1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" 2 "http://www.w3.org/TR/html4/strict.dtd"> 3<html> 4<head> 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"> 11</head> 12 13<body> 14 15<h1>LLVM Language Reference Manual</h1> 16<ol> 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> 21 <ol> 22 <li><a href="#modulestructure">Module Structure</a></li> 23 <li><a href="#linkage">Linkage Types</a> 24 <ol> 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> 41 </ol> 42 </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> 58 </ol> 59 </li> 60 <li><a href="#typesystem">Type System</a> 61 <ol> 62 <li><a href="#t_classifications">Type Classifications</a></li> 63 <li><a href="#t_primitive">Primitive Types</a> 64 <ol> 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> 71 </ol> 72 </li> 73 <li><a href="#t_derived">Derived Types</a> 74 <ol> 75 <li><a href="#t_aggregate">Aggregate Types</a> 76 <ol> 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> 81 </ol> 82 </li> 83 <li><a href="#t_function">Function Type</a></li> 84 <li><a href="#t_pointer">Pointer Type</a></li> 85 </ol> 86 </li> 87 </ol> 88 </li> 89 <li><a href="#constants">Constants</a> 90 <ol> 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> 98 </ol> 99 </li> 100 <li><a href="#othervalues">Other Values</a> 101 <ol> 102 <li><a href="#inlineasm">Inline Assembler Expressions</a></li> 103 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a> 104 <ol> 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> 109 </ol> 110 </li> 111 </ol> 112 </li> 113 <li><a href="#module_flags">Module Flags Metadata</a> 114 <ol> 115 <li><a href="#objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a></li> 116 </ol> 117 </li> 118 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a> 119 <ol> 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> 127 </ol> 128 </li> 129 <li><a href="#instref">Instruction Reference</a> 130 <ol> 131 <li><a href="#terminators">Terminator Instructions</a> 132 <ol> 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> 140 </ol> 141 </li> 142 <li><a href="#binaryops">Binary Operations</a> 143 <ol> 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> 156 </ol> 157 </li> 158 <li><a href="#bitwiseops">Bitwise Binary Operations</a> 159 <ol> 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> 166 </ol> 167 </li> 168 <li><a href="#vectorops">Vector Operations</a> 169 <ol> 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> 173 </ol> 174 </li> 175 <li><a href="#aggregateops">Aggregate Operations</a> 176 <ol> 177 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li> 178 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li> 179 </ol> 180 </li> 181 <li><a href="#memoryops">Memory Access and Addressing Operations</a> 182 <ol> 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> 190 </ol> 191 </li> 192 <li><a href="#convertops">Conversion Operations</a> 193 <ol> 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> 206 </ol> 207 </li> 208 <li><a href="#otherops">Other Operations</a> 209 <ol> 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> 217 </ol> 218 </li> 219 </ol> 220 </li> 221 <li><a href="#intrinsics">Intrinsic Functions</a> 222 <ol> 223 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a> 224 <ol> 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> 228 </ol> 229 </li> 230 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a> 231 <ol> 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> 235 </ol> 236 </li> 237 <li><a href="#int_codegen">Code Generator Intrinsics</a> 238 <ol> 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> 246 </ol> 247 </li> 248 <li><a href="#int_libc">Standard C Library Intrinsics</a> 249 <ol> 250 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li> 251 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li> 252 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li> 253 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li> 254 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li> 255 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li> 256 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li> 257 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li> 258 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li> 259 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li> 260 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li> 261 <li><a href="#int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a></li> 262 <li><a href="#int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a></li> 263 </ol> 264 </li> 265 <li><a href="#int_manip">Bit Manipulation Intrinsics</a> 266 <ol> 267 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li> 268 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li> 269 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li> 270 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li> 271 </ol> 272 </li> 273 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a> 274 <ol> 275 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li> 276 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li> 277 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li> 278 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li> 279 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li> 280 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li> 281 </ol> 282 </li> 283 <li><a href="#spec_arithmetic">Specialised Arithmetic Intrinsics</a> 284 <ol> 285 <li><a href="#fmuladd">'<tt>llvm.fmuladd</tt> Intrinsic</a></li> 286 </ol> 287 </li> 288 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a> 289 <ol> 290 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li> 291 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li> 292 </ol> 293 </li> 294 <li><a href="#int_debugger">Debugger intrinsics</a></li> 295 <li><a href="#int_eh">Exception Handling intrinsics</a></li> 296 <li><a href="#int_trampoline">Trampoline Intrinsics</a> 297 <ol> 298 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li> 299 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li> 300 </ol> 301 </li> 302 <li><a href="#int_memorymarkers">Memory Use Markers</a> 303 <ol> 304 <li><a href="#int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a></li> 305 <li><a href="#int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a></li> 306 <li><a href="#int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a></li> 307 <li><a href="#int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a></li> 308 </ol> 309 </li> 310 <li><a href="#int_general">General intrinsics</a> 311 <ol> 312 <li><a href="#int_var_annotation"> 313 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li> 314 <li><a href="#int_annotation"> 315 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li> 316 <li><a href="#int_trap"> 317 '<tt>llvm.trap</tt>' Intrinsic</a></li> 318 <li><a href="#int_debugtrap"> 319 '<tt>llvm.debugtrap</tt>' Intrinsic</a></li> 320 <li><a href="#int_stackprotector"> 321 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li> 322 <li><a href="#int_objectsize"> 323 '<tt>llvm.objectsize</tt>' Intrinsic</a></li> 324 <li><a href="#int_expect"> 325 '<tt>llvm.expect</tt>' Intrinsic</a></li> 326 <li><a href="#int_donothing"> 327 '<tt>llvm.donothing</tt>' Intrinsic</a></li> 328 </ol> 329 </li> 330 </ol> 331 </li> 332</ol> 333 334<div class="doc_author"> 335 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a> 336 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p> 337</div> 338 339<!-- *********************************************************************** --> 340<h2><a name="abstract">Abstract</a></h2> 341<!-- *********************************************************************** --> 342 343<div> 344 345<p>This document is a reference manual for the LLVM assembly language. LLVM is 346 a Static Single Assignment (SSA) based representation that provides type 347 safety, low-level operations, flexibility, and the capability of representing 348 'all' high-level languages cleanly. It is the common code representation 349 used throughout all phases of the LLVM compilation strategy.</p> 350 351</div> 352 353<!-- *********************************************************************** --> 354<h2><a name="introduction">Introduction</a></h2> 355<!-- *********************************************************************** --> 356 357<div> 358 359<p>The LLVM code representation is designed to be used in three different forms: 360 as an in-memory compiler IR, as an on-disk bitcode representation (suitable 361 for fast loading by a Just-In-Time compiler), and as a human readable 362 assembly language representation. This allows LLVM to provide a powerful 363 intermediate representation for efficient compiler transformations and 364 analysis, while providing a natural means to debug and visualize the 365 transformations. The three different forms of LLVM are all equivalent. This 366 document describes the human readable representation and notation.</p> 367 368<p>The LLVM representation aims to be light-weight and low-level while being 369 expressive, typed, and extensible at the same time. It aims to be a 370 "universal IR" of sorts, by being at a low enough level that high-level ideas 371 may be cleanly mapped to it (similar to how microprocessors are "universal 372 IR's", allowing many source languages to be mapped to them). By providing 373 type information, LLVM can be used as the target of optimizations: for 374 example, through pointer analysis, it can be proven that a C automatic 375 variable is never accessed outside of the current function, allowing it to 376 be promoted to a simple SSA value instead of a memory location.</p> 377 378<!-- _______________________________________________________________________ --> 379<h4> 380 <a name="wellformed">Well-Formedness</a> 381</h4> 382 383<div> 384 385<p>It is important to note that this document describes 'well formed' LLVM 386 assembly language. There is a difference between what the parser accepts and 387 what is considered 'well formed'. For example, the following instruction is 388 syntactically okay, but not well formed:</p> 389 390<pre class="doc_code"> 391%x = <a href="#i_add">add</a> i32 1, %x 392</pre> 393 394<p>because the definition of <tt>%x</tt> does not dominate all of its uses. The 395 LLVM infrastructure provides a verification pass that may be used to verify 396 that an LLVM module is well formed. This pass is automatically run by the 397 parser after parsing input assembly and by the optimizer before it outputs 398 bitcode. The violations pointed out by the verifier pass indicate bugs in 399 transformation passes or input to the parser.</p> 400 401</div> 402 403</div> 404 405<!-- Describe the typesetting conventions here. --> 406 407<!-- *********************************************************************** --> 408<h2><a name="identifiers">Identifiers</a></h2> 409<!-- *********************************************************************** --> 410 411<div> 412 413<p>LLVM identifiers come in two basic types: global and local. Global 414 identifiers (functions, global variables) begin with the <tt>'@'</tt> 415 character. Local identifiers (register names, types) begin with 416 the <tt>'%'</tt> character. Additionally, there are three different formats 417 for identifiers, for different purposes:</p> 418 419<ol> 420 <li>Named values are represented as a string of characters with their prefix. 421 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>, 422 <tt>%a.really.long.identifier</tt>. The actual regular expression used is 423 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require 424 other characters in their names can be surrounded with quotes. Special 425 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the 426 ASCII code for the character in hexadecimal. In this way, any character 427 can be used in a name value, even quotes themselves.</li> 428 429 <li>Unnamed values are represented as an unsigned numeric value with their 430 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li> 431 432 <li>Constants, which are described in a <a href="#constants">section about 433 constants</a>, below.</li> 434</ol> 435 436<p>LLVM requires that values start with a prefix for two reasons: Compilers 437 don't need to worry about name clashes with reserved words, and the set of 438 reserved words may be expanded in the future without penalty. Additionally, 439 unnamed identifiers allow a compiler to quickly come up with a temporary 440 variable without having to avoid symbol table conflicts.</p> 441 442<p>Reserved words in LLVM are very similar to reserved words in other 443 languages. There are keywords for different opcodes 444 ('<tt><a href="#i_add">add</a></tt>', 445 '<tt><a href="#i_bitcast">bitcast</a></tt>', 446 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names 447 ('<tt><a href="#t_void">void</a></tt>', 448 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These 449 reserved words cannot conflict with variable names, because none of them 450 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p> 451 452<p>Here is an example of LLVM code to multiply the integer variable 453 '<tt>%X</tt>' by 8:</p> 454 455<p>The easy way:</p> 456 457<pre class="doc_code"> 458%result = <a href="#i_mul">mul</a> i32 %X, 8 459</pre> 460 461<p>After strength reduction:</p> 462 463<pre class="doc_code"> 464%result = <a href="#i_shl">shl</a> i32 %X, i8 3 465</pre> 466 467<p>And the hard way:</p> 468 469<pre class="doc_code"> 470%0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i> 471%1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i> 472%result = <a href="#i_add">add</a> i32 %1, %1 473</pre> 474 475<p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important 476 lexical features of LLVM:</p> 477 478<ol> 479 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of 480 line.</li> 481 482 <li>Unnamed temporaries are created when the result of a computation is not 483 assigned to a named value.</li> 484 485 <li>Unnamed temporaries are numbered sequentially</li> 486</ol> 487 488<p>It also shows a convention that we follow in this document. When 489 demonstrating instructions, we will follow an instruction with a comment that 490 defines the type and name of value produced. Comments are shown in italic 491 text.</p> 492 493</div> 494 495<!-- *********************************************************************** --> 496<h2><a name="highlevel">High Level Structure</a></h2> 497<!-- *********************************************************************** --> 498<div> 499<!-- ======================================================================= --> 500<h3> 501 <a name="modulestructure">Module Structure</a> 502</h3> 503 504<div> 505 506<p>LLVM programs are composed of <tt>Module</tt>s, each of which is a 507 translation unit of the input programs. Each module consists of functions, 508 global variables, and symbol table entries. Modules may be combined together 509 with the LLVM linker, which merges function (and global variable) 510 definitions, resolves forward declarations, and merges symbol table 511 entries. Here is an example of the "hello world" module:</p> 512 513<pre class="doc_code"> 514<i>; Declare the string constant as a global constant.</i> 515<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" 516 517<i>; External declaration of the puts function</i> 518<a href="#functionstructure">declare</a> i32 @puts(i8* <a href="#nocapture">nocapture</a>) <a href="#fnattrs">nounwind</a> 519 520<i>; Definition of main function</i> 521define i32 @main() { <i>; i32()* </i> 522 <i>; Convert [13 x i8]* to i8 *...</i> 523 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.str, i64 0, i64 0 524 525 <i>; Call puts function to write out the string to stdout.</i> 526 <a href="#i_call">call</a> i32 @puts(i8* %cast210) 527 <a href="#i_ret">ret</a> i32 0 528} 529 530<i>; Named metadata</i> 531!1 = metadata !{i32 42} 532!foo = !{!1, null} 533</pre> 534 535<p>This example is made up of a <a href="#globalvars">global variable</a> named 536 "<tt>.str</tt>", an external declaration of the "<tt>puts</tt>" function, 537 a <a href="#functionstructure">function definition</a> for 538 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a> 539 "<tt>foo</tt>".</p> 540 541<p>In general, a module is made up of a list of global values (where both 542 functions and global variables are global values). Global values are 543 represented by a pointer to a memory location (in this case, a pointer to an 544 array of char, and a pointer to a function), and have one of the 545 following <a href="#linkage">linkage types</a>.</p> 546 547</div> 548 549<!-- ======================================================================= --> 550<h3> 551 <a name="linkage">Linkage Types</a> 552</h3> 553 554<div> 555 556<p>All Global Variables and Functions have one of the following types of 557 linkage:</p> 558 559<dl> 560 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt> 561 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible 562 by objects in the current module. In particular, linking code into a 563 module with an private global value may cause the private to be renamed as 564 necessary to avoid collisions. Because the symbol is private to the 565 module, all references can be updated. This doesn't show up in any symbol 566 table in the object file.</dd> 567 568 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt> 569 <dd>Similar to <tt>private</tt>, but the symbol is passed through the 570 assembler and evaluated by the linker. Unlike normal strong symbols, they 571 are removed by the linker from the final linked image (executable or 572 dynamic library).</dd> 573 574 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt> 575 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that 576 <tt>linker_private_weak</tt> symbols are subject to coalescing by the 577 linker. The symbols are removed by the linker from the final linked image 578 (executable or dynamic library).</dd> 579 580 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt> 581 <dd>Similar to private, but the value shows as a local symbol 582 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This 583 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd> 584 585 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt> 586 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted 587 into the object file corresponding to the LLVM module. They exist to 588 allow inlining and other optimizations to take place given knowledge of 589 the definition of the global, which is known to be somewhere outside the 590 module. Globals with <tt>available_externally</tt> linkage are allowed to 591 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>. 592 This linkage type is only allowed on definitions, not declarations.</dd> 593 594 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt> 595 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of 596 the same name when linkage occurs. This can be used to implement 597 some forms of inline functions, templates, or other code which must be 598 generated in each translation unit that uses it, but where the body may 599 be overridden with a more definitive definition later. Unreferenced 600 <tt>linkonce</tt> globals are allowed to be discarded. Note that 601 <tt>linkonce</tt> linkage does not actually allow the optimizer to 602 inline the body of this function into callers because it doesn't know if 603 this definition of the function is the definitive definition within the 604 program or whether it will be overridden by a stronger definition. 605 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>" 606 linkage.</dd> 607 608 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt> 609 <dd>"<tt>weak</tt>" linkage has the same merging semantics as 610 <tt>linkonce</tt> linkage, except that unreferenced globals with 611 <tt>weak</tt> linkage may not be discarded. This is used for globals that 612 are declared "weak" in C source code.</dd> 613 614 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt> 615 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but 616 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at 617 global scope. 618 Symbols with "<tt>common</tt>" linkage are merged in the same way as 619 <tt>weak symbols</tt>, and they may not be deleted if unreferenced. 620 <tt>common</tt> symbols may not have an explicit section, 621 must have a zero initializer, and may not be marked '<a 622 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not 623 have common linkage.</dd> 624 625 626 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt> 627 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of 628 pointer to array type. When two global variables with appending linkage 629 are linked together, the two global arrays are appended together. This is 630 the LLVM, typesafe, equivalent of having the system linker append together 631 "sections" with identical names when .o files are linked.</dd> 632 633 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt> 634 <dd>The semantics of this linkage follow the ELF object file model: the symbol 635 is weak until linked, if not linked, the symbol becomes null instead of 636 being an undefined reference.</dd> 637 638 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt> 639 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt> 640 <dd>Some languages allow differing globals to be merged, such as two functions 641 with different semantics. Other languages, such as <tt>C++</tt>, ensure 642 that only equivalent globals are ever merged (the "one definition rule" 643 — "ODR"). Such languages can use the <tt>linkonce_odr</tt> 644 and <tt>weak_odr</tt> linkage types to indicate that the global will only 645 be merged with equivalent globals. These linkage types are otherwise the 646 same as their non-<tt>odr</tt> versions.</dd> 647 648 <dt><tt><b><a name="linkage_linkonce_odr_auto_hide">linkonce_odr_auto_hide</a></b></tt></dt> 649 <dd>Similar to "<tt>linkonce_odr</tt>", but nothing in the translation unit 650 takes the address of this definition. For instance, functions that had an 651 inline definition, but the compiler decided not to inline it. 652 <tt>linkonce_odr_auto_hide</tt> may have only <tt>default</tt> visibility. 653 The symbols are removed by the linker from the final linked image 654 (executable or dynamic library).</dd> 655 656 <dt><tt><b><a name="linkage_external">external</a></b></tt></dt> 657 <dd>If none of the above identifiers are used, the global is externally 658 visible, meaning that it participates in linkage and can be used to 659 resolve external symbol references.</dd> 660</dl> 661 662<p>The next two types of linkage are targeted for Microsoft Windows platform 663 only. They are designed to support importing (exporting) symbols from (to) 664 DLLs (Dynamic Link Libraries).</p> 665 666<dl> 667 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt> 668 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function 669 or variable via a global pointer to a pointer that is set up by the DLL 670 exporting the symbol. On Microsoft Windows targets, the pointer name is 671 formed by combining <code>__imp_</code> and the function or variable 672 name.</dd> 673 674 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt> 675 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global 676 pointer to a pointer in a DLL, so that it can be referenced with the 677 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer 678 name is formed by combining <code>__imp_</code> and the function or 679 variable name.</dd> 680</dl> 681 682<p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if 683 another module defined a "<tt>.LC0</tt>" variable and was linked with this 684 one, one of the two would be renamed, preventing a collision. Since 685 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage 686 declarations), they are accessible outside of the current module.</p> 687 688<p>It is illegal for a function <i>declaration</i> to have any linkage type 689 other than <tt>external</tt>, <tt>dllimport</tt> 690 or <tt>extern_weak</tt>.</p> 691 692<p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt> 693 or <tt>weak_odr</tt> linkages.</p> 694 695</div> 696 697<!-- ======================================================================= --> 698<h3> 699 <a name="callingconv">Calling Conventions</a> 700</h3> 701 702<div> 703 704<p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a> 705 and <a href="#i_invoke">invokes</a> can all have an optional calling 706 convention specified for the call. The calling convention of any pair of 707 dynamic caller/callee must match, or the behavior of the program is 708 undefined. The following calling conventions are supported by LLVM, and more 709 may be added in the future:</p> 710 711<dl> 712 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt> 713 <dd>This calling convention (the default if no other calling convention is 714 specified) matches the target C calling conventions. This calling 715 convention supports varargs function calls and tolerates some mismatch in 716 the declared prototype and implemented declaration of the function (as 717 does normal C).</dd> 718 719 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt> 720 <dd>This calling convention attempts to make calls as fast as possible 721 (e.g. by passing things in registers). This calling convention allows the 722 target to use whatever tricks it wants to produce fast code for the 723 target, without having to conform to an externally specified ABI 724 (Application Binary Interface). 725 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized 726 when this or the GHC convention is used.</a> This calling convention 727 does not support varargs and requires the prototype of all callees to 728 exactly match the prototype of the function definition.</dd> 729 730 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt> 731 <dd>This calling convention attempts to make code in the caller as efficient 732 as possible under the assumption that the call is not commonly executed. 733 As such, these calls often preserve all registers so that the call does 734 not break any live ranges in the caller side. This calling convention 735 does not support varargs and requires the prototype of all callees to 736 exactly match the prototype of the function definition.</dd> 737 738 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt> 739 <dd>This calling convention has been implemented specifically for use by the 740 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>. 741 It passes everything in registers, going to extremes to achieve this by 742 disabling callee save registers. This calling convention should not be 743 used lightly but only for specific situations such as an alternative to 744 the <em>register pinning</em> performance technique often used when 745 implementing functional programming languages.At the moment only X86 746 supports this convention and it has the following limitations: 747 <ul> 748 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No 749 floating point types are supported.</li> 750 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and 751 6 floating point parameters.</li> 752 </ul> 753 This calling convention supports 754 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but 755 requires both the caller and callee are using it. 756 </dd> 757 758 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt> 759 <dd>Any calling convention may be specified by number, allowing 760 target-specific calling conventions to be used. Target specific calling 761 conventions start at 64.</dd> 762</dl> 763 764<p>More calling conventions can be added/defined on an as-needed basis, to 765 support Pascal conventions or any other well-known target-independent 766 convention.</p> 767 768</div> 769 770<!-- ======================================================================= --> 771<h3> 772 <a name="visibility">Visibility Styles</a> 773</h3> 774 775<div> 776 777<p>All Global Variables and Functions have one of the following visibility 778 styles:</p> 779 780<dl> 781 <dt><b>"<tt>default</tt>" - Default style</b>:</dt> 782 <dd>On targets that use the ELF object file format, default visibility means 783 that the declaration is visible to other modules and, in shared libraries, 784 means that the declared entity may be overridden. On Darwin, default 785 visibility means that the declaration is visible to other modules. Default 786 visibility corresponds to "external linkage" in the language.</dd> 787 788 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt> 789 <dd>Two declarations of an object with hidden visibility refer to the same 790 object if they are in the same shared object. Usually, hidden visibility 791 indicates that the symbol will not be placed into the dynamic symbol 792 table, so no other module (executable or shared library) can reference it 793 directly.</dd> 794 795 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt> 796 <dd>On ELF, protected visibility indicates that the symbol will be placed in 797 the dynamic symbol table, but that references within the defining module 798 will bind to the local symbol. That is, the symbol cannot be overridden by 799 another module.</dd> 800</dl> 801 802</div> 803 804<!-- ======================================================================= --> 805<h3> 806 <a name="namedtypes">Named Types</a> 807</h3> 808 809<div> 810 811<p>LLVM IR allows you to specify name aliases for certain types. This can make 812 it easier to read the IR and make the IR more condensed (particularly when 813 recursive types are involved). An example of a name specification is:</p> 814 815<pre class="doc_code"> 816%mytype = type { %mytype*, i32 } 817</pre> 818 819<p>You may give a name to any <a href="#typesystem">type</a> except 820 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type 821 is expected with the syntax "%mytype".</p> 822 823<p>Note that type names are aliases for the structural type that they indicate, 824 and that you can therefore specify multiple names for the same type. This 825 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR 826 uses structural typing, the name is not part of the type. When printing out 827 LLVM IR, the printer will pick <em>one name</em> to render all types of a 828 particular shape. This means that if you have code where two different 829 source types end up having the same LLVM type, that the dumper will sometimes 830 print the "wrong" or unexpected type. This is an important design point and 831 isn't going to change.</p> 832 833</div> 834 835<!-- ======================================================================= --> 836<h3> 837 <a name="globalvars">Global Variables</a> 838</h3> 839 840<div> 841 842<p>Global variables define regions of memory allocated at compilation time 843 instead of run-time. Global variables may optionally be initialized, may 844 have an explicit section to be placed in, and may have an optional explicit 845 alignment specified.</p> 846 847<p>A variable may be defined as <tt>thread_local</tt>, which 848 means that it will not be shared by threads (each thread will have a 849 separated copy of the variable). Not all targets support thread-local 850 variables. Optionally, a TLS model may be specified:</p> 851 852<dl> 853 <dt><b><tt>localdynamic</tt></b>:</dt> 854 <dd>For variables that are only used within the current shared library.</dd> 855 856 <dt><b><tt>initialexec</tt></b>:</dt> 857 <dd>For variables in modules that will not be loaded dynamically.</dd> 858 859 <dt><b><tt>localexec</tt></b>:</dt> 860 <dd>For variables defined in the executable and only used within it.</dd> 861</dl> 862 863<p>The models correspond to the ELF TLS models; see 864 <a href="http://people.redhat.com/drepper/tls.pdf">ELF 865 Handling For Thread-Local Storage</a> for more information on under which 866 circumstances the different models may be used. The target may choose a 867 different TLS model if the specified model is not supported, or if a better 868 choice of model can be made.</p> 869 870<p>A variable may be defined as a global 871 "constant," which indicates that the contents of the variable 872 will <b>never</b> be modified (enabling better optimization, allowing the 873 global data to be placed in the read-only section of an executable, etc). 874 Note that variables that need runtime initialization cannot be marked 875 "constant" as there is a store to the variable.</p> 876 877<p>LLVM explicitly allows <em>declarations</em> of global variables to be marked 878 constant, even if the final definition of the global is not. This capability 879 can be used to enable slightly better optimization of the program, but 880 requires the language definition to guarantee that optimizations based on the 881 'constantness' are valid for the translation units that do not include the 882 definition.</p> 883 884<p>As SSA values, global variables define pointer values that are in scope 885 (i.e. they dominate) all basic blocks in the program. Global variables 886 always define a pointer to their "content" type because they describe a 887 region of memory, and all memory objects in LLVM are accessed through 888 pointers.</p> 889 890<p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates 891 that the address is not significant, only the content. Constants marked 892 like this can be merged with other constants if they have the same 893 initializer. Note that a constant with significant address <em>can</em> 894 be merged with a <tt>unnamed_addr</tt> constant, the result being a 895 constant whose address is significant.</p> 896 897<p>A global variable may be declared to reside in a target-specific numbered 898 address space. For targets that support them, address spaces may affect how 899 optimizations are performed and/or what target instructions are used to 900 access the variable. The default address space is zero. The address space 901 qualifier must precede any other attributes.</p> 902 903<p>LLVM allows an explicit section to be specified for globals. If the target 904 supports it, it will emit globals to the section specified.</p> 905 906<p>An explicit alignment may be specified for a global, which must be a power 907 of 2. If not present, or if the alignment is set to zero, the alignment of 908 the global is set by the target to whatever it feels convenient. If an 909 explicit alignment is specified, the global is forced to have exactly that 910 alignment. Targets and optimizers are not allowed to over-align the global 911 if the global has an assigned section. In this case, the extra alignment 912 could be observable: for example, code could assume that the globals are 913 densely packed in their section and try to iterate over them as an array, 914 alignment padding would break this iteration.</p> 915 916<p>For example, the following defines a global in a numbered address space with 917 an initializer, section, and alignment:</p> 918 919<pre class="doc_code"> 920@G = addrspace(5) constant float 1.0, section "foo", align 4 921</pre> 922 923<p>The following example defines a thread-local global with 924 the <tt>initialexec</tt> TLS model:</p> 925 926<pre class="doc_code"> 927@G = thread_local(initialexec) global i32 0, align 4 928</pre> 929 930</div> 931 932 933<!-- ======================================================================= --> 934<h3> 935 <a name="functionstructure">Functions</a> 936</h3> 937 938<div> 939 940<p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an 941 optional <a href="#linkage">linkage type</a>, an optional 942 <a href="#visibility">visibility style</a>, an optional 943 <a href="#callingconv">calling convention</a>, 944 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional 945 <a href="#paramattrs">parameter attribute</a> for the return type, a function 946 name, a (possibly empty) argument list (each with optional 947 <a href="#paramattrs">parameter attributes</a>), optional 948 <a href="#fnattrs">function attributes</a>, an optional section, an optional 949 alignment, an optional <a href="#gc">garbage collector name</a>, an opening 950 curly brace, a list of basic blocks, and a closing curly brace.</p> 951 952<p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an 953 optional <a href="#linkage">linkage type</a>, an optional 954 <a href="#visibility">visibility style</a>, an optional 955 <a href="#callingconv">calling convention</a>, 956 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional 957 <a href="#paramattrs">parameter attribute</a> for the return type, a function 958 name, a possibly empty list of arguments, an optional alignment, and an 959 optional <a href="#gc">garbage collector name</a>.</p> 960 961<p>A function definition contains a list of basic blocks, forming the CFG 962 (Control Flow Graph) for the function. Each basic block may optionally start 963 with a label (giving the basic block a symbol table entry), contains a list 964 of instructions, and ends with a <a href="#terminators">terminator</a> 965 instruction (such as a branch or function return).</p> 966 967<p>The first basic block in a function is special in two ways: it is immediately 968 executed on entrance to the function, and it is not allowed to have 969 predecessor basic blocks (i.e. there can not be any branches to the entry 970 block of a function). Because the block can have no predecessors, it also 971 cannot have any <a href="#i_phi">PHI nodes</a>.</p> 972 973<p>LLVM allows an explicit section to be specified for functions. If the target 974 supports it, it will emit functions to the section specified.</p> 975 976<p>An explicit alignment may be specified for a function. If not present, or if 977 the alignment is set to zero, the alignment of the function is set by the 978 target to whatever it feels convenient. If an explicit alignment is 979 specified, the function is forced to have at least that much alignment. All 980 alignments must be a power of 2.</p> 981 982<p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not 983 be significant and two identical functions can be merged.</p> 984 985<h5>Syntax:</h5> 986<pre class="doc_code"> 987define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>] 988 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] 989 <ResultType> @<FunctionName> ([argument list]) 990 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N] 991 [<a href="#gc">gc</a>] { ... } 992</pre> 993 994</div> 995 996<!-- ======================================================================= --> 997<h3> 998 <a name="aliasstructure">Aliases</a> 999</h3> 1000 1001<div> 1002 1003<p>Aliases act as "second name" for the aliasee value (which can be either 1004 function, global variable, another alias or bitcast of global value). Aliases 1005 may have an optional <a href="#linkage">linkage type</a>, and an 1006 optional <a href="#visibility">visibility style</a>.</p> 1007 1008<h5>Syntax:</h5> 1009<pre class="doc_code"> 1010@<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee> 1011</pre> 1012 1013</div> 1014 1015<!-- ======================================================================= --> 1016<h3> 1017 <a name="namedmetadatastructure">Named Metadata</a> 1018</h3> 1019 1020<div> 1021 1022<p>Named metadata is a collection of metadata. <a href="#metadata">Metadata 1023 nodes</a> (but not metadata strings) are the only valid operands for 1024 a named metadata.</p> 1025 1026<h5>Syntax:</h5> 1027<pre class="doc_code"> 1028; Some unnamed metadata nodes, which are referenced by the named metadata. 1029!0 = metadata !{metadata !"zero"} 1030!1 = metadata !{metadata !"one"} 1031!2 = metadata !{metadata !"two"} 1032; A named metadata. 1033!name = !{!0, !1, !2} 1034</pre> 1035 1036</div> 1037 1038<!-- ======================================================================= --> 1039<h3> 1040 <a name="paramattrs">Parameter Attributes</a> 1041</h3> 1042 1043<div> 1044 1045<p>The return type and each parameter of a function type may have a set of 1046 <i>parameter attributes</i> associated with them. Parameter attributes are 1047 used to communicate additional information about the result or parameters of 1048 a function. Parameter attributes are considered to be part of the function, 1049 not of the function type, so functions with different parameter attributes 1050 can have the same function type.</p> 1051 1052<p>Parameter attributes are simple keywords that follow the type specified. If 1053 multiple parameter attributes are needed, they are space separated. For 1054 example:</p> 1055 1056<pre class="doc_code"> 1057declare i32 @printf(i8* noalias nocapture, ...) 1058declare i32 @atoi(i8 zeroext) 1059declare signext i8 @returns_signed_char() 1060</pre> 1061 1062<p>Note that any attributes for the function result (<tt>nounwind</tt>, 1063 <tt>readonly</tt>) come immediately after the argument list.</p> 1064 1065<p>Currently, only the following parameter attributes are defined:</p> 1066 1067<dl> 1068 <dt><tt><b>zeroext</b></tt></dt> 1069 <dd>This indicates to the code generator that the parameter or return value 1070 should be zero-extended to the extent required by the target's ABI (which 1071 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a 1072 parameter) or the callee (for a return value).</dd> 1073 1074 <dt><tt><b>signext</b></tt></dt> 1075 <dd>This indicates to the code generator that the parameter or return value 1076 should be sign-extended to the extent required by the target's ABI (which 1077 is usually 32-bits) by the caller (for a parameter) or the callee (for a 1078 return value).</dd> 1079 1080 <dt><tt><b>inreg</b></tt></dt> 1081 <dd>This indicates that this parameter or return value should be treated in a 1082 special target-dependent fashion during while emitting code for a function 1083 call or return (usually, by putting it in a register as opposed to memory, 1084 though some targets use it to distinguish between two different kinds of 1085 registers). Use of this attribute is target-specific.</dd> 1086 1087 <dt><tt><b><a name="byval">byval</a></b></tt></dt> 1088 <dd><p>This indicates that the pointer parameter should really be passed by 1089 value to the function. The attribute implies that a hidden copy of the 1090 pointee 1091 is made between the caller and the callee, so the callee is unable to 1092 modify the value in the caller. This attribute is only valid on LLVM 1093 pointer arguments. It is generally used to pass structs and arrays by 1094 value, but is also valid on pointers to scalars. The copy is considered 1095 to belong to the caller not the callee (for example, 1096 <tt><a href="#readonly">readonly</a></tt> functions should not write to 1097 <tt>byval</tt> parameters). This is not a valid attribute for return 1098 values.</p> 1099 1100 <p>The byval attribute also supports specifying an alignment with 1101 the align attribute. It indicates the alignment of the stack slot to 1102 form and the known alignment of the pointer specified to the call site. If 1103 the alignment is not specified, then the code generator makes a 1104 target-specific assumption.</p></dd> 1105 1106 <dt><tt><b><a name="sret">sret</a></b></tt></dt> 1107 <dd>This indicates that the pointer parameter specifies the address of a 1108 structure that is the return value of the function in the source program. 1109 This pointer must be guaranteed by the caller to be valid: loads and 1110 stores to the structure may be assumed by the callee to not to trap. This 1111 may only be applied to the first parameter. This is not a valid attribute 1112 for return values. </dd> 1113 1114 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt> 1115 <dd>This indicates that pointer values 1116 <a href="#pointeraliasing"><i>based</i></a> on the argument or return 1117 value do not alias pointer values which are not <i>based</i> on it, 1118 ignoring certain "irrelevant" dependencies. 1119 For a call to the parent function, dependencies between memory 1120 references from before or after the call and from those during the call 1121 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and 1122 return value used in that call. 1123 The caller shares the responsibility with the callee for ensuring that 1124 these requirements are met. 1125 For further details, please see the discussion of the NoAlias response in 1126 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br> 1127<br> 1128 Note that this definition of <tt>noalias</tt> is intentionally 1129 similar to the definition of <tt>restrict</tt> in C99 for function 1130 arguments, though it is slightly weaker. 1131<br> 1132 For function return values, C99's <tt>restrict</tt> is not meaningful, 1133 while LLVM's <tt>noalias</tt> is. 1134 </dd> 1135 1136 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt> 1137 <dd>This indicates that the callee does not make any copies of the pointer 1138 that outlive the callee itself. This is not a valid attribute for return 1139 values.</dd> 1140 1141 <dt><tt><b><a name="nest">nest</a></b></tt></dt> 1142 <dd>This indicates that the pointer parameter can be excised using the 1143 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid 1144 attribute for return values.</dd> 1145</dl> 1146 1147</div> 1148 1149<!-- ======================================================================= --> 1150<h3> 1151 <a name="gc">Garbage Collector Names</a> 1152</h3> 1153 1154<div> 1155 1156<p>Each function may specify a garbage collector name, which is simply a 1157 string:</p> 1158 1159<pre class="doc_code"> 1160define void @f() gc "name" { ... } 1161</pre> 1162 1163<p>The compiler declares the supported values of <i>name</i>. Specifying a 1164 collector which will cause the compiler to alter its output in order to 1165 support the named garbage collection algorithm.</p> 1166 1167</div> 1168 1169<!-- ======================================================================= --> 1170<h3> 1171 <a name="fnattrs">Function Attributes</a> 1172</h3> 1173 1174<div> 1175 1176<p>Function attributes are set to communicate additional information about a 1177 function. Function attributes are considered to be part of the function, not 1178 of the function type, so functions with different parameter attributes can 1179 have the same function type.</p> 1180 1181<p>Function attributes are simple keywords that follow the type specified. If 1182 multiple attributes are needed, they are space separated. For example:</p> 1183 1184<pre class="doc_code"> 1185define void @f() noinline { ... } 1186define void @f() alwaysinline { ... } 1187define void @f() alwaysinline optsize { ... } 1188define void @f() optsize { ... } 1189</pre> 1190 1191<dl> 1192 <dt><tt><b>address_safety</b></tt></dt> 1193 <dd>This attribute indicates that the address safety analysis 1194 is enabled for this function. </dd> 1195 1196 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt> 1197 <dd>This attribute indicates that, when emitting the prologue and epilogue, 1198 the backend should forcibly align the stack pointer. Specify the 1199 desired alignment, which must be a power of two, in parentheses. 1200 1201 <dt><tt><b>alwaysinline</b></tt></dt> 1202 <dd>This attribute indicates that the inliner should attempt to inline this 1203 function into callers whenever possible, ignoring any active inlining size 1204 threshold for this caller.</dd> 1205 1206 <dt><tt><b>nonlazybind</b></tt></dt> 1207 <dd>This attribute suppresses lazy symbol binding for the function. This 1208 may make calls to the function faster, at the cost of extra program 1209 startup time if the function is not called during program startup.</dd> 1210 1211 <dt><tt><b>inlinehint</b></tt></dt> 1212 <dd>This attribute indicates that the source code contained a hint that inlining 1213 this function is desirable (such as the "inline" keyword in C/C++). It 1214 is just a hint; it imposes no requirements on the inliner.</dd> 1215 1216 <dt><tt><b>naked</b></tt></dt> 1217 <dd>This attribute disables prologue / epilogue emission for the function. 1218 This can have very system-specific consequences.</dd> 1219 1220 <dt><tt><b>noimplicitfloat</b></tt></dt> 1221 <dd>This attributes disables implicit floating point instructions.</dd> 1222 1223 <dt><tt><b>noinline</b></tt></dt> 1224 <dd>This attribute indicates that the inliner should never inline this 1225 function in any situation. This attribute may not be used together with 1226 the <tt>alwaysinline</tt> attribute.</dd> 1227 1228 <dt><tt><b>noredzone</b></tt></dt> 1229 <dd>This attribute indicates that the code generator should not use a red 1230 zone, even if the target-specific ABI normally permits it.</dd> 1231 1232 <dt><tt><b>noreturn</b></tt></dt> 1233 <dd>This function attribute indicates that the function never returns 1234 normally. This produces undefined behavior at runtime if the function 1235 ever does dynamically return.</dd> 1236 1237 <dt><tt><b>nounwind</b></tt></dt> 1238 <dd>This function attribute indicates that the function never returns with an 1239 unwind or exceptional control flow. If the function does unwind, its 1240 runtime behavior is undefined.</dd> 1241 1242 <dt><tt><b>optsize</b></tt></dt> 1243 <dd>This attribute suggests that optimization passes and code generator passes 1244 make choices that keep the code size of this function low, and otherwise 1245 do optimizations specifically to reduce code size.</dd> 1246 1247 <dt><tt><b>readnone</b></tt></dt> 1248 <dd>This attribute indicates that the function computes its result (or decides 1249 to unwind an exception) based strictly on its arguments, without 1250 dereferencing any pointer arguments or otherwise accessing any mutable 1251 state (e.g. memory, control registers, etc) visible to caller functions. 1252 It does not write through any pointer arguments 1253 (including <tt><a href="#byval">byval</a></tt> arguments) and never 1254 changes any state visible to callers. This means that it cannot unwind 1255 exceptions by calling the <tt>C++</tt> exception throwing methods.</dd> 1256 1257 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt> 1258 <dd>This attribute indicates that the function does not write through any 1259 pointer arguments (including <tt><a href="#byval">byval</a></tt> 1260 arguments) or otherwise modify any state (e.g. memory, control registers, 1261 etc) visible to caller functions. It may dereference pointer arguments 1262 and read state that may be set in the caller. A readonly function always 1263 returns the same value (or unwinds an exception identically) when called 1264 with the same set of arguments and global state. It cannot unwind an 1265 exception by calling the <tt>C++</tt> exception throwing methods.</dd> 1266 1267 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt> 1268 <dd>This attribute indicates that this function can return twice. The 1269 C <code>setjmp</code> is an example of such a function. The compiler 1270 disables some optimizations (like tail calls) in the caller of these 1271 functions.</dd> 1272 1273 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt> 1274 <dd>This attribute indicates that the function should emit a stack smashing 1275 protector. It is in the form of a "canary"—a random value placed on 1276 the stack before the local variables that's checked upon return from the 1277 function to see if it has been overwritten. A heuristic is used to 1278 determine if a function needs stack protectors or not.<br> 1279<br> 1280 If a function that has an <tt>ssp</tt> attribute is inlined into a 1281 function that doesn't have an <tt>ssp</tt> attribute, then the resulting 1282 function will have an <tt>ssp</tt> attribute.</dd> 1283 1284 <dt><tt><b>sspreq</b></tt></dt> 1285 <dd>This attribute indicates that the function should <em>always</em> emit a 1286 stack smashing protector. This overrides 1287 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br> 1288<br> 1289 If a function that has an <tt>sspreq</tt> attribute is inlined into a 1290 function that doesn't have an <tt>sspreq</tt> attribute or which has 1291 an <tt>ssp</tt> attribute, then the resulting function will have 1292 an <tt>sspreq</tt> attribute.</dd> 1293 1294 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt> 1295 <dd>This attribute indicates that the ABI being targeted requires that 1296 an unwind table entry be produce for this function even if we can 1297 show that no exceptions passes by it. This is normally the case for 1298 the ELF x86-64 abi, but it can be disabled for some compilation 1299 units.</dd> 1300</dl> 1301 1302</div> 1303 1304<!-- ======================================================================= --> 1305<h3> 1306 <a name="moduleasm">Module-Level Inline Assembly</a> 1307</h3> 1308 1309<div> 1310 1311<p>Modules may contain "module-level inline asm" blocks, which corresponds to 1312 the GCC "file scope inline asm" blocks. These blocks are internally 1313 concatenated by LLVM and treated as a single unit, but may be separated in 1314 the <tt>.ll</tt> file if desired. The syntax is very simple:</p> 1315 1316<pre class="doc_code"> 1317module asm "inline asm code goes here" 1318module asm "more can go here" 1319</pre> 1320 1321<p>The strings can contain any character by escaping non-printable characters. 1322 The escape sequence used is simply "\xx" where "xx" is the two digit hex code 1323 for the number.</p> 1324 1325<p>The inline asm code is simply printed to the machine code .s file when 1326 assembly code is generated.</p> 1327 1328</div> 1329 1330<!-- ======================================================================= --> 1331<h3> 1332 <a name="datalayout">Data Layout</a> 1333</h3> 1334 1335<div> 1336 1337<p>A module may specify a target specific data layout string that specifies how 1338 data is to be laid out in memory. The syntax for the data layout is 1339 simply:</p> 1340 1341<pre class="doc_code"> 1342target datalayout = "<i>layout specification</i>" 1343</pre> 1344 1345<p>The <i>layout specification</i> consists of a list of specifications 1346 separated by the minus sign character ('-'). Each specification starts with 1347 a letter and may include other information after the letter to define some 1348 aspect of the data layout. The specifications accepted are as follows:</p> 1349 1350<dl> 1351 <dt><tt>E</tt></dt> 1352 <dd>Specifies that the target lays out data in big-endian form. That is, the 1353 bits with the most significance have the lowest address location.</dd> 1354 1355 <dt><tt>e</tt></dt> 1356 <dd>Specifies that the target lays out data in little-endian form. That is, 1357 the bits with the least significance have the lowest address 1358 location.</dd> 1359 1360 <dt><tt>S<i>size</i></tt></dt> 1361 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion 1362 of stack variables is limited to the natural stack alignment to avoid 1363 dynamic stack realignment. The stack alignment must be a multiple of 1364 8-bits. If omitted, the natural stack alignment defaults to "unspecified", 1365 which does not prevent any alignment promotions.</dd> 1366 1367 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt> 1368 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and 1369 <i>preferred</i> alignments. All sizes are in bits. Specifying 1370 the <i>pref</i> alignment is optional. If omitted, the 1371 preceding <tt>:</tt> should be omitted too.</dd> 1372 1373 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt> 1374 <dd>This specifies the alignment for an integer type of a given bit 1375 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd> 1376 1377 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt> 1378 <dd>This specifies the alignment for a vector type of a given bit 1379 <i>size</i>.</dd> 1380 1381 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt> 1382 <dd>This specifies the alignment for a floating point type of a given bit 1383 <i>size</i>. Only values of <i>size</i> that are supported by the target 1384 will work. 32 (float) and 64 (double) are supported on all targets; 1385 80 or 128 (different flavors of long double) are also supported on some 1386 targets. 1387 1388 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt> 1389 <dd>This specifies the alignment for an aggregate type of a given bit 1390 <i>size</i>.</dd> 1391 1392 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt> 1393 <dd>This specifies the alignment for a stack object of a given bit 1394 <i>size</i>.</dd> 1395 1396 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt> 1397 <dd>This specifies a set of native integer widths for the target CPU 1398 in bits. For example, it might contain "n32" for 32-bit PowerPC, 1399 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of 1400 this set are considered to support most general arithmetic 1401 operations efficiently.</dd> 1402</dl> 1403 1404<p>When constructing the data layout for a given target, LLVM starts with a 1405 default set of specifications which are then (possibly) overridden by the 1406 specifications in the <tt>datalayout</tt> keyword. The default specifications 1407 are given in this list:</p> 1408 1409<ul> 1410 <li><tt>E</tt> - big endian</li> 1411 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li> 1412 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li> 1413 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li> 1414 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li> 1415 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li> 1416 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred 1417 alignment of 64-bits</li> 1418 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li> 1419 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li> 1420 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li> 1421 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li> 1422 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li> 1423 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li> 1424</ul> 1425 1426<p>When LLVM is determining the alignment for a given type, it uses the 1427 following rules:</p> 1428 1429<ol> 1430 <li>If the type sought is an exact match for one of the specifications, that 1431 specification is used.</li> 1432 1433 <li>If no match is found, and the type sought is an integer type, then the 1434 smallest integer type that is larger than the bitwidth of the sought type 1435 is used. If none of the specifications are larger than the bitwidth then 1436 the largest integer type is used. For example, given the default 1437 specifications above, the i7 type will use the alignment of i8 (next 1438 largest) while both i65 and i256 will use the alignment of i64 (largest 1439 specified).</li> 1440 1441 <li>If no match is found, and the type sought is a vector type, then the 1442 largest vector type that is smaller than the sought vector type will be 1443 used as a fall back. This happens because <128 x double> can be 1444 implemented in terms of 64 <2 x double>, for example.</li> 1445</ol> 1446 1447<p>The function of the data layout string may not be what you expect. Notably, 1448 this is not a specification from the frontend of what alignment the code 1449 generator should use.</p> 1450 1451<p>Instead, if specified, the target data layout is required to match what the 1452 ultimate <em>code generator</em> expects. This string is used by the 1453 mid-level optimizers to 1454 improve code, and this only works if it matches what the ultimate code 1455 generator uses. If you would like to generate IR that does not embed this 1456 target-specific detail into the IR, then you don't have to specify the 1457 string. This will disable some optimizations that require precise layout 1458 information, but this also prevents those optimizations from introducing 1459 target specificity into the IR.</p> 1460 1461 1462 1463</div> 1464 1465<!-- ======================================================================= --> 1466<h3> 1467 <a name="pointeraliasing">Pointer Aliasing Rules</a> 1468</h3> 1469 1470<div> 1471 1472<p>Any memory access must be done through a pointer value associated 1473with an address range of the memory access, otherwise the behavior 1474is undefined. Pointer values are associated with address ranges 1475according to the following rules:</p> 1476 1477<ul> 1478 <li>A pointer value is associated with the addresses associated with 1479 any value it is <i>based</i> on. 1480 <li>An address of a global variable is associated with the address 1481 range of the variable's storage.</li> 1482 <li>The result value of an allocation instruction is associated with 1483 the address range of the allocated storage.</li> 1484 <li>A null pointer in the default address-space is associated with 1485 no address.</li> 1486 <li>An integer constant other than zero or a pointer value returned 1487 from a function not defined within LLVM may be associated with address 1488 ranges allocated through mechanisms other than those provided by 1489 LLVM. Such ranges shall not overlap with any ranges of addresses 1490 allocated by mechanisms provided by LLVM.</li> 1491</ul> 1492 1493<p>A pointer value is <i>based</i> on another pointer value according 1494 to the following rules:</p> 1495 1496<ul> 1497 <li>A pointer value formed from a 1498 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation 1499 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li> 1500 <li>The result value of a 1501 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand 1502 of the <tt>bitcast</tt>.</li> 1503 <li>A pointer value formed by an 1504 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all 1505 pointer values that contribute (directly or indirectly) to the 1506 computation of the pointer's value.</li> 1507 <li>The "<i>based</i> on" relationship is transitive.</li> 1508</ul> 1509 1510<p>Note that this definition of <i>"based"</i> is intentionally 1511 similar to the definition of <i>"based"</i> in C99, though it is 1512 slightly weaker.</p> 1513 1514<p>LLVM IR does not associate types with memory. The result type of a 1515<tt><a href="#i_load">load</a></tt> merely indicates the size and 1516alignment of the memory from which to load, as well as the 1517interpretation of the value. The first operand type of a 1518<tt><a href="#i_store">store</a></tt> similarly only indicates the size 1519and alignment of the store.</p> 1520 1521<p>Consequently, type-based alias analysis, aka TBAA, aka 1522<tt>-fstrict-aliasing</tt>, is not applicable to general unadorned 1523LLVM IR. <a href="#metadata">Metadata</a> may be used to encode 1524additional information which specialized optimization passes may use 1525to implement type-based alias analysis.</p> 1526 1527</div> 1528 1529<!-- ======================================================================= --> 1530<h3> 1531 <a name="volatile">Volatile Memory Accesses</a> 1532</h3> 1533 1534<div> 1535 1536<p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a 1537href="#i_store"><tt>store</tt></a>s, and <a 1538href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>. 1539The optimizers must not change the number of volatile operations or change their 1540order of execution relative to other volatile operations. The optimizers 1541<i>may</i> change the order of volatile operations relative to non-volatile 1542operations. This is not Java's "volatile" and has no cross-thread 1543synchronization behavior.</p> 1544 1545</div> 1546 1547<!-- ======================================================================= --> 1548<h3> 1549 <a name="memmodel">Memory Model for Concurrent Operations</a> 1550</h3> 1551 1552<div> 1553 1554<p>The LLVM IR does not define any way to start parallel threads of execution 1555or to register signal handlers. Nonetheless, there are platform-specific 1556ways to create them, and we define LLVM IR's behavior in their presence. This 1557model is inspired by the C++0x memory model.</p> 1558 1559<p>For a more informal introduction to this model, see the 1560<a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>. 1561 1562<p>We define a <i>happens-before</i> partial order as the least partial order 1563that</p> 1564<ul> 1565 <li>Is a superset of single-thread program order, and</li> 1566 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from 1567 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced 1568 by platform-specific techniques, like pthread locks, thread 1569 creation, thread joining, etc., and by atomic instructions. 1570 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>). 1571 </li> 1572</ul> 1573 1574<p>Note that program order does not introduce <i>happens-before</i> edges 1575between a thread and signals executing inside that thread.</p> 1576 1577<p>Every (defined) read operation (load instructions, memcpy, atomic 1578loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by 1579(defined) write operations (store instructions, atomic 1580stores/read-modify-writes, memcpy, etc.). For the purposes of this section, 1581initialized globals are considered to have a write of the initializer which is 1582atomic and happens before any other read or write of the memory in question. 1583For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see 1584any write to the same byte, except:</p> 1585 1586<ul> 1587 <li>If <var>write<sub>1</sub></var> happens before 1588 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens 1589 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var> 1590 does not see <var>write<sub>1</sub></var>. 1591 <li>If <var>R<sub>byte</sub></var> happens before 1592 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not 1593 see <var>write<sub>3</sub></var>. 1594</ul> 1595 1596<p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows: 1597<ul> 1598 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile 1599 is supposed to give guarantees which can support 1600 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to 1601 addresses which do not behave like normal memory. It does not generally 1602 provide cross-thread synchronization.) 1603 <li>Otherwise, if there is no write to the same byte that happens before 1604 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns 1605 <tt>undef</tt> for that byte. 1606 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write, 1607 <var>R<sub>byte</sub></var> returns the value written by that 1608 write.</li> 1609 <li>Otherwise, if <var>R</var> is atomic, and all the writes 1610 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the 1611 values written. See the <a href="#ordering">Atomic Memory Ordering 1612 Constraints</a> section for additional constraints on how the choice 1613 is made. 1614 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li> 1615</ul> 1616 1617<p><var>R</var> returns the value composed of the series of bytes it read. 1618This implies that some bytes within the value may be <tt>undef</tt> 1619<b>without</b> the entire value being <tt>undef</tt>. Note that this only 1620defines the semantics of the operation; it doesn't mean that targets will 1621emit more than one instruction to read the series of bytes.</p> 1622 1623<p>Note that in cases where none of the atomic intrinsics are used, this model 1624places only one restriction on IR transformations on top of what is required 1625for single-threaded execution: introducing a store to a byte which might not 1626otherwise be stored is not allowed in general. (Specifically, in the case 1627where another thread might write to and read from an address, introducing a 1628store can change a load that may see exactly one write into a load that may 1629see multiple writes.)</p> 1630 1631<!-- FIXME: This model assumes all targets where concurrency is relevant have 1632a byte-size store which doesn't affect adjacent bytes. As far as I can tell, 1633none of the backends currently in the tree fall into this category; however, 1634there might be targets which care. If there are, we want a paragraph 1635like the following: 1636 1637Targets may specify that stores narrower than a certain width are not 1638available; on such a target, for the purposes of this model, treat any 1639non-atomic write with an alignment or width less than the minimum width 1640as if it writes to the relevant surrounding bytes. 1641--> 1642 1643</div> 1644 1645<!-- ======================================================================= --> 1646<h3> 1647 <a name="ordering">Atomic Memory Ordering Constraints</a> 1648</h3> 1649 1650<div> 1651 1652<p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>, 1653<a href="#i_atomicrmw"><code>atomicrmw</code></a>, 1654<a href="#i_fence"><code>fence</code></a>, 1655<a href="#i_load"><code>atomic load</code></a>, and 1656<a href="#i_store"><code>atomic store</code></a>) take an ordering parameter 1657that determines which other atomic instructions on the same address they 1658<i>synchronize with</i>. These semantics are borrowed from Java and C++0x, 1659but are somewhat more colloquial. If these descriptions aren't precise enough, 1660check those specs (see spec references in the 1661<a href="Atomics.html#introduction">atomics guide</a>). 1662<a href="#i_fence"><code>fence</code></a> instructions 1663treat these orderings somewhat differently since they don't take an address. 1664See that instruction's documentation for details.</p> 1665 1666<p>For a simpler introduction to the ordering constraints, see the 1667<a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p> 1668 1669<dl> 1670<dt><code>unordered</code></dt> 1671<dd>The set of values that can be read is governed by the happens-before 1672partial order. A value cannot be read unless some operation wrote it. 1673This is intended to provide a guarantee strong enough to model Java's 1674non-volatile shared variables. This ordering cannot be specified for 1675read-modify-write operations; it is not strong enough to make them atomic 1676in any interesting way.</dd> 1677<dt><code>monotonic</code></dt> 1678<dd>In addition to the guarantees of <code>unordered</code>, there is a single 1679total order for modifications by <code>monotonic</code> operations on each 1680address. All modification orders must be compatible with the happens-before 1681order. There is no guarantee that the modification orders can be combined to 1682a global total order for the whole program (and this often will not be 1683possible). The read in an atomic read-modify-write operation 1684(<a href="#i_cmpxchg"><code>cmpxchg</code></a> and 1685<a href="#i_atomicrmw"><code>atomicrmw</code></a>) 1686reads the value in the modification order immediately before the value it 1687writes. If one atomic read happens before another atomic read of the same 1688address, the later read must see the same value or a later value in the 1689address's modification order. This disallows reordering of 1690<code>monotonic</code> (or stronger) operations on the same address. If an 1691address is written <code>monotonic</code>ally by one thread, and other threads 1692<code>monotonic</code>ally read that address repeatedly, the other threads must 1693eventually see the write. This corresponds to the C++0x/C1x 1694<code>memory_order_relaxed</code>.</dd> 1695<dt><code>acquire</code></dt> 1696<dd>In addition to the guarantees of <code>monotonic</code>, 1697a <i>synchronizes-with</i> edge may be formed with a <code>release</code> 1698operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd> 1699<dt><code>release</code></dt> 1700<dd>In addition to the guarantees of <code>monotonic</code>, if this operation 1701writes a value which is subsequently read by an <code>acquire</code> operation, 1702it <i>synchronizes-with</i> that operation. (This isn't a complete 1703description; see the C++0x definition of a release sequence.) This corresponds 1704to the C++0x/C1x <code>memory_order_release</code>.</dd> 1705<dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an 1706<code>acquire</code> and <code>release</code> operation on its address. 1707This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd> 1708<dt><code>seq_cst</code> (sequentially consistent)</dt><dd> 1709<dd>In addition to the guarantees of <code>acq_rel</code> 1710(<code>acquire</code> for an operation which only reads, <code>release</code> 1711for an operation which only writes), there is a global total order on all 1712sequentially-consistent operations on all addresses, which is consistent with 1713the <i>happens-before</i> partial order and with the modification orders of 1714all the affected addresses. Each sequentially-consistent read sees the last 1715preceding write to the same address in this global order. This corresponds 1716to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd> 1717</dl> 1718 1719<p id="singlethread">If an atomic operation is marked <code>singlethread</code>, 1720it only <i>synchronizes with</i> or participates in modification and seq_cst 1721total orderings with other operations running in the same thread (for example, 1722in signal handlers).</p> 1723 1724</div> 1725 1726</div> 1727 1728<!-- *********************************************************************** --> 1729<h2><a name="typesystem">Type System</a></h2> 1730<!-- *********************************************************************** --> 1731 1732<div> 1733 1734<p>The LLVM type system is one of the most important features of the 1735 intermediate representation. Being typed enables a number of optimizations 1736 to be performed on the intermediate representation directly, without having 1737 to do extra analyses on the side before the transformation. A strong type 1738 system makes it easier to read the generated code and enables novel analyses 1739 and transformations that are not feasible to perform on normal three address 1740 code representations.</p> 1741 1742<!-- ======================================================================= --> 1743<h3> 1744 <a name="t_classifications">Type Classifications</a> 1745</h3> 1746 1747<div> 1748 1749<p>The types fall into a few useful classifications:</p> 1750 1751<table border="1" cellspacing="0" cellpadding="4"> 1752 <tbody> 1753 <tr><th>Classification</th><th>Types</th></tr> 1754 <tr> 1755 <td><a href="#t_integer">integer</a></td> 1756 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td> 1757 </tr> 1758 <tr> 1759 <td><a href="#t_floating">floating point</a></td> 1760 <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td> 1761 </tr> 1762 <tr> 1763 <td><a name="t_firstclass">first class</a></td> 1764 <td><a href="#t_integer">integer</a>, 1765 <a href="#t_floating">floating point</a>, 1766 <a href="#t_pointer">pointer</a>, 1767 <a href="#t_vector">vector</a>, 1768 <a href="#t_struct">structure</a>, 1769 <a href="#t_array">array</a>, 1770 <a href="#t_label">label</a>, 1771 <a href="#t_metadata">metadata</a>. 1772 </td> 1773 </tr> 1774 <tr> 1775 <td><a href="#t_primitive">primitive</a></td> 1776 <td><a href="#t_label">label</a>, 1777 <a href="#t_void">void</a>, 1778 <a href="#t_integer">integer</a>, 1779 <a href="#t_floating">floating point</a>, 1780 <a href="#t_x86mmx">x86mmx</a>, 1781 <a href="#t_metadata">metadata</a>.</td> 1782 </tr> 1783 <tr> 1784 <td><a href="#t_derived">derived</a></td> 1785 <td><a href="#t_array">array</a>, 1786 <a href="#t_function">function</a>, 1787 <a href="#t_pointer">pointer</a>, 1788 <a href="#t_struct">structure</a>, 1789 <a href="#t_vector">vector</a>, 1790 <a href="#t_opaque">opaque</a>. 1791 </td> 1792 </tr> 1793 </tbody> 1794</table> 1795 1796<p>The <a href="#t_firstclass">first class</a> types are perhaps the most 1797 important. Values of these types are the only ones which can be produced by 1798 instructions.</p> 1799 1800</div> 1801 1802<!-- ======================================================================= --> 1803<h3> 1804 <a name="t_primitive">Primitive Types</a> 1805</h3> 1806 1807<div> 1808 1809<p>The primitive types are the fundamental building blocks of the LLVM 1810 system.</p> 1811 1812<!-- _______________________________________________________________________ --> 1813<h4> 1814 <a name="t_integer">Integer Type</a> 1815</h4> 1816 1817<div> 1818 1819<h5>Overview:</h5> 1820<p>The integer type is a very simple type that simply specifies an arbitrary 1821 bit width for the integer type desired. Any bit width from 1 bit to 1822 2<sup>23</sup>-1 (about 8 million) can be specified.</p> 1823 1824<h5>Syntax:</h5> 1825<pre> 1826 iN 1827</pre> 1828 1829<p>The number of bits the integer will occupy is specified by the <tt>N</tt> 1830 value.</p> 1831 1832<h5>Examples:</h5> 1833<table class="layout"> 1834 <tr class="layout"> 1835 <td class="left"><tt>i1</tt></td> 1836 <td class="left">a single-bit integer.</td> 1837 </tr> 1838 <tr class="layout"> 1839 <td class="left"><tt>i32</tt></td> 1840 <td class="left">a 32-bit integer.</td> 1841 </tr> 1842 <tr class="layout"> 1843 <td class="left"><tt>i1942652</tt></td> 1844 <td class="left">a really big integer of over 1 million bits.</td> 1845 </tr> 1846</table> 1847 1848</div> 1849 1850<!-- _______________________________________________________________________ --> 1851<h4> 1852 <a name="t_floating">Floating Point Types</a> 1853</h4> 1854 1855<div> 1856 1857<table> 1858 <tbody> 1859 <tr><th>Type</th><th>Description</th></tr> 1860 <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr> 1861 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr> 1862 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr> 1863 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr> 1864 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr> 1865 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr> 1866 </tbody> 1867</table> 1868 1869</div> 1870 1871<!-- _______________________________________________________________________ --> 1872<h4> 1873 <a name="t_x86mmx">X86mmx Type</a> 1874</h4> 1875 1876<div> 1877 1878<h5>Overview:</h5> 1879<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> 1880 1881<h5>Syntax:</h5> 1882<pre> 1883 x86mmx 1884</pre> 1885 1886</div> 1887 1888<!-- _______________________________________________________________________ --> 1889<h4> 1890 <a name="t_void">Void Type</a> 1891</h4> 1892 1893<div> 1894 1895<h5>Overview:</h5> 1896<p>The void type does not represent any value and has no size.</p> 1897 1898<h5>Syntax:</h5> 1899<pre> 1900 void 1901</pre> 1902 1903</div> 1904 1905<!-- _______________________________________________________________________ --> 1906<h4> 1907 <a name="t_label">Label Type</a> 1908</h4> 1909 1910<div> 1911 1912<h5>Overview:</h5> 1913<p>The label type represents code labels.</p> 1914 1915<h5>Syntax:</h5> 1916<pre> 1917 label 1918</pre> 1919 1920</div> 1921 1922<!-- _______________________________________________________________________ --> 1923<h4> 1924 <a name="t_metadata">Metadata Type</a> 1925</h4> 1926 1927<div> 1928 1929<h5>Overview:</h5> 1930<p>The metadata type represents embedded metadata. No derived types may be 1931 created from metadata except for <a href="#t_function">function</a> 1932 arguments. 1933 1934<h5>Syntax:</h5> 1935<pre> 1936 metadata 1937</pre> 1938 1939</div> 1940 1941</div> 1942 1943<!-- ======================================================================= --> 1944<h3> 1945 <a name="t_derived">Derived Types</a> 1946</h3> 1947 1948<div> 1949 1950<p>The real power in LLVM comes from the derived types in the system. This is 1951 what allows a programmer to represent arrays, functions, pointers, and other 1952 useful types. Each of these types contain one or more element types which 1953 may be a primitive type, or another derived type. For example, it is 1954 possible to have a two dimensional array, using an array as the element type 1955 of another array.</p> 1956 1957<!-- _______________________________________________________________________ --> 1958<h4> 1959 <a name="t_aggregate">Aggregate Types</a> 1960</h4> 1961 1962<div> 1963 1964<p>Aggregate Types are a subset of derived types that can contain multiple 1965 member types. <a href="#t_array">Arrays</a> and 1966 <a href="#t_struct">structs</a> are aggregate types. 1967 <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p> 1968 1969</div> 1970 1971<!-- _______________________________________________________________________ --> 1972<h4> 1973 <a name="t_array">Array Type</a> 1974</h4> 1975 1976<div> 1977 1978<h5>Overview:</h5> 1979<p>The array type is a very simple derived type that arranges elements 1980 sequentially in memory. The array type requires a size (number of elements) 1981 and an underlying data type.</p> 1982 1983<h5>Syntax:</h5> 1984<pre> 1985 [<# elements> x <elementtype>] 1986</pre> 1987 1988<p>The number of elements is a constant integer value; <tt>elementtype</tt> may 1989 be any type with a size.</p> 1990 1991<h5>Examples:</h5> 1992<table class="layout"> 1993 <tr class="layout"> 1994 <td class="left"><tt>[40 x i32]</tt></td> 1995 <td class="left">Array of 40 32-bit integer values.</td> 1996 </tr> 1997 <tr class="layout"> 1998 <td class="left"><tt>[41 x i32]</tt></td> 1999 <td class="left">Array of 41 32-bit integer values.</td> 2000 </tr> 2001 <tr class="layout"> 2002 <td class="left"><tt>[4 x i8]</tt></td> 2003 <td class="left">Array of 4 8-bit integer values.</td> 2004 </tr> 2005</table> 2006<p>Here are some examples of multidimensional arrays:</p> 2007<table class="layout"> 2008 <tr class="layout"> 2009 <td class="left"><tt>[3 x [4 x i32]]</tt></td> 2010 <td class="left">3x4 array of 32-bit integer values.</td> 2011 </tr> 2012 <tr class="layout"> 2013 <td class="left"><tt>[12 x [10 x float]]</tt></td> 2014 <td class="left">12x10 array of single precision floating point values.</td> 2015 </tr> 2016 <tr class="layout"> 2017 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td> 2018 <td class="left">2x3x4 array of 16-bit integer values.</td> 2019 </tr> 2020</table> 2021 2022<p>There is no restriction on indexing beyond the end of the array implied by 2023 a static type (though there are restrictions on indexing beyond the bounds 2024 of an allocated object in some cases). This means that single-dimension 2025 'variable sized array' addressing can be implemented in LLVM with a zero 2026 length array type. An implementation of 'pascal style arrays' in LLVM could 2027 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p> 2028 2029</div> 2030 2031<!-- _______________________________________________________________________ --> 2032<h4> 2033 <a name="t_function">Function Type</a> 2034</h4> 2035 2036<div> 2037 2038<h5>Overview:</h5> 2039<p>The function type can be thought of as a function signature. It consists of 2040 a return type and a list of formal parameter types. The return type of a 2041 function type is a first class type or a void type.</p> 2042 2043<h5>Syntax:</h5> 2044<pre> 2045 <returntype> (<parameter list>) 2046</pre> 2047 2048<p>...where '<tt><parameter list></tt>' is a comma-separated list of type 2049 specifiers. Optionally, the parameter list may include a type <tt>...</tt>, 2050 which indicates that the function takes a variable number of arguments. 2051 Variable argument functions can access their arguments with 2052 the <a href="#int_varargs">variable argument handling intrinsic</a> 2053 functions. '<tt><returntype></tt>' is any type except 2054 <a href="#t_label">label</a>.</p> 2055 2056<h5>Examples:</h5> 2057<table class="layout"> 2058 <tr class="layout"> 2059 <td class="left"><tt>i32 (i32)</tt></td> 2060 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt> 2061 </td> 2062 </tr><tr class="layout"> 2063 <td class="left"><tt>float (i16, i32 *) * 2064 </tt></td> 2065 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes 2066 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>, 2067 returning <tt>float</tt>. 2068 </td> 2069 </tr><tr class="layout"> 2070 <td class="left"><tt>i32 (i8*, ...)</tt></td> 2071 <td class="left">A vararg function that takes at least one 2072 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C), 2073 which returns an integer. This is the signature for <tt>printf</tt> in 2074 LLVM. 2075 </td> 2076 </tr><tr class="layout"> 2077 <td class="left"><tt>{i32, i32} (i32)</tt></td> 2078 <td class="left">A function taking an <tt>i32</tt>, returning a 2079 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values 2080 </td> 2081 </tr> 2082</table> 2083 2084</div> 2085 2086<!-- _______________________________________________________________________ --> 2087<h4> 2088 <a name="t_struct">Structure Type</a> 2089</h4> 2090 2091<div> 2092 2093<h5>Overview:</h5> 2094<p>The structure type is used to represent a collection of data members together 2095 in memory. The elements of a structure may be any type that has a size.</p> 2096 2097<p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>' 2098 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field 2099 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction. 2100 Structures in registers are accessed using the 2101 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and 2102 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p> 2103 2104<p>Structures may optionally be "packed" structures, which indicate that the 2105 alignment of the struct is one byte, and that there is no padding between 2106 the elements. In non-packed structs, padding between field types is inserted 2107 as defined by the TargetData string in the module, which is required to match 2108 what the underlying code generator expects.</p> 2109 2110<p>Structures can either be "literal" or "identified". A literal structure is 2111 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified 2112 types are always defined at the top level with a name. Literal types are 2113 uniqued by their contents and can never be recursive or opaque since there is 2114 no way to write one. Identified types can be recursive, can be opaqued, and are 2115 never uniqued. 2116</p> 2117 2118<h5>Syntax:</h5> 2119<pre> 2120 %T1 = type { <type list> } <i>; Identified normal struct type</i> 2121 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i> 2122</pre> 2123 2124<h5>Examples:</h5> 2125<table class="layout"> 2126 <tr class="layout"> 2127 <td class="left"><tt>{ i32, i32, i32 }</tt></td> 2128 <td class="left">A triple of three <tt>i32</tt> values</td> 2129 </tr> 2130 <tr class="layout"> 2131 <td class="left"><tt>{ float, i32 (i32) * }</tt></td> 2132 <td class="left">A pair, where the first element is a <tt>float</tt> and the 2133 second element is a <a href="#t_pointer">pointer</a> to a 2134 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning 2135 an <tt>i32</tt>.</td> 2136 </tr> 2137 <tr class="layout"> 2138 <td class="left"><tt><{ i8, i32 }></tt></td> 2139 <td class="left">A packed struct known to be 5 bytes in size.</td> 2140 </tr> 2141</table> 2142 2143</div> 2144 2145<!-- _______________________________________________________________________ --> 2146<h4> 2147 <a name="t_opaque">Opaque Structure Types</a> 2148</h4> 2149 2150<div> 2151 2152<h5>Overview:</h5> 2153<p>Opaque structure types are used to represent named structure types that do 2154 not have a body specified. This corresponds (for example) to the C notion of 2155 a forward declared structure.</p> 2156 2157<h5>Syntax:</h5> 2158<pre> 2159 %X = type opaque 2160 %52 = type opaque 2161</pre> 2162 2163<h5>Examples:</h5> 2164<table class="layout"> 2165 <tr class="layout"> 2166 <td class="left"><tt>opaque</tt></td> 2167 <td class="left">An opaque type.</td> 2168 </tr> 2169</table> 2170 2171</div> 2172 2173 2174 2175<!-- _______________________________________________________________________ --> 2176<h4> 2177 <a name="t_pointer">Pointer Type</a> 2178</h4> 2179 2180<div> 2181 2182<h5>Overview:</h5> 2183<p>The pointer type is used to specify memory locations. 2184 Pointers are commonly used to reference objects in memory.</p> 2185 2186<p>Pointer types may have an optional address space attribute defining the 2187 numbered address space where the pointed-to object resides. The default 2188 address space is number zero. The semantics of non-zero address 2189 spaces are target-specific.</p> 2190 2191<p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it 2192 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p> 2193 2194<h5>Syntax:</h5> 2195<pre> 2196 <type> * 2197</pre> 2198 2199<h5>Examples:</h5> 2200<table class="layout"> 2201 <tr class="layout"> 2202 <td class="left"><tt>[4 x i32]*</tt></td> 2203 <td class="left">A <a href="#t_pointer">pointer</a> to <a 2204 href="#t_array">array</a> of four <tt>i32</tt> values.</td> 2205 </tr> 2206 <tr class="layout"> 2207 <td class="left"><tt>i32 (i32*) *</tt></td> 2208 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a 2209 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an 2210 <tt>i32</tt>.</td> 2211 </tr> 2212 <tr class="layout"> 2213 <td class="left"><tt>i32 addrspace(5)*</tt></td> 2214 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value 2215 that resides in address space #5.</td> 2216 </tr> 2217</table> 2218 2219</div> 2220 2221<!-- _______________________________________________________________________ --> 2222<h4> 2223 <a name="t_vector">Vector Type</a> 2224</h4> 2225 2226<div> 2227 2228<h5>Overview:</h5> 2229<p>A vector type is a simple derived type that represents a vector of elements. 2230 Vector types are used when multiple primitive data are operated in parallel 2231 using a single instruction (SIMD). A vector type requires a size (number of 2232 elements) and an underlying primitive data type. Vector types are considered 2233 <a href="#t_firstclass">first class</a>.</p> 2234 2235<h5>Syntax:</h5> 2236<pre> 2237 < <# elements> x <elementtype> > 2238</pre> 2239 2240<p>The number of elements is a constant integer value larger than 0; elementtype 2241 may be any integer or floating point type, or a pointer to these types. 2242 Vectors of size zero are not allowed. </p> 2243 2244<h5>Examples:</h5> 2245<table class="layout"> 2246 <tr class="layout"> 2247 <td class="left"><tt><4 x i32></tt></td> 2248 <td class="left">Vector of 4 32-bit integer values.</td> 2249 </tr> 2250 <tr class="layout"> 2251 <td class="left"><tt><8 x float></tt></td> 2252 <td class="left">Vector of 8 32-bit floating-point values.</td> 2253 </tr> 2254 <tr class="layout"> 2255 <td class="left"><tt><2 x i64></tt></td> 2256 <td class="left">Vector of 2 64-bit integer values.</td> 2257 </tr> 2258 <tr class="layout"> 2259 <td class="left"><tt><4 x i64*></tt></td> 2260 <td class="left">Vector of 4 pointers to 64-bit integer values.</td> 2261 </tr> 2262</table> 2263 2264</div> 2265 2266</div> 2267 2268</div> 2269 2270<!-- *********************************************************************** --> 2271<h2><a name="constants">Constants</a></h2> 2272<!-- *********************************************************************** --> 2273 2274<div> 2275 2276<p>LLVM has several different basic types of constants. This section describes 2277 them all and their syntax.</p> 2278 2279<!-- ======================================================================= --> 2280<h3> 2281 <a name="simpleconstants">Simple Constants</a> 2282</h3> 2283 2284<div> 2285 2286<dl> 2287 <dt><b>Boolean constants</b></dt> 2288 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid 2289 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd> 2290 2291 <dt><b>Integer constants</b></dt> 2292 <dd>Standard integers (such as '4') are constants of 2293 the <a href="#t_integer">integer</a> type. Negative numbers may be used 2294 with integer types.</dd> 2295 2296 <dt><b>Floating point constants</b></dt> 2297 <dd>Floating point constants use standard decimal notation (e.g. 123.421), 2298 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal 2299 notation (see below). The assembler requires the exact decimal value of a 2300 floating-point constant. For example, the assembler accepts 1.25 but 2301 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point 2302 constants must have a <a href="#t_floating">floating point</a> type. </dd> 2303 2304 <dt><b>Null pointer constants</b></dt> 2305 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant 2306 and must be of <a href="#t_pointer">pointer type</a>.</dd> 2307</dl> 2308 2309<p>The one non-intuitive notation for constants is the hexadecimal form of 2310 floating point constants. For example, the form '<tt>double 2311 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) 2312 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point 2313 constants are required (and the only time that they are generated by the 2314 disassembler) is when a floating point constant must be emitted but it cannot 2315 be represented as a decimal floating point number in a reasonable number of 2316 digits. For example, NaN's, infinities, and other special values are 2317 represented in their IEEE hexadecimal format so that assembly and disassembly 2318 do not cause any bits to change in the constants.</p> 2319 2320<p>When using the hexadecimal form, constants of types half, float, and double are 2321 represented using the 16-digit form shown above (which matches the IEEE754 2322 representation for double); half and float values must, however, be exactly 2323 representable as IEE754 half and single precision, respectively. 2324 Hexadecimal format is always used 2325 for long double, and there are three forms of long double. The 80-bit format 2326 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits. 2327 The 128-bit format used by PowerPC (two adjacent doubles) is represented 2328 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format 2329 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no 2330 currently supported target uses this format. Long doubles will only work if 2331 they match the long double format on your target. The IEEE 16-bit format 2332 (half precision) is represented by <tt>0xH</tt> followed by 4 hexadecimal 2333 digits. All hexadecimal formats are big-endian (sign bit at the left).</p> 2334 2335<p>There are no constants of type x86mmx.</p> 2336</div> 2337 2338<!-- ======================================================================= --> 2339<h3> 2340<a name="aggregateconstants"></a> <!-- old anchor --> 2341<a name="complexconstants">Complex Constants</a> 2342</h3> 2343 2344<div> 2345 2346<p>Complex constants are a (potentially recursive) combination of simple 2347 constants and smaller complex constants.</p> 2348 2349<dl> 2350 <dt><b>Structure constants</b></dt> 2351 <dd>Structure constants are represented with notation similar to structure 2352 type definitions (a comma separated list of elements, surrounded by braces 2353 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>", 2354 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". 2355 Structure constants must have <a href="#t_struct">structure type</a>, and 2356 the number and types of elements must match those specified by the 2357 type.</dd> 2358 2359 <dt><b>Array constants</b></dt> 2360 <dd>Array constants are represented with notation similar to array type 2361 definitions (a comma separated list of elements, surrounded by square 2362 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 2363 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and 2364 the number and types of elements must match those specified by the 2365 type.</dd> 2366 2367 <dt><b>Vector constants</b></dt> 2368 <dd>Vector constants are represented with notation similar to vector type 2369 definitions (a comma separated list of elements, surrounded by 2370 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 2371 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must 2372 have <a href="#t_vector">vector type</a>, and the number and types of 2373 elements must match those specified by the type.</dd> 2374 2375 <dt><b>Zero initialization</b></dt> 2376 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a 2377 value to zero of <em>any</em> type, including scalar and 2378 <a href="#t_aggregate">aggregate</a> types. 2379 This is often used to avoid having to print large zero initializers 2380 (e.g. for large arrays) and is always exactly equivalent to using explicit 2381 zero initializers.</dd> 2382 2383 <dt><b>Metadata node</b></dt> 2384 <dd>A metadata node is a structure-like constant with 2385 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{ 2386 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to 2387 be interpreted as part of the instruction stream, metadata is a place to 2388 attach additional information such as debug info.</dd> 2389</dl> 2390 2391</div> 2392 2393<!-- ======================================================================= --> 2394<h3> 2395 <a name="globalconstants">Global Variable and Function Addresses</a> 2396</h3> 2397 2398<div> 2399 2400<p>The addresses of <a href="#globalvars">global variables</a> 2401 and <a href="#functionstructure">functions</a> are always implicitly valid 2402 (link-time) constants. These constants are explicitly referenced when 2403 the <a href="#identifiers">identifier for the global</a> is used and always 2404 have <a href="#t_pointer">pointer</a> type. For example, the following is a 2405 legal LLVM file:</p> 2406 2407<pre class="doc_code"> 2408@X = global i32 17 2409@Y = global i32 42 2410@Z = global [2 x i32*] [ i32* @X, i32* @Y ] 2411</pre> 2412 2413</div> 2414 2415<!-- ======================================================================= --> 2416<h3> 2417 <a name="undefvalues">Undefined Values</a> 2418</h3> 2419 2420<div> 2421 2422<p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and 2423 indicates that the user of the value may receive an unspecified bit-pattern. 2424 Undefined values may be of any type (other than '<tt>label</tt>' 2425 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p> 2426 2427<p>Undefined values are useful because they indicate to the compiler that the 2428 program is well defined no matter what value is used. This gives the 2429 compiler more freedom to optimize. Here are some examples of (potentially 2430 surprising) transformations that are valid (in pseudo IR):</p> 2431 2432 2433<pre class="doc_code"> 2434 %A = add %X, undef 2435 %B = sub %X, undef 2436 %C = xor %X, undef 2437Safe: 2438 %A = undef 2439 %B = undef 2440 %C = undef 2441</pre> 2442 2443<p>This is safe because all of the output bits are affected by the undef bits. 2444 Any output bit can have a zero or one depending on the input bits.</p> 2445 2446<pre class="doc_code"> 2447 %A = or %X, undef 2448 %B = and %X, undef 2449Safe: 2450 %A = -1 2451 %B = 0 2452Unsafe: 2453 %A = undef 2454 %B = undef 2455</pre> 2456 2457<p>These logical operations have bits that are not always affected by the input. 2458 For example, if <tt>%X</tt> has a zero bit, then the output of the 2459 '<tt>and</tt>' operation will always be a zero for that bit, no matter what 2460 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to 2461 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'. 2462 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be 2463 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that 2464 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be 2465 set, allowing the '<tt>or</tt>' to be folded to -1.</p> 2466 2467<pre class="doc_code"> 2468 %A = select undef, %X, %Y 2469 %B = select undef, 42, %Y 2470 %C = select %X, %Y, undef 2471Safe: 2472 %A = %X (or %Y) 2473 %B = 42 (or %Y) 2474 %C = %Y 2475Unsafe: 2476 %A = undef 2477 %B = undef 2478 %C = undef 2479</pre> 2480 2481<p>This set of examples shows that undefined '<tt>select</tt>' (and conditional 2482 branch) conditions can go <em>either way</em>, but they have to come from one 2483 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and 2484 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would 2485 have to have a cleared low bit. However, in the <tt>%C</tt> example, the 2486 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the 2487 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be 2488 eliminated.</p> 2489 2490<pre class="doc_code"> 2491 %A = xor undef, undef 2492 2493 %B = undef 2494 %C = xor %B, %B 2495 2496 %D = undef 2497 %E = icmp lt %D, 4 2498 %F = icmp gte %D, 4 2499 2500Safe: 2501 %A = undef 2502 %B = undef 2503 %C = undef 2504 %D = undef 2505 %E = undef 2506 %F = undef 2507</pre> 2508 2509<p>This example points out that two '<tt>undef</tt>' operands are not 2510 necessarily the same. This can be surprising to people (and also matches C 2511 semantics) where they assume that "<tt>X^X</tt>" is always zero, even 2512 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the 2513 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change 2514 its value over its "live range". This is true because the variable doesn't 2515 actually <em>have a live range</em>. Instead, the value is logically read 2516 from arbitrary registers that happen to be around when needed, so the value 2517 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt> 2518 need to have the same semantics or the core LLVM "replace all uses with" 2519 concept would not hold.</p> 2520 2521<pre class="doc_code"> 2522 %A = fdiv undef, %X 2523 %B = fdiv %X, undef 2524Safe: 2525 %A = undef 2526b: unreachable 2527</pre> 2528 2529<p>These examples show the crucial difference between an <em>undefined 2530 value</em> and <em>undefined behavior</em>. An undefined value (like 2531 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that 2532 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because 2533 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently) 2534 defined on SNaN's. However, in the second example, we can make a more 2535 aggressive assumption: because the <tt>undef</tt> is allowed to be an 2536 arbitrary value, we are allowed to assume that it could be zero. Since a 2537 divide by zero has <em>undefined behavior</em>, we are allowed to assume that 2538 the operation does not execute at all. This allows us to delete the divide and 2539 all code after it. Because the undefined operation "can't happen", the 2540 optimizer can assume that it occurs in dead code.</p> 2541 2542<pre class="doc_code"> 2543a: store undef -> %X 2544b: store %X -> undef 2545Safe: 2546a: <deleted> 2547b: unreachable 2548</pre> 2549 2550<p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an 2551 undefined value can be assumed to not have any effect; we can assume that the 2552 value is overwritten with bits that happen to match what was already there. 2553 However, a store <em>to</em> an undefined location could clobber arbitrary 2554 memory, therefore, it has undefined behavior.</p> 2555 2556</div> 2557 2558<!-- ======================================================================= --> 2559<h3> 2560 <a name="poisonvalues">Poison Values</a> 2561</h3> 2562 2563<div> 2564 2565<p>Poison values are similar to <a href="#undefvalues">undef values</a>, however 2566 they also represent the fact that an instruction or constant expression which 2567 cannot evoke side effects has nevertheless detected a condition which results 2568 in undefined behavior.</p> 2569 2570<p>There is currently no way of representing a poison value in the IR; they 2571 only exist when produced by operations such as 2572 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p> 2573 2574<p>Poison value behavior is defined in terms of value <i>dependence</i>:</p> 2575 2576<ul> 2577<li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on 2578 their operands.</li> 2579 2580<li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding 2581 to their dynamic predecessor basic block.</li> 2582 2583<li>Function arguments depend on the corresponding actual argument values in 2584 the dynamic callers of their functions.</li> 2585 2586<li><a href="#i_call"><tt>Call</tt></a> instructions depend on the 2587 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer 2588 control back to them.</li> 2589 2590<li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the 2591 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>, 2592 or exception-throwing call instructions that dynamically transfer control 2593 back to them.</li> 2594 2595<li>Non-volatile loads and stores depend on the most recent stores to all of the 2596 referenced memory addresses, following the order in the IR 2597 (including loads and stores implied by intrinsics such as 2598 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li> 2599 2600<!-- TODO: In the case of multiple threads, this only applies if the store 2601 "happens-before" the load or store. --> 2602 2603<!-- TODO: floating-point exception state --> 2604 2605<li>An instruction with externally visible side effects depends on the most 2606 recent preceding instruction with externally visible side effects, following 2607 the order in the IR. (This includes 2608 <a href="#volatile">volatile operations</a>.)</li> 2609 2610<li>An instruction <i>control-depends</i> on a 2611 <a href="#terminators">terminator instruction</a> 2612 if the terminator instruction has multiple successors and the instruction 2613 is always executed when control transfers to one of the successors, and 2614 may not be executed when control is transferred to another.</li> 2615 2616<li>Additionally, an instruction also <i>control-depends</i> on a terminator 2617 instruction if the set of instructions it otherwise depends on would be 2618 different if the terminator had transferred control to a different 2619 successor.</li> 2620 2621<li>Dependence is transitive.</li> 2622 2623</ul> 2624 2625<p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>, 2626 with the additional affect that any instruction which has a <i>dependence</i> 2627 on a poison value has undefined behavior.</p> 2628 2629<p>Here are some examples:</p> 2630 2631<pre class="doc_code"> 2632entry: 2633 %poison = sub nuw i32 0, 1 ; Results in a poison value. 2634 %still_poison = and i32 %poison, 0 ; 0, but also poison. 2635 %poison_yet_again = getelementptr i32* @h, i32 %still_poison 2636 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned 2637 2638 store i32 %poison, i32* @g ; Poison value stored to memory. 2639 %poison2 = load i32* @g ; Poison value loaded back from memory. 2640 2641 store volatile i32 %poison, i32* @g ; External observation; undefined behavior. 2642 2643 %narrowaddr = bitcast i32* @g to i16* 2644 %wideaddr = bitcast i32* @g to i64* 2645 %poison3 = load i16* %narrowaddr ; Returns a poison value. 2646 %poison4 = load i64* %wideaddr ; Returns a poison value. 2647 2648 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value. 2649 br i1 %cmp, label %true, label %end ; Branch to either destination. 2650 2651true: 2652 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so 2653 ; it has undefined behavior. 2654 br label %end 2655 2656end: 2657 %p = phi i32 [ 0, %entry ], [ 1, %true ] 2658 ; Both edges into this PHI are 2659 ; control-dependent on %cmp, so this 2660 ; always results in a poison value. 2661 2662 store volatile i32 0, i32* @g ; This would depend on the store in %true 2663 ; if %cmp is true, or the store in %entry 2664 ; otherwise, so this is undefined behavior. 2665 2666 br i1 %cmp, label %second_true, label %second_end 2667 ; The same branch again, but this time the 2668 ; true block doesn't have side effects. 2669 2670second_true: 2671 ; No side effects! 2672 ret void 2673 2674second_end: 2675 store volatile i32 0, i32* @g ; This time, the instruction always depends 2676 ; on the store in %end. Also, it is 2677 ; control-equivalent to %end, so this is 2678 ; well-defined (ignoring earlier undefined 2679 ; behavior in this example). 2680</pre> 2681 2682</div> 2683 2684<!-- ======================================================================= --> 2685<h3> 2686 <a name="blockaddress">Addresses of Basic Blocks</a> 2687</h3> 2688 2689<div> 2690 2691<p><b><tt>blockaddress(@function, %block)</tt></b></p> 2692 2693<p>The '<tt>blockaddress</tt>' constant computes the address of the specified 2694 basic block in the specified function, and always has an i8* type. Taking 2695 the address of the entry block is illegal.</p> 2696 2697<p>This value only has defined behavior when used as an operand to the 2698 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for 2699 comparisons against null. Pointer equality tests between labels addresses 2700 results in undefined behavior — though, again, comparison against null 2701 is ok, and no label is equal to the null pointer. This may be passed around 2702 as an opaque pointer sized value as long as the bits are not inspected. This 2703 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so 2704 long as the original value is reconstituted before the <tt>indirectbr</tt> 2705 instruction.</p> 2706 2707<p>Finally, some targets may provide defined semantics when using the value as 2708 the operand to an inline assembly, but that is target specific.</p> 2709 2710</div> 2711 2712 2713<!-- ======================================================================= --> 2714<h3> 2715 <a name="constantexprs">Constant Expressions</a> 2716</h3> 2717 2718<div> 2719 2720<p>Constant expressions are used to allow expressions involving other constants 2721 to be used as constants. Constant expressions may be of 2722 any <a href="#t_firstclass">first class</a> type and may involve any LLVM 2723 operation that does not have side effects (e.g. load and call are not 2724 supported). The following is the syntax for constant expressions:</p> 2725 2726<dl> 2727 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt> 2728 <dd>Truncate a constant to another type. The bit size of CST must be larger 2729 than the bit size of TYPE. Both types must be integers.</dd> 2730 2731 <dt><b><tt>zext (CST to TYPE)</tt></b></dt> 2732 <dd>Zero extend a constant to another type. The bit size of CST must be 2733 smaller than the bit size of TYPE. Both types must be integers.</dd> 2734 2735 <dt><b><tt>sext (CST to TYPE)</tt></b></dt> 2736 <dd>Sign extend a constant to another type. The bit size of CST must be 2737 smaller than the bit size of TYPE. Both types must be integers.</dd> 2738 2739 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt> 2740 <dd>Truncate a floating point constant to another floating point type. The 2741 size of CST must be larger than the size of TYPE. Both types must be 2742 floating point.</dd> 2743 2744 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt> 2745 <dd>Floating point extend a constant to another type. The size of CST must be 2746 smaller or equal to the size of TYPE. Both types must be floating 2747 point.</dd> 2748 2749 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt> 2750 <dd>Convert a floating point constant to the corresponding unsigned integer 2751 constant. TYPE must be a scalar or vector integer type. CST must be of 2752 scalar or vector floating point type. Both CST and TYPE must be scalars, 2753 or vectors of the same number of elements. If the value won't fit in the 2754 integer type, the results are undefined.</dd> 2755 2756 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt> 2757 <dd>Convert a floating point constant to the corresponding signed integer 2758 constant. TYPE must be a scalar or vector integer type. CST must be of 2759 scalar or vector floating point type. Both CST and TYPE must be scalars, 2760 or vectors of the same number of elements. If the value won't fit in the 2761 integer type, the results are undefined.</dd> 2762 2763 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt> 2764 <dd>Convert an unsigned integer constant to the corresponding floating point 2765 constant. TYPE must be a scalar or vector floating point type. CST must be 2766 of scalar or vector integer type. Both CST and TYPE must be scalars, or 2767 vectors of the same number of elements. If the value won't fit in the 2768 floating point type, the results are undefined.</dd> 2769 2770 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt> 2771 <dd>Convert a signed integer constant to the corresponding floating point 2772 constant. TYPE must be a scalar or vector floating point type. CST must be 2773 of scalar or vector integer type. Both CST and TYPE must be scalars, or 2774 vectors of the same number of elements. If the value won't fit in the 2775 floating point type, the results are undefined.</dd> 2776 2777 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt> 2778 <dd>Convert a pointer typed constant to the corresponding integer constant 2779 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer 2780 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to 2781 make it fit in <tt>TYPE</tt>.</dd> 2782 2783 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt> 2784 <dd>Convert an integer constant to a pointer constant. TYPE must be a pointer 2785 type. CST must be of integer type. The CST value is zero extended, 2786 truncated, or unchanged to make it fit in a pointer size. This one is 2787 <i>really</i> dangerous!</dd> 2788 2789 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt> 2790 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands 2791 are the same as those for the <a href="#i_bitcast">bitcast 2792 instruction</a>.</dd> 2793 2794 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt> 2795 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt> 2796 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on 2797 constants. As with the <a href="#i_getelementptr">getelementptr</a> 2798 instruction, the index list may have zero or more indexes, which are 2799 required to make sense for the type of "CSTPTR".</dd> 2800 2801 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt> 2802 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd> 2803 2804 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt> 2805 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd> 2806 2807 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt> 2808 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd> 2809 2810 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt> 2811 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on 2812 constants.</dd> 2813 2814 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt> 2815 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on 2816 constants.</dd> 2817 2818 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt> 2819 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on 2820 constants.</dd> 2821 2822 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt> 2823 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on 2824 constants. The index list is interpreted in a similar manner as indices in 2825 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one 2826 index value must be specified.</dd> 2827 2828 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt> 2829 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on 2830 constants. The index list is interpreted in a similar manner as indices in 2831 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one 2832 index value must be specified.</dd> 2833 2834 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt> 2835 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may 2836 be any of the <a href="#binaryops">binary</a> 2837 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints 2838 on operands are the same as those for the corresponding instruction 2839 (e.g. no bitwise operations on floating point values are allowed).</dd> 2840</dl> 2841 2842</div> 2843 2844</div> 2845 2846<!-- *********************************************************************** --> 2847<h2><a name="othervalues">Other Values</a></h2> 2848<!-- *********************************************************************** --> 2849<div> 2850<!-- ======================================================================= --> 2851<h3> 2852<a name="inlineasm">Inline Assembler Expressions</a> 2853</h3> 2854 2855<div> 2856 2857<p>LLVM supports inline assembler expressions (as opposed 2858 to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of 2859 a special value. This value represents the inline assembler as a string 2860 (containing the instructions to emit), a list of operand constraints (stored 2861 as a string), a flag that indicates whether or not the inline asm 2862 expression has side effects, and a flag indicating whether the function 2863 containing the asm needs to align its stack conservatively. An example 2864 inline assembler expression is:</p> 2865 2866<pre class="doc_code"> 2867i32 (i32) asm "bswap $0", "=r,r" 2868</pre> 2869 2870<p>Inline assembler expressions may <b>only</b> be used as the callee operand of 2871 a <a href="#i_call"><tt>call</tt></a> or an 2872 <a href="#i_invoke"><tt>invoke</tt></a> instruction. 2873 Thus, typically we have:</p> 2874 2875<pre class="doc_code"> 2876%X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y) 2877</pre> 2878 2879<p>Inline asms with side effects not visible in the constraint list must be 2880 marked as having side effects. This is done through the use of the 2881 '<tt>sideeffect</tt>' keyword, like so:</p> 2882 2883<pre class="doc_code"> 2884call void asm sideeffect "eieio", ""() 2885</pre> 2886 2887<p>In some cases inline asms will contain code that will not work unless the 2888 stack is aligned in some way, such as calls or SSE instructions on x86, 2889 yet will not contain code that does that alignment within the asm. 2890 The compiler should make conservative assumptions about what the asm might 2891 contain and should generate its usual stack alignment code in the prologue 2892 if the '<tt>alignstack</tt>' keyword is present:</p> 2893 2894<pre class="doc_code"> 2895call void asm alignstack "eieio", ""() 2896</pre> 2897 2898<p>Inline asms also support using non-standard assembly dialects. The assumed 2899 dialect is ATT. When the '<tt>inteldialect</tt>' keyword is present, the 2900 inline asm is using the Intel dialect. Currently, ATT and Intel are the 2901 only supported dialects. An example is:</p> 2902 2903<pre class="doc_code"> 2904call void asm inteldialect "eieio", ""() 2905</pre> 2906 2907<p>If multiple keywords appear the '<tt>sideeffect</tt>' keyword must come 2908 first, the '<tt>alignstack</tt>' keyword second and the 2909 '<tt>inteldialect</tt>' keyword last.</p> 2910 2911<!-- 2912<p>TODO: The format of the asm and constraints string still need to be 2913 documented here. Constraints on what can be done (e.g. duplication, moving, 2914 etc need to be documented). This is probably best done by reference to 2915 another document that covers inline asm from a holistic perspective.</p> 2916 --> 2917 2918<!-- _______________________________________________________________________ --> 2919<h4> 2920 <a name="inlineasm_md">Inline Asm Metadata</a> 2921</h4> 2922 2923<div> 2924 2925<p>The call instructions that wrap inline asm nodes may have a 2926 "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant 2927 integers. If present, the code generator will use the integer as the 2928 location cookie value when report errors through the <tt>LLVMContext</tt> 2929 error reporting mechanisms. This allows a front-end to correlate backend 2930 errors that occur with inline asm back to the source code that produced it. 2931 For example:</p> 2932 2933<pre class="doc_code"> 2934call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b> 2935... 2936!42 = !{ i32 1234567 } 2937</pre> 2938 2939<p>It is up to the front-end to make sense of the magic numbers it places in the 2940 IR. If the MDNode contains multiple constants, the code generator will use 2941 the one that corresponds to the line of the asm that the error occurs on.</p> 2942 2943</div> 2944 2945</div> 2946 2947<!-- ======================================================================= --> 2948<h3> 2949 <a name="metadata">Metadata Nodes and Metadata Strings</a> 2950</h3> 2951 2952<div> 2953 2954<p>LLVM IR allows metadata to be attached to instructions in the program that 2955 can convey extra information about the code to the optimizers and code 2956 generator. One example application of metadata is source-level debug 2957 information. There are two metadata primitives: strings and nodes. All 2958 metadata has the <tt>metadata</tt> type and is identified in syntax by a 2959 preceding exclamation point ('<tt>!</tt>').</p> 2960 2961<p>A metadata string is a string surrounded by double quotes. It can contain 2962 any character by escaping non-printable characters with "<tt>\xx</tt>" where 2963 "<tt>xx</tt>" is the two digit hex code. For example: 2964 "<tt>!"test\00"</tt>".</p> 2965 2966<p>Metadata nodes are represented with notation similar to structure constants 2967 (a comma separated list of elements, surrounded by braces and preceded by an 2968 exclamation point). Metadata nodes can have any values as their operand. For 2969 example:</p> 2970 2971<div class="doc_code"> 2972<pre> 2973!{ metadata !"test\00", i32 10} 2974</pre> 2975</div> 2976 2977<p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of 2978 metadata nodes, which can be looked up in the module symbol table. For 2979 example:</p> 2980 2981<div class="doc_code"> 2982<pre> 2983!foo = metadata !{!4, !3} 2984</pre> 2985</div> 2986 2987<p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt> 2988 function is using two metadata arguments:</p> 2989 2990<div class="doc_code"> 2991<pre> 2992call void @llvm.dbg.value(metadata !24, i64 0, metadata !25) 2993</pre> 2994</div> 2995 2996<p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is 2997 attached to the <tt>add</tt> instruction using the <tt>!dbg</tt> 2998 identifier:</p> 2999 3000<div class="doc_code"> 3001<pre> 3002%indvar.next = add i64 %indvar, 1, !dbg !21 3003</pre> 3004</div> 3005 3006<p>More information about specific metadata nodes recognized by the optimizers 3007 and code generator is found below.</p> 3008 3009<!-- _______________________________________________________________________ --> 3010<h4> 3011 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a> 3012</h4> 3013 3014<div> 3015 3016<p>In LLVM IR, memory does not have types, so LLVM's own type system is not 3017 suitable for doing TBAA. Instead, metadata is added to the IR to describe 3018 a type system of a higher level language. This can be used to implement 3019 typical C/C++ TBAA, but it can also be used to implement custom alias 3020 analysis behavior for other languages.</p> 3021 3022<p>The current metadata format is very simple. TBAA metadata nodes have up to 3023 three fields, e.g.:</p> 3024 3025<div class="doc_code"> 3026<pre> 3027!0 = metadata !{ metadata !"an example type tree" } 3028!1 = metadata !{ metadata !"int", metadata !0 } 3029!2 = metadata !{ metadata !"float", metadata !0 } 3030!3 = metadata !{ metadata !"const float", metadata !2, i64 1 } 3031</pre> 3032</div> 3033 3034<p>The first field is an identity field. It can be any value, usually 3035 a metadata string, which uniquely identifies the type. The most important 3036 name in the tree is the name of the root node. Two trees with 3037 different root node names are entirely disjoint, even if they 3038 have leaves with common names.</p> 3039 3040<p>The second field identifies the type's parent node in the tree, or 3041 is null or omitted for a root node. A type is considered to alias 3042 all of its descendants and all of its ancestors in the tree. Also, 3043 a type is considered to alias all types in other trees, so that 3044 bitcode produced from multiple front-ends is handled conservatively.</p> 3045 3046<p>If the third field is present, it's an integer which if equal to 1 3047 indicates that the type is "constant" (meaning 3048 <tt>pointsToConstantMemory</tt> should return true; see 3049 <a href="AliasAnalysis.html#OtherItfs">other useful 3050 <tt>AliasAnalysis</tt> methods</a>).</p> 3051 3052</div> 3053 3054<!-- _______________________________________________________________________ --> 3055<h4> 3056 <a name="tbaa.struct">'<tt>tbaa.struct</tt>' Metadata</a> 3057</h4> 3058 3059<div> 3060 3061<p>The <a href="#int_memcpy"><tt>llvm.memcpy</tt></a> is often used to implement 3062aggregate assignment operations in C and similar languages, however it is 3063defined to copy a contiguous region of memory, which is more than strictly 3064necessary for aggregate types which contain holes due to padding. Also, it 3065doesn't contain any TBAA information about the fields of the aggregate.</p> 3066 3067<p><tt>!tbaa.struct</tt> metadata can describe which memory subregions in a memcpy 3068are padding and what the TBAA tags of the struct are.</p> 3069 3070<p>The current metadata format is very simple. <tt>!tbaa.struct</tt> metadata nodes 3071 are a list of operands which are in conceptual groups of three. For each 3072 group of three, the first operand gives the byte offset of a field in bytes, 3073 the second gives its size in bytes, and the third gives its 3074 tbaa tag. e.g.:</p> 3075 3076<div class="doc_code"> 3077<pre> 3078!4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 } 3079</pre> 3080</div> 3081 3082<p>This describes a struct with two fields. The first is at offset 0 bytes 3083 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes 3084 and has size 4 bytes and has tbaa tag !2.</p> 3085 3086<p>Note that the fields need not be contiguous. In this example, there is a 3087 4 byte gap between the two fields. This gap represents padding which 3088 does not carry useful data and need not be preserved.</p> 3089 3090</div> 3091 3092<!-- _______________________________________________________________________ --> 3093<h4> 3094 <a name="fpmath">'<tt>fpmath</tt>' Metadata</a> 3095</h4> 3096 3097<div> 3098 3099<p><tt>fpmath</tt> metadata may be attached to any instruction of floating point 3100 type. It can be used to express the maximum acceptable error in the result of 3101 that instruction, in ULPs, thus potentially allowing the compiler to use a 3102 more efficient but less accurate method of computing it. ULP is defined as 3103 follows:</p> 3104 3105<blockquote> 3106 3107<p>If <tt>x</tt> is a real number that lies between two finite consecutive 3108 floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one 3109 of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the 3110 distance between the two non-equal finite floating-point numbers nearest 3111 <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p> 3112 3113</blockquote> 3114 3115<p>The metadata node shall consist of a single positive floating point number 3116 representing the maximum relative error, for example:</p> 3117 3118<div class="doc_code"> 3119<pre> 3120!0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs 3121</pre> 3122</div> 3123 3124</div> 3125 3126<!-- _______________________________________________________________________ --> 3127<h4> 3128 <a name="range">'<tt>range</tt>' Metadata</a> 3129</h4> 3130 3131<div> 3132<p><tt>range</tt> metadata may be attached only to loads of integer types. It 3133 expresses the possible ranges the loaded value is in. The ranges are 3134 represented with a flattened list of integers. The loaded value is known to 3135 be in the union of the ranges defined by each consecutive pair. Each pair 3136 has the following properties:</p> 3137<ul> 3138 <li>The type must match the type loaded by the instruction.</li> 3139 <li>The pair <tt>a,b</tt> represents the range <tt>[a,b)</tt>.</li> 3140 <li>Both <tt>a</tt> and <tt>b</tt> are constants.</li> 3141 <li>The range is allowed to wrap.</li> 3142 <li>The range should not represent the full or empty set. That is, 3143 <tt>a!=b</tt>. </li> 3144</ul> 3145<p> In addition, the pairs must be in signed order of the lower bound and 3146 they must be non-contiguous.</p> 3147 3148<p>Examples:</p> 3149<div class="doc_code"> 3150<pre> 3151 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1 3152 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1 3153 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5 3154 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5 3155... 3156!0 = metadata !{ i8 0, i8 2 } 3157!1 = metadata !{ i8 255, i8 2 } 3158!2 = metadata !{ i8 0, i8 2, i8 3, i8 6 } 3159!3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 } 3160</pre> 3161</div> 3162</div> 3163</div> 3164 3165</div> 3166 3167<!-- *********************************************************************** --> 3168<h2> 3169 <a name="module_flags">Module Flags Metadata</a> 3170</h2> 3171<!-- *********************************************************************** --> 3172 3173<div> 3174 3175<p>Information about the module as a whole is difficult to convey to LLVM's 3176 subsystems. The LLVM IR isn't sufficient to transmit this 3177 information. The <tt>llvm.module.flags</tt> named metadata exists in order to 3178 facilitate this. These flags are in the form of key / value pairs — 3179 much like a dictionary — making it easy for any subsystem who cares 3180 about a flag to look it up.</p> 3181 3182<p>The <tt>llvm.module.flags</tt> metadata contains a list of metadata 3183 triplets. Each triplet has the following form:</p> 3184 3185<ul> 3186 <li>The first element is a <i>behavior</i> flag, which specifies the behavior 3187 when two (or more) modules are merged together, and it encounters two (or 3188 more) metadata with the same ID. The supported behaviors are described 3189 below.</li> 3190 3191 <li>The second element is a metadata string that is a unique ID for the 3192 metadata. How each ID is interpreted is documented below.</li> 3193 3194 <li>The third element is the value of the flag.</li> 3195</ul> 3196 3197<p>When two (or more) modules are merged together, the resulting 3198 <tt>llvm.module.flags</tt> metadata is the union of the 3199 modules' <tt>llvm.module.flags</tt> metadata. The only exception being a flag 3200 with the <i>Override</i> behavior, which may override another flag's value 3201 (see below).</p> 3202 3203<p>The following behaviors are supported:</p> 3204 3205<table border="1" cellspacing="0" cellpadding="4"> 3206 <tbody> 3207 <tr> 3208 <th>Value</th> 3209 <th>Behavior</th> 3210 </tr> 3211 <tr> 3212 <td>1</td> 3213 <td align="left"> 3214 <dl> 3215 <dt><b>Error</b></dt> 3216 <dd>Emits an error if two values disagree. It is an error to have an ID 3217 with both an Error and a Warning behavior.</dd> 3218 </dl> 3219 </td> 3220 </tr> 3221 <tr> 3222 <td>2</td> 3223 <td align="left"> 3224 <dl> 3225 <dt><b>Warning</b></dt> 3226 <dd>Emits a warning if two values disagree.</dd> 3227 </dl> 3228 </td> 3229 </tr> 3230 <tr> 3231 <td>3</td> 3232 <td align="left"> 3233 <dl> 3234 <dt><b>Require</b></dt> 3235 <dd>Emits an error when the specified value is not present or doesn't 3236 have the specified value. It is an error for two (or more) 3237 <tt>llvm.module.flags</tt> with the same ID to have the Require 3238 behavior but different values. There may be multiple Require flags 3239 per ID.</dd> 3240 </dl> 3241 </td> 3242 </tr> 3243 <tr> 3244 <td>4</td> 3245 <td align="left"> 3246 <dl> 3247 <dt><b>Override</b></dt> 3248 <dd>Uses the specified value if the two values disagree. It is an 3249 error for two (or more) <tt>llvm.module.flags</tt> with the same 3250 ID to have the Override behavior but different values.</dd> 3251 </dl> 3252 </td> 3253 </tr> 3254 </tbody> 3255</table> 3256 3257<p>An example of module flags:</p> 3258 3259<pre class="doc_code"> 3260!0 = metadata !{ i32 1, metadata !"foo", i32 1 } 3261!1 = metadata !{ i32 4, metadata !"bar", i32 37 } 3262!2 = metadata !{ i32 2, metadata !"qux", i32 42 } 3263!3 = metadata !{ i32 3, metadata !"qux", 3264 metadata !{ 3265 metadata !"foo", i32 1 3266 } 3267} 3268!llvm.module.flags = !{ !0, !1, !2, !3 } 3269</pre> 3270 3271<ul> 3272 <li><p>Metadata <tt>!0</tt> has the ID <tt>!"foo"</tt> and the value '1'. The 3273 behavior if two or more <tt>!"foo"</tt> flags are seen is to emit an 3274 error if their values are not equal.</p></li> 3275 3276 <li><p>Metadata <tt>!1</tt> has the ID <tt>!"bar"</tt> and the value '37'. The 3277 behavior if two or more <tt>!"bar"</tt> flags are seen is to use the 3278 value '37' if their values are not equal.</p></li> 3279 3280 <li><p>Metadata <tt>!2</tt> has the ID <tt>!"qux"</tt> and the value '42'. The 3281 behavior if two or more <tt>!"qux"</tt> flags are seen is to emit a 3282 warning if their values are not equal.</p></li> 3283 3284 <li><p>Metadata <tt>!3</tt> has the ID <tt>!"qux"</tt> and the value:</p> 3285 3286<pre class="doc_code"> 3287metadata !{ metadata !"foo", i32 1 } 3288</pre> 3289 3290 <p>The behavior is to emit an error if the <tt>llvm.module.flags</tt> does 3291 not contain a flag with the ID <tt>!"foo"</tt> that has the value 3292 '1'. If two or more <tt>!"qux"</tt> flags exist, then they must have 3293 the same value or an error will be issued.</p></li> 3294</ul> 3295 3296 3297<!-- ======================================================================= --> 3298<h3> 3299<a name="objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a> 3300</h3> 3301 3302<div> 3303 3304<p>On the Mach-O platform, Objective-C stores metadata about garbage collection 3305 in a special section called "image info". The metadata consists of a version 3306 number and a bitmask specifying what types of garbage collection are 3307 supported (if any) by the file. If two or more modules are linked together 3308 their garbage collection metadata needs to be merged rather than appended 3309 together.</p> 3310 3311<p>The Objective-C garbage collection module flags metadata consists of the 3312 following key-value pairs:</p> 3313 3314<table border="1" cellspacing="0" cellpadding="4"> 3315 <col width="30%"> 3316 <tbody> 3317 <tr> 3318 <th>Key</th> 3319 <th>Value</th> 3320 </tr> 3321 <tr> 3322 <td><tt>Objective-C Version</tt></td> 3323 <td align="left"><b>[Required]</b> — The Objective-C ABI 3324 version. Valid values are 1 and 2.</td> 3325 </tr> 3326 <tr> 3327 <td><tt>Objective-C Image Info Version</tt></td> 3328 <td align="left"><b>[Required]</b> — The version of the image info 3329 section. Currently always 0.</td> 3330 </tr> 3331 <tr> 3332 <td><tt>Objective-C Image Info Section</tt></td> 3333 <td align="left"><b>[Required]</b> — The section to place the 3334 metadata. Valid values are <tt>"__OBJC, __image_info, regular"</tt> for 3335 Objective-C ABI version 1, and <tt>"__DATA,__objc_imageinfo, regular, 3336 no_dead_strip"</tt> for Objective-C ABI version 2.</td> 3337 </tr> 3338 <tr> 3339 <td><tt>Objective-C Garbage Collection</tt></td> 3340 <td align="left"><b>[Required]</b> — Specifies whether garbage 3341 collection is supported or not. Valid values are 0, for no garbage 3342 collection, and 2, for garbage collection supported.</td> 3343 </tr> 3344 <tr> 3345 <td><tt>Objective-C GC Only</tt></td> 3346 <td align="left"><b>[Optional]</b> — Specifies that only garbage 3347 collection is supported. If present, its value must be 6. This flag 3348 requires that the <tt>Objective-C Garbage Collection</tt> flag have the 3349 value 2.</td> 3350 </tr> 3351 </tbody> 3352</table> 3353 3354<p>Some important flag interactions:</p> 3355 3356<ul> 3357 <li>If a module with <tt>Objective-C Garbage Collection</tt> set to 0 is 3358 merged with a module with <tt>Objective-C Garbage Collection</tt> set to 3359 2, then the resulting module has the <tt>Objective-C Garbage 3360 Collection</tt> flag set to 0.</li> 3361 3362 <li>A module with <tt>Objective-C Garbage Collection</tt> set to 0 cannot be 3363 merged with a module with <tt>Objective-C GC Only</tt> set to 6.</li> 3364</ul> 3365 3366</div> 3367 3368</div> 3369 3370<!-- *********************************************************************** --> 3371<h2> 3372 <a name="intrinsic_globals">Intrinsic Global Variables</a> 3373</h2> 3374<!-- *********************************************************************** --> 3375<div> 3376<p>LLVM has a number of "magic" global variables that contain data that affect 3377code generation or other IR semantics. These are documented here. All globals 3378of this sort should have a section specified as "<tt>llvm.metadata</tt>". This 3379section and all globals that start with "<tt>llvm.</tt>" are reserved for use 3380by LLVM.</p> 3381 3382<!-- ======================================================================= --> 3383<h3> 3384<a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a> 3385</h3> 3386 3387<div> 3388 3389<p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a 3390href="#linkage_appending">appending linkage</a>. This array contains a list of 3391pointers to global variables and functions which may optionally have a pointer 3392cast formed of bitcast or getelementptr. For example, a legal use of it is:</p> 3393 3394<div class="doc_code"> 3395<pre> 3396@X = global i8 4 3397@Y = global i32 123 3398 3399@llvm.used = appending global [2 x i8*] [ 3400 i8* @X, 3401 i8* bitcast (i32* @Y to i8*) 3402], section "llvm.metadata" 3403</pre> 3404</div> 3405 3406<p>If a global variable appears in the <tt>@llvm.used</tt> list, then the 3407 compiler, assembler, and linker are required to treat the symbol as if there 3408 is a reference to the global that it cannot see. For example, if a variable 3409 has internal linkage and no references other than that from 3410 the <tt>@llvm.used</tt> list, it cannot be deleted. This is commonly used to 3411 represent references from inline asms and other things the compiler cannot 3412 "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p> 3413 3414<p>On some targets, the code generator must emit a directive to the assembler or 3415 object file to prevent the assembler and linker from molesting the 3416 symbol.</p> 3417 3418</div> 3419 3420<!-- ======================================================================= --> 3421<h3> 3422 <a name="intg_compiler_used"> 3423 The '<tt>llvm.compiler.used</tt>' Global Variable 3424 </a> 3425</h3> 3426 3427<div> 3428 3429<p>The <tt>@llvm.compiler.used</tt> directive is the same as the 3430 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from 3431 touching the symbol. On targets that support it, this allows an intelligent 3432 linker to optimize references to the symbol without being impeded as it would 3433 be by <tt>@llvm.used</tt>.</p> 3434 3435<p>This is a rare construct that should only be used in rare circumstances, and 3436 should not be exposed to source languages.</p> 3437 3438</div> 3439 3440<!-- ======================================================================= --> 3441<h3> 3442<a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a> 3443</h3> 3444 3445<div> 3446 3447<div class="doc_code"> 3448<pre> 3449%0 = type { i32, void ()* } 3450@llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }] 3451</pre> 3452</div> 3453 3454<p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor 3455 functions and associated priorities. The functions referenced by this array 3456 will be called in ascending order of priority (i.e. lowest first) when the 3457 module is loaded. The order of functions with the same priority is not 3458 defined.</p> 3459 3460</div> 3461 3462<!-- ======================================================================= --> 3463<h3> 3464<a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a> 3465</h3> 3466 3467<div> 3468 3469<div class="doc_code"> 3470<pre> 3471%0 = type { i32, void ()* } 3472@llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }] 3473</pre> 3474</div> 3475 3476<p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions 3477 and associated priorities. The functions referenced by this array will be 3478 called in descending order of priority (i.e. highest first) when the module 3479 is loaded. The order of functions with the same priority is not defined.</p> 3480 3481</div> 3482 3483</div> 3484 3485<!-- *********************************************************************** --> 3486<h2><a name="instref">Instruction Reference</a></h2> 3487<!-- *********************************************************************** --> 3488 3489<div> 3490 3491<p>The LLVM instruction set consists of several different classifications of 3492 instructions: <a href="#terminators">terminator 3493 instructions</a>, <a href="#binaryops">binary instructions</a>, 3494 <a href="#bitwiseops">bitwise binary instructions</a>, 3495 <a href="#memoryops">memory instructions</a>, and 3496 <a href="#otherops">other instructions</a>.</p> 3497 3498<!-- ======================================================================= --> 3499<h3> 3500 <a name="terminators">Terminator Instructions</a> 3501</h3> 3502 3503<div> 3504 3505<p>As mentioned <a href="#functionstructure">previously</a>, every basic block 3506 in a program ends with a "Terminator" instruction, which indicates which 3507 block should be executed after the current block is finished. These 3508 terminator instructions typically yield a '<tt>void</tt>' value: they produce 3509 control flow, not values (the one exception being the 3510 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p> 3511 3512<p>The terminator instructions are: 3513 '<a href="#i_ret"><tt>ret</tt></a>', 3514 '<a href="#i_br"><tt>br</tt></a>', 3515 '<a href="#i_switch"><tt>switch</tt></a>', 3516 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>', 3517 '<a href="#i_invoke"><tt>invoke</tt></a>', 3518 '<a href="#i_resume"><tt>resume</tt></a>', and 3519 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p> 3520 3521<!-- _______________________________________________________________________ --> 3522<h4> 3523 <a name="i_ret">'<tt>ret</tt>' Instruction</a> 3524</h4> 3525 3526<div> 3527 3528<h5>Syntax:</h5> 3529<pre> 3530 ret <type> <value> <i>; Return a value from a non-void function</i> 3531 ret void <i>; Return from void function</i> 3532</pre> 3533 3534<h5>Overview:</h5> 3535<p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally 3536 a value) from a function back to the caller.</p> 3537 3538<p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a 3539 value and then causes control flow, and one that just causes control flow to 3540 occur.</p> 3541 3542<h5>Arguments:</h5> 3543<p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the 3544 return value. The type of the return value must be a 3545 '<a href="#t_firstclass">first class</a>' type.</p> 3546 3547<p>A function is not <a href="#wellformed">well formed</a> if it it has a 3548 non-void return type and contains a '<tt>ret</tt>' instruction with no return 3549 value or a return value with a type that does not match its type, or if it 3550 has a void return type and contains a '<tt>ret</tt>' instruction with a 3551 return value.</p> 3552 3553<h5>Semantics:</h5> 3554<p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to 3555 the calling function's context. If the caller is a 3556 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the 3557 instruction after the call. If the caller was an 3558 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at 3559 the beginning of the "normal" destination block. If the instruction returns 3560 a value, that value shall set the call or invoke instruction's return 3561 value.</p> 3562 3563<h5>Example:</h5> 3564<pre> 3565 ret i32 5 <i>; Return an integer value of 5</i> 3566 ret void <i>; Return from a void function</i> 3567 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i> 3568</pre> 3569 3570</div> 3571<!-- _______________________________________________________________________ --> 3572<h4> 3573 <a name="i_br">'<tt>br</tt>' Instruction</a> 3574</h4> 3575 3576<div> 3577 3578<h5>Syntax:</h5> 3579<pre> 3580 br i1 <cond>, label <iftrue>, label <iffalse> 3581 br label <dest> <i>; Unconditional branch</i> 3582</pre> 3583 3584<h5>Overview:</h5> 3585<p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a 3586 different basic block in the current function. There are two forms of this 3587 instruction, corresponding to a conditional branch and an unconditional 3588 branch.</p> 3589 3590<h5>Arguments:</h5> 3591<p>The conditional branch form of the '<tt>br</tt>' instruction takes a single 3592 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form 3593 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a 3594 target.</p> 3595 3596<h5>Semantics:</h5> 3597<p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>' 3598 argument is evaluated. If the value is <tt>true</tt>, control flows to the 3599 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>, 3600 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p> 3601 3602<h5>Example:</h5> 3603<pre> 3604Test: 3605 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b 3606 br i1 %cond, label %IfEqual, label %IfUnequal 3607IfEqual: 3608 <a href="#i_ret">ret</a> i32 1 3609IfUnequal: 3610 <a href="#i_ret">ret</a> i32 0 3611</pre> 3612 3613</div> 3614 3615<!-- _______________________________________________________________________ --> 3616<h4> 3617 <a name="i_switch">'<tt>switch</tt>' Instruction</a> 3618</h4> 3619 3620<div> 3621 3622<h5>Syntax:</h5> 3623<pre> 3624 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ] 3625</pre> 3626 3627<h5>Overview:</h5> 3628<p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of 3629 several different places. It is a generalization of the '<tt>br</tt>' 3630 instruction, allowing a branch to occur to one of many possible 3631 destinations.</p> 3632 3633<h5>Arguments:</h5> 3634<p>The '<tt>switch</tt>' instruction uses three parameters: an integer 3635 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, 3636 and an array of pairs of comparison value constants and '<tt>label</tt>'s. 3637 The table is not allowed to contain duplicate constant entries.</p> 3638 3639<h5>Semantics:</h5> 3640<p>The <tt>switch</tt> instruction specifies a table of values and 3641 destinations. When the '<tt>switch</tt>' instruction is executed, this table 3642 is searched for the given value. If the value is found, control flow is 3643 transferred to the corresponding destination; otherwise, control flow is 3644 transferred to the default destination.</p> 3645 3646<h5>Implementation:</h5> 3647<p>Depending on properties of the target machine and the particular 3648 <tt>switch</tt> instruction, this instruction may be code generated in 3649 different ways. For example, it could be generated as a series of chained 3650 conditional branches or with a lookup table.</p> 3651 3652<h5>Example:</h5> 3653<pre> 3654 <i>; Emulate a conditional br instruction</i> 3655 %Val = <a href="#i_zext">zext</a> i1 %value to i32 3656 switch i32 %Val, label %truedest [ i32 0, label %falsedest ] 3657 3658 <i>; Emulate an unconditional br instruction</i> 3659 switch i32 0, label %dest [ ] 3660 3661 <i>; Implement a jump table:</i> 3662 switch i32 %val, label %otherwise [ i32 0, label %onzero 3663 i32 1, label %onone 3664 i32 2, label %ontwo ] 3665</pre> 3666 3667</div> 3668 3669 3670<!-- _______________________________________________________________________ --> 3671<h4> 3672 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a> 3673</h4> 3674 3675<div> 3676 3677<h5>Syntax:</h5> 3678<pre> 3679 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ] 3680</pre> 3681 3682<h5>Overview:</h5> 3683 3684<p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label 3685 within the current function, whose address is specified by 3686 "<tt>address</tt>". Address must be derived from a <a 3687 href="#blockaddress">blockaddress</a> constant.</p> 3688 3689<h5>Arguments:</h5> 3690 3691<p>The '<tt>address</tt>' argument is the address of the label to jump to. The 3692 rest of the arguments indicate the full set of possible destinations that the 3693 address may point to. Blocks are allowed to occur multiple times in the 3694 destination list, though this isn't particularly useful.</p> 3695 3696<p>This destination list is required so that dataflow analysis has an accurate 3697 understanding of the CFG.</p> 3698 3699<h5>Semantics:</h5> 3700 3701<p>Control transfers to the block specified in the address argument. All 3702 possible destination blocks must be listed in the label list, otherwise this 3703 instruction has undefined behavior. This implies that jumps to labels 3704 defined in other functions have undefined behavior as well.</p> 3705 3706<h5>Implementation:</h5> 3707 3708<p>This is typically implemented with a jump through a register.</p> 3709 3710<h5>Example:</h5> 3711<pre> 3712 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ] 3713</pre> 3714 3715</div> 3716 3717 3718<!-- _______________________________________________________________________ --> 3719<h4> 3720 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a> 3721</h4> 3722 3723<div> 3724 3725<h5>Syntax:</h5> 3726<pre> 3727 <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>] 3728 to label <normal label> unwind label <exception label> 3729</pre> 3730 3731<h5>Overview:</h5> 3732<p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified 3733 function, with the possibility of control flow transfer to either the 3734 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee 3735 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction, 3736 control flow will return to the "normal" label. If the callee (or any 3737 indirect callees) returns via the "<a href="#i_resume"><tt>resume</tt></a>" 3738 instruction or other exception handling mechanism, control is interrupted and 3739 continued at the dynamically nearest "exception" label.</p> 3740 3741<p>The '<tt>exception</tt>' label is a 3742 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the 3743 exception. As such, '<tt>exception</tt>' label is required to have the 3744 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains 3745 the information about the behavior of the program after unwinding 3746 happens, as its first non-PHI instruction. The restrictions on the 3747 "<tt>landingpad</tt>" instruction's tightly couples it to the 3748 "<tt>invoke</tt>" instruction, so that the important information contained 3749 within the "<tt>landingpad</tt>" instruction can't be lost through normal 3750 code motion.</p> 3751 3752<h5>Arguments:</h5> 3753<p>This instruction requires several arguments:</p> 3754 3755<ol> 3756 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling 3757 convention</a> the call should use. If none is specified, the call 3758 defaults to using C calling conventions.</li> 3759 3760 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for 3761 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and 3762 '<tt>inreg</tt>' attributes are valid here.</li> 3763 3764 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to 3765 function value being invoked. In most cases, this is a direct function 3766 invocation, but indirect <tt>invoke</tt>s are just as possible, branching 3767 off an arbitrary pointer to function value.</li> 3768 3769 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a 3770 function to be invoked. </li> 3771 3772 <li>'<tt>function args</tt>': argument list whose types match the function 3773 signature argument types and parameter attributes. All arguments must be 3774 of <a href="#t_firstclass">first class</a> type. If the function 3775 signature indicates the function accepts a variable number of arguments, 3776 the extra arguments can be specified.</li> 3777 3778 <li>'<tt>normal label</tt>': the label reached when the called function 3779 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li> 3780 3781 <li>'<tt>exception label</tt>': the label reached when a callee returns via 3782 the <a href="#i_resume"><tt>resume</tt></a> instruction or other exception 3783 handling mechanism.</li> 3784 3785 <li>The optional <a href="#fnattrs">function attributes</a> list. Only 3786 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and 3787 '<tt>readnone</tt>' attributes are valid here.</li> 3788</ol> 3789 3790<h5>Semantics:</h5> 3791<p>This instruction is designed to operate as a standard 3792 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The 3793 primary difference is that it establishes an association with a label, which 3794 is used by the runtime library to unwind the stack.</p> 3795 3796<p>This instruction is used in languages with destructors to ensure that proper 3797 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown 3798 exception. Additionally, this is important for implementation of 3799 '<tt>catch</tt>' clauses in high-level languages that support them.</p> 3800 3801<p>For the purposes of the SSA form, the definition of the value returned by the 3802 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current 3803 block to the "normal" label. If the callee unwinds then no return value is 3804 available.</p> 3805 3806<h5>Example:</h5> 3807<pre> 3808 %retval = invoke i32 @Test(i32 15) to label %Continue 3809 unwind label %TestCleanup <i>; {i32}:retval set</i> 3810 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue 3811 unwind label %TestCleanup <i>; {i32}:retval set</i> 3812</pre> 3813 3814</div> 3815 3816 <!-- _______________________________________________________________________ --> 3817 3818<h4> 3819 <a name="i_resume">'<tt>resume</tt>' Instruction</a> 3820</h4> 3821 3822<div> 3823 3824<h5>Syntax:</h5> 3825<pre> 3826 resume <type> <value> 3827</pre> 3828 3829<h5>Overview:</h5> 3830<p>The '<tt>resume</tt>' instruction is a terminator instruction that has no 3831 successors.</p> 3832 3833<h5>Arguments:</h5> 3834<p>The '<tt>resume</tt>' instruction requires one argument, which must have the 3835 same type as the result of any '<tt>landingpad</tt>' instruction in the same 3836 function.</p> 3837 3838<h5>Semantics:</h5> 3839<p>The '<tt>resume</tt>' instruction resumes propagation of an existing 3840 (in-flight) exception whose unwinding was interrupted with 3841 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p> 3842 3843<h5>Example:</h5> 3844<pre> 3845 resume { i8*, i32 } %exn 3846</pre> 3847 3848</div> 3849 3850<!-- _______________________________________________________________________ --> 3851 3852<h4> 3853 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a> 3854</h4> 3855 3856<div> 3857 3858<h5>Syntax:</h5> 3859<pre> 3860 unreachable 3861</pre> 3862 3863<h5>Overview:</h5> 3864<p>The '<tt>unreachable</tt>' instruction has no defined semantics. This 3865 instruction is used to inform the optimizer that a particular portion of the 3866 code is not reachable. This can be used to indicate that the code after a 3867 no-return function cannot be reached, and other facts.</p> 3868 3869<h5>Semantics:</h5> 3870<p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p> 3871 3872</div> 3873 3874</div> 3875 3876<!-- ======================================================================= --> 3877<h3> 3878 <a name="binaryops">Binary Operations</a> 3879</h3> 3880 3881<div> 3882 3883<p>Binary operators are used to do most of the computation in a program. They 3884 require two operands of the same type, execute an operation on them, and 3885 produce a single value. The operands might represent multiple data, as is 3886 the case with the <a href="#t_vector">vector</a> data type. The result value 3887 has the same type as its operands.</p> 3888 3889<p>There are several different binary operators:</p> 3890 3891<!-- _______________________________________________________________________ --> 3892<h4> 3893 <a name="i_add">'<tt>add</tt>' Instruction</a> 3894</h4> 3895 3896<div> 3897 3898<h5>Syntax:</h5> 3899<pre> 3900 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3901 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3902 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3903 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3904</pre> 3905 3906<h5>Overview:</h5> 3907<p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p> 3908 3909<h5>Arguments:</h5> 3910<p>The two arguments to the '<tt>add</tt>' instruction must 3911 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of 3912 integer values. Both arguments must have identical types.</p> 3913 3914<h5>Semantics:</h5> 3915<p>The value produced is the integer sum of the two operands.</p> 3916 3917<p>If the sum has unsigned overflow, the result returned is the mathematical 3918 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p> 3919 3920<p>Because LLVM integers use a two's complement representation, this instruction 3921 is appropriate for both signed and unsigned integers.</p> 3922 3923<p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap" 3924 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or 3925 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt> 3926 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow, 3927 respectively, occurs.</p> 3928 3929<h5>Example:</h5> 3930<pre> 3931 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i> 3932</pre> 3933 3934</div> 3935 3936<!-- _______________________________________________________________________ --> 3937<h4> 3938 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a> 3939</h4> 3940 3941<div> 3942 3943<h5>Syntax:</h5> 3944<pre> 3945 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3946</pre> 3947 3948<h5>Overview:</h5> 3949<p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p> 3950 3951<h5>Arguments:</h5> 3952<p>The two arguments to the '<tt>fadd</tt>' instruction must be 3953 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of 3954 floating point values. Both arguments must have identical types.</p> 3955 3956<h5>Semantics:</h5> 3957<p>The value produced is the floating point sum of the two operands.</p> 3958 3959<h5>Example:</h5> 3960<pre> 3961 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i> 3962</pre> 3963 3964</div> 3965 3966<!-- _______________________________________________________________________ --> 3967<h4> 3968 <a name="i_sub">'<tt>sub</tt>' Instruction</a> 3969</h4> 3970 3971<div> 3972 3973<h5>Syntax:</h5> 3974<pre> 3975 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3976 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3977 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3978 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3979</pre> 3980 3981<h5>Overview:</h5> 3982<p>The '<tt>sub</tt>' instruction returns the difference of its two 3983 operands.</p> 3984 3985<p>Note that the '<tt>sub</tt>' instruction is used to represent the 3986 '<tt>neg</tt>' instruction present in most other intermediate 3987 representations.</p> 3988 3989<h5>Arguments:</h5> 3990<p>The two arguments to the '<tt>sub</tt>' instruction must 3991 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of 3992 integer values. Both arguments must have identical types.</p> 3993 3994<h5>Semantics:</h5> 3995<p>The value produced is the integer difference of the two operands.</p> 3996 3997<p>If the difference has unsigned overflow, the result returned is the 3998 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the 3999 result.</p> 4000 4001<p>Because LLVM integers use a two's complement representation, this instruction 4002 is appropriate for both signed and unsigned integers.</p> 4003 4004<p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap" 4005 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or 4006 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt> 4007 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow, 4008 respectively, occurs.</p> 4009 4010<h5>Example:</h5> 4011<pre> 4012 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i> 4013 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i> 4014</pre> 4015 4016</div> 4017 4018<!-- _______________________________________________________________________ --> 4019<h4> 4020 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a> 4021</h4> 4022 4023<div> 4024 4025<h5>Syntax:</h5> 4026<pre> 4027 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4028</pre> 4029 4030<h5>Overview:</h5> 4031<p>The '<tt>fsub</tt>' instruction returns the difference of its two 4032 operands.</p> 4033 4034<p>Note that the '<tt>fsub</tt>' instruction is used to represent the 4035 '<tt>fneg</tt>' instruction present in most other intermediate 4036 representations.</p> 4037 4038<h5>Arguments:</h5> 4039<p>The two arguments to the '<tt>fsub</tt>' instruction must be 4040 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of 4041 floating point values. Both arguments must have identical types.</p> 4042 4043<h5>Semantics:</h5> 4044<p>The value produced is the floating point difference of the two operands.</p> 4045 4046<h5>Example:</h5> 4047<pre> 4048 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i> 4049 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i> 4050</pre> 4051 4052</div> 4053 4054<!-- _______________________________________________________________________ --> 4055<h4> 4056 <a name="i_mul">'<tt>mul</tt>' Instruction</a> 4057</h4> 4058 4059<div> 4060 4061<h5>Syntax:</h5> 4062<pre> 4063 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4064 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4065 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4066 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4067</pre> 4068 4069<h5>Overview:</h5> 4070<p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p> 4071 4072<h5>Arguments:</h5> 4073<p>The two arguments to the '<tt>mul</tt>' instruction must 4074 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of 4075 integer values. Both arguments must have identical types.</p> 4076 4077<h5>Semantics:</h5> 4078<p>The value produced is the integer product of the two operands.</p> 4079 4080<p>If the result of the multiplication has unsigned overflow, the result 4081 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit 4082 width of the result.</p> 4083 4084<p>Because LLVM integers use a two's complement representation, and the result 4085 is the same width as the operands, this instruction returns the correct 4086 result for both signed and unsigned integers. If a full product 4087 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should 4088 be sign-extended or zero-extended as appropriate to the width of the full 4089 product.</p> 4090 4091<p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap" 4092 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or 4093 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt> 4094 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow, 4095 respectively, occurs.</p> 4096 4097<h5>Example:</h5> 4098<pre> 4099 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i> 4100</pre> 4101 4102</div> 4103 4104<!-- _______________________________________________________________________ --> 4105<h4> 4106 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a> 4107</h4> 4108 4109<div> 4110 4111<h5>Syntax:</h5> 4112<pre> 4113 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4114</pre> 4115 4116<h5>Overview:</h5> 4117<p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p> 4118 4119<h5>Arguments:</h5> 4120<p>The two arguments to the '<tt>fmul</tt>' instruction must be 4121 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of 4122 floating point values. Both arguments must have identical types.</p> 4123 4124<h5>Semantics:</h5> 4125<p>The value produced is the floating point product of the two operands.</p> 4126 4127<h5>Example:</h5> 4128<pre> 4129 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i> 4130</pre> 4131 4132</div> 4133 4134<!-- _______________________________________________________________________ --> 4135<h4> 4136 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a> 4137</h4> 4138 4139<div> 4140 4141<h5>Syntax:</h5> 4142<pre> 4143 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4144 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4145</pre> 4146 4147<h5>Overview:</h5> 4148<p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p> 4149 4150<h5>Arguments:</h5> 4151<p>The two arguments to the '<tt>udiv</tt>' instruction must be 4152 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4153 values. Both arguments must have identical types.</p> 4154 4155<h5>Semantics:</h5> 4156<p>The value produced is the unsigned integer quotient of the two operands.</p> 4157 4158<p>Note that unsigned integer division and signed integer division are distinct 4159 operations; for signed integer division, use '<tt>sdiv</tt>'.</p> 4160 4161<p>Division by zero leads to undefined behavior.</p> 4162 4163<p>If the <tt>exact</tt> keyword is present, the result value of the 4164 <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a 4165 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p> 4166 4167 4168<h5>Example:</h5> 4169<pre> 4170 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i> 4171</pre> 4172 4173</div> 4174 4175<!-- _______________________________________________________________________ --> 4176<h4> 4177 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a> 4178</h4> 4179 4180<div> 4181 4182<h5>Syntax:</h5> 4183<pre> 4184 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4185 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4186</pre> 4187 4188<h5>Overview:</h5> 4189<p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p> 4190 4191<h5>Arguments:</h5> 4192<p>The two arguments to the '<tt>sdiv</tt>' instruction must be 4193 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4194 values. Both arguments must have identical types.</p> 4195 4196<h5>Semantics:</h5> 4197<p>The value produced is the signed integer quotient of the two operands rounded 4198 towards zero.</p> 4199 4200<p>Note that signed integer division and unsigned integer division are distinct 4201 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p> 4202 4203<p>Division by zero leads to undefined behavior. Overflow also leads to 4204 undefined behavior; this is a rare case, but can occur, for example, by doing 4205 a 32-bit division of -2147483648 by -1.</p> 4206 4207<p>If the <tt>exact</tt> keyword is present, the result value of the 4208 <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would 4209 be rounded.</p> 4210 4211<h5>Example:</h5> 4212<pre> 4213 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i> 4214</pre> 4215 4216</div> 4217 4218<!-- _______________________________________________________________________ --> 4219<h4> 4220 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a> 4221</h4> 4222 4223<div> 4224 4225<h5>Syntax:</h5> 4226<pre> 4227 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4228</pre> 4229 4230<h5>Overview:</h5> 4231<p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p> 4232 4233<h5>Arguments:</h5> 4234<p>The two arguments to the '<tt>fdiv</tt>' instruction must be 4235 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of 4236 floating point values. Both arguments must have identical types.</p> 4237 4238<h5>Semantics:</h5> 4239<p>The value produced is the floating point quotient of the two operands.</p> 4240 4241<h5>Example:</h5> 4242<pre> 4243 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i> 4244</pre> 4245 4246</div> 4247 4248<!-- _______________________________________________________________________ --> 4249<h4> 4250 <a name="i_urem">'<tt>urem</tt>' Instruction</a> 4251</h4> 4252 4253<div> 4254 4255<h5>Syntax:</h5> 4256<pre> 4257 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4258</pre> 4259 4260<h5>Overview:</h5> 4261<p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned 4262 division of its two arguments.</p> 4263 4264<h5>Arguments:</h5> 4265<p>The two arguments to the '<tt>urem</tt>' instruction must be 4266 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4267 values. Both arguments must have identical types.</p> 4268 4269<h5>Semantics:</h5> 4270<p>This instruction returns the unsigned integer <i>remainder</i> of a division. 4271 This instruction always performs an unsigned division to get the 4272 remainder.</p> 4273 4274<p>Note that unsigned integer remainder and signed integer remainder are 4275 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p> 4276 4277<p>Taking the remainder of a division by zero leads to undefined behavior.</p> 4278 4279<h5>Example:</h5> 4280<pre> 4281 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i> 4282</pre> 4283 4284</div> 4285 4286<!-- _______________________________________________________________________ --> 4287<h4> 4288 <a name="i_srem">'<tt>srem</tt>' Instruction</a> 4289</h4> 4290 4291<div> 4292 4293<h5>Syntax:</h5> 4294<pre> 4295 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4296</pre> 4297 4298<h5>Overview:</h5> 4299<p>The '<tt>srem</tt>' instruction returns the remainder from the signed 4300 division of its two operands. This instruction can also take 4301 <a href="#t_vector">vector</a> versions of the values in which case the 4302 elements must be integers.</p> 4303 4304<h5>Arguments:</h5> 4305<p>The two arguments to the '<tt>srem</tt>' instruction must be 4306 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4307 values. Both arguments must have identical types.</p> 4308 4309<h5>Semantics:</h5> 4310<p>This instruction returns the <i>remainder</i> of a division (where the result 4311 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the 4312 <i>modulo</i> operator (where the result is either zero or has the same sign 4313 as the divisor, <tt>op2</tt>) of a value. 4314 For more information about the difference, 4315 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The 4316 Math Forum</a>. For a table of how this is implemented in various languages, 4317 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation"> 4318 Wikipedia: modulo operation</a>.</p> 4319 4320<p>Note that signed integer remainder and unsigned integer remainder are 4321 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p> 4322 4323<p>Taking the remainder of a division by zero leads to undefined behavior. 4324 Overflow also leads to undefined behavior; this is a rare case, but can 4325 occur, for example, by taking the remainder of a 32-bit division of 4326 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule 4327 lets srem be implemented using instructions that return both the result of 4328 the division and the remainder.)</p> 4329 4330<h5>Example:</h5> 4331<pre> 4332 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i> 4333</pre> 4334 4335</div> 4336 4337<!-- _______________________________________________________________________ --> 4338<h4> 4339 <a name="i_frem">'<tt>frem</tt>' Instruction</a> 4340</h4> 4341 4342<div> 4343 4344<h5>Syntax:</h5> 4345<pre> 4346 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4347</pre> 4348 4349<h5>Overview:</h5> 4350<p>The '<tt>frem</tt>' instruction returns the remainder from the division of 4351 its two operands.</p> 4352 4353<h5>Arguments:</h5> 4354<p>The two arguments to the '<tt>frem</tt>' instruction must be 4355 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of 4356 floating point values. Both arguments must have identical types.</p> 4357 4358<h5>Semantics:</h5> 4359<p>This instruction returns the <i>remainder</i> of a division. The remainder 4360 has the same sign as the dividend.</p> 4361 4362<h5>Example:</h5> 4363<pre> 4364 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i> 4365</pre> 4366 4367</div> 4368 4369</div> 4370 4371<!-- ======================================================================= --> 4372<h3> 4373 <a name="bitwiseops">Bitwise Binary Operations</a> 4374</h3> 4375 4376<div> 4377 4378<p>Bitwise binary operators are used to do various forms of bit-twiddling in a 4379 program. They are generally very efficient instructions and can commonly be 4380 strength reduced from other instructions. They require two operands of the 4381 same type, execute an operation on them, and produce a single value. The 4382 resulting value is the same type as its operands.</p> 4383 4384<!-- _______________________________________________________________________ --> 4385<h4> 4386 <a name="i_shl">'<tt>shl</tt>' Instruction</a> 4387</h4> 4388 4389<div> 4390 4391<h5>Syntax:</h5> 4392<pre> 4393 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4394 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4395 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4396 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4397</pre> 4398 4399<h5>Overview:</h5> 4400<p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left 4401 a specified number of bits.</p> 4402 4403<h5>Arguments:</h5> 4404<p>Both arguments to the '<tt>shl</tt>' instruction must be the 4405 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of 4406 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p> 4407 4408<h5>Semantics:</h5> 4409<p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 4410 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt> 4411 is (statically or dynamically) negative or equal to or larger than the number 4412 of bits in <tt>op1</tt>, the result is undefined. If the arguments are 4413 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding 4414 shift amount in <tt>op2</tt>.</p> 4415 4416<p>If the <tt>nuw</tt> keyword is present, then the shift produces a 4417 <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits. If 4418 the <tt>nsw</tt> keyword is present, then the shift produces a 4419 <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree 4420 with the resultant sign bit. As such, NUW/NSW have the same semantics as 4421 they would if the shift were expressed as a mul instruction with the same 4422 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p> 4423 4424<h5>Example:</h5> 4425<pre> 4426 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i> 4427 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i> 4428 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i> 4429 <result> = shl i32 1, 32 <i>; undefined</i> 4430 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i> 4431</pre> 4432 4433</div> 4434 4435<!-- _______________________________________________________________________ --> 4436<h4> 4437 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a> 4438</h4> 4439 4440<div> 4441 4442<h5>Syntax:</h5> 4443<pre> 4444 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4445 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4446</pre> 4447 4448<h5>Overview:</h5> 4449<p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first 4450 operand shifted to the right a specified number of bits with zero fill.</p> 4451 4452<h5>Arguments:</h5> 4453<p>Both arguments to the '<tt>lshr</tt>' instruction must be the same 4454 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4455 type. '<tt>op2</tt>' is treated as an unsigned value.</p> 4456 4457<h5>Semantics:</h5> 4458<p>This instruction always performs a logical shift right operation. The most 4459 significant bits of the result will be filled with zero bits after the shift. 4460 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the 4461 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are 4462 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding 4463 shift amount in <tt>op2</tt>.</p> 4464 4465<p>If the <tt>exact</tt> keyword is present, the result value of the 4466 <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits 4467 shifted out are non-zero.</p> 4468 4469 4470<h5>Example:</h5> 4471<pre> 4472 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i> 4473 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i> 4474 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i> 4475 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i> 4476 <result> = lshr i32 1, 32 <i>; undefined</i> 4477 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i> 4478</pre> 4479 4480</div> 4481 4482<!-- _______________________________________________________________________ --> 4483<h4> 4484 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a> 4485</h4> 4486 4487<div> 4488 4489<h5>Syntax:</h5> 4490<pre> 4491 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4492 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4493</pre> 4494 4495<h5>Overview:</h5> 4496<p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first 4497 operand shifted to the right a specified number of bits with sign 4498 extension.</p> 4499 4500<h5>Arguments:</h5> 4501<p>Both arguments to the '<tt>ashr</tt>' instruction must be the same 4502 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4503 type. '<tt>op2</tt>' is treated as an unsigned value.</p> 4504 4505<h5>Semantics:</h5> 4506<p>This instruction always performs an arithmetic shift right operation, The 4507 most significant bits of the result will be filled with the sign bit 4508 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or 4509 larger than the number of bits in <tt>op1</tt>, the result is undefined. If 4510 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by 4511 the corresponding shift amount in <tt>op2</tt>.</p> 4512 4513<p>If the <tt>exact</tt> keyword is present, the result value of the 4514 <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits 4515 shifted out are non-zero.</p> 4516 4517<h5>Example:</h5> 4518<pre> 4519 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i> 4520 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i> 4521 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i> 4522 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i> 4523 <result> = ashr i32 1, 32 <i>; undefined</i> 4524 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i> 4525</pre> 4526 4527</div> 4528 4529<!-- _______________________________________________________________________ --> 4530<h4> 4531 <a name="i_and">'<tt>and</tt>' Instruction</a> 4532</h4> 4533 4534<div> 4535 4536<h5>Syntax:</h5> 4537<pre> 4538 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4539</pre> 4540 4541<h5>Overview:</h5> 4542<p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two 4543 operands.</p> 4544 4545<h5>Arguments:</h5> 4546<p>The two arguments to the '<tt>and</tt>' instruction must be 4547 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4548 values. Both arguments must have identical types.</p> 4549 4550<h5>Semantics:</h5> 4551<p>The truth table used for the '<tt>and</tt>' instruction is:</p> 4552 4553<table border="1" cellspacing="0" cellpadding="4"> 4554 <tbody> 4555 <tr> 4556 <th>In0</th> 4557 <th>In1</th> 4558 <th>Out</th> 4559 </tr> 4560 <tr> 4561 <td>0</td> 4562 <td>0</td> 4563 <td>0</td> 4564 </tr> 4565 <tr> 4566 <td>0</td> 4567 <td>1</td> 4568 <td>0</td> 4569 </tr> 4570 <tr> 4571 <td>1</td> 4572 <td>0</td> 4573 <td>0</td> 4574 </tr> 4575 <tr> 4576 <td>1</td> 4577 <td>1</td> 4578 <td>1</td> 4579 </tr> 4580 </tbody> 4581</table> 4582 4583<h5>Example:</h5> 4584<pre> 4585 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i> 4586 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i> 4587 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i> 4588</pre> 4589</div> 4590<!-- _______________________________________________________________________ --> 4591<h4> 4592 <a name="i_or">'<tt>or</tt>' Instruction</a> 4593</h4> 4594 4595<div> 4596 4597<h5>Syntax:</h5> 4598<pre> 4599 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4600</pre> 4601 4602<h5>Overview:</h5> 4603<p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its 4604 two operands.</p> 4605 4606<h5>Arguments:</h5> 4607<p>The two arguments to the '<tt>or</tt>' instruction must be 4608 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4609 values. Both arguments must have identical types.</p> 4610 4611<h5>Semantics:</h5> 4612<p>The truth table used for the '<tt>or</tt>' instruction is:</p> 4613 4614<table border="1" cellspacing="0" cellpadding="4"> 4615 <tbody> 4616 <tr> 4617 <th>In0</th> 4618 <th>In1</th> 4619 <th>Out</th> 4620 </tr> 4621 <tr> 4622 <td>0</td> 4623 <td>0</td> 4624 <td>0</td> 4625 </tr> 4626 <tr> 4627 <td>0</td> 4628 <td>1</td> 4629 <td>1</td> 4630 </tr> 4631 <tr> 4632 <td>1</td> 4633 <td>0</td> 4634 <td>1</td> 4635 </tr> 4636 <tr> 4637 <td>1</td> 4638 <td>1</td> 4639 <td>1</td> 4640 </tr> 4641 </tbody> 4642</table> 4643 4644<h5>Example:</h5> 4645<pre> 4646 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i> 4647 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i> 4648 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i> 4649</pre> 4650 4651</div> 4652 4653<!-- _______________________________________________________________________ --> 4654<h4> 4655 <a name="i_xor">'<tt>xor</tt>' Instruction</a> 4656</h4> 4657 4658<div> 4659 4660<h5>Syntax:</h5> 4661<pre> 4662 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4663</pre> 4664 4665<h5>Overview:</h5> 4666<p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of 4667 its two operands. The <tt>xor</tt> is used to implement the "one's 4668 complement" operation, which is the "~" operator in C.</p> 4669 4670<h5>Arguments:</h5> 4671<p>The two arguments to the '<tt>xor</tt>' instruction must be 4672 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4673 values. Both arguments must have identical types.</p> 4674 4675<h5>Semantics:</h5> 4676<p>The truth table used for the '<tt>xor</tt>' instruction is:</p> 4677 4678<table border="1" cellspacing="0" cellpadding="4"> 4679 <tbody> 4680 <tr> 4681 <th>In0</th> 4682 <th>In1</th> 4683 <th>Out</th> 4684 </tr> 4685 <tr> 4686 <td>0</td> 4687 <td>0</td> 4688 <td>0</td> 4689 </tr> 4690 <tr> 4691 <td>0</td> 4692 <td>1</td> 4693 <td>1</td> 4694 </tr> 4695 <tr> 4696 <td>1</td> 4697 <td>0</td> 4698 <td>1</td> 4699 </tr> 4700 <tr> 4701 <td>1</td> 4702 <td>1</td> 4703 <td>0</td> 4704 </tr> 4705 </tbody> 4706</table> 4707 4708<h5>Example:</h5> 4709<pre> 4710 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i> 4711 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i> 4712 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i> 4713 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i> 4714</pre> 4715 4716</div> 4717 4718</div> 4719 4720<!-- ======================================================================= --> 4721<h3> 4722 <a name="vectorops">Vector Operations</a> 4723</h3> 4724 4725<div> 4726 4727<p>LLVM supports several instructions to represent vector operations in a 4728 target-independent manner. These instructions cover the element-access and 4729 vector-specific operations needed to process vectors effectively. While LLVM 4730 does directly support these vector operations, many sophisticated algorithms 4731 will want to use target-specific intrinsics to take full advantage of a 4732 specific target.</p> 4733 4734<!-- _______________________________________________________________________ --> 4735<h4> 4736 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a> 4737</h4> 4738 4739<div> 4740 4741<h5>Syntax:</h5> 4742<pre> 4743 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i> 4744</pre> 4745 4746<h5>Overview:</h5> 4747<p>The '<tt>extractelement</tt>' instruction extracts a single scalar element 4748 from a vector at a specified index.</p> 4749 4750 4751<h5>Arguments:</h5> 4752<p>The first operand of an '<tt>extractelement</tt>' instruction is a value 4753 of <a href="#t_vector">vector</a> type. The second operand is an index 4754 indicating the position from which to extract the element. The index may be 4755 a variable.</p> 4756 4757<h5>Semantics:</h5> 4758<p>The result is a scalar of the same type as the element type of 4759 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of 4760 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the 4761 results are undefined.</p> 4762 4763<h5>Example:</h5> 4764<pre> 4765 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i> 4766</pre> 4767 4768</div> 4769 4770<!-- _______________________________________________________________________ --> 4771<h4> 4772 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a> 4773</h4> 4774 4775<div> 4776 4777<h5>Syntax:</h5> 4778<pre> 4779 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i> 4780</pre> 4781 4782<h5>Overview:</h5> 4783<p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a 4784 vector at a specified index.</p> 4785 4786<h5>Arguments:</h5> 4787<p>The first operand of an '<tt>insertelement</tt>' instruction is a value 4788 of <a href="#t_vector">vector</a> type. The second operand is a scalar value 4789 whose type must equal the element type of the first operand. The third 4790 operand is an index indicating the position at which to insert the value. 4791 The index may be a variable.</p> 4792 4793<h5>Semantics:</h5> 4794<p>The result is a vector of the same type as <tt>val</tt>. Its element values 4795 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the 4796 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the 4797 results are undefined.</p> 4798 4799<h5>Example:</h5> 4800<pre> 4801 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i> 4802</pre> 4803 4804</div> 4805 4806<!-- _______________________________________________________________________ --> 4807<h4> 4808 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a> 4809</h4> 4810 4811<div> 4812 4813<h5>Syntax:</h5> 4814<pre> 4815 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i> 4816</pre> 4817 4818<h5>Overview:</h5> 4819<p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements 4820 from two input vectors, returning a vector with the same element type as the 4821 input and length that is the same as the shuffle mask.</p> 4822 4823<h5>Arguments:</h5> 4824<p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors 4825 with the same type. The third argument is a shuffle mask whose 4826 element type is always 'i32'. The result of the instruction is a vector 4827 whose length is the same as the shuffle mask and whose element type is the 4828 same as the element type of the first two operands.</p> 4829 4830<p>The shuffle mask operand is required to be a constant vector with either 4831 constant integer or undef values.</p> 4832 4833<h5>Semantics:</h5> 4834<p>The elements of the two input vectors are numbered from left to right across 4835 both of the vectors. The shuffle mask operand specifies, for each element of 4836 the result vector, which element of the two input vectors the result element 4837 gets. The element selector may be undef (meaning "don't care") and the 4838 second operand may be undef if performing a shuffle from only one vector.</p> 4839 4840<h5>Example:</h5> 4841<pre> 4842 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2, 4843 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i> 4844 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef, 4845 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle. 4846 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef, 4847 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> 4848 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2, 4849 <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> 4850</pre> 4851 4852</div> 4853 4854</div> 4855 4856<!-- ======================================================================= --> 4857<h3> 4858 <a name="aggregateops">Aggregate Operations</a> 4859</h3> 4860 4861<div> 4862 4863<p>LLVM supports several instructions for working with 4864 <a href="#t_aggregate">aggregate</a> values.</p> 4865 4866<!-- _______________________________________________________________________ --> 4867<h4> 4868 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a> 4869</h4> 4870 4871<div> 4872 4873<h5>Syntax:</h5> 4874<pre> 4875 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}* 4876</pre> 4877 4878<h5>Overview:</h5> 4879<p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field 4880 from an <a href="#t_aggregate">aggregate</a> value.</p> 4881 4882<h5>Arguments:</h5> 4883<p>The first operand of an '<tt>extractvalue</tt>' instruction is a value 4884 of <a href="#t_struct">struct</a> or 4885 <a href="#t_array">array</a> type. The operands are constant indices to 4886 specify which value to extract in a similar manner as indices in a 4887 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p> 4888 <p>The major differences to <tt>getelementptr</tt> indexing are:</p> 4889 <ul> 4890 <li>Since the value being indexed is not a pointer, the first index is 4891 omitted and assumed to be zero.</li> 4892 <li>At least one index must be specified.</li> 4893 <li>Not only struct indices but also array indices must be in 4894 bounds.</li> 4895 </ul> 4896 4897<h5>Semantics:</h5> 4898<p>The result is the value at the position in the aggregate specified by the 4899 index operands.</p> 4900 4901<h5>Example:</h5> 4902<pre> 4903 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i> 4904</pre> 4905 4906</div> 4907 4908<!-- _______________________________________________________________________ --> 4909<h4> 4910 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a> 4911</h4> 4912 4913<div> 4914 4915<h5>Syntax:</h5> 4916<pre> 4917 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i> 4918</pre> 4919 4920<h5>Overview:</h5> 4921<p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field 4922 in an <a href="#t_aggregate">aggregate</a> value.</p> 4923 4924<h5>Arguments:</h5> 4925<p>The first operand of an '<tt>insertvalue</tt>' instruction is a value 4926 of <a href="#t_struct">struct</a> or 4927 <a href="#t_array">array</a> type. The second operand is a first-class 4928 value to insert. The following operands are constant indices indicating 4929 the position at which to insert the value in a similar manner as indices in a 4930 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The 4931 value to insert must have the same type as the value identified by the 4932 indices.</p> 4933 4934<h5>Semantics:</h5> 4935<p>The result is an aggregate of the same type as <tt>val</tt>. Its value is 4936 that of <tt>val</tt> except that the value at the position specified by the 4937 indices is that of <tt>elt</tt>.</p> 4938 4939<h5>Example:</h5> 4940<pre> 4941 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i> 4942 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i> 4943 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i> 4944</pre> 4945 4946</div> 4947 4948</div> 4949 4950<!-- ======================================================================= --> 4951<h3> 4952 <a name="memoryops">Memory Access and Addressing Operations</a> 4953</h3> 4954 4955<div> 4956 4957<p>A key design point of an SSA-based representation is how it represents 4958 memory. In LLVM, no memory locations are in SSA form, which makes things 4959 very simple. This section describes how to read, write, and allocate 4960 memory in LLVM.</p> 4961 4962<!-- _______________________________________________________________________ --> 4963<h4> 4964 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a> 4965</h4> 4966 4967<div> 4968 4969<h5>Syntax:</h5> 4970<pre> 4971 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i> 4972</pre> 4973 4974<h5>Overview:</h5> 4975<p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the 4976 currently executing function, to be automatically released when this function 4977 returns to its caller. The object is always allocated in the generic address 4978 space (address space zero).</p> 4979 4980<h5>Arguments:</h5> 4981<p>The '<tt>alloca</tt>' instruction 4982 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the 4983 runtime stack, returning a pointer of the appropriate type to the program. 4984 If "NumElements" is specified, it is the number of elements allocated, 4985 otherwise "NumElements" is defaulted to be one. If a constant alignment is 4986 specified, the value result of the allocation is guaranteed to be aligned to 4987 at least that boundary. If not specified, or if zero, the target can choose 4988 to align the allocation on any convenient boundary compatible with the 4989 type.</p> 4990 4991<p>'<tt>type</tt>' may be any sized type.</p> 4992 4993<h5>Semantics:</h5> 4994<p>Memory is allocated; a pointer is returned. The operation is undefined if 4995 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d 4996 memory is automatically released when the function returns. The 4997 '<tt>alloca</tt>' instruction is commonly used to represent automatic 4998 variables that must have an address available. When the function returns 4999 (either with the <tt><a href="#i_ret">ret</a></tt> 5000 or <tt><a href="#i_resume">resume</a></tt> instructions), the memory is 5001 reclaimed. Allocating zero bytes is legal, but the result is undefined. 5002 The order in which memory is allocated (ie., which way the stack grows) is 5003 not specified.</p> 5004 5005<p> 5006 5007<h5>Example:</h5> 5008<pre> 5009 %ptr = alloca i32 <i>; yields {i32*}:ptr</i> 5010 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i> 5011 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i> 5012 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i> 5013</pre> 5014 5015</div> 5016 5017<!-- _______________________________________________________________________ --> 5018<h4> 5019 <a name="i_load">'<tt>load</tt>' Instruction</a> 5020</h4> 5021 5022<div> 5023 5024<h5>Syntax:</h5> 5025<pre> 5026 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>] 5027 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment> 5028 !<index> = !{ i32 1 } 5029</pre> 5030 5031<h5>Overview:</h5> 5032<p>The '<tt>load</tt>' instruction is used to read from memory.</p> 5033 5034<h5>Arguments:</h5> 5035<p>The argument to the '<tt>load</tt>' instruction specifies the memory address 5036 from which to load. The pointer must point to 5037 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is 5038 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the 5039 number or order of execution of this <tt>load</tt> with other <a 5040 href="#volatile">volatile operations</a>.</p> 5041 5042<p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra 5043 <a href="#ordering">ordering</a> and optional <code>singlethread</code> 5044 argument. The <code>release</code> and <code>acq_rel</code> orderings are 5045 not valid on <code>load</code> instructions. Atomic loads produce <a 5046 href="#memorymodel">defined</a> results when they may see multiple atomic 5047 stores. The type of the pointee must be an integer type whose bit width 5048 is a power of two greater than or equal to eight and less than or equal 5049 to a target-specific size limit. <code>align</code> must be explicitly 5050 specified on atomic loads, and the load has undefined behavior if the 5051 alignment is not set to a value which is at least the size in bytes of 5052 the pointee. <code>!nontemporal</code> does not have any defined semantics 5053 for atomic loads.</p> 5054 5055<p>The optional constant <tt>align</tt> argument specifies the alignment of the 5056 operation (that is, the alignment of the memory address). A value of 0 or an 5057 omitted <tt>align</tt> argument means that the operation has the preferential 5058 alignment for the target. It is the responsibility of the code emitter to 5059 ensure that the alignment information is correct. Overestimating the 5060 alignment results in undefined behavior. Underestimating the alignment may 5061 produce less efficient code. An alignment of 1 is always safe.</p> 5062 5063<p>The optional <tt>!nontemporal</tt> metadata must reference a single 5064 metatadata name <index> corresponding to a metadata node with 5065 one <tt>i32</tt> entry of value 1. The existence of 5066 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer 5067 and code generator that this load is not expected to be reused in the cache. 5068 The code generator may select special instructions to save cache bandwidth, 5069 such as the <tt>MOVNT</tt> instruction on x86.</p> 5070 5071<p>The optional <tt>!invariant.load</tt> metadata must reference a single 5072 metatadata name <index> corresponding to a metadata node with no 5073 entries. The existence of the <tt>!invariant.load</tt> metatadata on the 5074 instruction tells the optimizer and code generator that this load address 5075 points to memory which does not change value during program execution. 5076 The optimizer may then move this load around, for example, by hoisting it 5077 out of loops using loop invariant code motion.</p> 5078 5079<h5>Semantics:</h5> 5080<p>The location of memory pointed to is loaded. If the value being loaded is of 5081 scalar type then the number of bytes read does not exceed the minimum number 5082 of bytes needed to hold all bits of the type. For example, loading an 5083 <tt>i24</tt> reads at most three bytes. When loading a value of a type like 5084 <tt>i20</tt> with a size that is not an integral number of bytes, the result 5085 is undefined if the value was not originally written using a store of the 5086 same type.</p> 5087 5088<h5>Examples:</h5> 5089<pre> 5090 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i> 5091 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i> 5092 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i> 5093</pre> 5094 5095</div> 5096 5097<!-- _______________________________________________________________________ --> 5098<h4> 5099 <a name="i_store">'<tt>store</tt>' Instruction</a> 5100</h4> 5101 5102<div> 5103 5104<h5>Syntax:</h5> 5105<pre> 5106 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i> 5107 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i> 5108</pre> 5109 5110<h5>Overview:</h5> 5111<p>The '<tt>store</tt>' instruction is used to write to memory.</p> 5112 5113<h5>Arguments:</h5> 5114<p>There are two arguments to the '<tt>store</tt>' instruction: a value to store 5115 and an address at which to store it. The type of the 5116 '<tt><pointer></tt>' operand must be a pointer to 5117 the <a href="#t_firstclass">first class</a> type of the 5118 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as 5119 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or 5120 order of execution of this <tt>store</tt> with other <a 5121 href="#volatile">volatile operations</a>.</p> 5122 5123<p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra 5124 <a href="#ordering">ordering</a> and optional <code>singlethread</code> 5125 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't 5126 valid on <code>store</code> instructions. Atomic loads produce <a 5127 href="#memorymodel">defined</a> results when they may see multiple atomic 5128 stores. The type of the pointee must be an integer type whose bit width 5129 is a power of two greater than or equal to eight and less than or equal 5130 to a target-specific size limit. <code>align</code> must be explicitly 5131 specified on atomic stores, and the store has undefined behavior if the 5132 alignment is not set to a value which is at least the size in bytes of 5133 the pointee. <code>!nontemporal</code> does not have any defined semantics 5134 for atomic stores.</p> 5135 5136<p>The optional constant "align" argument specifies the alignment of the 5137 operation (that is, the alignment of the memory address). A value of 0 or an 5138 omitted "align" argument means that the operation has the preferential 5139 alignment for the target. It is the responsibility of the code emitter to 5140 ensure that the alignment information is correct. Overestimating the 5141 alignment results in an undefined behavior. Underestimating the alignment may 5142 produce less efficient code. An alignment of 1 is always safe.</p> 5143 5144<p>The optional !nontemporal metadata must reference a single metatadata 5145 name <index> corresponding to a metadata node with one i32 entry of 5146 value 1. The existence of the !nontemporal metatadata on the 5147 instruction tells the optimizer and code generator that this load is 5148 not expected to be reused in the cache. The code generator may 5149 select special instructions to save cache bandwidth, such as the 5150 MOVNT instruction on x86.</p> 5151 5152 5153<h5>Semantics:</h5> 5154<p>The contents of memory are updated to contain '<tt><value></tt>' at the 5155 location specified by the '<tt><pointer></tt>' operand. If 5156 '<tt><value></tt>' is of scalar type then the number of bytes written 5157 does not exceed the minimum number of bytes needed to hold all bits of the 5158 type. For example, storing an <tt>i24</tt> writes at most three bytes. When 5159 writing a value of a type like <tt>i20</tt> with a size that is not an 5160 integral number of bytes, it is unspecified what happens to the extra bits 5161 that do not belong to the type, but they will typically be overwritten.</p> 5162 5163<h5>Example:</h5> 5164<pre> 5165 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i> 5166 store i32 3, i32* %ptr <i>; yields {void}</i> 5167 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i> 5168</pre> 5169 5170</div> 5171 5172<!-- _______________________________________________________________________ --> 5173<h4> 5174<a name="i_fence">'<tt>fence</tt>' Instruction</a> 5175</h4> 5176 5177<div> 5178 5179<h5>Syntax:</h5> 5180<pre> 5181 fence [singlethread] <ordering> <i>; yields {void}</i> 5182</pre> 5183 5184<h5>Overview:</h5> 5185<p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges 5186between operations.</p> 5187 5188<h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a 5189href="#ordering">ordering</a> argument which defines what 5190<i>synchronizes-with</i> edges they add. They can only be given 5191<code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and 5192<code>seq_cst</code> orderings.</p> 5193 5194<h5>Semantics:</h5> 5195<p>A fence <var>A</var> which has (at least) <code>release</code> ordering 5196semantics <i>synchronizes with</i> a fence <var>B</var> with (at least) 5197<code>acquire</code> ordering semantics if and only if there exist atomic 5198operations <var>X</var> and <var>Y</var>, both operating on some atomic object 5199<var>M</var>, such that <var>A</var> is sequenced before <var>X</var>, 5200<var>X</var> modifies <var>M</var> (either directly or through some side effect 5201of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before 5202<var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a 5203<i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather 5204than an explicit <code>fence</code>, one (but not both) of the atomic operations 5205<var>X</var> or <var>Y</var> might provide a <code>release</code> or 5206<code>acquire</code> (resp.) ordering constraint and still 5207<i>synchronize-with</i> the explicit <code>fence</code> and establish the 5208<i>happens-before</i> edge.</p> 5209 5210<p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to 5211having both <code>acquire</code> and <code>release</code> semantics specified 5212above, participates in the global program order of other <code>seq_cst</code> 5213operations and/or fences.</p> 5214 5215<p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument 5216specifies that the fence only synchronizes with other fences in the same 5217thread. (This is useful for interacting with signal handlers.)</p> 5218 5219<h5>Example:</h5> 5220<pre> 5221 fence acquire <i>; yields {void}</i> 5222 fence singlethread seq_cst <i>; yields {void}</i> 5223</pre> 5224 5225</div> 5226 5227<!-- _______________________________________________________________________ --> 5228<h4> 5229<a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a> 5230</h4> 5231 5232<div> 5233 5234<h5>Syntax:</h5> 5235<pre> 5236 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i> 5237</pre> 5238 5239<h5>Overview:</h5> 5240<p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory. 5241It loads a value in memory and compares it to a given value. If they are 5242equal, it stores a new value into the memory.</p> 5243 5244<h5>Arguments:</h5> 5245<p>There are three arguments to the '<code>cmpxchg</code>' instruction: an 5246address to operate on, a value to compare to the value currently be at that 5247address, and a new value to place at that address if the compared values are 5248equal. The type of '<var><cmp></var>' must be an integer type whose 5249bit width is a power of two greater than or equal to eight and less than 5250or equal to a target-specific size limit. '<var><cmp></var>' and 5251'<var><new></var>' must have the same type, and the type of 5252'<var><pointer></var>' must be a pointer to that type. If the 5253<code>cmpxchg</code> is marked as <code>volatile</code>, then the 5254optimizer is not allowed to modify the number or order of execution 5255of this <code>cmpxchg</code> with other <a href="#volatile">volatile 5256operations</a>.</p> 5257 5258<!-- FIXME: Extend allowed types. --> 5259 5260<p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this 5261<code>cmpxchg</code> synchronizes with other atomic operations.</p> 5262 5263<p>The optional "<code>singlethread</code>" argument declares that the 5264<code>cmpxchg</code> is only atomic with respect to code (usually signal 5265handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the 5266cmpxchg is atomic with respect to all other code in the system.</p> 5267 5268<p>The pointer passed into cmpxchg must have alignment greater than or equal to 5269the size in memory of the operand. 5270 5271<h5>Semantics:</h5> 5272<p>The contents of memory at the location specified by the 5273'<tt><pointer></tt>' operand is read and compared to 5274'<tt><cmp></tt>'; if the read value is the equal, 5275'<tt><new></tt>' is written. The original value at the location 5276is returned. 5277 5278<p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the 5279purpose of identifying <a href="#release_sequence">release sequences</a>. A 5280failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering 5281parameter determined by dropping any <code>release</code> part of the 5282<code>cmpxchg</code>'s ordering.</p> 5283 5284<!-- 5285FIXME: Is compare_exchange_weak() necessary? (Consider after we've done 5286optimization work on ARM.) 5287 5288FIXME: Is a weaker ordering constraint on failure helpful in practice? 5289--> 5290 5291<h5>Example:</h5> 5292<pre> 5293entry: 5294 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i> 5295 <a href="#i_br">br</a> label %loop 5296 5297loop: 5298 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop] 5299 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp 5300 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i> 5301 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old 5302 <a href="#i_br">br</a> i1 %success, label %done, label %loop 5303 5304done: 5305 ... 5306</pre> 5307 5308</div> 5309 5310<!-- _______________________________________________________________________ --> 5311<h4> 5312<a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a> 5313</h4> 5314 5315<div> 5316 5317<h5>Syntax:</h5> 5318<pre> 5319 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i> 5320</pre> 5321 5322<h5>Overview:</h5> 5323<p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p> 5324 5325<h5>Arguments:</h5> 5326<p>There are three arguments to the '<code>atomicrmw</code>' instruction: an 5327operation to apply, an address whose value to modify, an argument to the 5328operation. The operation must be one of the following keywords:</p> 5329<ul> 5330 <li>xchg</li> 5331 <li>add</li> 5332 <li>sub</li> 5333 <li>and</li> 5334 <li>nand</li> 5335 <li>or</li> 5336 <li>xor</li> 5337 <li>max</li> 5338 <li>min</li> 5339 <li>umax</li> 5340 <li>umin</li> 5341</ul> 5342 5343<p>The type of '<var><value></var>' must be an integer type whose 5344bit width is a power of two greater than or equal to eight and less than 5345or equal to a target-specific size limit. The type of the 5346'<code><pointer></code>' operand must be a pointer to that type. 5347If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the 5348optimizer is not allowed to modify the number or order of execution of this 5349<code>atomicrmw</code> with other <a href="#volatile">volatile 5350 operations</a>.</p> 5351 5352<!-- FIXME: Extend allowed types. --> 5353 5354<h5>Semantics:</h5> 5355<p>The contents of memory at the location specified by the 5356'<tt><pointer></tt>' operand are atomically read, modified, and written 5357back. The original value at the location is returned. The modification is 5358specified by the <var>operation</var> argument:</p> 5359 5360<ul> 5361 <li>xchg: <code>*ptr = val</code></li> 5362 <li>add: <code>*ptr = *ptr + val</code></li> 5363 <li>sub: <code>*ptr = *ptr - val</code></li> 5364 <li>and: <code>*ptr = *ptr & val</code></li> 5365 <li>nand: <code>*ptr = ~(*ptr & val)</code></li> 5366 <li>or: <code>*ptr = *ptr | val</code></li> 5367 <li>xor: <code>*ptr = *ptr ^ val</code></li> 5368 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li> 5369 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li> 5370 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li> 5371 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li> 5372</ul> 5373 5374<h5>Example:</h5> 5375<pre> 5376 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i> 5377</pre> 5378 5379</div> 5380 5381<!-- _______________________________________________________________________ --> 5382<h4> 5383 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a> 5384</h4> 5385 5386<div> 5387 5388<h5>Syntax:</h5> 5389<pre> 5390 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}* 5391 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}* 5392 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx 5393</pre> 5394 5395<h5>Overview:</h5> 5396<p>The '<tt>getelementptr</tt>' instruction is used to get the address of a 5397 subelement of an <a href="#t_aggregate">aggregate</a> data structure. 5398 It performs address calculation only and does not access memory.</p> 5399 5400<h5>Arguments:</h5> 5401<p>The first argument is always a pointer or a vector of pointers, 5402 and forms the basis of the 5403 calculation. The remaining arguments are indices that indicate which of the 5404 elements of the aggregate object are indexed. The interpretation of each 5405 index is dependent on the type being indexed into. The first index always 5406 indexes the pointer value given as the first argument, the second index 5407 indexes a value of the type pointed to (not necessarily the value directly 5408 pointed to, since the first index can be non-zero), etc. The first type 5409 indexed into must be a pointer value, subsequent types can be arrays, 5410 vectors, and structs. Note that subsequent types being indexed into 5411 can never be pointers, since that would require loading the pointer before 5412 continuing calculation.</p> 5413 5414<p>The type of each index argument depends on the type it is indexing into. 5415 When indexing into a (optionally packed) structure, only <tt>i32</tt> 5416 integer <b>constants</b> are allowed. When indexing into an array, pointer 5417 or vector, integers of any width are allowed, and they are not required to be 5418 constant. These integers are treated as signed values where relevant.</p> 5419 5420<p>For example, let's consider a C code fragment and how it gets compiled to 5421 LLVM:</p> 5422 5423<pre class="doc_code"> 5424struct RT { 5425 char A; 5426 int B[10][20]; 5427 char C; 5428}; 5429struct ST { 5430 int X; 5431 double Y; 5432 struct RT Z; 5433}; 5434 5435int *foo(struct ST *s) { 5436 return &s[1].Z.B[5][13]; 5437} 5438</pre> 5439 5440<p>The LLVM code generated by Clang is:</p> 5441 5442<pre class="doc_code"> 5443%struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 } 5444%struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT } 5445 5446define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp { 5447entry: 5448 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13 5449 ret i32* %arrayidx 5450} 5451</pre> 5452 5453<h5>Semantics:</h5> 5454<p>In the example above, the first index is indexing into the 5455 '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a 5456 '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a 5457 structure. The second index indexes into the third element of the structure, 5458 yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>' 5459 type, another structure. The third index indexes into the second element of 5460 the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The 5461 two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>' 5462 type. The '<tt>getelementptr</tt>' instruction returns a pointer to this 5463 element, thus computing a value of '<tt>i32*</tt>' type.</p> 5464 5465<p>Note that it is perfectly legal to index partially through a structure, 5466 returning a pointer to an inner element. Because of this, the LLVM code for 5467 the given testcase is equivalent to:</p> 5468 5469<pre class="doc_code"> 5470define i32* @foo(%struct.ST* %s) { 5471 %t1 = getelementptr %struct.ST* %s, i32 1 <i>; yields %struct.ST*:%t1</i> 5472 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 <i>; yields %struct.RT*:%t2</i> 5473 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i> 5474 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i> 5475 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i> 5476 ret i32* %t5 5477} 5478</pre> 5479 5480<p>If the <tt>inbounds</tt> keyword is present, the result value of the 5481 <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the 5482 base pointer is not an <i>in bounds</i> address of an allocated object, 5483 or if any of the addresses that would be formed by successive addition of 5484 the offsets implied by the indices to the base address with infinitely 5485 precise signed arithmetic are not an <i>in bounds</i> address of that 5486 allocated object. The <i>in bounds</i> addresses for an allocated object 5487 are all the addresses that point into the object, plus the address one 5488 byte past the end. 5489 In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword 5490 applies to each of the computations element-wise. </p> 5491 5492<p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to 5493 the base address with silently-wrapping two's complement arithmetic. If the 5494 offsets have a different width from the pointer, they are sign-extended or 5495 truncated to the width of the pointer. The result value of the 5496 <tt>getelementptr</tt> may be outside the object pointed to by the base 5497 pointer. The result value may not necessarily be used to access memory 5498 though, even if it happens to point into allocated storage. See the 5499 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more 5500 information.</p> 5501 5502<p>The getelementptr instruction is often confusing. For some more insight into 5503 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p> 5504 5505<h5>Example:</h5> 5506<pre> 5507 <i>; yields [12 x i8]*:aptr</i> 5508 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1 5509 <i>; yields i8*:vptr</i> 5510 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1 5511 <i>; yields i8*:eptr</i> 5512 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1 5513 <i>; yields i32*:iptr</i> 5514 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0 5515</pre> 5516 5517<p>In cases where the pointer argument is a vector of pointers, only a 5518 single index may be used, and the number of vector elements has to be 5519 the same. For example: </p> 5520<pre class="doc_code"> 5521 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets, 5522</pre> 5523 5524</div> 5525 5526</div> 5527 5528<!-- ======================================================================= --> 5529<h3> 5530 <a name="convertops">Conversion Operations</a> 5531</h3> 5532 5533<div> 5534 5535<p>The instructions in this category are the conversion instructions (casting) 5536 which all take a single operand and a type. They perform various bit 5537 conversions on the operand.</p> 5538 5539<!-- _______________________________________________________________________ --> 5540<h4> 5541 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a> 5542</h4> 5543 5544<div> 5545 5546<h5>Syntax:</h5> 5547<pre> 5548 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i> 5549</pre> 5550 5551<h5>Overview:</h5> 5552<p>The '<tt>trunc</tt>' instruction truncates its operand to the 5553 type <tt>ty2</tt>.</p> 5554 5555<h5>Arguments:</h5> 5556<p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to. 5557 Both types must be of <a href="#t_integer">integer</a> types, or vectors 5558 of the same number of integers. 5559 The bit size of the <tt>value</tt> must be larger than 5560 the bit size of the destination type, <tt>ty2</tt>. 5561 Equal sized types are not allowed.</p> 5562 5563<h5>Semantics:</h5> 5564<p>The '<tt>trunc</tt>' instruction truncates the high order bits 5565 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the 5566 source size must be larger than the destination size, <tt>trunc</tt> cannot 5567 be a <i>no-op cast</i>. It will always truncate bits.</p> 5568 5569<h5>Example:</h5> 5570<pre> 5571 %X = trunc i32 257 to i8 <i>; yields i8:1</i> 5572 %Y = trunc i32 123 to i1 <i>; yields i1:true</i> 5573 %Z = trunc i32 122 to i1 <i>; yields i1:false</i> 5574 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i> 5575</pre> 5576 5577</div> 5578 5579<!-- _______________________________________________________________________ --> 5580<h4> 5581 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a> 5582</h4> 5583 5584<div> 5585 5586<h5>Syntax:</h5> 5587<pre> 5588 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i> 5589</pre> 5590 5591<h5>Overview:</h5> 5592<p>The '<tt>zext</tt>' instruction zero extends its operand to type 5593 <tt>ty2</tt>.</p> 5594 5595 5596<h5>Arguments:</h5> 5597<p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to. 5598 Both types must be of <a href="#t_integer">integer</a> types, or vectors 5599 of the same number of integers. 5600 The bit size of the <tt>value</tt> must be smaller than 5601 the bit size of the destination type, 5602 <tt>ty2</tt>.</p> 5603 5604<h5>Semantics:</h5> 5605<p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero 5606 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p> 5607 5608<p>When zero extending from i1, the result will always be either 0 or 1.</p> 5609 5610<h5>Example:</h5> 5611<pre> 5612 %X = zext i32 257 to i64 <i>; yields i64:257</i> 5613 %Y = zext i1 true to i32 <i>; yields i32:1</i> 5614 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i> 5615</pre> 5616 5617</div> 5618 5619<!-- _______________________________________________________________________ --> 5620<h4> 5621 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a> 5622</h4> 5623 5624<div> 5625 5626<h5>Syntax:</h5> 5627<pre> 5628 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i> 5629</pre> 5630 5631<h5>Overview:</h5> 5632<p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p> 5633 5634<h5>Arguments:</h5> 5635<p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to. 5636 Both types must be of <a href="#t_integer">integer</a> types, or vectors 5637 of the same number of integers. 5638 The bit size of the <tt>value</tt> must be smaller than 5639 the bit size of the destination type, 5640 <tt>ty2</tt>.</p> 5641 5642<h5>Semantics:</h5> 5643<p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign 5644 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size 5645 of the type <tt>ty2</tt>.</p> 5646 5647<p>When sign extending from i1, the extension always results in -1 or 0.</p> 5648 5649<h5>Example:</h5> 5650<pre> 5651 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i> 5652 %Y = sext i1 true to i32 <i>; yields i32:-1</i> 5653 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i> 5654</pre> 5655 5656</div> 5657 5658<!-- _______________________________________________________________________ --> 5659<h4> 5660 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a> 5661</h4> 5662 5663<div> 5664 5665<h5>Syntax:</h5> 5666<pre> 5667 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i> 5668</pre> 5669 5670<h5>Overview:</h5> 5671<p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type 5672 <tt>ty2</tt>.</p> 5673 5674<h5>Arguments:</h5> 5675<p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating 5676 point</a> value to cast and a <a href="#t_floating">floating point</a> type 5677 to cast it to. The size of <tt>value</tt> must be larger than the size of 5678 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a 5679 <i>no-op cast</i>.</p> 5680 5681<h5>Semantics:</h5> 5682<p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger 5683 <a href="#t_floating">floating point</a> type to a smaller 5684 <a href="#t_floating">floating point</a> type. If the value cannot fit 5685 within the destination type, <tt>ty2</tt>, then the results are 5686 undefined.</p> 5687 5688<h5>Example:</h5> 5689<pre> 5690 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i> 5691 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i> 5692</pre> 5693 5694</div> 5695 5696<!-- _______________________________________________________________________ --> 5697<h4> 5698 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a> 5699</h4> 5700 5701<div> 5702 5703<h5>Syntax:</h5> 5704<pre> 5705 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i> 5706</pre> 5707 5708<h5>Overview:</h5> 5709<p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger 5710 floating point value.</p> 5711 5712<h5>Arguments:</h5> 5713<p>The '<tt>fpext</tt>' instruction takes a 5714 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and 5715 a <a href="#t_floating">floating point</a> type to cast it to. The source 5716 type must be smaller than the destination type.</p> 5717 5718<h5>Semantics:</h5> 5719<p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller 5720 <a href="#t_floating">floating point</a> type to a larger 5721 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be 5722 used to make a <i>no-op cast</i> because it always changes bits. Use 5723 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p> 5724 5725<h5>Example:</h5> 5726<pre> 5727 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i> 5728 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i> 5729</pre> 5730 5731</div> 5732 5733<!-- _______________________________________________________________________ --> 5734<h4> 5735 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a> 5736</h4> 5737 5738<div> 5739 5740<h5>Syntax:</h5> 5741<pre> 5742 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i> 5743</pre> 5744 5745<h5>Overview:</h5> 5746<p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its 5747 unsigned integer equivalent of type <tt>ty2</tt>.</p> 5748 5749<h5>Arguments:</h5> 5750<p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a 5751 scalar or vector <a href="#t_floating">floating point</a> value, and a type 5752 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> 5753 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a 5754 vector integer type with the same number of elements as <tt>ty</tt></p> 5755 5756<h5>Semantics:</h5> 5757<p>The '<tt>fptoui</tt>' instruction converts its 5758 <a href="#t_floating">floating point</a> operand into the nearest (rounding 5759 towards zero) unsigned integer value. If the value cannot fit 5760 in <tt>ty2</tt>, the results are undefined.</p> 5761 5762<h5>Example:</h5> 5763<pre> 5764 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i> 5765 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i> 5766 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i> 5767</pre> 5768 5769</div> 5770 5771<!-- _______________________________________________________________________ --> 5772<h4> 5773 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a> 5774</h4> 5775 5776<div> 5777 5778<h5>Syntax:</h5> 5779<pre> 5780 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i> 5781</pre> 5782 5783<h5>Overview:</h5> 5784<p>The '<tt>fptosi</tt>' instruction converts 5785 <a href="#t_floating">floating point</a> <tt>value</tt> to 5786 type <tt>ty2</tt>.</p> 5787 5788<h5>Arguments:</h5> 5789<p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a 5790 scalar or vector <a href="#t_floating">floating point</a> value, and a type 5791 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> 5792 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a 5793 vector integer type with the same number of elements as <tt>ty</tt></p> 5794 5795<h5>Semantics:</h5> 5796<p>The '<tt>fptosi</tt>' instruction converts its 5797 <a href="#t_floating">floating point</a> operand into the nearest (rounding 5798 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>, 5799 the results are undefined.</p> 5800 5801<h5>Example:</h5> 5802<pre> 5803 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i> 5804 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i> 5805 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i> 5806</pre> 5807 5808</div> 5809 5810<!-- _______________________________________________________________________ --> 5811<h4> 5812 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a> 5813</h4> 5814 5815<div> 5816 5817<h5>Syntax:</h5> 5818<pre> 5819 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i> 5820</pre> 5821 5822<h5>Overview:</h5> 5823<p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned 5824 integer and converts that value to the <tt>ty2</tt> type.</p> 5825 5826<h5>Arguments:</h5> 5827<p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a 5828 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast 5829 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a> 5830 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector 5831 floating point type with the same number of elements as <tt>ty</tt></p> 5832 5833<h5>Semantics:</h5> 5834<p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned 5835 integer quantity and converts it to the corresponding floating point 5836 value. If the value cannot fit in the floating point value, the results are 5837 undefined.</p> 5838 5839<h5>Example:</h5> 5840<pre> 5841 %X = uitofp i32 257 to float <i>; yields float:257.0</i> 5842 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i> 5843</pre> 5844 5845</div> 5846 5847<!-- _______________________________________________________________________ --> 5848<h4> 5849 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a> 5850</h4> 5851 5852<div> 5853 5854<h5>Syntax:</h5> 5855<pre> 5856 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i> 5857</pre> 5858 5859<h5>Overview:</h5> 5860<p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer 5861 and converts that value to the <tt>ty2</tt> type.</p> 5862 5863<h5>Arguments:</h5> 5864<p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a 5865 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast 5866 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a> 5867 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector 5868 floating point type with the same number of elements as <tt>ty</tt></p> 5869 5870<h5>Semantics:</h5> 5871<p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer 5872 quantity and converts it to the corresponding floating point value. If the 5873 value cannot fit in the floating point value, the results are undefined.</p> 5874 5875<h5>Example:</h5> 5876<pre> 5877 %X = sitofp i32 257 to float <i>; yields float:257.0</i> 5878 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i> 5879</pre> 5880 5881</div> 5882 5883<!-- _______________________________________________________________________ --> 5884<h4> 5885 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a> 5886</h4> 5887 5888<div> 5889 5890<h5>Syntax:</h5> 5891<pre> 5892 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i> 5893</pre> 5894 5895<h5>Overview:</h5> 5896<p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of 5897 pointers <tt>value</tt> to 5898 the integer (or vector of integers) type <tt>ty2</tt>.</p> 5899 5900<h5>Arguments:</h5> 5901<p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which 5902 must be a a value of type <a href="#t_pointer">pointer</a> or a vector of 5903 pointers, and a type to cast it to 5904 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector 5905 of integers type.</p> 5906 5907<h5>Semantics:</h5> 5908<p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type 5909 <tt>ty2</tt> by interpreting the pointer value as an integer and either 5910 truncating or zero extending that value to the size of the integer type. If 5911 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If 5912 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they 5913 are the same size, then nothing is done (<i>no-op cast</i>) other than a type 5914 change.</p> 5915 5916<h5>Example:</h5> 5917<pre> 5918 %X = ptrtoint i32* %P to i8 <i>; yields truncation on 32-bit architecture</i> 5919 %Y = ptrtoint i32* %P to i64 <i>; yields zero extension on 32-bit architecture</i> 5920 %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> 5921</pre> 5922 5923</div> 5924 5925<!-- _______________________________________________________________________ --> 5926<h4> 5927 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a> 5928</h4> 5929 5930<div> 5931 5932<h5>Syntax:</h5> 5933<pre> 5934 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i> 5935</pre> 5936 5937<h5>Overview:</h5> 5938<p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a 5939 pointer type, <tt>ty2</tt>.</p> 5940 5941<h5>Arguments:</h5> 5942<p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a> 5943 value to cast, and a type to cast it to, which must be a 5944 <a href="#t_pointer">pointer</a> type.</p> 5945 5946<h5>Semantics:</h5> 5947<p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type 5948 <tt>ty2</tt> by applying either a zero extension or a truncation depending on 5949 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the 5950 size of a pointer then a truncation is done. If <tt>value</tt> is smaller 5951 than the size of a pointer then a zero extension is done. If they are the 5952 same size, nothing is done (<i>no-op cast</i>).</p> 5953 5954<h5>Example:</h5> 5955<pre> 5956 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i> 5957 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i> 5958 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i> 5959 %Z = inttoptr <4 x i32> %G to <4 x i8*><i>; yields truncation of vector G to four pointers</i> 5960</pre> 5961 5962</div> 5963 5964<!-- _______________________________________________________________________ --> 5965<h4> 5966 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a> 5967</h4> 5968 5969<div> 5970 5971<h5>Syntax:</h5> 5972<pre> 5973 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i> 5974</pre> 5975 5976<h5>Overview:</h5> 5977<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type 5978 <tt>ty2</tt> without changing any bits.</p> 5979 5980<h5>Arguments:</h5> 5981<p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a 5982 non-aggregate first class value, and a type to cast it to, which must also be 5983 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes 5984 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be 5985 identical. If the source type is a pointer, the destination type must also be 5986 a pointer. This instruction supports bitwise conversion of vectors to 5987 integers and to vectors of other types (as long as they have the same 5988 size).</p> 5989 5990<h5>Semantics:</h5> 5991<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type 5992 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with 5993 this conversion. The conversion is done as if the <tt>value</tt> had been 5994 stored to memory and read back as type <tt>ty2</tt>. 5995 Pointer (or vector of pointers) types may only be converted to other pointer 5996 (or vector of pointers) types with this instruction. To convert 5997 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or 5998 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p> 5999 6000<h5>Example:</h5> 6001<pre> 6002 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i> 6003 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i> 6004 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i> 6005 %Z = bitcast <2 x i32*> %V to <2 x i64*> <i>; yields <2 x i64*></i> 6006</pre> 6007 6008</div> 6009 6010</div> 6011 6012<!-- ======================================================================= --> 6013<h3> 6014 <a name="otherops">Other Operations</a> 6015</h3> 6016 6017<div> 6018 6019<p>The instructions in this category are the "miscellaneous" instructions, which 6020 defy better classification.</p> 6021 6022<!-- _______________________________________________________________________ --> 6023<h4> 6024 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a> 6025</h4> 6026 6027<div> 6028 6029<h5>Syntax:</h5> 6030<pre> 6031 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i> 6032</pre> 6033 6034<h5>Overview:</h5> 6035<p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of 6036 boolean values based on comparison of its two integer, integer vector, 6037 pointer, or pointer vector operands.</p> 6038 6039<h5>Arguments:</h5> 6040<p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is 6041 the condition code indicating the kind of comparison to perform. It is not a 6042 value, just a keyword. The possible condition code are:</p> 6043 6044<ol> 6045 <li><tt>eq</tt>: equal</li> 6046 <li><tt>ne</tt>: not equal </li> 6047 <li><tt>ugt</tt>: unsigned greater than</li> 6048 <li><tt>uge</tt>: unsigned greater or equal</li> 6049 <li><tt>ult</tt>: unsigned less than</li> 6050 <li><tt>ule</tt>: unsigned less or equal</li> 6051 <li><tt>sgt</tt>: signed greater than</li> 6052 <li><tt>sge</tt>: signed greater or equal</li> 6053 <li><tt>slt</tt>: signed less than</li> 6054 <li><tt>sle</tt>: signed less or equal</li> 6055</ol> 6056 6057<p>The remaining two arguments must be <a href="#t_integer">integer</a> or 6058 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a> 6059 typed. They must also be identical types.</p> 6060 6061<h5>Semantics:</h5> 6062<p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the 6063 condition code given as <tt>cond</tt>. The comparison performed always yields 6064 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt> 6065 result, as follows:</p> 6066 6067<ol> 6068 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal, 6069 <tt>false</tt> otherwise. No sign interpretation is necessary or 6070 performed.</li> 6071 6072 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal, 6073 <tt>false</tt> otherwise. No sign interpretation is necessary or 6074 performed.</li> 6075 6076 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields 6077 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li> 6078 6079 <li><tt>uge</tt>: interprets the operands as unsigned values and yields 6080 <tt>true</tt> if <tt>op1</tt> is greater than or equal 6081 to <tt>op2</tt>.</li> 6082 6083 <li><tt>ult</tt>: interprets the operands as unsigned values and yields 6084 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li> 6085 6086 <li><tt>ule</tt>: interprets the operands as unsigned values and yields 6087 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li> 6088 6089 <li><tt>sgt</tt>: interprets the operands as signed values and yields 6090 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li> 6091 6092 <li><tt>sge</tt>: interprets the operands as signed values and yields 6093 <tt>true</tt> if <tt>op1</tt> is greater than or equal 6094 to <tt>op2</tt>.</li> 6095 6096 <li><tt>slt</tt>: interprets the operands as signed values and yields 6097 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li> 6098 6099 <li><tt>sle</tt>: interprets the operands as signed values and yields 6100 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li> 6101</ol> 6102 6103<p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer 6104 values are compared as if they were integers.</p> 6105 6106<p>If the operands are integer vectors, then they are compared element by 6107 element. The result is an <tt>i1</tt> vector with the same number of elements 6108 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p> 6109 6110<h5>Example:</h5> 6111<pre> 6112 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i> 6113 <result> = icmp ne float* %X, %X <i>; yields: result=false</i> 6114 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i> 6115 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i> 6116 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i> 6117 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i> 6118</pre> 6119 6120<p>Note that the code generator does not yet support vector types with 6121 the <tt>icmp</tt> instruction.</p> 6122 6123</div> 6124 6125<!-- _______________________________________________________________________ --> 6126<h4> 6127 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a> 6128</h4> 6129 6130<div> 6131 6132<h5>Syntax:</h5> 6133<pre> 6134 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i> 6135</pre> 6136 6137<h5>Overview:</h5> 6138<p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean 6139 values based on comparison of its operands.</p> 6140 6141<p>If the operands are floating point scalars, then the result type is a boolean 6142(<a href="#t_integer"><tt>i1</tt></a>).</p> 6143 6144<p>If the operands are floating point vectors, then the result type is a vector 6145 of boolean with the same number of elements as the operands being 6146 compared.</p> 6147 6148<h5>Arguments:</h5> 6149<p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is 6150 the condition code indicating the kind of comparison to perform. It is not a 6151 value, just a keyword. The possible condition code are:</p> 6152 6153<ol> 6154 <li><tt>false</tt>: no comparison, always returns false</li> 6155 <li><tt>oeq</tt>: ordered and equal</li> 6156 <li><tt>ogt</tt>: ordered and greater than </li> 6157 <li><tt>oge</tt>: ordered and greater than or equal</li> 6158 <li><tt>olt</tt>: ordered and less than </li> 6159 <li><tt>ole</tt>: ordered and less than or equal</li> 6160 <li><tt>one</tt>: ordered and not equal</li> 6161 <li><tt>ord</tt>: ordered (no nans)</li> 6162 <li><tt>ueq</tt>: unordered or equal</li> 6163 <li><tt>ugt</tt>: unordered or greater than </li> 6164 <li><tt>uge</tt>: unordered or greater than or equal</li> 6165 <li><tt>ult</tt>: unordered or less than </li> 6166 <li><tt>ule</tt>: unordered or less than or equal</li> 6167 <li><tt>une</tt>: unordered or not equal</li> 6168 <li><tt>uno</tt>: unordered (either nans)</li> 6169 <li><tt>true</tt>: no comparison, always returns true</li> 6170</ol> 6171 6172<p><i>Ordered</i> means that neither operand is a QNAN while 6173 <i>unordered</i> means that either operand may be a QNAN.</p> 6174 6175<p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either 6176 a <a href="#t_floating">floating point</a> type or 6177 a <a href="#t_vector">vector</a> of floating point type. They must have 6178 identical types.</p> 6179 6180<h5>Semantics:</h5> 6181<p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt> 6182 according to the condition code given as <tt>cond</tt>. If the operands are 6183 vectors, then the vectors are compared element by element. Each comparison 6184 performed always yields an <a href="#t_integer">i1</a> result, as 6185 follows:</p> 6186 6187<ol> 6188 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li> 6189 6190 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and 6191 <tt>op1</tt> is equal to <tt>op2</tt>.</li> 6192 6193 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and 6194 <tt>op1</tt> is greater than <tt>op2</tt>.</li> 6195 6196 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and 6197 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li> 6198 6199 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and 6200 <tt>op1</tt> is less than <tt>op2</tt>.</li> 6201 6202 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and 6203 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li> 6204 6205 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and 6206 <tt>op1</tt> is not equal to <tt>op2</tt>.</li> 6207 6208 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li> 6209 6210 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or 6211 <tt>op1</tt> is equal to <tt>op2</tt>.</li> 6212 6213 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or 6214 <tt>op1</tt> is greater than <tt>op2</tt>.</li> 6215 6216 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or 6217 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li> 6218 6219 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or 6220 <tt>op1</tt> is less than <tt>op2</tt>.</li> 6221 6222 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or 6223 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li> 6224 6225 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or 6226 <tt>op1</tt> is not equal to <tt>op2</tt>.</li> 6227 6228 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li> 6229 6230 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li> 6231</ol> 6232 6233<h5>Example:</h5> 6234<pre> 6235 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i> 6236 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i> 6237 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i> 6238 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i> 6239</pre> 6240 6241<p>Note that the code generator does not yet support vector types with 6242 the <tt>fcmp</tt> instruction.</p> 6243 6244</div> 6245 6246<!-- _______________________________________________________________________ --> 6247<h4> 6248 <a name="i_phi">'<tt>phi</tt>' Instruction</a> 6249</h4> 6250 6251<div> 6252 6253<h5>Syntax:</h5> 6254<pre> 6255 <result> = phi <ty> [ <val0>, <label0>], ... 6256</pre> 6257 6258<h5>Overview:</h5> 6259<p>The '<tt>phi</tt>' instruction is used to implement the φ node in the 6260 SSA graph representing the function.</p> 6261 6262<h5>Arguments:</h5> 6263<p>The type of the incoming values is specified with the first type field. After 6264 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with 6265 one pair for each predecessor basic block of the current block. Only values 6266 of <a href="#t_firstclass">first class</a> type may be used as the value 6267 arguments to the PHI node. Only labels may be used as the label 6268 arguments.</p> 6269 6270<p>There must be no non-phi instructions between the start of a basic block and 6271 the PHI instructions: i.e. PHI instructions must be first in a basic 6272 block.</p> 6273 6274<p>For the purposes of the SSA form, the use of each incoming value is deemed to 6275 occur on the edge from the corresponding predecessor block to the current 6276 block (but after any definition of an '<tt>invoke</tt>' instruction's return 6277 value on the same edge).</p> 6278 6279<h5>Semantics:</h5> 6280<p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value 6281 specified by the pair corresponding to the predecessor basic block that 6282 executed just prior to the current block.</p> 6283 6284<h5>Example:</h5> 6285<pre> 6286Loop: ; Infinite loop that counts from 0 on up... 6287 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ] 6288 %nextindvar = add i32 %indvar, 1 6289 br label %Loop 6290</pre> 6291 6292</div> 6293 6294<!-- _______________________________________________________________________ --> 6295<h4> 6296 <a name="i_select">'<tt>select</tt>' Instruction</a> 6297</h4> 6298 6299<div> 6300 6301<h5>Syntax:</h5> 6302<pre> 6303 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i> 6304 6305 <i>selty</i> is either i1 or {<N x i1>} 6306</pre> 6307 6308<h5>Overview:</h5> 6309<p>The '<tt>select</tt>' instruction is used to choose one value based on a 6310 condition, without branching.</p> 6311 6312 6313<h5>Arguments:</h5> 6314<p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1' 6315 values indicating the condition, and two values of the 6316 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are 6317 vectors and the condition is a scalar, then entire vectors are selected, not 6318 individual elements.</p> 6319 6320<h5>Semantics:</h5> 6321<p>If the condition is an i1 and it evaluates to 1, the instruction returns the 6322 first value argument; otherwise, it returns the second value argument.</p> 6323 6324<p>If the condition is a vector of i1, then the value arguments must be vectors 6325 of the same size, and the selection is done element by element.</p> 6326 6327<h5>Example:</h5> 6328<pre> 6329 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i> 6330</pre> 6331 6332</div> 6333 6334<!-- _______________________________________________________________________ --> 6335<h4> 6336 <a name="i_call">'<tt>call</tt>' Instruction</a> 6337</h4> 6338 6339<div> 6340 6341<h5>Syntax:</h5> 6342<pre> 6343 <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>] 6344</pre> 6345 6346<h5>Overview:</h5> 6347<p>The '<tt>call</tt>' instruction represents a simple function call.</p> 6348 6349<h5>Arguments:</h5> 6350<p>This instruction requires several arguments:</p> 6351 6352<ol> 6353 <li>The optional "tail" marker indicates that the callee function does not 6354 access any allocas or varargs in the caller. Note that calls may be 6355 marked "tail" even if they do not occur before 6356 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is 6357 present, the function call is eligible for tail call optimization, 6358 but <a href="CodeGenerator.html#tailcallopt">might not in fact be 6359 optimized into a jump</a>. The code generator may optimize calls marked 6360 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt"> 6361 sibling call optimization</a> when the caller and callee have 6362 matching signatures, or 2) forced tail call optimization when the 6363 following extra requirements are met: 6364 <ul> 6365 <li>Caller and callee both have the calling 6366 convention <tt>fastcc</tt>.</li> 6367 <li>The call is in tail position (ret immediately follows call and ret 6368 uses value of call or is void).</li> 6369 <li>Option <tt>-tailcallopt</tt> is enabled, 6370 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li> 6371 <li><a href="CodeGenerator.html#tailcallopt">Platform specific 6372 constraints are met.</a></li> 6373 </ul> 6374 </li> 6375 6376 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling 6377 convention</a> the call should use. If none is specified, the call 6378 defaults to using C calling conventions. The calling convention of the 6379 call must match the calling convention of the target function, or else the 6380 behavior is undefined.</li> 6381 6382 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for 6383 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and 6384 '<tt>inreg</tt>' attributes are valid here.</li> 6385 6386 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the 6387 type of the return value. Functions that return no value are marked 6388 <tt><a href="#t_void">void</a></tt>.</li> 6389 6390 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value 6391 being invoked. The argument types must match the types implied by this 6392 signature. This type can be omitted if the function is not varargs and if 6393 the function type does not return a pointer to a function.</li> 6394 6395 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to 6396 be invoked. In most cases, this is a direct function invocation, but 6397 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer 6398 to function value.</li> 6399 6400 <li>'<tt>function args</tt>': argument list whose types match the function 6401 signature argument types and parameter attributes. All arguments must be 6402 of <a href="#t_firstclass">first class</a> type. If the function 6403 signature indicates the function accepts a variable number of arguments, 6404 the extra arguments can be specified.</li> 6405 6406 <li>The optional <a href="#fnattrs">function attributes</a> list. Only 6407 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and 6408 '<tt>readnone</tt>' attributes are valid here.</li> 6409</ol> 6410 6411<h5>Semantics:</h5> 6412<p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to 6413 a specified function, with its incoming arguments bound to the specified 6414 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called 6415 function, control flow continues with the instruction after the function 6416 call, and the return value of the function is bound to the result 6417 argument.</p> 6418 6419<h5>Example:</h5> 6420<pre> 6421 %retval = call i32 @test(i32 %argc) 6422 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i> 6423 %X = tail call i32 @foo() <i>; yields i32</i> 6424 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i> 6425 call void %foo(i8 97 signext) 6426 6427 %struct.A = type { i32, i8 } 6428 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i> 6429 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i> 6430 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i> 6431 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i> 6432 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i> 6433</pre> 6434 6435<p>llvm treats calls to some functions with names and arguments that match the 6436standard C99 library as being the C99 library functions, and may perform 6437optimizations or generate code for them under that assumption. This is 6438something we'd like to change in the future to provide better support for 6439freestanding environments and non-C-based languages.</p> 6440 6441</div> 6442 6443<!-- _______________________________________________________________________ --> 6444<h4> 6445 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a> 6446</h4> 6447 6448<div> 6449 6450<h5>Syntax:</h5> 6451<pre> 6452 <resultval> = va_arg <va_list*> <arglist>, <argty> 6453</pre> 6454 6455<h5>Overview:</h5> 6456<p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through 6457 the "variable argument" area of a function call. It is used to implement the 6458 <tt>va_arg</tt> macro in C.</p> 6459 6460<h5>Arguments:</h5> 6461<p>This instruction takes a <tt>va_list*</tt> value and the type of the 6462 argument. It returns a value of the specified argument type and increments 6463 the <tt>va_list</tt> to point to the next argument. The actual type 6464 of <tt>va_list</tt> is target specific.</p> 6465 6466<h5>Semantics:</h5> 6467<p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type 6468 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point 6469 to the next argument. For more information, see the variable argument 6470 handling <a href="#int_varargs">Intrinsic Functions</a>.</p> 6471 6472<p>It is legal for this instruction to be called in a function which does not 6473 take a variable number of arguments, for example, the <tt>vfprintf</tt> 6474 function.</p> 6475 6476<p><tt>va_arg</tt> is an LLVM instruction instead of 6477 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an 6478 argument.</p> 6479 6480<h5>Example:</h5> 6481<p>See the <a href="#int_varargs">variable argument processing</a> section.</p> 6482 6483<p>Note that the code generator does not yet fully support va_arg on many 6484 targets. Also, it does not currently support va_arg with aggregate types on 6485 any target.</p> 6486 6487</div> 6488 6489<!-- _______________________________________________________________________ --> 6490<h4> 6491 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a> 6492</h4> 6493 6494<div> 6495 6496<h5>Syntax:</h5> 6497<pre> 6498 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+ 6499 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>* 6500 6501 <clause> := catch <type> <value> 6502 <clause> := filter <array constant type> <array constant> 6503</pre> 6504 6505<h5>Overview:</h5> 6506<p>The '<tt>landingpad</tt>' instruction is used by 6507 <a href="ExceptionHandling.html#overview">LLVM's exception handling 6508 system</a> to specify that a basic block is a landing pad — one where 6509 the exception lands, and corresponds to the code found in the 6510 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It 6511 defines values supplied by the personality function (<tt>pers_fn</tt>) upon 6512 re-entry to the function. The <tt>resultval</tt> has the 6513 type <tt>resultty</tt>.</p> 6514 6515<h5>Arguments:</h5> 6516<p>This instruction takes a <tt>pers_fn</tt> value. This is the personality 6517 function associated with the unwinding mechanism. The optional 6518 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p> 6519 6520<p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt> 6521 or <tt>filter</tt> — and contains the global variable representing the 6522 "type" that may be caught or filtered respectively. Unlike the 6523 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as 6524 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot 6525 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em> 6526 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p> 6527 6528<h5>Semantics:</h5> 6529<p>The '<tt>landingpad</tt>' instruction defines the values which are set by the 6530 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and 6531 therefore the "result type" of the <tt>landingpad</tt> instruction. As with 6532 calling conventions, how the personality function results are represented in 6533 LLVM IR is target specific.</p> 6534 6535<p>The clauses are applied in order from top to bottom. If two 6536 <tt>landingpad</tt> instructions are merged together through inlining, the 6537 clauses from the calling function are appended to the list of clauses. 6538 When the call stack is being unwound due to an exception being thrown, the 6539 exception is compared against each <tt>clause</tt> in turn. If it doesn't 6540 match any of the clauses, and the <tt>cleanup</tt> flag is not set, then 6541 unwinding continues further up the call stack.</p> 6542 6543<p>The <tt>landingpad</tt> instruction has several restrictions:</p> 6544 6545<ul> 6546 <li>A landing pad block is a basic block which is the unwind destination of an 6547 '<tt>invoke</tt>' instruction.</li> 6548 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its 6549 first non-PHI instruction.</li> 6550 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing 6551 pad block.</li> 6552 <li>A basic block that is not a landing pad block may not include a 6553 '<tt>landingpad</tt>' instruction.</li> 6554 <li>All '<tt>landingpad</tt>' instructions in a function must have the same 6555 personality function.</li> 6556</ul> 6557 6558<h5>Example:</h5> 6559<pre> 6560 ;; A landing pad which can catch an integer. 6561 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0 6562 catch i8** @_ZTIi 6563 ;; A landing pad that is a cleanup. 6564 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0 6565 cleanup 6566 ;; A landing pad which can catch an integer and can only throw a double. 6567 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0 6568 catch i8** @_ZTIi 6569 filter [1 x i8**] [@_ZTId] 6570</pre> 6571 6572</div> 6573 6574</div> 6575 6576</div> 6577 6578<!-- *********************************************************************** --> 6579<h2><a name="intrinsics">Intrinsic Functions</a></h2> 6580<!-- *********************************************************************** --> 6581 6582<div> 6583 6584<p>LLVM supports the notion of an "intrinsic function". These functions have 6585 well known names and semantics and are required to follow certain 6586 restrictions. Overall, these intrinsics represent an extension mechanism for 6587 the LLVM language that does not require changing all of the transformations 6588 in LLVM when adding to the language (or the bitcode reader/writer, the 6589 parser, etc...).</p> 6590 6591<p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This 6592 prefix is reserved in LLVM for intrinsic names; thus, function names may not 6593 begin with this prefix. Intrinsic functions must always be external 6594 functions: you cannot define the body of intrinsic functions. Intrinsic 6595 functions may only be used in call or invoke instructions: it is illegal to 6596 take the address of an intrinsic function. Additionally, because intrinsic 6597 functions are part of the LLVM language, it is required if any are added that 6598 they be documented here.</p> 6599 6600<p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a 6601 family of functions that perform the same operation but on different data 6602 types. Because LLVM can represent over 8 million different integer types, 6603 overloading is used commonly to allow an intrinsic function to operate on any 6604 integer type. One or more of the argument types or the result type can be 6605 overloaded to accept any integer type. Argument types may also be defined as 6606 exactly matching a previous argument's type or the result type. This allows 6607 an intrinsic function which accepts multiple arguments, but needs all of them 6608 to be of the same type, to only be overloaded with respect to a single 6609 argument or the result.</p> 6610 6611<p>Overloaded intrinsics will have the names of its overloaded argument types 6612 encoded into its function name, each preceded by a period. Only those types 6613 which are overloaded result in a name suffix. Arguments whose type is matched 6614 against another type do not. For example, the <tt>llvm.ctpop</tt> function 6615 can take an integer of any width and returns an integer of exactly the same 6616 integer width. This leads to a family of functions such as 6617 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 6618 %val)</tt>. Only one type, the return type, is overloaded, and only one type 6619 suffix is required. Because the argument's type is matched against the return 6620 type, it does not require its own name suffix.</p> 6621 6622<p>To learn how to add an intrinsic function, please see the 6623 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p> 6624 6625<!-- ======================================================================= --> 6626<h3> 6627 <a name="int_varargs">Variable Argument Handling Intrinsics</a> 6628</h3> 6629 6630<div> 6631 6632<p>Variable argument support is defined in LLVM with 6633 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three 6634 intrinsic functions. These functions are related to the similarly named 6635 macros defined in the <tt><stdarg.h></tt> header file.</p> 6636 6637<p>All of these functions operate on arguments that use a target-specific value 6638 type "<tt>va_list</tt>". The LLVM assembly language reference manual does 6639 not define what this type is, so all transformations should be prepared to 6640 handle these functions regardless of the type used.</p> 6641 6642<p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a> 6643 instruction and the variable argument handling intrinsic functions are 6644 used.</p> 6645 6646<pre class="doc_code"> 6647define i32 @test(i32 %X, ...) { 6648 ; Initialize variable argument processing 6649 %ap = alloca i8* 6650 %ap2 = bitcast i8** %ap to i8* 6651 call void @llvm.va_start(i8* %ap2) 6652 6653 ; Read a single integer argument 6654 %tmp = va_arg i8** %ap, i32 6655 6656 ; Demonstrate usage of llvm.va_copy and llvm.va_end 6657 %aq = alloca i8* 6658 %aq2 = bitcast i8** %aq to i8* 6659 call void @llvm.va_copy(i8* %aq2, i8* %ap2) 6660 call void @llvm.va_end(i8* %aq2) 6661 6662 ; Stop processing of arguments. 6663 call void @llvm.va_end(i8* %ap2) 6664 ret i32 %tmp 6665} 6666 6667declare void @llvm.va_start(i8*) 6668declare void @llvm.va_copy(i8*, i8*) 6669declare void @llvm.va_end(i8*) 6670</pre> 6671 6672<!-- _______________________________________________________________________ --> 6673<h4> 6674 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a> 6675</h4> 6676 6677 6678<div> 6679 6680<h5>Syntax:</h5> 6681<pre> 6682 declare void %llvm.va_start(i8* <arglist>) 6683</pre> 6684 6685<h5>Overview:</h5> 6686<p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt> 6687 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p> 6688 6689<h5>Arguments:</h5> 6690<p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p> 6691 6692<h5>Semantics:</h5> 6693<p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt> 6694 macro available in C. In a target-dependent way, it initializes 6695 the <tt>va_list</tt> element to which the argument points, so that the next 6696 call to <tt>va_arg</tt> will produce the first variable argument passed to 6697 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not 6698 need to know the last argument of the function as the compiler can figure 6699 that out.</p> 6700 6701</div> 6702 6703<!-- _______________________________________________________________________ --> 6704<h4> 6705 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a> 6706</h4> 6707 6708<div> 6709 6710<h5>Syntax:</h5> 6711<pre> 6712 declare void @llvm.va_end(i8* <arglist>) 6713</pre> 6714 6715<h5>Overview:</h5> 6716<p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>, 6717 which has been initialized previously 6718 with <tt><a href="#int_va_start">llvm.va_start</a></tt> 6719 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p> 6720 6721<h5>Arguments:</h5> 6722<p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p> 6723 6724<h5>Semantics:</h5> 6725<p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt> 6726 macro available in C. In a target-dependent way, it destroys 6727 the <tt>va_list</tt> element to which the argument points. Calls 6728 to <a href="#int_va_start"><tt>llvm.va_start</tt></a> 6729 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly 6730 with calls to <tt>llvm.va_end</tt>.</p> 6731 6732</div> 6733 6734<!-- _______________________________________________________________________ --> 6735<h4> 6736 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a> 6737</h4> 6738 6739<div> 6740 6741<h5>Syntax:</h5> 6742<pre> 6743 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>) 6744</pre> 6745 6746<h5>Overview:</h5> 6747<p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position 6748 from the source argument list to the destination argument list.</p> 6749 6750<h5>Arguments:</h5> 6751<p>The first argument is a pointer to a <tt>va_list</tt> element to initialize. 6752 The second argument is a pointer to a <tt>va_list</tt> element to copy 6753 from.</p> 6754 6755<h5>Semantics:</h5> 6756<p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> 6757 macro available in C. In a target-dependent way, it copies the 6758 source <tt>va_list</tt> element into the destination <tt>va_list</tt> 6759 element. This intrinsic is necessary because 6760 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be 6761 arbitrarily complex and require, for example, memory allocation.</p> 6762 6763</div> 6764 6765</div> 6766 6767<!-- ======================================================================= --> 6768<h3> 6769 <a name="int_gc">Accurate Garbage Collection Intrinsics</a> 6770</h3> 6771 6772<div> 6773 6774<p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage 6775Collection</a> (GC) requires the implementation and generation of these 6776intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC 6777roots on the stack</a>, as well as garbage collector implementations that 6778require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> 6779barriers. Front-ends for type-safe garbage collected languages should generate 6780these intrinsics to make use of the LLVM garbage collectors. For more details, 6781see <a href="GarbageCollection.html">Accurate Garbage Collection with 6782LLVM</a>.</p> 6783 6784<p>The garbage collection intrinsics only operate on objects in the generic 6785 address space (address space zero).</p> 6786 6787<!-- _______________________________________________________________________ --> 6788<h4> 6789 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a> 6790</h4> 6791 6792<div> 6793 6794<h5>Syntax:</h5> 6795<pre> 6796 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata) 6797</pre> 6798 6799<h5>Overview:</h5> 6800<p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to 6801 the code generator, and allows some metadata to be associated with it.</p> 6802 6803<h5>Arguments:</h5> 6804<p>The first argument specifies the address of a stack object that contains the 6805 root pointer. The second pointer (which must be either a constant or a 6806 global value address) contains the meta-data to be associated with the 6807 root.</p> 6808 6809<h5>Semantics:</h5> 6810<p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc" 6811 location. At compile-time, the code generator generates information to allow 6812 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>' 6813 intrinsic may only be used in a function which <a href="#gc">specifies a GC 6814 algorithm</a>.</p> 6815 6816</div> 6817 6818<!-- _______________________________________________________________________ --> 6819<h4> 6820 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a> 6821</h4> 6822 6823<div> 6824 6825<h5>Syntax:</h5> 6826<pre> 6827 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr) 6828</pre> 6829 6830<h5>Overview:</h5> 6831<p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap 6832 locations, allowing garbage collector implementations that require read 6833 barriers.</p> 6834 6835<h5>Arguments:</h5> 6836<p>The second argument is the address to read from, which should be an address 6837 allocated from the garbage collector. The first object is a pointer to the 6838 start of the referenced object, if needed by the language runtime (otherwise 6839 null).</p> 6840 6841<h5>Semantics:</h5> 6842<p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load 6843 instruction, but may be replaced with substantially more complex code by the 6844 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic 6845 may only be used in a function which <a href="#gc">specifies a GC 6846 algorithm</a>.</p> 6847 6848</div> 6849 6850<!-- _______________________________________________________________________ --> 6851<h4> 6852 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a> 6853</h4> 6854 6855<div> 6856 6857<h5>Syntax:</h5> 6858<pre> 6859 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2) 6860</pre> 6861 6862<h5>Overview:</h5> 6863<p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap 6864 locations, allowing garbage collector implementations that require write 6865 barriers (such as generational or reference counting collectors).</p> 6866 6867<h5>Arguments:</h5> 6868<p>The first argument is the reference to store, the second is the start of the 6869 object to store it to, and the third is the address of the field of Obj to 6870 store to. If the runtime does not require a pointer to the object, Obj may 6871 be null.</p> 6872 6873<h5>Semantics:</h5> 6874<p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store 6875 instruction, but may be replaced with substantially more complex code by the 6876 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic 6877 may only be used in a function which <a href="#gc">specifies a GC 6878 algorithm</a>.</p> 6879 6880</div> 6881 6882</div> 6883 6884<!-- ======================================================================= --> 6885<h3> 6886 <a name="int_codegen">Code Generator Intrinsics</a> 6887</h3> 6888 6889<div> 6890 6891<p>These intrinsics are provided by LLVM to expose special features that may 6892 only be implemented with code generator support.</p> 6893 6894<!-- _______________________________________________________________________ --> 6895<h4> 6896 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a> 6897</h4> 6898 6899<div> 6900 6901<h5>Syntax:</h5> 6902<pre> 6903 declare i8 *@llvm.returnaddress(i32 <level>) 6904</pre> 6905 6906<h5>Overview:</h5> 6907<p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a 6908 target-specific value indicating the return address of the current function 6909 or one of its callers.</p> 6910 6911<h5>Arguments:</h5> 6912<p>The argument to this intrinsic indicates which function to return the address 6913 for. Zero indicates the calling function, one indicates its caller, etc. 6914 The argument is <b>required</b> to be a constant integer value.</p> 6915 6916<h5>Semantics:</h5> 6917<p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer 6918 indicating the return address of the specified call frame, or zero if it 6919 cannot be identified. The value returned by this intrinsic is likely to be 6920 incorrect or 0 for arguments other than zero, so it should only be used for 6921 debugging purposes.</p> 6922 6923<p>Note that calling this intrinsic does not prevent function inlining or other 6924 aggressive transformations, so the value returned may not be that of the 6925 obvious source-language caller.</p> 6926 6927</div> 6928 6929<!-- _______________________________________________________________________ --> 6930<h4> 6931 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a> 6932</h4> 6933 6934<div> 6935 6936<h5>Syntax:</h5> 6937<pre> 6938 declare i8* @llvm.frameaddress(i32 <level>) 6939</pre> 6940 6941<h5>Overview:</h5> 6942<p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the 6943 target-specific frame pointer value for the specified stack frame.</p> 6944 6945<h5>Arguments:</h5> 6946<p>The argument to this intrinsic indicates which function to return the frame 6947 pointer for. Zero indicates the calling function, one indicates its caller, 6948 etc. The argument is <b>required</b> to be a constant integer value.</p> 6949 6950<h5>Semantics:</h5> 6951<p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer 6952 indicating the frame address of the specified call frame, or zero if it 6953 cannot be identified. The value returned by this intrinsic is likely to be 6954 incorrect or 0 for arguments other than zero, so it should only be used for 6955 debugging purposes.</p> 6956 6957<p>Note that calling this intrinsic does not prevent function inlining or other 6958 aggressive transformations, so the value returned may not be that of the 6959 obvious source-language caller.</p> 6960 6961</div> 6962 6963<!-- _______________________________________________________________________ --> 6964<h4> 6965 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a> 6966</h4> 6967 6968<div> 6969 6970<h5>Syntax:</h5> 6971<pre> 6972 declare i8* @llvm.stacksave() 6973</pre> 6974 6975<h5>Overview:</h5> 6976<p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state 6977 of the function stack, for use 6978 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is 6979 useful for implementing language features like scoped automatic variable 6980 sized arrays in C99.</p> 6981 6982<h5>Semantics:</h5> 6983<p>This intrinsic returns a opaque pointer value that can be passed 6984 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When 6985 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved 6986 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack 6987 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. 6988 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the 6989 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p> 6990 6991</div> 6992 6993<!-- _______________________________________________________________________ --> 6994<h4> 6995 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a> 6996</h4> 6997 6998<div> 6999 7000<h5>Syntax:</h5> 7001<pre> 7002 declare void @llvm.stackrestore(i8* %ptr) 7003</pre> 7004 7005<h5>Overview:</h5> 7006<p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of 7007 the function stack to the state it was in when the 7008 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic 7009 executed. This is useful for implementing language features like scoped 7010 automatic variable sized arrays in C99.</p> 7011 7012<h5>Semantics:</h5> 7013<p>See the description 7014 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p> 7015 7016</div> 7017 7018<!-- _______________________________________________________________________ --> 7019<h4> 7020 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a> 7021</h4> 7022 7023<div> 7024 7025<h5>Syntax:</h5> 7026<pre> 7027 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>) 7028</pre> 7029 7030<h5>Overview:</h5> 7031<p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to 7032 insert a prefetch instruction if supported; otherwise, it is a noop. 7033 Prefetches have no effect on the behavior of the program but can change its 7034 performance characteristics.</p> 7035 7036<h5>Arguments:</h5> 7037<p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the 7038 specifier determining if the fetch should be for a read (0) or write (1), 7039 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no 7040 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt> 7041 specifies whether the prefetch is performed on the data (1) or instruction (0) 7042 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments 7043 must be constant integers.</p> 7044 7045<h5>Semantics:</h5> 7046<p>This intrinsic does not modify the behavior of the program. In particular, 7047 prefetches cannot trap and do not produce a value. On targets that support 7048 this intrinsic, the prefetch can provide hints to the processor cache for 7049 better performance.</p> 7050 7051</div> 7052 7053<!-- _______________________________________________________________________ --> 7054<h4> 7055 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a> 7056</h4> 7057 7058<div> 7059 7060<h5>Syntax:</h5> 7061<pre> 7062 declare void @llvm.pcmarker(i32 <id>) 7063</pre> 7064 7065<h5>Overview:</h5> 7066<p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program 7067 Counter (PC) in a region of code to simulators and other tools. The method 7068 is target specific, but it is expected that the marker will use exported 7069 symbols to transmit the PC of the marker. The marker makes no guarantees 7070 that it will remain with any specific instruction after optimizations. It is 7071 possible that the presence of a marker will inhibit optimizations. The 7072 intended use is to be inserted after optimizations to allow correlations of 7073 simulation runs.</p> 7074 7075<h5>Arguments:</h5> 7076<p><tt>id</tt> is a numerical id identifying the marker.</p> 7077 7078<h5>Semantics:</h5> 7079<p>This intrinsic does not modify the behavior of the program. Backends that do 7080 not support this intrinsic may ignore it.</p> 7081 7082</div> 7083 7084<!-- _______________________________________________________________________ --> 7085<h4> 7086 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a> 7087</h4> 7088 7089<div> 7090 7091<h5>Syntax:</h5> 7092<pre> 7093 declare i64 @llvm.readcyclecounter() 7094</pre> 7095 7096<h5>Overview:</h5> 7097<p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle 7098 counter register (or similar low latency, high accuracy clocks) on those 7099 targets that support it. On X86, it should map to RDTSC. On Alpha, it 7100 should map to RPCC. As the backing counters overflow quickly (on the order 7101 of 9 seconds on alpha), this should only be used for small timings.</p> 7102 7103<h5>Semantics:</h5> 7104<p>When directly supported, reading the cycle counter should not modify any 7105 memory. Implementations are allowed to either return a application specific 7106 value or a system wide value. On backends without support, this is lowered 7107 to a constant 0.</p> 7108 7109</div> 7110 7111</div> 7112 7113<!-- ======================================================================= --> 7114<h3> 7115 <a name="int_libc">Standard C Library Intrinsics</a> 7116</h3> 7117 7118<div> 7119 7120<p>LLVM provides intrinsics for a few important standard C library functions. 7121 These intrinsics allow source-language front-ends to pass information about 7122 the alignment of the pointer arguments to the code generator, providing 7123 opportunity for more efficient code generation.</p> 7124 7125<!-- _______________________________________________________________________ --> 7126<h4> 7127 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a> 7128</h4> 7129 7130<div> 7131 7132<h5>Syntax:</h5> 7133<p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any 7134 integer bit width and for different address spaces. Not all targets support 7135 all bit widths however.</p> 7136 7137<pre> 7138 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>, 7139 i32 <len>, i32 <align>, i1 <isvolatile>) 7140 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>, 7141 i64 <len>, i32 <align>, i1 <isvolatile>) 7142</pre> 7143 7144<h5>Overview:</h5> 7145<p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the 7146 source location to the destination location.</p> 7147 7148<p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt> 7149 intrinsics do not return a value, takes extra alignment/isvolatile arguments 7150 and the pointers can be in specified address spaces.</p> 7151 7152<h5>Arguments:</h5> 7153 7154<p>The first argument is a pointer to the destination, the second is a pointer 7155 to the source. The third argument is an integer argument specifying the 7156 number of bytes to copy, the fourth argument is the alignment of the 7157 source and destination locations, and the fifth is a boolean indicating a 7158 volatile access.</p> 7159 7160<p>If the call to this intrinsic has an alignment value that is not 0 or 1, 7161 then the caller guarantees that both the source and destination pointers are 7162 aligned to that boundary.</p> 7163 7164<p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the 7165 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>. 7166 The detailed access behavior is not very cleanly specified and it is unwise 7167 to depend on it.</p> 7168 7169<h5>Semantics:</h5> 7170 7171<p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the 7172 source location to the destination location, which are not allowed to 7173 overlap. It copies "len" bytes of memory over. If the argument is known to 7174 be aligned to some boundary, this can be specified as the fourth argument, 7175 otherwise it should be set to 0 or 1.</p> 7176 7177</div> 7178 7179<!-- _______________________________________________________________________ --> 7180<h4> 7181 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a> 7182</h4> 7183 7184<div> 7185 7186<h5>Syntax:</h5> 7187<p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit 7188 width and for different address space. Not all targets support all bit 7189 widths however.</p> 7190 7191<pre> 7192 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>, 7193 i32 <len>, i32 <align>, i1 <isvolatile>) 7194 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>, 7195 i64 <len>, i32 <align>, i1 <isvolatile>) 7196</pre> 7197 7198<h5>Overview:</h5> 7199<p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the 7200 source location to the destination location. It is similar to the 7201 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to 7202 overlap.</p> 7203 7204<p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt> 7205 intrinsics do not return a value, takes extra alignment/isvolatile arguments 7206 and the pointers can be in specified address spaces.</p> 7207 7208<h5>Arguments:</h5> 7209 7210<p>The first argument is a pointer to the destination, the second is a pointer 7211 to the source. The third argument is an integer argument specifying the 7212 number of bytes to copy, the fourth argument is the alignment of the 7213 source and destination locations, and the fifth is a boolean indicating a 7214 volatile access.</p> 7215 7216<p>If the call to this intrinsic has an alignment value that is not 0 or 1, 7217 then the caller guarantees that the source and destination pointers are 7218 aligned to that boundary.</p> 7219 7220<p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the 7221 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>. 7222 The detailed access behavior is not very cleanly specified and it is unwise 7223 to depend on it.</p> 7224 7225<h5>Semantics:</h5> 7226 7227<p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the 7228 source location to the destination location, which may overlap. It copies 7229 "len" bytes of memory over. If the argument is known to be aligned to some 7230 boundary, this can be specified as the fourth argument, otherwise it should 7231 be set to 0 or 1.</p> 7232 7233</div> 7234 7235<!-- _______________________________________________________________________ --> 7236<h4> 7237 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a> 7238</h4> 7239 7240<div> 7241 7242<h5>Syntax:</h5> 7243<p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit 7244 width and for different address spaces. However, not all targets support all 7245 bit widths.</p> 7246 7247<pre> 7248 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>, 7249 i32 <len>, i32 <align>, i1 <isvolatile>) 7250 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>, 7251 i64 <len>, i32 <align>, i1 <isvolatile>) 7252</pre> 7253 7254<h5>Overview:</h5> 7255<p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a 7256 particular byte value.</p> 7257 7258<p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt> 7259 intrinsic does not return a value and takes extra alignment/volatile 7260 arguments. Also, the destination can be in an arbitrary address space.</p> 7261 7262<h5>Arguments:</h5> 7263<p>The first argument is a pointer to the destination to fill, the second is the 7264 byte value with which to fill it, the third argument is an integer argument 7265 specifying the number of bytes to fill, and the fourth argument is the known 7266 alignment of the destination location.</p> 7267 7268<p>If the call to this intrinsic has an alignment value that is not 0 or 1, 7269 then the caller guarantees that the destination pointer is aligned to that 7270 boundary.</p> 7271 7272<p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the 7273 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>. 7274 The detailed access behavior is not very cleanly specified and it is unwise 7275 to depend on it.</p> 7276 7277<h5>Semantics:</h5> 7278<p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting 7279 at the destination location. If the argument is known to be aligned to some 7280 boundary, this can be specified as the fourth argument, otherwise it should 7281 be set to 0 or 1.</p> 7282 7283</div> 7284 7285<!-- _______________________________________________________________________ --> 7286<h4> 7287 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a> 7288</h4> 7289 7290<div> 7291 7292<h5>Syntax:</h5> 7293<p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any 7294 floating point or vector of floating point type. Not all targets support all 7295 types however.</p> 7296 7297<pre> 7298 declare float @llvm.sqrt.f32(float %Val) 7299 declare double @llvm.sqrt.f64(double %Val) 7300 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val) 7301 declare fp128 @llvm.sqrt.f128(fp128 %Val) 7302 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val) 7303</pre> 7304 7305<h5>Overview:</h5> 7306<p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand, 7307 returning the same value as the libm '<tt>sqrt</tt>' functions would. 7308 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined 7309 behavior for negative numbers other than -0.0 (which allows for better 7310 optimization, because there is no need to worry about errno being 7311 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p> 7312 7313<h5>Arguments:</h5> 7314<p>The argument and return value are floating point numbers of the same 7315 type.</p> 7316 7317<h5>Semantics:</h5> 7318<p>This function returns the sqrt of the specified operand if it is a 7319 nonnegative floating point number.</p> 7320 7321</div> 7322 7323<!-- _______________________________________________________________________ --> 7324<h4> 7325 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a> 7326</h4> 7327 7328<div> 7329 7330<h5>Syntax:</h5> 7331<p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any 7332 floating point or vector of floating point type. Not all targets support all 7333 types however.</p> 7334 7335<pre> 7336 declare float @llvm.powi.f32(float %Val, i32 %power) 7337 declare double @llvm.powi.f64(double %Val, i32 %power) 7338 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power) 7339 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power) 7340 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power) 7341</pre> 7342 7343<h5>Overview:</h5> 7344<p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the 7345 specified (positive or negative) power. The order of evaluation of 7346 multiplications is not defined. When a vector of floating point type is 7347 used, the second argument remains a scalar integer value.</p> 7348 7349<h5>Arguments:</h5> 7350<p>The second argument is an integer power, and the first is a value to raise to 7351 that power.</p> 7352 7353<h5>Semantics:</h5> 7354<p>This function returns the first value raised to the second power with an 7355 unspecified sequence of rounding operations.</p> 7356 7357</div> 7358 7359<!-- _______________________________________________________________________ --> 7360<h4> 7361 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a> 7362</h4> 7363 7364<div> 7365 7366<h5>Syntax:</h5> 7367<p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any 7368 floating point or vector of floating point type. Not all targets support all 7369 types however.</p> 7370 7371<pre> 7372 declare float @llvm.sin.f32(float %Val) 7373 declare double @llvm.sin.f64(double %Val) 7374 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val) 7375 declare fp128 @llvm.sin.f128(fp128 %Val) 7376 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val) 7377</pre> 7378 7379<h5>Overview:</h5> 7380<p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p> 7381 7382<h5>Arguments:</h5> 7383<p>The argument and return value are floating point numbers of the same 7384 type.</p> 7385 7386<h5>Semantics:</h5> 7387<p>This function returns the sine of the specified operand, returning the same 7388 values as the libm <tt>sin</tt> functions would, and handles error conditions 7389 in the same way.</p> 7390 7391</div> 7392 7393<!-- _______________________________________________________________________ --> 7394<h4> 7395 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a> 7396</h4> 7397 7398<div> 7399 7400<h5>Syntax:</h5> 7401<p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any 7402 floating point or vector of floating point type. Not all targets support all 7403 types however.</p> 7404 7405<pre> 7406 declare float @llvm.cos.f32(float %Val) 7407 declare double @llvm.cos.f64(double %Val) 7408 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val) 7409 declare fp128 @llvm.cos.f128(fp128 %Val) 7410 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val) 7411</pre> 7412 7413<h5>Overview:</h5> 7414<p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p> 7415 7416<h5>Arguments:</h5> 7417<p>The argument and return value are floating point numbers of the same 7418 type.</p> 7419 7420<h5>Semantics:</h5> 7421<p>This function returns the cosine of the specified operand, returning the same 7422 values as the libm <tt>cos</tt> functions would, and handles error conditions 7423 in the same way.</p> 7424 7425</div> 7426 7427<!-- _______________________________________________________________________ --> 7428<h4> 7429 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a> 7430</h4> 7431 7432<div> 7433 7434<h5>Syntax:</h5> 7435<p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any 7436 floating point or vector of floating point type. Not all targets support all 7437 types however.</p> 7438 7439<pre> 7440 declare float @llvm.pow.f32(float %Val, float %Power) 7441 declare double @llvm.pow.f64(double %Val, double %Power) 7442 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power) 7443 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power) 7444 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power) 7445</pre> 7446 7447<h5>Overview:</h5> 7448<p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the 7449 specified (positive or negative) power.</p> 7450 7451<h5>Arguments:</h5> 7452<p>The second argument is a floating point power, and the first is a value to 7453 raise to that power.</p> 7454 7455<h5>Semantics:</h5> 7456<p>This function returns the first value raised to the second power, returning 7457 the same values as the libm <tt>pow</tt> functions would, and handles error 7458 conditions in the same way.</p> 7459 7460</div> 7461 7462<!-- _______________________________________________________________________ --> 7463<h4> 7464 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a> 7465</h4> 7466 7467<div> 7468 7469<h5>Syntax:</h5> 7470<p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any 7471 floating point or vector of floating point type. Not all targets support all 7472 types however.</p> 7473 7474<pre> 7475 declare float @llvm.exp.f32(float %Val) 7476 declare double @llvm.exp.f64(double %Val) 7477 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val) 7478 declare fp128 @llvm.exp.f128(fp128 %Val) 7479 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val) 7480</pre> 7481 7482<h5>Overview:</h5> 7483<p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p> 7484 7485<h5>Arguments:</h5> 7486<p>The argument and return value are floating point numbers of the same 7487 type.</p> 7488 7489<h5>Semantics:</h5> 7490<p>This function returns the same values as the libm <tt>exp</tt> functions 7491 would, and handles error conditions in the same way.</p> 7492 7493</div> 7494 7495<!-- _______________________________________________________________________ --> 7496<h4> 7497 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a> 7498</h4> 7499 7500<div> 7501 7502<h5>Syntax:</h5> 7503<p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any 7504 floating point or vector of floating point type. Not all targets support all 7505 types however.</p> 7506 7507<pre> 7508 declare float @llvm.log.f32(float %Val) 7509 declare double @llvm.log.f64(double %Val) 7510 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val) 7511 declare fp128 @llvm.log.f128(fp128 %Val) 7512 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val) 7513</pre> 7514 7515<h5>Overview:</h5> 7516<p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p> 7517 7518<h5>Arguments:</h5> 7519<p>The argument and return value are floating point numbers of the same 7520 type.</p> 7521 7522<h5>Semantics:</h5> 7523<p>This function returns the same values as the libm <tt>log</tt> functions 7524 would, and handles error conditions in the same way.</p> 7525 7526</div> 7527 7528<!-- _______________________________________________________________________ --> 7529<h4> 7530 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a> 7531</h4> 7532 7533<div> 7534 7535<h5>Syntax:</h5> 7536<p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any 7537 floating point or vector of floating point type. Not all targets support all 7538 types however.</p> 7539 7540<pre> 7541 declare float @llvm.fma.f32(float %a, float %b, float %c) 7542 declare double @llvm.fma.f64(double %a, double %b, double %c) 7543 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c) 7544 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c) 7545 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c) 7546</pre> 7547 7548<h5>Overview:</h5> 7549<p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add 7550 operation.</p> 7551 7552<h5>Arguments:</h5> 7553<p>The argument and return value are floating point numbers of the same 7554 type.</p> 7555 7556<h5>Semantics:</h5> 7557<p>This function returns the same values as the libm <tt>fma</tt> functions 7558 would.</p> 7559 7560</div> 7561 7562<!-- _______________________________________________________________________ --> 7563<h4> 7564 <a name="int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a> 7565</h4> 7566 7567<div> 7568 7569<h5>Syntax:</h5> 7570<p>This is an overloaded intrinsic. You can use <tt>llvm.fabs</tt> on any 7571 floating point or vector of floating point type. Not all targets support all 7572 types however.</p> 7573 7574<pre> 7575 declare float @llvm.fabs.f32(float %Val) 7576 declare double @llvm.fabs.f64(double %Val) 7577 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val) 7578 declare fp128 @llvm.fabs.f128(fp128 %Val) 7579 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val) 7580</pre> 7581 7582<h5>Overview:</h5> 7583<p>The '<tt>llvm.fabs.*</tt>' intrinsics return the absolute value of 7584 the operand.</p> 7585 7586<h5>Arguments:</h5> 7587<p>The argument and return value are floating point numbers of the same 7588 type.</p> 7589 7590<h5>Semantics:</h5> 7591<p>This function returns the same values as the libm <tt>fabs</tt> functions 7592 would, and handles error conditions in the same way.</p> 7593 7594</div> 7595 7596<!-- _______________________________________________________________________ --> 7597<h4> 7598 <a name="int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a> 7599</h4> 7600 7601<div> 7602 7603<h5>Syntax:</h5> 7604<p>This is an overloaded intrinsic. You can use <tt>llvm.floor</tt> on any 7605 floating point or vector of floating point type. Not all targets support all 7606 types however.</p> 7607 7608<pre> 7609 declare float @llvm.floor.f32(float %Val) 7610 declare double @llvm.floor.f64(double %Val) 7611 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val) 7612 declare fp128 @llvm.floor.f128(fp128 %Val) 7613 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val) 7614</pre> 7615 7616<h5>Overview:</h5> 7617<p>The '<tt>llvm.floor.*</tt>' intrinsics return the floor of 7618 the operand.</p> 7619 7620<h5>Arguments:</h5> 7621<p>The argument and return value are floating point numbers of the same 7622 type.</p> 7623 7624<h5>Semantics:</h5> 7625<p>This function returns the same values as the libm <tt>floor</tt> functions 7626 would, and handles error conditions in the same way.</p> 7627 7628</div> 7629 7630</div> 7631 7632<!-- ======================================================================= --> 7633<h3> 7634 <a name="int_manip">Bit Manipulation Intrinsics</a> 7635</h3> 7636 7637<div> 7638 7639<p>LLVM provides intrinsics for a few important bit manipulation operations. 7640 These allow efficient code generation for some algorithms.</p> 7641 7642<!-- _______________________________________________________________________ --> 7643<h4> 7644 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a> 7645</h4> 7646 7647<div> 7648 7649<h5>Syntax:</h5> 7650<p>This is an overloaded intrinsic function. You can use bswap on any integer 7651 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p> 7652 7653<pre> 7654 declare i16 @llvm.bswap.i16(i16 <id>) 7655 declare i32 @llvm.bswap.i32(i32 <id>) 7656 declare i64 @llvm.bswap.i64(i64 <id>) 7657</pre> 7658 7659<h5>Overview:</h5> 7660<p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer 7661 values with an even number of bytes (positive multiple of 16 bits). These 7662 are useful for performing operations on data that is not in the target's 7663 native byte order.</p> 7664 7665<h5>Semantics:</h5> 7666<p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high 7667 and low byte of the input i16 swapped. Similarly, 7668 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four 7669 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1, 7670 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order. 7671 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics 7672 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and 7673 more, respectively).</p> 7674 7675</div> 7676 7677<!-- _______________________________________________________________________ --> 7678<h4> 7679 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a> 7680</h4> 7681 7682<div> 7683 7684<h5>Syntax:</h5> 7685<p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit 7686 width, or on any vector with integer elements. Not all targets support all 7687 bit widths or vector types, however.</p> 7688 7689<pre> 7690 declare i8 @llvm.ctpop.i8(i8 <src>) 7691 declare i16 @llvm.ctpop.i16(i16 <src>) 7692 declare i32 @llvm.ctpop.i32(i32 <src>) 7693 declare i64 @llvm.ctpop.i64(i64 <src>) 7694 declare i256 @llvm.ctpop.i256(i256 <src>) 7695 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>) 7696</pre> 7697 7698<h5>Overview:</h5> 7699<p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set 7700 in a value.</p> 7701 7702<h5>Arguments:</h5> 7703<p>The only argument is the value to be counted. The argument may be of any 7704 integer type, or a vector with integer elements. 7705 The return type must match the argument type.</p> 7706 7707<h5>Semantics:</h5> 7708<p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each 7709 element of a vector.</p> 7710 7711</div> 7712 7713<!-- _______________________________________________________________________ --> 7714<h4> 7715 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a> 7716</h4> 7717 7718<div> 7719 7720<h5>Syntax:</h5> 7721<p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any 7722 integer bit width, or any vector whose elements are integers. Not all 7723 targets support all bit widths or vector types, however.</p> 7724 7725<pre> 7726 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>) 7727 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>) 7728 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>) 7729 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>) 7730 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>) 7731 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>) 7732</pre> 7733 7734<h5>Overview:</h5> 7735<p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of 7736 leading zeros in a variable.</p> 7737 7738<h5>Arguments:</h5> 7739<p>The first argument is the value to be counted. This argument may be of any 7740 integer type, or a vectory with integer element type. The return type 7741 must match the first argument type.</p> 7742 7743<p>The second argument must be a constant and is a flag to indicate whether the 7744 intrinsic should ensure that a zero as the first argument produces a defined 7745 result. Historically some architectures did not provide a defined result for 7746 zero values as efficiently, and many algorithms are now predicated on 7747 avoiding zero-value inputs.</p> 7748 7749<h5>Semantics:</h5> 7750<p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) 7751 zeros in a variable, or within each element of the vector. 7752 If <tt>src == 0</tt> then the result is the size in bits of the type of 7753 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise. 7754 For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p> 7755 7756</div> 7757 7758<!-- _______________________________________________________________________ --> 7759<h4> 7760 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a> 7761</h4> 7762 7763<div> 7764 7765<h5>Syntax:</h5> 7766<p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any 7767 integer bit width, or any vector of integer elements. Not all targets 7768 support all bit widths or vector types, however.</p> 7769 7770<pre> 7771 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>) 7772 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>) 7773 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>) 7774 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>) 7775 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>) 7776 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>) 7777</pre> 7778 7779<h5>Overview:</h5> 7780<p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of 7781 trailing zeros.</p> 7782 7783<h5>Arguments:</h5> 7784<p>The first argument is the value to be counted. This argument may be of any 7785 integer type, or a vectory with integer element type. The return type 7786 must match the first argument type.</p> 7787 7788<p>The second argument must be a constant and is a flag to indicate whether the 7789 intrinsic should ensure that a zero as the first argument produces a defined 7790 result. Historically some architectures did not provide a defined result for 7791 zero values as efficiently, and many algorithms are now predicated on 7792 avoiding zero-value inputs.</p> 7793 7794<h5>Semantics:</h5> 7795<p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) 7796 zeros in a variable, or within each element of a vector. 7797 If <tt>src == 0</tt> then the result is the size in bits of the type of 7798 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise. 7799 For example, <tt>llvm.cttz(2) = 1</tt>.</p> 7800 7801</div> 7802 7803</div> 7804 7805<!-- ======================================================================= --> 7806<h3> 7807 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a> 7808</h3> 7809 7810<div> 7811 7812<p>LLVM provides intrinsics for some arithmetic with overflow operations.</p> 7813 7814<!-- _______________________________________________________________________ --> 7815<h4> 7816 <a name="int_sadd_overflow"> 7817 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics 7818 </a> 7819</h4> 7820 7821<div> 7822 7823<h5>Syntax:</h5> 7824<p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt> 7825 on any integer bit width.</p> 7826 7827<pre> 7828 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b) 7829 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b) 7830 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b) 7831</pre> 7832 7833<h5>Overview:</h5> 7834<p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform 7835 a signed addition of the two arguments, and indicate whether an overflow 7836 occurred during the signed summation.</p> 7837 7838<h5>Arguments:</h5> 7839<p>The arguments (%a and %b) and the first element of the result structure may 7840 be of integer types of any bit width, but they must have the same bit 7841 width. The second element of the result structure must be of 7842 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will 7843 undergo signed addition.</p> 7844 7845<h5>Semantics:</h5> 7846<p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform 7847 a signed addition of the two variables. They return a structure — the 7848 first element of which is the signed summation, and the second element of 7849 which is a bit specifying if the signed summation resulted in an 7850 overflow.</p> 7851 7852<h5>Examples:</h5> 7853<pre> 7854 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b) 7855 %sum = extractvalue {i32, i1} %res, 0 7856 %obit = extractvalue {i32, i1} %res, 1 7857 br i1 %obit, label %overflow, label %normal 7858</pre> 7859 7860</div> 7861 7862<!-- _______________________________________________________________________ --> 7863<h4> 7864 <a name="int_uadd_overflow"> 7865 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics 7866 </a> 7867</h4> 7868 7869<div> 7870 7871<h5>Syntax:</h5> 7872<p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt> 7873 on any integer bit width.</p> 7874 7875<pre> 7876 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b) 7877 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 7878 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b) 7879</pre> 7880 7881<h5>Overview:</h5> 7882<p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform 7883 an unsigned addition of the two arguments, and indicate whether a carry 7884 occurred during the unsigned summation.</p> 7885 7886<h5>Arguments:</h5> 7887<p>The arguments (%a and %b) and the first element of the result structure may 7888 be of integer types of any bit width, but they must have the same bit 7889 width. The second element of the result structure must be of 7890 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will 7891 undergo unsigned addition.</p> 7892 7893<h5>Semantics:</h5> 7894<p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform 7895 an unsigned addition of the two arguments. They return a structure — 7896 the first element of which is the sum, and the second element of which is a 7897 bit specifying if the unsigned summation resulted in a carry.</p> 7898 7899<h5>Examples:</h5> 7900<pre> 7901 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 7902 %sum = extractvalue {i32, i1} %res, 0 7903 %obit = extractvalue {i32, i1} %res, 1 7904 br i1 %obit, label %carry, label %normal 7905</pre> 7906 7907</div> 7908 7909<!-- _______________________________________________________________________ --> 7910<h4> 7911 <a name="int_ssub_overflow"> 7912 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics 7913 </a> 7914</h4> 7915 7916<div> 7917 7918<h5>Syntax:</h5> 7919<p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt> 7920 on any integer bit width.</p> 7921 7922<pre> 7923 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b) 7924 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b) 7925 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b) 7926</pre> 7927 7928<h5>Overview:</h5> 7929<p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform 7930 a signed subtraction of the two arguments, and indicate whether an overflow 7931 occurred during the signed subtraction.</p> 7932 7933<h5>Arguments:</h5> 7934<p>The arguments (%a and %b) and the first element of the result structure may 7935 be of integer types of any bit width, but they must have the same bit 7936 width. The second element of the result structure must be of 7937 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will 7938 undergo signed subtraction.</p> 7939 7940<h5>Semantics:</h5> 7941<p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform 7942 a signed subtraction of the two arguments. They return a structure — 7943 the first element of which is the subtraction, and the second element of 7944 which is a bit specifying if the signed subtraction resulted in an 7945 overflow.</p> 7946 7947<h5>Examples:</h5> 7948<pre> 7949 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b) 7950 %sum = extractvalue {i32, i1} %res, 0 7951 %obit = extractvalue {i32, i1} %res, 1 7952 br i1 %obit, label %overflow, label %normal 7953</pre> 7954 7955</div> 7956 7957<!-- _______________________________________________________________________ --> 7958<h4> 7959 <a name="int_usub_overflow"> 7960 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics 7961 </a> 7962</h4> 7963 7964<div> 7965 7966<h5>Syntax:</h5> 7967<p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt> 7968 on any integer bit width.</p> 7969 7970<pre> 7971 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b) 7972 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b) 7973 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b) 7974</pre> 7975 7976<h5>Overview:</h5> 7977<p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform 7978 an unsigned subtraction of the two arguments, and indicate whether an 7979 overflow occurred during the unsigned subtraction.</p> 7980 7981<h5>Arguments:</h5> 7982<p>The arguments (%a and %b) and the first element of the result structure may 7983 be of integer types of any bit width, but they must have the same bit 7984 width. The second element of the result structure must be of 7985 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will 7986 undergo unsigned subtraction.</p> 7987 7988<h5>Semantics:</h5> 7989<p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform 7990 an unsigned subtraction of the two arguments. They return a structure — 7991 the first element of which is the subtraction, and the second element of 7992 which is a bit specifying if the unsigned subtraction resulted in an 7993 overflow.</p> 7994 7995<h5>Examples:</h5> 7996<pre> 7997 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b) 7998 %sum = extractvalue {i32, i1} %res, 0 7999 %obit = extractvalue {i32, i1} %res, 1 8000 br i1 %obit, label %overflow, label %normal 8001</pre> 8002 8003</div> 8004 8005<!-- _______________________________________________________________________ --> 8006<h4> 8007 <a name="int_smul_overflow"> 8008 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics 8009 </a> 8010</h4> 8011 8012<div> 8013 8014<h5>Syntax:</h5> 8015<p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt> 8016 on any integer bit width.</p> 8017 8018<pre> 8019 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b) 8020 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b) 8021 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b) 8022</pre> 8023 8024<h5>Overview:</h5> 8025 8026<p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform 8027 a signed multiplication of the two arguments, and indicate whether an 8028 overflow occurred during the signed multiplication.</p> 8029 8030<h5>Arguments:</h5> 8031<p>The arguments (%a and %b) and the first element of the result structure may 8032 be of integer types of any bit width, but they must have the same bit 8033 width. The second element of the result structure must be of 8034 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will 8035 undergo signed multiplication.</p> 8036 8037<h5>Semantics:</h5> 8038<p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform 8039 a signed multiplication of the two arguments. They return a structure — 8040 the first element of which is the multiplication, and the second element of 8041 which is a bit specifying if the signed multiplication resulted in an 8042 overflow.</p> 8043 8044<h5>Examples:</h5> 8045<pre> 8046 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b) 8047 %sum = extractvalue {i32, i1} %res, 0 8048 %obit = extractvalue {i32, i1} %res, 1 8049 br i1 %obit, label %overflow, label %normal 8050</pre> 8051 8052</div> 8053 8054<!-- _______________________________________________________________________ --> 8055<h4> 8056 <a name="int_umul_overflow"> 8057 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics 8058 </a> 8059</h4> 8060 8061<div> 8062 8063<h5>Syntax:</h5> 8064<p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt> 8065 on any integer bit width.</p> 8066 8067<pre> 8068 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b) 8069 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b) 8070 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b) 8071</pre> 8072 8073<h5>Overview:</h5> 8074<p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform 8075 a unsigned multiplication of the two arguments, and indicate whether an 8076 overflow occurred during the unsigned multiplication.</p> 8077 8078<h5>Arguments:</h5> 8079<p>The arguments (%a and %b) and the first element of the result structure may 8080 be of integer types of any bit width, but they must have the same bit 8081 width. The second element of the result structure must be of 8082 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will 8083 undergo unsigned multiplication.</p> 8084 8085<h5>Semantics:</h5> 8086<p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform 8087 an unsigned multiplication of the two arguments. They return a structure 8088 — the first element of which is the multiplication, and the second 8089 element of which is a bit specifying if the unsigned multiplication resulted 8090 in an overflow.</p> 8091 8092<h5>Examples:</h5> 8093<pre> 8094 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b) 8095 %sum = extractvalue {i32, i1} %res, 0 8096 %obit = extractvalue {i32, i1} %res, 1 8097 br i1 %obit, label %overflow, label %normal 8098</pre> 8099 8100</div> 8101 8102</div> 8103 8104<!-- ======================================================================= --> 8105<h3> 8106 <a name="spec_arithmetic">Specialised Arithmetic Intrinsics</a> 8107</h3> 8108 8109<!-- _______________________________________________________________________ --> 8110 8111<h4> 8112 <a name="fmuladd">'<tt>llvm.fmuladd.*</tt>' Intrinsic</a> 8113</h4> 8114 8115<div> 8116 8117<h5>Syntax:</h5> 8118<pre> 8119 declare float @llvm.fmuladd.f32(float %a, float %b, float %c) 8120 declare double @llvm.fmuladd.f64(double %a, double %b, double %c) 8121</pre> 8122 8123<h5>Overview:</h5> 8124<p>The '<tt>llvm.fmuladd.*</tt>' intrinsic functions represent multiply-add 8125expressions that can be fused if the code generator determines that the fused 8126expression would be legal and efficient.</p> 8127 8128<h5>Arguments:</h5> 8129<p>The '<tt>llvm.fmuladd.*</tt>' intrinsics each take three arguments: two 8130multiplicands, a and b, and an addend c.</p> 8131 8132<h5>Semantics:</h5> 8133<p>The expression:</p> 8134<pre> 8135 %0 = call float @llvm.fmuladd.f32(%a, %b, %c) 8136</pre> 8137<p>is equivalent to the expression a * b + c, except that rounding will not be 8138performed between the multiplication and addition steps if the code generator 8139fuses the operations. Fusion is not guaranteed, even if the target platform 8140supports it. If a fused multiply-add is required the corresponding llvm.fma.* 8141intrinsic function should be used instead.</p> 8142 8143<h5>Examples:</h5> 8144<pre> 8145 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c 8146</pre> 8147 8148</div> 8149 8150<!-- ======================================================================= --> 8151<h3> 8152 <a name="int_fp16">Half Precision Floating Point Intrinsics</a> 8153</h3> 8154 8155<div> 8156 8157<p>For most target platforms, half precision floating point is a storage-only 8158 format. This means that it is 8159 a dense encoding (in memory) but does not support computation in the 8160 format.</p> 8161 8162<p>This means that code must first load the half-precision floating point 8163 value as an i16, then convert it to float with <a 8164 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>. 8165 Computation can then be performed on the float value (including extending to 8166 double etc). To store the value back to memory, it is first converted to 8167 float if needed, then converted to i16 with 8168 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then 8169 storing as an i16 value.</p> 8170 8171<!-- _______________________________________________________________________ --> 8172<h4> 8173 <a name="int_convert_to_fp16"> 8174 '<tt>llvm.convert.to.fp16</tt>' Intrinsic 8175 </a> 8176</h4> 8177 8178<div> 8179 8180<h5>Syntax:</h5> 8181<pre> 8182 declare i16 @llvm.convert.to.fp16(f32 %a) 8183</pre> 8184 8185<h5>Overview:</h5> 8186<p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs 8187 a conversion from single precision floating point format to half precision 8188 floating point format.</p> 8189 8190<h5>Arguments:</h5> 8191<p>The intrinsic function contains single argument - the value to be 8192 converted.</p> 8193 8194<h5>Semantics:</h5> 8195<p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs 8196 a conversion from single precision floating point format to half precision 8197 floating point format. The return value is an <tt>i16</tt> which 8198 contains the converted number.</p> 8199 8200<h5>Examples:</h5> 8201<pre> 8202 %res = call i16 @llvm.convert.to.fp16(f32 %a) 8203 store i16 %res, i16* @x, align 2 8204</pre> 8205 8206</div> 8207 8208<!-- _______________________________________________________________________ --> 8209<h4> 8210 <a name="int_convert_from_fp16"> 8211 '<tt>llvm.convert.from.fp16</tt>' Intrinsic 8212 </a> 8213</h4> 8214 8215<div> 8216 8217<h5>Syntax:</h5> 8218<pre> 8219 declare f32 @llvm.convert.from.fp16(i16 %a) 8220</pre> 8221 8222<h5>Overview:</h5> 8223<p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs 8224 a conversion from half precision floating point format to single precision 8225 floating point format.</p> 8226 8227<h5>Arguments:</h5> 8228<p>The intrinsic function contains single argument - the value to be 8229 converted.</p> 8230 8231<h5>Semantics:</h5> 8232<p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a 8233 conversion from half single precision floating point format to single 8234 precision floating point format. The input half-float value is represented by 8235 an <tt>i16</tt> value.</p> 8236 8237<h5>Examples:</h5> 8238<pre> 8239 %a = load i16* @x, align 2 8240 %res = call f32 @llvm.convert.from.fp16(i16 %a) 8241</pre> 8242 8243</div> 8244 8245</div> 8246 8247<!-- ======================================================================= --> 8248<h3> 8249 <a name="int_debugger">Debugger Intrinsics</a> 8250</h3> 8251 8252<div> 8253 8254<p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> 8255 prefix), are described in 8256 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source 8257 Level Debugging</a> document.</p> 8258 8259</div> 8260 8261<!-- ======================================================================= --> 8262<h3> 8263 <a name="int_eh">Exception Handling Intrinsics</a> 8264</h3> 8265 8266<div> 8267 8268<p>The LLVM exception handling intrinsics (which all start with 8269 <tt>llvm.eh.</tt> prefix), are described in 8270 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception 8271 Handling</a> document.</p> 8272 8273</div> 8274 8275<!-- ======================================================================= --> 8276<h3> 8277 <a name="int_trampoline">Trampoline Intrinsics</a> 8278</h3> 8279 8280<div> 8281 8282<p>These intrinsics make it possible to excise one parameter, marked with 8283 the <a href="#nest"><tt>nest</tt></a> attribute, from a function. 8284 The result is a callable 8285 function pointer lacking the nest parameter - the caller does not need to 8286 provide a value for it. Instead, the value to use is stored in advance in a 8287 "trampoline", a block of memory usually allocated on the stack, which also 8288 contains code to splice the nest value into the argument list. This is used 8289 to implement the GCC nested function address extension.</p> 8290 8291<p>For example, if the function is 8292 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function 8293 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as 8294 follows:</p> 8295 8296<pre class="doc_code"> 8297 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86 8298 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0 8299 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval) 8300 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1) 8301 %fp = bitcast i8* %p to i32 (i32, i32)* 8302</pre> 8303 8304<p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent 8305 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p> 8306 8307<!-- _______________________________________________________________________ --> 8308<h4> 8309 <a name="int_it"> 8310 '<tt>llvm.init.trampoline</tt>' Intrinsic 8311 </a> 8312</h4> 8313 8314<div> 8315 8316<h5>Syntax:</h5> 8317<pre> 8318 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>) 8319</pre> 8320 8321<h5>Overview:</h5> 8322<p>This fills the memory pointed to by <tt>tramp</tt> with executable code, 8323 turning it into a trampoline.</p> 8324 8325<h5>Arguments:</h5> 8326<p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all 8327 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and 8328 sufficiently aligned block of memory; this memory is written to by the 8329 intrinsic. Note that the size and the alignment are target-specific - LLVM 8330 currently provides no portable way of determining them, so a front-end that 8331 generates this intrinsic needs to have some target-specific knowledge. 8332 The <tt>func</tt> argument must hold a function bitcast to 8333 an <tt>i8*</tt>.</p> 8334 8335<h5>Semantics:</h5> 8336<p>The block of memory pointed to by <tt>tramp</tt> is filled with target 8337 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be 8338 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer 8339 which can be <a href="#int_trampoline">bitcast (to a new function) and 8340 called</a>. The new function's signature is the same as that of 8341 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute 8342 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of 8343 pointer type. Calling the new function is equivalent to calling <tt>func</tt> 8344 with the same argument list, but with <tt>nval</tt> used for the missing 8345 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the 8346 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call 8347 to the returned function pointer is undefined.</p> 8348</div> 8349 8350<!-- _______________________________________________________________________ --> 8351<h4> 8352 <a name="int_at"> 8353 '<tt>llvm.adjust.trampoline</tt>' Intrinsic 8354 </a> 8355</h4> 8356 8357<div> 8358 8359<h5>Syntax:</h5> 8360<pre> 8361 declare i8* @llvm.adjust.trampoline(i8* <tramp>) 8362</pre> 8363 8364<h5>Overview:</h5> 8365<p>This performs any required machine-specific adjustment to the address of a 8366 trampoline (passed as <tt>tramp</tt>).</p> 8367 8368<h5>Arguments:</h5> 8369<p><tt>tramp</tt> must point to a block of memory which already has trampoline code 8370 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt> 8371 </a>.</p> 8372 8373<h5>Semantics:</h5> 8374<p>On some architectures the address of the code to be executed needs to be 8375 different to the address where the trampoline is actually stored. This 8376 intrinsic returns the executable address corresponding to <tt>tramp</tt> 8377 after performing the required machine specific adjustments. 8378 The pointer returned can then be <a href="#int_trampoline"> bitcast and 8379 executed</a>. 8380</p> 8381 8382</div> 8383 8384</div> 8385 8386<!-- ======================================================================= --> 8387<h3> 8388 <a name="int_memorymarkers">Memory Use Markers</a> 8389</h3> 8390 8391<div> 8392 8393<p>This class of intrinsics exists to information about the lifetime of memory 8394 objects and ranges where variables are immutable.</p> 8395 8396<!-- _______________________________________________________________________ --> 8397<h4> 8398 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a> 8399</h4> 8400 8401<div> 8402 8403<h5>Syntax:</h5> 8404<pre> 8405 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>) 8406</pre> 8407 8408<h5>Overview:</h5> 8409<p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory 8410 object's lifetime.</p> 8411 8412<h5>Arguments:</h5> 8413<p>The first argument is a constant integer representing the size of the 8414 object, or -1 if it is variable sized. The second argument is a pointer to 8415 the object.</p> 8416 8417<h5>Semantics:</h5> 8418<p>This intrinsic indicates that before this point in the code, the value of the 8419 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to 8420 never be used and has an undefined value. A load from the pointer that 8421 precedes this intrinsic can be replaced with 8422 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p> 8423 8424</div> 8425 8426<!-- _______________________________________________________________________ --> 8427<h4> 8428 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a> 8429</h4> 8430 8431<div> 8432 8433<h5>Syntax:</h5> 8434<pre> 8435 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>) 8436</pre> 8437 8438<h5>Overview:</h5> 8439<p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory 8440 object's lifetime.</p> 8441 8442<h5>Arguments:</h5> 8443<p>The first argument is a constant integer representing the size of the 8444 object, or -1 if it is variable sized. The second argument is a pointer to 8445 the object.</p> 8446 8447<h5>Semantics:</h5> 8448<p>This intrinsic indicates that after this point in the code, the value of the 8449 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to 8450 never be used and has an undefined value. Any stores into the memory object 8451 following this intrinsic may be removed as dead. 8452 8453</div> 8454 8455<!-- _______________________________________________________________________ --> 8456<h4> 8457 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a> 8458</h4> 8459 8460<div> 8461 8462<h5>Syntax:</h5> 8463<pre> 8464 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>) 8465</pre> 8466 8467<h5>Overview:</h5> 8468<p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of 8469 a memory object will not change.</p> 8470 8471<h5>Arguments:</h5> 8472<p>The first argument is a constant integer representing the size of the 8473 object, or -1 if it is variable sized. The second argument is a pointer to 8474 the object.</p> 8475 8476<h5>Semantics:</h5> 8477<p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses 8478 the return value, the referenced memory location is constant and 8479 unchanging.</p> 8480 8481</div> 8482 8483<!-- _______________________________________________________________________ --> 8484<h4> 8485 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a> 8486</h4> 8487 8488<div> 8489 8490<h5>Syntax:</h5> 8491<pre> 8492 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>) 8493</pre> 8494 8495<h5>Overview:</h5> 8496<p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of 8497 a memory object are mutable.</p> 8498 8499<h5>Arguments:</h5> 8500<p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic. 8501 The second argument is a constant integer representing the size of the 8502 object, or -1 if it is variable sized and the third argument is a pointer 8503 to the object.</p> 8504 8505<h5>Semantics:</h5> 8506<p>This intrinsic indicates that the memory is mutable again.</p> 8507 8508</div> 8509 8510</div> 8511 8512<!-- ======================================================================= --> 8513<h3> 8514 <a name="int_general">General Intrinsics</a> 8515</h3> 8516 8517<div> 8518 8519<p>This class of intrinsics is designed to be generic and has no specific 8520 purpose.</p> 8521 8522<!-- _______________________________________________________________________ --> 8523<h4> 8524 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a> 8525</h4> 8526 8527<div> 8528 8529<h5>Syntax:</h5> 8530<pre> 8531 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>) 8532</pre> 8533 8534<h5>Overview:</h5> 8535<p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p> 8536 8537<h5>Arguments:</h5> 8538<p>The first argument is a pointer to a value, the second is a pointer to a 8539 global string, the third is a pointer to a global string which is the source 8540 file name, and the last argument is the line number.</p> 8541 8542<h5>Semantics:</h5> 8543<p>This intrinsic allows annotation of local variables with arbitrary strings. 8544 This can be useful for special purpose optimizations that want to look for 8545 these annotations. These have no other defined use; they are ignored by code 8546 generation and optimization.</p> 8547 8548</div> 8549 8550<!-- _______________________________________________________________________ --> 8551<h4> 8552 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a> 8553</h4> 8554 8555<div> 8556 8557<h5>Syntax:</h5> 8558<p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on 8559 any integer bit width.</p> 8560 8561<pre> 8562 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>) 8563 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>) 8564 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>) 8565 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>) 8566 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>) 8567</pre> 8568 8569<h5>Overview:</h5> 8570<p>The '<tt>llvm.annotation</tt>' intrinsic.</p> 8571 8572<h5>Arguments:</h5> 8573<p>The first argument is an integer value (result of some expression), the 8574 second is a pointer to a global string, the third is a pointer to a global 8575 string which is the source file name, and the last argument is the line 8576 number. It returns the value of the first argument.</p> 8577 8578<h5>Semantics:</h5> 8579<p>This intrinsic allows annotations to be put on arbitrary expressions with 8580 arbitrary strings. This can be useful for special purpose optimizations that 8581 want to look for these annotations. These have no other defined use; they 8582 are ignored by code generation and optimization.</p> 8583 8584</div> 8585 8586<!-- _______________________________________________________________________ --> 8587<h4> 8588 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a> 8589</h4> 8590 8591<div> 8592 8593<h5>Syntax:</h5> 8594<pre> 8595 declare void @llvm.trap() noreturn nounwind 8596</pre> 8597 8598<h5>Overview:</h5> 8599<p>The '<tt>llvm.trap</tt>' intrinsic.</p> 8600 8601<h5>Arguments:</h5> 8602<p>None.</p> 8603 8604<h5>Semantics:</h5> 8605<p>This intrinsic is lowered to the target dependent trap instruction. If the 8606 target does not have a trap instruction, this intrinsic will be lowered to 8607 a call of the <tt>abort()</tt> function.</p> 8608 8609</div> 8610 8611<!-- _______________________________________________________________________ --> 8612<h4> 8613 <a name="int_debugtrap">'<tt>llvm.debugtrap</tt>' Intrinsic</a> 8614</h4> 8615 8616<div> 8617 8618<h5>Syntax:</h5> 8619<pre> 8620 declare void @llvm.debugtrap() nounwind 8621</pre> 8622 8623<h5>Overview:</h5> 8624<p>The '<tt>llvm.debugtrap</tt>' intrinsic.</p> 8625 8626<h5>Arguments:</h5> 8627<p>None.</p> 8628 8629<h5>Semantics:</h5> 8630<p>This intrinsic is lowered to code which is intended to cause an execution 8631 trap with the intention of requesting the attention of a debugger.</p> 8632 8633</div> 8634 8635<!-- _______________________________________________________________________ --> 8636<h4> 8637 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a> 8638</h4> 8639 8640<div> 8641 8642<h5>Syntax:</h5> 8643<pre> 8644 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>) 8645</pre> 8646 8647<h5>Overview:</h5> 8648<p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and 8649 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to 8650 ensure that it is placed on the stack before local variables.</p> 8651 8652<h5>Arguments:</h5> 8653<p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer 8654 arguments. The first argument is the value loaded from the stack 8655 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt> 8656 that has enough space to hold the value of the guard.</p> 8657 8658<h5>Semantics:</h5> 8659<p>This intrinsic causes the prologue/epilogue inserter to force the position of 8660 the <tt>AllocaInst</tt> stack slot to be before local variables on the 8661 stack. This is to ensure that if a local variable on the stack is 8662 overwritten, it will destroy the value of the guard. When the function exits, 8663 the guard on the stack is checked against the original guard. If they are 8664 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt> 8665 function.</p> 8666 8667</div> 8668 8669<!-- _______________________________________________________________________ --> 8670<h4> 8671 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a> 8672</h4> 8673 8674<div> 8675 8676<h5>Syntax:</h5> 8677<pre> 8678 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>) 8679 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>) 8680</pre> 8681 8682<h5>Overview:</h5> 8683<p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to 8684 the optimizers to determine at compile time whether a) an operation (like 8685 memcpy) will overflow a buffer that corresponds to an object, or b) that a 8686 runtime check for overflow isn't necessary. An object in this context means 8687 an allocation of a specific class, structure, array, or other object.</p> 8688 8689<h5>Arguments:</h5> 8690<p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first 8691 argument is a pointer to or into the <tt>object</tt>. The second argument 8692 is a boolean and determines whether <tt>llvm.objectsize</tt> returns 0 (if 8693 true) or -1 (if false) when the object size is unknown. 8694 The second argument only accepts constants.</p> 8695 8696<h5>Semantics:</h5> 8697<p>The <tt>llvm.objectsize</tt> intrinsic is lowered to a constant representing 8698 the size of the object concerned. If the size cannot be determined at compile 8699 time, <tt>llvm.objectsize</tt> returns <tt>i32/i64 -1 or 0</tt> 8700 (depending on the <tt>min</tt> argument).</p> 8701 8702</div> 8703<!-- _______________________________________________________________________ --> 8704<h4> 8705 <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a> 8706</h4> 8707 8708<div> 8709 8710<h5>Syntax:</h5> 8711<pre> 8712 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>) 8713 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>) 8714</pre> 8715 8716<h5>Overview:</h5> 8717<p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the 8718 most probable) value of <tt>val</tt>, which can be used by optimizers.</p> 8719 8720<h5>Arguments:</h5> 8721<p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first 8722 argument is a value. The second argument is an expected value, this needs to 8723 be a constant value, variables are not allowed.</p> 8724 8725<h5>Semantics:</h5> 8726<p>This intrinsic is lowered to the <tt>val</tt>.</p> 8727</div> 8728 8729<!-- _______________________________________________________________________ --> 8730<h4> 8731 <a name="int_donothing">'<tt>llvm.donothing</tt>' Intrinsic</a> 8732</h4> 8733 8734<div> 8735 8736<h5>Syntax:</h5> 8737<pre> 8738 declare void @llvm.donothing() nounwind readnone 8739</pre> 8740 8741<h5>Overview:</h5> 8742<p>The <tt>llvm.donothing</tt> intrinsic doesn't perform any operation. It's the 8743only intrinsic that can be called with an invoke instruction.</p> 8744 8745<h5>Arguments:</h5> 8746<p>None.</p> 8747 8748<h5>Semantics:</h5> 8749<p>This intrinsic does nothing, and it's removed by optimizers and ignored by 8750codegen.</p> 8751</div> 8752 8753</div> 8754 8755</div> 8756<!-- *********************************************************************** --> 8757<hr> 8758<address> 8759 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img 8760 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a> 8761 <a href="http://validator.w3.org/check/referer"><img 8762 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a> 8763 8764 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br> 8765 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br> 8766 Last modified: $Date$ 8767</address> 8768 8769</body> 8770</html> 8771