single precision
8602 floating point format.</p>
8605 <p>The intrinsic function contains single argument - the value to be
8609 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
8610 conversion from half single precision floating point format to single
8611 precision floating point format. The input half-float value is represented by
8612 an <tt>i16</tt> value.</p>
8616 %a = load i16* @x, align 2
8617 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8624 <!-- ======================================================================= -->
8626 <a name="int_debugger">Debugger Intrinsics</a>
8631 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
8632 prefix), are described in
8633 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
8634 Level Debugging</a> document.</p>
8638 <!-- ======================================================================= -->
8640 <a name="int_eh">Exception Handling Intrinsics</a>
8645 <p>The LLVM exception handling intrinsics (which all start with
8646 <tt>llvm.eh.</tt> prefix), are described in
8647 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
8648 Handling</a> document.</p>
8652 <!-- ======================================================================= -->
8654 <a name="int_trampoline">Trampoline Intrinsics</a>
8659 <p>These intrinsics make it possible to excise one parameter, marked with
8660 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
8661 The result is a callable
8662 function pointer lacking the nest parameter - the caller does not need to
8663 provide a value for it. Instead, the value to use is stored in advance in a
8664 "trampoline", a block of memory usually allocated on the stack, which also
8665 contains code to splice the nest value into the argument list. This is used
8666 to implement the GCC nested function address extension.</p>
8668 <p>For example, if the function is
8669 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
8670 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
8673 <pre class="doc_code">
8674 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8675 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8676 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8677 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8678 %fp = bitcast i8* %p to i32 (i32, i32)*
8681 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
8682 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
8684 <!-- _______________________________________________________________________ -->
8687 '<tt>llvm.init.trampoline</tt>' Intrinsic
8695 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8699 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
8700 turning it into a trampoline.</p>
8703 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
8704 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
8705 sufficiently aligned block of memory; this memory is written to by the
8706 intrinsic. Note that the size and the alignment are target-specific - LLVM
8707 currently provides no portable way of determining them, so a front-end that
8708 generates this intrinsic needs to have some target-specific knowledge.
8709 The <tt>func</tt> argument must hold a function bitcast to
8710 an <tt>i8*</tt>.</p>
8713 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
8714 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
8715 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
8716 which can be <a href="#int_trampoline">bitcast (to a new function) and
8717 called</a>. The new function's signature is the same as that of
8718 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
8719 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
8720 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
8721 with the same argument list, but with <tt>nval</tt> used for the missing
8722 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
8723 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
8724 to the returned function pointer is undefined.</p>
8727 <!-- _______________________________________________________________________ -->
8730 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
8738 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8742 <p>This performs any required machine-specific adjustment to the address of a
8743 trampoline (passed as <tt>tramp</tt>).</p>
8746 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
8747 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
8751 <p>On some architectures the address of the code to be executed needs to be
8752 different to the address where the trampoline is actually stored. This
8753 intrinsic returns the executable address corresponding to <tt>tramp</tt>
8754 after performing the required machine specific adjustments.
8755 The pointer returned can then be <a href="#int_trampoline"> bitcast and
8763 <!-- ======================================================================= -->
8765 <a name="int_memorymarkers">Memory Use Markers</a>
8770 <p>This class of intrinsics exists to information about the lifetime of memory
8771 objects and ranges where variables are immutable.</p>
8773 <!-- _______________________________________________________________________ -->
8775 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8782 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8786 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8787 object's lifetime.</p>
8790 <p>The first argument is a constant integer representing the size of the
8791 object, or -1 if it is variable sized. The second argument is a pointer to
8795 <p>This intrinsic indicates that before this point in the code, the value of the
8796 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8797 never be used and has an undefined value. A load from the pointer that
8798 precedes this intrinsic can be replaced with
8799 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8803 <!-- _______________________________________________________________________ -->
8805 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8812 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8816 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8817 object's lifetime.</p>
8820 <p>The first argument is a constant integer representing the size of the
8821 object, or -1 if it is variable sized. The second argument is a pointer to
8825 <p>This intrinsic indicates that after this point in the code, the value of the
8826 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8827 never be used and has an undefined value. Any stores into the memory object
8828 following this intrinsic may be removed as dead.
8832 <!-- _______________________________________________________________________ -->
8834 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8841 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8845 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8846 a memory object will not change.</p>
8849 <p>The first argument is a constant integer representing the size of the
8850 object, or -1 if it is variable sized. The second argument is a pointer to
8854 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8855 the return value, the referenced memory location is constant and
8860 <!-- _______________________________________________________________________ -->
8862 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8869 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8873 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8874 a memory object are mutable.</p>
8877 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8878 The second argument is a constant integer representing the size of the
8879 object, or -1 if it is variable sized and the third argument is a pointer
8883 <p>This intrinsic indicates that the memory is mutable again.</p>
8889 <!-- ======================================================================= -->
8891 <a name="int_general">General Intrinsics</a>
8896 <p>This class of intrinsics is designed to be generic and has no specific
8899 <!-- _______________________________________________________________________ -->
8901 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8908 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8912 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8915 <p>The first argument is a pointer to a value, the second is a pointer to a
8916 global string, the third is a pointer to a global string which is the source
8917 file name, and the last argument is the line number.</p>
8920 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8921 This can be useful for special purpose optimizations that want to look for
8922 these annotations. These have no other defined use; they are ignored by code
8923 generation and optimization.</p>
8927 <!-- _______________________________________________________________________ -->
8929 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8935 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8936 any integer bit width.</p>
8939 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8940 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8941 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8942 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8943 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8947 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8950 <p>The first argument is an integer value (result of some expression), the
8951 second is a pointer to a global string, the third is a pointer to a global
8952 string which is the source file name, and the last argument is the line
8953 number. It returns the value of the first argument.</p>
8956 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8957 arbitrary strings. This can be useful for special purpose optimizations that
8958 want to look for these annotations. These have no other defined use; they
8959 are ignored by code generation and optimization.</p>
8963 <!-- _______________________________________________________________________ -->
8965 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8972 declare void @llvm.trap() noreturn nounwind
8976 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8982 <p>This intrinsic is lowered to the target dependent trap instruction. If the
8983 target does not have a trap instruction, this intrinsic will be lowered to
8984 a call of the <tt>abort()</tt> function.</p>
8988 <!-- _______________________________________________________________________ -->
8990 <a name="int_debugtrap">'<tt>llvm.debugtrap</tt>' Intrinsic</a>
8997 declare void @llvm.debugtrap() nounwind
9001 <p>The '<tt>llvm.debugtrap</tt>' intrinsic.</p>
9007 <p>This intrinsic is lowered to code which is intended to cause an execution
9008 trap with the intention of requesting the attention of a debugger.</p>
9012 <!-- _______________________________________________________________________ -->
9014 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
9021 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
9025 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
9026 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
9027 ensure that it is placed on the stack before local variables.</p>
9030 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
9031 arguments. The first argument is the value loaded from the stack
9032 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
9033 that has enough space to hold the value of the guard.</p>
9036 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
9037 the <tt>AllocaInst</tt> stack slot to be before local variables on the
9038 stack. This is to ensure that if a local variable on the stack is
9039 overwritten, it will destroy the value of the guard. When the function exits,
9040 the guard on the stack is checked against the original guard. If they are
9041 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
9046 <!-- _______________________________________________________________________ -->
9048 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
9055 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
9056 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
9060 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
9061 the optimizers to determine at compile time whether a) an operation (like
9062 memcpy) will overflow a buffer that corresponds to an object, or b) that a
9063 runtime check for overflow isn't necessary. An object in this context means
9064 an allocation of a specific class, structure, array, or other object.</p>
9067 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
9068 argument is a pointer to or into the <tt>object</tt>. The second argument
9069 is a boolean and determines whether <tt>llvm.objectsize</tt> returns 0 (if
9070 true) or -1 (if false) when the object size is unknown.
9071 The second argument only accepts constants.</p>
9074 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to a constant representing
9075 the size of the object concerned. If the size cannot be determined at compile
9076 time, <tt>llvm.objectsize</tt> returns <tt>i32/i64 -1 or 0</tt>
9077 (depending on the <tt>min</tt> argument).</p>
9080 <!-- _______________________________________________________________________ -->
9082 <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a>
9089 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
9090 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
9094 <p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the
9095 most probable) value of <tt>val</tt>, which can be used by optimizers.</p>
9098 <p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first
9099 argument is a value. The second argument is an expected value, this needs to
9100 be a constant value, variables are not allowed.</p>
9103 <p>This intrinsic is lowered to the <tt>val</tt>.</p>
9106 <!-- _______________________________________________________________________ -->
9108 <a name="int_donothing">'<tt>llvm.donothing</tt>' Intrinsic</a>
9115 declare void @llvm.donothing() nounwind readnone
9119 <p>The <tt>llvm.donothing</tt> intrinsic doesn't perform any operation. It's the
9120 only intrinsic that can be called with an invoke instruction.</p>
9126 <p>This intrinsic does nothing, and it's removed by optimizers and ignored by
9133 <!-- *********************************************************************** -->
9136 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
9137 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
9138 <a href="http://validator.w3.org/check/referer"><img
9139 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
9141 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
9142 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
9143 Last modified: $Date$