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="#fpmath">'<tt>fpmath</tt>' Metadata</a></li> 107 <li><a href="#range">'<tt>range</tt>' Metadata</a></li> 108 </ol> 109 </li> 110 </ol> 111 </li> 112 <li><a href="#module_flags">Module Flags Metadata</a> 113 <ol> 114 <li><a href="#objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a></li> 115 </ol> 116 </li> 117 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a> 118 <ol> 119 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li> 120 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>' 121 Global Variable</a></li> 122 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>' 123 Global Variable</a></li> 124 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>' 125 Global Variable</a></li> 126 </ol> 127 </li> 128 <li><a href="#instref">Instruction Reference</a> 129 <ol> 130 <li><a href="#terminators">Terminator Instructions</a> 131 <ol> 132 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li> 133 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li> 134 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li> 135 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li> 136 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li> 137 <li><a href="#i_resume">'<tt>resume</tt>' Instruction</a></li> 138 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li> 139 </ol> 140 </li> 141 <li><a href="#binaryops">Binary Operations</a> 142 <ol> 143 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li> 144 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li> 145 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li> 146 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li> 147 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li> 148 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li> 149 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li> 150 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li> 151 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li> 152 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li> 153 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li> 154 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li> 155 </ol> 156 </li> 157 <li><a href="#bitwiseops">Bitwise Binary Operations</a> 158 <ol> 159 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li> 160 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li> 161 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li> 162 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li> 163 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li> 164 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li> 165 </ol> 166 </li> 167 <li><a href="#vectorops">Vector Operations</a> 168 <ol> 169 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li> 170 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li> 171 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li> 172 </ol> 173 </li> 174 <li><a href="#aggregateops">Aggregate Operations</a> 175 <ol> 176 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li> 177 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li> 178 </ol> 179 </li> 180 <li><a href="#memoryops">Memory Access and Addressing Operations</a> 181 <ol> 182 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li> 183 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li> 184 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li> 185 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li> 186 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li> 187 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li> 188 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li> 189 </ol> 190 </li> 191 <li><a href="#convertops">Conversion Operations</a> 192 <ol> 193 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li> 194 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li> 195 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li> 196 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li> 197 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li> 198 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li> 199 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li> 200 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li> 201 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li> 202 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li> 203 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li> 204 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li> 205 </ol> 206 </li> 207 <li><a href="#otherops">Other Operations</a> 208 <ol> 209 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li> 210 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li> 211 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li> 212 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li> 213 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li> 214 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li> 215 <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li> 216 </ol> 217 </li> 218 </ol> 219 </li> 220 <li><a href="#intrinsics">Intrinsic Functions</a> 221 <ol> 222 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a> 223 <ol> 224 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li> 225 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li> 226 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li> 227 </ol> 228 </li> 229 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a> 230 <ol> 231 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li> 232 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li> 233 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li> 234 </ol> 235 </li> 236 <li><a href="#int_codegen">Code Generator Intrinsics</a> 237 <ol> 238 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li> 239 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li> 240 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li> 241 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li> 242 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li> 243 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li> 244 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li> 245 </ol> 246 </li> 247 <li><a href="#int_libc">Standard C Library Intrinsics</a> 248 <ol> 249 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li> 250 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li> 251 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li> 252 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li> 253 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li> 254 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li> 255 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li> 256 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li> 257 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li> 258 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li> 259 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li> 260 <li><a href="#int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a></li> 261 <li><a href="#int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a></li> 262 </ol> 263 </li> 264 <li><a href="#int_manip">Bit Manipulation Intrinsics</a> 265 <ol> 266 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li> 267 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li> 268 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li> 269 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li> 270 </ol> 271 </li> 272 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a> 273 <ol> 274 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li> 275 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li> 276 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li> 277 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li> 278 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li> 279 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li> 280 </ol> 281 </li> 282 <li><a href="#spec_arithmetic">Specialised Arithmetic Intrinsics</a> 283 <ol> 284 <li><a href="#fmuladd">'<tt>llvm.fmuladd</tt> Intrinsic</a></li> 285 </ol> 286 </li> 287 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a> 288 <ol> 289 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li> 290 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li> 291 </ol> 292 </li> 293 <li><a href="#int_debugger">Debugger intrinsics</a></li> 294 <li><a href="#int_eh">Exception Handling intrinsics</a></li> 295 <li><a href="#int_trampoline">Trampoline Intrinsics</a> 296 <ol> 297 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li> 298 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li> 299 </ol> 300 </li> 301 <li><a href="#int_memorymarkers">Memory Use Markers</a> 302 <ol> 303 <li><a href="#int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a></li> 304 <li><a href="#int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a></li> 305 <li><a href="#int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a></li> 306 <li><a href="#int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a></li> 307 </ol> 308 </li> 309 <li><a href="#int_general">General intrinsics</a> 310 <ol> 311 <li><a href="#int_var_annotation"> 312 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li> 313 <li><a href="#int_annotation"> 314 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li> 315 <li><a href="#int_trap"> 316 '<tt>llvm.trap</tt>' Intrinsic</a></li> 317 <li><a href="#int_debugtrap"> 318 '<tt>llvm.debugtrap</tt>' Intrinsic</a></li> 319 <li><a href="#int_stackprotector"> 320 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li> 321 <li><a href="#int_objectsize"> 322 '<tt>llvm.objectsize</tt>' Intrinsic</a></li> 323 <li><a href="#int_expect"> 324 '<tt>llvm.expect</tt>' Intrinsic</a></li> 325 <li><a href="#int_donothing"> 326 '<tt>llvm.donothing</tt>' Intrinsic</a></li> 327 </ol> 328 </li> 329 </ol> 330 </li> 331 </ol> 332 333 <div class="doc_author"> 334 <p>Written by <a href="mailto:sabre (a] nondot.org">Chris Lattner</a> 335 and <a href="mailto:vadve (a] cs.uiuc.edu">Vikram Adve</a></p> 336 </div> 337 338 <!-- *********************************************************************** --> 339 <h2><a name="abstract">Abstract</a></h2> 340 <!-- *********************************************************************** --> 341 342 <div> 343 344 <p>This document is a reference manual for the LLVM assembly language. LLVM is 345 a Static Single Assignment (SSA) based representation that provides type 346 safety, low-level operations, flexibility, and the capability of representing 347 'all' high-level languages cleanly. It is the common code representation 348 used throughout all phases of the LLVM compilation strategy.</p> 349 350 </div> 351 352 <!-- *********************************************************************** --> 353 <h2><a name="introduction">Introduction</a></h2> 354 <!-- *********************************************************************** --> 355 356 <div> 357 358 <p>The LLVM code representation is designed to be used in three different forms: 359 as an in-memory compiler IR, as an on-disk bitcode representation (suitable 360 for fast loading by a Just-In-Time compiler), and as a human readable 361 assembly language representation. This allows LLVM to provide a powerful 362 intermediate representation for efficient compiler transformations and 363 analysis, while providing a natural means to debug and visualize the 364 transformations. The three different forms of LLVM are all equivalent. This 365 document describes the human readable representation and notation.</p> 366 367 <p>The LLVM representation aims to be light-weight and low-level while being 368 expressive, typed, and extensible at the same time. It aims to be a 369 "universal IR" of sorts, by being at a low enough level that high-level ideas 370 may be cleanly mapped to it (similar to how microprocessors are "universal 371 IR's", allowing many source languages to be mapped to them). By providing 372 type information, LLVM can be used as the target of optimizations: for 373 example, through pointer analysis, it can be proven that a C automatic 374 variable is never accessed outside of the current function, allowing it to 375 be promoted to a simple SSA value instead of a memory location.</p> 376 377 <!-- _______________________________________________________________________ --> 378 <h4> 379 <a name="wellformed">Well-Formedness</a> 380 </h4> 381 382 <div> 383 384 <p>It is important to note that this document describes 'well formed' LLVM 385 assembly language. There is a difference between what the parser accepts and 386 what is considered 'well formed'. For example, the following instruction is 387 syntactically okay, but not well formed:</p> 388 389 <pre class="doc_code"> 390 %x = <a href="#i_add">add</a> i32 1, %x 391 </pre> 392 393 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The 394 LLVM infrastructure provides a verification pass that may be used to verify 395 that an LLVM module is well formed. This pass is automatically run by the 396 parser after parsing input assembly and by the optimizer before it outputs 397 bitcode. The violations pointed out by the verifier pass indicate bugs in 398 transformation passes or input to the parser.</p> 399 400 </div> 401 402 </div> 403 404 <!-- Describe the typesetting conventions here. --> 405 406 <!-- *********************************************************************** --> 407 <h2><a name="identifiers">Identifiers</a></h2> 408 <!-- *********************************************************************** --> 409 410 <div> 411 412 <p>LLVM identifiers come in two basic types: global and local. Global 413 identifiers (functions, global variables) begin with the <tt>'@'</tt> 414 character. Local identifiers (register names, types) begin with 415 the <tt>'%'</tt> character. Additionally, there are three different formats 416 for identifiers, for different purposes:</p> 417 418 <ol> 419 <li>Named values are represented as a string of characters with their prefix. 420 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>, 421 <tt>%a.really.long.identifier</tt>. The actual regular expression used is 422 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require 423 other characters in their names can be surrounded with quotes. Special 424 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the 425 ASCII code for the character in hexadecimal. In this way, any character 426 can be used in a name value, even quotes themselves.</li> 427 428 <li>Unnamed values are represented as an unsigned numeric value with their 429 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li> 430 431 <li>Constants, which are described in a <a href="#constants">section about 432 constants</a>, below.</li> 433 </ol> 434 435 <p>LLVM requires that values start with a prefix for two reasons: Compilers 436 don't need to worry about name clashes with reserved words, and the set of 437 reserved words may be expanded in the future without penalty. Additionally, 438 unnamed identifiers allow a compiler to quickly come up with a temporary 439 variable without having to avoid symbol table conflicts.</p> 440 441 <p>Reserved words in LLVM are very similar to reserved words in other 442 languages. There are keywords for different opcodes 443 ('<tt><a href="#i_add">add</a></tt>', 444 '<tt><a href="#i_bitcast">bitcast</a></tt>', 445 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names 446 ('<tt><a href="#t_void">void</a></tt>', 447 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These 448 reserved words cannot conflict with variable names, because none of them 449 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p> 450 451 <p>Here is an example of LLVM code to multiply the integer variable 452 '<tt>%X</tt>' by 8:</p> 453 454 <p>The easy way:</p> 455 456 <pre class="doc_code"> 457 %result = <a href="#i_mul">mul</a> i32 %X, 8 458 </pre> 459 460 <p>After strength reduction:</p> 461 462 <pre class="doc_code"> 463 %result = <a href="#i_shl">shl</a> i32 %X, i8 3 464 </pre> 465 466 <p>And the hard way:</p> 467 468 <pre class="doc_code"> 469 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i> 470 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i> 471 %result = <a href="#i_add">add</a> i32 %1, %1 472 </pre> 473 474 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important 475 lexical features of LLVM:</p> 476 477 <ol> 478 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of 479 line.</li> 480 481 <li>Unnamed temporaries are created when the result of a computation is not 482 assigned to a named value.</li> 483 484 <li>Unnamed temporaries are numbered sequentially</li> 485 </ol> 486 487 <p>It also shows a convention that we follow in this document. When 488 demonstrating instructions, we will follow an instruction with a comment that 489 defines the type and name of value produced. Comments are shown in italic 490 text.</p> 491 492 </div> 493 494 <!-- *********************************************************************** --> 495 <h2><a name="highlevel">High Level Structure</a></h2> 496 <!-- *********************************************************************** --> 497 <div> 498 <!-- ======================================================================= --> 499 <h3> 500 <a name="modulestructure">Module Structure</a> 501 </h3> 502 503 <div> 504 505 <p>LLVM programs are composed of <tt>Module</tt>s, each of which is a 506 translation unit of the input programs. Each module consists of functions, 507 global variables, and symbol table entries. Modules may be combined together 508 with the LLVM linker, which merges function (and global variable) 509 definitions, resolves forward declarations, and merges symbol table 510 entries. Here is an example of the "hello world" module:</p> 511 512 <pre class="doc_code"> 513 <i>; Declare the string constant as a global constant.</i> 514 <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" 515 516 <i>; External declaration of the puts function</i> 517 <a href="#functionstructure">declare</a> i32 @puts(i8* <a href="#nocapture">nocapture</a>) <a href="#fnattrs">nounwind</a> 518 519 <i>; Definition of main function</i> 520 define i32 @main() { <i>; i32()* </i> 521 <i>; Convert [13 x i8]* to i8 *...</i> 522 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.str, i64 0, i64 0 523 524 <i>; Call puts function to write out the string to stdout.</i> 525 <a href="#i_call">call</a> i32 @puts(i8* %cast210) 526 <a href="#i_ret">ret</a> i32 0 527 } 528 529 <i>; Named metadata</i> 530 !1 = metadata !{i32 42} 531 !foo = !{!1, null} 532 </pre> 533 534 <p>This example is made up of a <a href="#globalvars">global variable</a> named 535 "<tt>.str</tt>", an external declaration of the "<tt>puts</tt>" function, 536 a <a href="#functionstructure">function definition</a> for 537 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a> 538 "<tt>foo</tt>".</p> 539 540 <p>In general, a module is made up of a list of global values (where both 541 functions and global variables are global values). Global values are 542 represented by a pointer to a memory location (in this case, a pointer to an 543 array of char, and a pointer to a function), and have one of the 544 following <a href="#linkage">linkage types</a>.</p> 545 546 </div> 547 548 <!-- ======================================================================= --> 549 <h3> 550 <a name="linkage">Linkage Types</a> 551 </h3> 552 553 <div> 554 555 <p>All Global Variables and Functions have one of the following types of 556 linkage:</p> 557 558 <dl> 559 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt> 560 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible 561 by objects in the current module. In particular, linking code into a 562 module with an private global value may cause the private to be renamed as 563 necessary to avoid collisions. Because the symbol is private to the 564 module, all references can be updated. This doesn't show up in any symbol 565 table in the object file.</dd> 566 567 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt> 568 <dd>Similar to <tt>private</tt>, but the symbol is passed through the 569 assembler and evaluated by the linker. Unlike normal strong symbols, they 570 are removed by the linker from the final linked image (executable or 571 dynamic library).</dd> 572 573 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt> 574 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that 575 <tt>linker_private_weak</tt> symbols are subject to coalescing by the 576 linker. The symbols are removed by the linker from the final linked image 577 (executable or dynamic library).</dd> 578 579 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt> 580 <dd>Similar to private, but the value shows as a local symbol 581 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This 582 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd> 583 584 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt> 585 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted 586 into the object file corresponding to the LLVM module. They exist to 587 allow inlining and other optimizations to take place given knowledge of 588 the definition of the global, which is known to be somewhere outside the 589 module. Globals with <tt>available_externally</tt> linkage are allowed to 590 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>. 591 This linkage type is only allowed on definitions, not declarations.</dd> 592 593 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt> 594 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of 595 the same name when linkage occurs. This can be used to implement 596 some forms of inline functions, templates, or other code which must be 597 generated in each translation unit that uses it, but where the body may 598 be overridden with a more definitive definition later. Unreferenced 599 <tt>linkonce</tt> globals are allowed to be discarded. Note that 600 <tt>linkonce</tt> linkage does not actually allow the optimizer to 601 inline the body of this function into callers because it doesn't know if 602 this definition of the function is the definitive definition within the 603 program or whether it will be overridden by a stronger definition. 604 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>" 605 linkage.</dd> 606 607 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt> 608 <dd>"<tt>weak</tt>" linkage has the same merging semantics as 609 <tt>linkonce</tt> linkage, except that unreferenced globals with 610 <tt>weak</tt> linkage may not be discarded. This is used for globals that 611 are declared "weak" in C source code.</dd> 612 613 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt> 614 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but 615 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at 616 global scope. 617 Symbols with "<tt>common</tt>" linkage are merged in the same way as 618 <tt>weak symbols</tt>, and they may not be deleted if unreferenced. 619 <tt>common</tt> symbols may not have an explicit section, 620 must have a zero initializer, and may not be marked '<a 621 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not 622 have common linkage.</dd> 623 624 625 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt> 626 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of 627 pointer to array type. When two global variables with appending linkage 628 are linked together, the two global arrays are appended together. This is 629 the LLVM, typesafe, equivalent of having the system linker append together 630 "sections" with identical names when .o files are linked.</dd> 631 632 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt> 633 <dd>The semantics of this linkage follow the ELF object file model: the symbol 634 is weak until linked, if not linked, the symbol becomes null instead of 635 being an undefined reference.</dd> 636 637 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt> 638 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt> 639 <dd>Some languages allow differing globals to be merged, such as two functions 640 with different semantics. Other languages, such as <tt>C++</tt>, ensure 641 that only equivalent globals are ever merged (the "one definition rule" 642 — "ODR"). Such languages can use the <tt>linkonce_odr</tt> 643 and <tt>weak_odr</tt> linkage types to indicate that the global will only 644 be merged with equivalent globals. These linkage types are otherwise the 645 same as their non-<tt>odr</tt> versions.</dd> 646 647 <dt><tt><b><a name="linkage_linkonce_odr_auto_hide">linkonce_odr_auto_hide</a></b></tt></dt> 648 <dd>Similar to "<tt>linkonce_odr</tt>", but nothing in the translation unit 649 takes the address of this definition. For instance, functions that had an 650 inline definition, but the compiler decided not to inline it. 651 <tt>linkonce_odr_auto_hide</tt> may have only <tt>default</tt> visibility. 652 The symbols are removed by the linker from the final linked image 653 (executable or dynamic library).</dd> 654 655 <dt><tt><b><a name="linkage_external">external</a></b></tt></dt> 656 <dd>If none of the above identifiers are used, the global is externally 657 visible, meaning that it participates in linkage and can be used to 658 resolve external symbol references.</dd> 659 </dl> 660 661 <p>The next two types of linkage are targeted for Microsoft Windows platform 662 only. They are designed to support importing (exporting) symbols from (to) 663 DLLs (Dynamic Link Libraries).</p> 664 665 <dl> 666 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt> 667 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function 668 or variable via a global pointer to a pointer that is set up by the DLL 669 exporting the symbol. On Microsoft Windows targets, the pointer name is 670 formed by combining <code>__imp_</code> and the function or variable 671 name.</dd> 672 673 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt> 674 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global 675 pointer to a pointer in a DLL, so that it can be referenced with the 676 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer 677 name is formed by combining <code>__imp_</code> and the function or 678 variable name.</dd> 679 </dl> 680 681 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if 682 another module defined a "<tt>.LC0</tt>" variable and was linked with this 683 one, one of the two would be renamed, preventing a collision. Since 684 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage 685 declarations), they are accessible outside of the current module.</p> 686 687 <p>It is illegal for a function <i>declaration</i> to have any linkage type 688 other than <tt>external</tt>, <tt>dllimport</tt> 689 or <tt>extern_weak</tt>.</p> 690 691 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt> 692 or <tt>weak_odr</tt> linkages.</p> 693 694 </div> 695 696 <!-- ======================================================================= --> 697 <h3> 698 <a name="callingconv">Calling Conventions</a> 699 </h3> 700 701 <div> 702 703 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a> 704 and <a href="#i_invoke">invokes</a> can all have an optional calling 705 convention specified for the call. The calling convention of any pair of 706 dynamic caller/callee must match, or the behavior of the program is 707 undefined. The following calling conventions are supported by LLVM, and more 708 may be added in the future:</p> 709 710 <dl> 711 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt> 712 <dd>This calling convention (the default if no other calling convention is 713 specified) matches the target C calling conventions. This calling 714 convention supports varargs function calls and tolerates some mismatch in 715 the declared prototype and implemented declaration of the function (as 716 does normal C).</dd> 717 718 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt> 719 <dd>This calling convention attempts to make calls as fast as possible 720 (e.g. by passing things in registers). This calling convention allows the 721 target to use whatever tricks it wants to produce fast code for the 722 target, without having to conform to an externally specified ABI 723 (Application Binary Interface). 724 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized 725 when this or the GHC convention is used.</a> This calling convention 726 does not support varargs and requires the prototype of all callees to 727 exactly match the prototype of the function definition.</dd> 728 729 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt> 730 <dd>This calling convention attempts to make code in the caller as efficient 731 as possible under the assumption that the call is not commonly executed. 732 As such, these calls often preserve all registers so that the call does 733 not break any live ranges in the caller side. This calling convention 734 does not support varargs and requires the prototype of all callees to 735 exactly match the prototype of the function definition.</dd> 736 737 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt> 738 <dd>This calling convention has been implemented specifically for use by the 739 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>. 740 It passes everything in registers, going to extremes to achieve this by 741 disabling callee save registers. This calling convention should not be 742 used lightly but only for specific situations such as an alternative to 743 the <em>register pinning</em> performance technique often used when 744 implementing functional programming languages.At the moment only X86 745 supports this convention and it has the following limitations: 746 <ul> 747 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No 748 floating point types are supported.</li> 749 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and 750 6 floating point parameters.</li> 751 </ul> 752 This calling convention supports 753 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but 754 requires both the caller and callee are using it. 755 </dd> 756 757 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt> 758 <dd>Any calling convention may be specified by number, allowing 759 target-specific calling conventions to be used. Target specific calling 760 conventions start at 64.</dd> 761 </dl> 762 763 <p>More calling conventions can be added/defined on an as-needed basis, to 764 support Pascal conventions or any other well-known target-independent 765 convention.</p> 766 767 </div> 768 769 <!-- ======================================================================= --> 770 <h3> 771 <a name="visibility">Visibility Styles</a> 772 </h3> 773 774 <div> 775 776 <p>All Global Variables and Functions have one of the following visibility 777 styles:</p> 778 779 <dl> 780 <dt><b>"<tt>default</tt>" - Default style</b>:</dt> 781 <dd>On targets that use the ELF object file format, default visibility means 782 that the declaration is visible to other modules and, in shared libraries, 783 means that the declared entity may be overridden. On Darwin, default 784 visibility means that the declaration is visible to other modules. Default 785 visibility corresponds to "external linkage" in the language.</dd> 786 787 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt> 788 <dd>Two declarations of an object with hidden visibility refer to the same 789 object if they are in the same shared object. Usually, hidden visibility 790 indicates that the symbol will not be placed into the dynamic symbol 791 table, so no other module (executable or shared library) can reference it 792 directly.</dd> 793 794 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt> 795 <dd>On ELF, protected visibility indicates that the symbol will be placed in 796 the dynamic symbol table, but that references within the defining module 797 will bind to the local symbol. That is, the symbol cannot be overridden by 798 another module.</dd> 799 </dl> 800 801 </div> 802 803 <!-- ======================================================================= --> 804 <h3> 805 <a name="namedtypes">Named Types</a> 806 </h3> 807 808 <div> 809 810 <p>LLVM IR allows you to specify name aliases for certain types. This can make 811 it easier to read the IR and make the IR more condensed (particularly when 812 recursive types are involved). An example of a name specification is:</p> 813 814 <pre class="doc_code"> 815 %mytype = type { %mytype*, i32 } 816 </pre> 817 818 <p>You may give a name to any <a href="#typesystem">type</a> except 819 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type 820 is expected with the syntax "%mytype".</p> 821 822 <p>Note that type names are aliases for the structural type that they indicate, 823 and that you can therefore specify multiple names for the same type. This 824 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR 825 uses structural typing, the name is not part of the type. When printing out 826 LLVM IR, the printer will pick <em>one name</em> to render all types of a 827 particular shape. This means that if you have code where two different 828 source types end up having the same LLVM type, that the dumper will sometimes 829 print the "wrong" or unexpected type. This is an important design point and 830 isn't going to change.</p> 831 832 </div> 833 834 <!-- ======================================================================= --> 835 <h3> 836 <a name="globalvars">Global Variables</a> 837 </h3> 838 839 <div> 840 841 <p>Global variables define regions of memory allocated at compilation time 842 instead of run-time. Global variables may optionally be initialized, may 843 have an explicit section to be placed in, and may have an optional explicit 844 alignment specified.</p> 845 846 <p>A variable may be defined as <tt>thread_local</tt>, which 847 means that it will not be shared by threads (each thread will have a 848 separated copy of the variable). Not all targets support thread-local 849 variables. Optionally, a TLS model may be specified:</p> 850 851 <dl> 852 <dt><b><tt>localdynamic</tt></b>:</dt> 853 <dd>For variables that are only used within the current shared library.</dd> 854 855 <dt><b><tt>initialexec</tt></b>:</dt> 856 <dd>For variables in modules that will not be loaded dynamically.</dd> 857 858 <dt><b><tt>localexec</tt></b>:</dt> 859 <dd>For variables defined in the executable and only used within it.</dd> 860 </dl> 861 862 <p>The models correspond to the ELF TLS models; see 863 <a href="http://people.redhat.com/drepper/tls.pdf">ELF 864 Handling For Thread-Local Storage</a> for more information on under which 865 circumstances the different models may be used. The target may choose a 866 different TLS model if the specified model is not supported, or if a better 867 choice of model can be made.</p> 868 869 <p>A variable may be defined as a global 870 "constant," which indicates that the contents of the variable 871 will <b>never</b> be modified (enabling better optimization, allowing the 872 global data to be placed in the read-only section of an executable, etc). 873 Note that variables that need runtime initialization cannot be marked 874 "constant" as there is a store to the variable.</p> 875 876 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked 877 constant, even if the final definition of the global is not. This capability 878 can be used to enable slightly better optimization of the program, but 879 requires the language definition to guarantee that optimizations based on the 880 'constantness' are valid for the translation units that do not include the 881 definition.</p> 882 883 <p>As SSA values, global variables define pointer values that are in scope 884 (i.e. they dominate) all basic blocks in the program. Global variables 885 always define a pointer to their "content" type because they describe a 886 region of memory, and all memory objects in LLVM are accessed through 887 pointers.</p> 888 889 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates 890 that the address is not significant, only the content. Constants marked 891 like this can be merged with other constants if they have the same 892 initializer. Note that a constant with significant address <em>can</em> 893 be merged with a <tt>unnamed_addr</tt> constant, the result being a 894 constant whose address is significant.</p> 895 896 <p>A global variable may be declared to reside in a target-specific numbered 897 address space. For targets that support them, address spaces may affect how 898 optimizations are performed and/or what target instructions are used to 899 access the variable. The default address space is zero. The address space 900 qualifier must precede any other attributes.</p> 901 902 <p>LLVM allows an explicit section to be specified for globals. If the target 903 supports it, it will emit globals to the section specified.</p> 904 905 <p>An explicit alignment may be specified for a global, which must be a power 906 of 2. If not present, or if the alignment is set to zero, the alignment of 907 the global is set by the target to whatever it feels convenient. If an 908 explicit alignment is specified, the global is forced to have exactly that 909 alignment. Targets and optimizers are not allowed to over-align the global 910 if the global has an assigned section. In this case, the extra alignment 911 could be observable: for example, code could assume that the globals are 912 densely packed in their section and try to iterate over them as an array, 913 alignment padding would break this iteration.</p> 914 915 <p>For example, the following defines a global in a numbered address space with 916 an initializer, section, and alignment:</p> 917 918 <pre class="doc_code"> 919 @G = addrspace(5) constant float 1.0, section "foo", align 4 920 </pre> 921 922 <p>The following example defines a thread-local global with 923 the <tt>initialexec</tt> TLS model:</p> 924 925 <pre class="doc_code"> 926 @G = thread_local(initialexec) global i32 0, align 4 927 </pre> 928 929 </div> 930 931 932 <!-- ======================================================================= --> 933 <h3> 934 <a name="functionstructure">Functions</a> 935 </h3> 936 937 <div> 938 939 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an 940 optional <a href="#linkage">linkage type</a>, an optional 941 <a href="#visibility">visibility style</a>, an optional 942 <a href="#callingconv">calling convention</a>, 943 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional 944 <a href="#paramattrs">parameter attribute</a> for the return type, a function 945 name, a (possibly empty) argument list (each with optional 946 <a href="#paramattrs">parameter attributes</a>), optional 947 <a href="#fnattrs">function attributes</a>, an optional section, an optional 948 alignment, an optional <a href="#gc">garbage collector name</a>, an opening 949 curly brace, a list of basic blocks, and a closing curly brace.</p> 950 951 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an 952 optional <a href="#linkage">linkage type</a>, an optional 953 <a href="#visibility">visibility style</a>, an optional 954 <a href="#callingconv">calling convention</a>, 955 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional 956 <a href="#paramattrs">parameter attribute</a> for the return type, a function 957 name, a possibly empty list of arguments, an optional alignment, and an 958 optional <a href="#gc">garbage collector name</a>.</p> 959 960 <p>A function definition contains a list of basic blocks, forming the CFG 961 (Control Flow Graph) for the function. Each basic block may optionally start 962 with a label (giving the basic block a symbol table entry), contains a list 963 of instructions, and ends with a <a href="#terminators">terminator</a> 964 instruction (such as a branch or function return).</p> 965 966 <p>The first basic block in a function is special in two ways: it is immediately 967 executed on entrance to the function, and it is not allowed to have 968 predecessor basic blocks (i.e. there can not be any branches to the entry 969 block of a function). Because the block can have no predecessors, it also 970 cannot have any <a href="#i_phi">PHI nodes</a>.</p> 971 972 <p>LLVM allows an explicit section to be specified for functions. If the target 973 supports it, it will emit functions to the section specified.</p> 974 975 <p>An explicit alignment may be specified for a function. If not present, or if 976 the alignment is set to zero, the alignment of the function is set by the 977 target to whatever it feels convenient. If an explicit alignment is 978 specified, the function is forced to have at least that much alignment. All 979 alignments must be a power of 2.</p> 980 981 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not 982 be significant and two identical functions can be merged.</p> 983 984 <h5>Syntax:</h5> 985 <pre class="doc_code"> 986 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>] 987 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] 988 <ResultType> @<FunctionName> ([argument list]) 989 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N] 990 [<a href="#gc">gc</a>] { ... } 991 </pre> 992 993 </div> 994 995 <!-- ======================================================================= --> 996 <h3> 997 <a name="aliasstructure">Aliases</a> 998 </h3> 999 1000 <div> 1001 1002 <p>Aliases act as "second name" for the aliasee value (which can be either 1003 function, global variable, another alias or bitcast of global value). Aliases 1004 may have an optional <a href="#linkage">linkage type</a>, and an 1005 optional <a href="#visibility">visibility style</a>.</p> 1006 1007 <h5>Syntax:</h5> 1008 <pre class="doc_code"> 1009 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee> 1010 </pre> 1011 1012 </div> 1013 1014 <!-- ======================================================================= --> 1015 <h3> 1016 <a name="namedmetadatastructure">Named Metadata</a> 1017 </h3> 1018 1019 <div> 1020 1021 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata 1022 nodes</a> (but not metadata strings) are the only valid operands for 1023 a named metadata.</p> 1024 1025 <h5>Syntax:</h5> 1026 <pre class="doc_code"> 1027 ; Some unnamed metadata nodes, which are referenced by the named metadata. 1028 !0 = metadata !{metadata !"zero"} 1029 !1 = metadata !{metadata !"one"} 1030 !2 = metadata !{metadata !"two"} 1031 ; A named metadata. 1032 !name = !{!0, !1, !2} 1033 </pre> 1034 1035 </div> 1036 1037 <!-- ======================================================================= --> 1038 <h3> 1039 <a name="paramattrs">Parameter Attributes</a> 1040 </h3> 1041 1042 <div> 1043 1044 <p>The return type and each parameter of a function type may have a set of 1045 <i>parameter attributes</i> associated with them. Parameter attributes are 1046 used to communicate additional information about the result or parameters of 1047 a function. Parameter attributes are considered to be part of the function, 1048 not of the function type, so functions with different parameter attributes 1049 can have the same function type.</p> 1050 1051 <p>Parameter attributes are simple keywords that follow the type specified. If 1052 multiple parameter attributes are needed, they are space separated. For 1053 example:</p> 1054 1055 <pre class="doc_code"> 1056 declare i32 @printf(i8* noalias nocapture, ...) 1057 declare i32 @atoi(i8 zeroext) 1058 declare signext i8 @returns_signed_char() 1059 </pre> 1060 1061 <p>Note that any attributes for the function result (<tt>nounwind</tt>, 1062 <tt>readonly</tt>) come immediately after the argument list.</p> 1063 1064 <p>Currently, only the following parameter attributes are defined:</p> 1065 1066 <dl> 1067 <dt><tt><b>zeroext</b></tt></dt> 1068 <dd>This indicates to the code generator that the parameter or return value 1069 should be zero-extended to the extent required by the target's ABI (which 1070 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a 1071 parameter) or the callee (for a return value).</dd> 1072 1073 <dt><tt><b>signext</b></tt></dt> 1074 <dd>This indicates to the code generator that the parameter or return value 1075 should be sign-extended to the extent required by the target's ABI (which 1076 is usually 32-bits) by the caller (for a parameter) or the callee (for a 1077 return value).</dd> 1078 1079 <dt><tt><b>inreg</b></tt></dt> 1080 <dd>This indicates that this parameter or return value should be treated in a 1081 special target-dependent fashion during while emitting code for a function 1082 call or return (usually, by putting it in a register as opposed to memory, 1083 though some targets use it to distinguish between two different kinds of 1084 registers). Use of this attribute is target-specific.</dd> 1085 1086 <dt><tt><b><a name="byval">byval</a></b></tt></dt> 1087 <dd><p>This indicates that the pointer parameter should really be passed by 1088 value to the function. The attribute implies that a hidden copy of the 1089 pointee 1090 is made between the caller and the callee, so the callee is unable to 1091 modify the value in the caller. This attribute is only valid on LLVM 1092 pointer arguments. It is generally used to pass structs and arrays by 1093 value, but is also valid on pointers to scalars. The copy is considered 1094 to belong to the caller not the callee (for example, 1095 <tt><a href="#readonly">readonly</a></tt> functions should not write to 1096 <tt>byval</tt> parameters). This is not a valid attribute for return 1097 values.</p> 1098 1099 <p>The byval attribute also supports specifying an alignment with 1100 the align attribute. It indicates the alignment of the stack slot to 1101 form and the known alignment of the pointer specified to the call site. If 1102 the alignment is not specified, then the code generator makes a 1103 target-specific assumption.</p></dd> 1104 1105 <dt><tt><b><a name="sret">sret</a></b></tt></dt> 1106 <dd>This indicates that the pointer parameter specifies the address of a 1107 structure that is the return value of the function in the source program. 1108 This pointer must be guaranteed by the caller to be valid: loads and 1109 stores to the structure may be assumed by the callee to not to trap. This 1110 may only be applied to the first parameter. This is not a valid attribute 1111 for return values. </dd> 1112 1113 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt> 1114 <dd>This indicates that pointer values 1115 <a href="#pointeraliasing"><i>based</i></a> on the argument or return 1116 value do not alias pointer values which are not <i>based</i> on it, 1117 ignoring certain "irrelevant" dependencies. 1118 For a call to the parent function, dependencies between memory 1119 references from before or after the call and from those during the call 1120 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and 1121 return value used in that call. 1122 The caller shares the responsibility with the callee for ensuring that 1123 these requirements are met. 1124 For further details, please see the discussion of the NoAlias response in 1125 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br> 1126 <br> 1127 Note that this definition of <tt>noalias</tt> is intentionally 1128 similar to the definition of <tt>restrict</tt> in C99 for function 1129 arguments, though it is slightly weaker. 1130 <br> 1131 For function return values, C99's <tt>restrict</tt> is not meaningful, 1132 while LLVM's <tt>noalias</tt> is. 1133 </dd> 1134 1135 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt> 1136 <dd>This indicates that the callee does not make any copies of the pointer 1137 that outlive the callee itself. This is not a valid attribute for return 1138 values.</dd> 1139 1140 <dt><tt><b><a name="nest">nest</a></b></tt></dt> 1141 <dd>This indicates that the pointer parameter can be excised using the 1142 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid 1143 attribute for return values.</dd> 1144 </dl> 1145 1146 </div> 1147 1148 <!-- ======================================================================= --> 1149 <h3> 1150 <a name="gc">Garbage Collector Names</a> 1151 </h3> 1152 1153 <div> 1154 1155 <p>Each function may specify a garbage collector name, which is simply a 1156 string:</p> 1157 1158 <pre class="doc_code"> 1159 define void @f() gc "name" { ... } 1160 </pre> 1161 1162 <p>The compiler declares the supported values of <i>name</i>. Specifying a 1163 collector which will cause the compiler to alter its output in order to 1164 support the named garbage collection algorithm.</p> 1165 1166 </div> 1167 1168 <!-- ======================================================================= --> 1169 <h3> 1170 <a name="fnattrs">Function Attributes</a> 1171 </h3> 1172 1173 <div> 1174 1175 <p>Function attributes are set to communicate additional information about a 1176 function. Function attributes are considered to be part of the function, not 1177 of the function type, so functions with different parameter attributes can 1178 have the same function type.</p> 1179 1180 <p>Function attributes are simple keywords that follow the type specified. If 1181 multiple attributes are needed, they are space separated. For example:</p> 1182 1183 <pre class="doc_code"> 1184 define void @f() noinline { ... } 1185 define void @f() alwaysinline { ... } 1186 define void @f() alwaysinline optsize { ... } 1187 define void @f() optsize { ... } 1188 </pre> 1189 1190 <dl> 1191 <dt><tt><b>address_safety</b></tt></dt> 1192 <dd>This attribute indicates that the address safety analysis 1193 is enabled for this function. </dd> 1194 1195 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt> 1196 <dd>This attribute indicates that, when emitting the prologue and epilogue, 1197 the backend should forcibly align the stack pointer. Specify the 1198 desired alignment, which must be a power of two, in parentheses. 1199 1200 <dt><tt><b>alwaysinline</b></tt></dt> 1201 <dd>This attribute indicates that the inliner should attempt to inline this 1202 function into callers whenever possible, ignoring any active inlining size 1203 threshold for this caller.</dd> 1204 1205 <dt><tt><b>nonlazybind</b></tt></dt> 1206 <dd>This attribute suppresses lazy symbol binding for the function. This 1207 may make calls to the function faster, at the cost of extra program 1208 startup time if the function is not called during program startup.</dd> 1209 1210 <dt><tt><b>inlinehint</b></tt></dt> 1211 <dd>This attribute indicates that the source code contained a hint that inlining 1212 this function is desirable (such as the "inline" keyword in C/C++). It 1213 is just a hint; it imposes no requirements on the inliner.</dd> 1214 1215 <dt><tt><b>naked</b></tt></dt> 1216 <dd>This attribute disables prologue / epilogue emission for the function. 1217 This can have very system-specific consequences.</dd> 1218 1219 <dt><tt><b>noimplicitfloat</b></tt></dt> 1220 <dd>This attributes disables implicit floating point instructions.</dd> 1221 1222 <dt><tt><b>noinline</b></tt></dt> 1223 <dd>This attribute indicates that the inliner should never inline this 1224 function in any situation. This attribute may not be used together with 1225 the <tt>alwaysinline</tt> attribute.</dd> 1226 1227 <dt><tt><b>noredzone</b></tt></dt> 1228 <dd>This attribute indicates that the code generator should not use a red 1229 zone, even if the target-specific ABI normally permits it.</dd> 1230 1231 <dt><tt><b>noreturn</b></tt></dt> 1232 <dd>This function attribute indicates that the function never returns 1233 normally. This produces undefined behavior at runtime if the function 1234 ever does dynamically return.</dd> 1235 1236 <dt><tt><b>nounwind</b></tt></dt> 1237 <dd>This function attribute indicates that the function never returns with an 1238 unwind or exceptional control flow. If the function does unwind, its 1239 runtime behavior is undefined.</dd> 1240 1241 <dt><tt><b>optsize</b></tt></dt> 1242 <dd>This attribute suggests that optimization passes and code generator passes 1243 make choices that keep the code size of this function low, and otherwise 1244 do optimizations specifically to reduce code size.</dd> 1245 1246 <dt><tt><b>readnone</b></tt></dt> 1247 <dd>This attribute indicates that the function computes its result (or decides 1248 to unwind an exception) based strictly on its arguments, without 1249 dereferencing any pointer arguments or otherwise accessing any mutable 1250 state (e.g. memory, control registers, etc) visible to caller functions. 1251 It does not write through any pointer arguments 1252 (including <tt><a href="#byval">byval</a></tt> arguments) and never 1253 changes any state visible to callers. This means that it cannot unwind 1254 exceptions by calling the <tt>C++</tt> exception throwing methods.</dd> 1255 1256 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt> 1257 <dd>This attribute indicates that the function does not write through any 1258 pointer arguments (including <tt><a href="#byval">byval</a></tt> 1259 arguments) or otherwise modify any state (e.g. memory, control registers, 1260 etc) visible to caller functions. It may dereference pointer arguments 1261 and read state that may be set in the caller. A readonly function always 1262 returns the same value (or unwinds an exception identically) when called 1263 with the same set of arguments and global state. It cannot unwind an 1264 exception by calling the <tt>C++</tt> exception throwing methods.</dd> 1265 1266 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt> 1267 <dd>This attribute indicates that this function can return twice. The 1268 C <code>setjmp</code> is an example of such a function. The compiler 1269 disables some optimizations (like tail calls) in the caller of these 1270 functions.</dd> 1271 1272 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt> 1273 <dd>This attribute indicates that the function should emit a stack smashing 1274 protector. It is in the form of a "canary"—a random value placed on 1275 the stack before the local variables that's checked upon return from the 1276 function to see if it has been overwritten. A heuristic is used to 1277 determine if a function needs stack protectors or not.<br> 1278 <br> 1279 If a function that has an <tt>ssp</tt> attribute is inlined into a 1280 function that doesn't have an <tt>ssp</tt> attribute, then the resulting 1281 function will have an <tt>ssp</tt> attribute.</dd> 1282 1283 <dt><tt><b>sspreq</b></tt></dt> 1284 <dd>This attribute indicates that the function should <em>always</em> emit a 1285 stack smashing protector. This overrides 1286 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br> 1287 <br> 1288 If a function that has an <tt>sspreq</tt> attribute is inlined into a 1289 function that doesn't have an <tt>sspreq</tt> attribute or which has 1290 an <tt>ssp</tt> attribute, then the resulting function will have 1291 an <tt>sspreq</tt> attribute.</dd> 1292 1293 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt> 1294 <dd>This attribute indicates that the ABI being targeted requires that 1295 an unwind table entry be produce for this function even if we can 1296 show that no exceptions passes by it. This is normally the case for 1297 the ELF x86-64 abi, but it can be disabled for some compilation 1298 units.</dd> 1299 </dl> 1300 1301 </div> 1302 1303 <!-- ======================================================================= --> 1304 <h3> 1305 <a name="moduleasm">Module-Level Inline Assembly</a> 1306 </h3> 1307 1308 <div> 1309 1310 <p>Modules may contain "module-level inline asm" blocks, which corresponds to 1311 the GCC "file scope inline asm" blocks. These blocks are internally 1312 concatenated by LLVM and treated as a single unit, but may be separated in 1313 the <tt>.ll</tt> file if desired. The syntax is very simple:</p> 1314 1315 <pre class="doc_code"> 1316 module asm "inline asm code goes here" 1317 module asm "more can go here" 1318 </pre> 1319 1320 <p>The strings can contain any character by escaping non-printable characters. 1321 The escape sequence used is simply "\xx" where "xx" is the two digit hex code 1322 for the number.</p> 1323 1324 <p>The inline asm code is simply printed to the machine code .s file when 1325 assembly code is generated.</p> 1326 1327 </div> 1328 1329 <!-- ======================================================================= --> 1330 <h3> 1331 <a name="datalayout">Data Layout</a> 1332 </h3> 1333 1334 <div> 1335 1336 <p>A module may specify a target specific data layout string that specifies how 1337 data is to be laid out in memory. The syntax for the data layout is 1338 simply:</p> 1339 1340 <pre class="doc_code"> 1341 target datalayout = "<i>layout specification</i>" 1342 </pre> 1343 1344 <p>The <i>layout specification</i> consists of a list of specifications 1345 separated by the minus sign character ('-'). Each specification starts with 1346 a letter and may include other information after the letter to define some 1347 aspect of the data layout. The specifications accepted are as follows:</p> 1348 1349 <dl> 1350 <dt><tt>E</tt></dt> 1351 <dd>Specifies that the target lays out data in big-endian form. That is, the 1352 bits with the most significance have the lowest address location.</dd> 1353 1354 <dt><tt>e</tt></dt> 1355 <dd>Specifies that the target lays out data in little-endian form. That is, 1356 the bits with the least significance have the lowest address 1357 location.</dd> 1358 1359 <dt><tt>S<i>size</i></tt></dt> 1360 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion 1361 of stack variables is limited to the natural stack alignment to avoid 1362 dynamic stack realignment. The stack alignment must be a multiple of 1363 8-bits. If omitted, the natural stack alignment defaults to "unspecified", 1364 which does not prevent any alignment promotions.</dd> 1365 1366 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt> 1367 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and 1368 <i>preferred</i> alignments. All sizes are in bits. Specifying 1369 the <i>pref</i> alignment is optional. If omitted, the 1370 preceding <tt>:</tt> should be omitted too.</dd> 1371 1372 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt> 1373 <dd>This specifies the alignment for an integer type of a given bit 1374 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd> 1375 1376 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt> 1377 <dd>This specifies the alignment for a vector type of a given bit 1378 <i>size</i>.</dd> 1379 1380 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt> 1381 <dd>This specifies the alignment for a floating point type of a given bit 1382 <i>size</i>. Only values of <i>size</i> that are supported by the target 1383 will work. 32 (float) and 64 (double) are supported on all targets; 1384 80 or 128 (different flavors of long double) are also supported on some 1385 targets. 1386 1387 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt> 1388 <dd>This specifies the alignment for an aggregate type of a given bit 1389 <i>size</i>.</dd> 1390 1391 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt> 1392 <dd>This specifies the alignment for a stack object of a given bit 1393 <i>size</i>.</dd> 1394 1395 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt> 1396 <dd>This specifies a set of native integer widths for the target CPU 1397 in bits. For example, it might contain "n32" for 32-bit PowerPC, 1398 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of 1399 this set are considered to support most general arithmetic 1400 operations efficiently.</dd> 1401 </dl> 1402 1403 <p>When constructing the data layout for a given target, LLVM starts with a 1404 default set of specifications which are then (possibly) overridden by the 1405 specifications in the <tt>datalayout</tt> keyword. The default specifications 1406 are given in this list:</p> 1407 1408 <ul> 1409 <li><tt>E</tt> - big endian</li> 1410 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li> 1411 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li> 1412 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li> 1413 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li> 1414 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li> 1415 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred 1416 alignment of 64-bits</li> 1417 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li> 1418 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li> 1419 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li> 1420 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li> 1421 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li> 1422 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li> 1423 </ul> 1424 1425 <p>When LLVM is determining the alignment for a given type, it uses the 1426 following rules:</p> 1427 1428 <ol> 1429 <li>If the type sought is an exact match for one of the specifications, that 1430 specification is used.</li> 1431 1432 <li>If no match is found, and the type sought is an integer type, then the 1433 smallest integer type that is larger than the bitwidth of the sought type 1434 is used. If none of the specifications are larger than the bitwidth then 1435 the largest integer type is used. For example, given the default 1436 specifications above, the i7 type will use the alignment of i8 (next 1437 largest) while both i65 and i256 will use the alignment of i64 (largest 1438 specified).</li> 1439 1440 <li>If no match is found, and the type sought is a vector type, then the 1441 largest vector type that is smaller than the sought vector type will be 1442 used as a fall back. This happens because <128 x double> can be 1443 implemented in terms of 64 <2 x double>, for example.</li> 1444 </ol> 1445 1446 <p>The function of the data layout string may not be what you expect. Notably, 1447 this is not a specification from the frontend of what alignment the code 1448 generator should use.</p> 1449 1450 <p>Instead, if specified, the target data layout is required to match what the 1451 ultimate <em>code generator</em> expects. This string is used by the 1452 mid-level optimizers to 1453 improve code, and this only works if it matches what the ultimate code 1454 generator uses. If you would like to generate IR that does not embed this 1455 target-specific detail into the IR, then you don't have to specify the 1456 string. This will disable some optimizations that require precise layout 1457 information, but this also prevents those optimizations from introducing 1458 target specificity into the IR.</p> 1459 1460 1461 1462 </div> 1463 1464 <!-- ======================================================================= --> 1465 <h3> 1466 <a name="pointeraliasing">Pointer Aliasing Rules</a> 1467 </h3> 1468 1469 <div> 1470 1471 <p>Any memory access must be done through a pointer value associated 1472 with an address range of the memory access, otherwise the behavior 1473 is undefined. Pointer values are associated with address ranges 1474 according to the following rules:</p> 1475 1476 <ul> 1477 <li>A pointer value is associated with the addresses associated with 1478 any value it is <i>based</i> on. 1479 <li>An address of a global variable is associated with the address 1480 range of the variable's storage.</li> 1481 <li>The result value of an allocation instruction is associated with 1482 the address range of the allocated storage.</li> 1483 <li>A null pointer in the default address-space is associated with 1484 no address.</li> 1485 <li>An integer constant other than zero or a pointer value returned 1486 from a function not defined within LLVM may be associated with address 1487 ranges allocated through mechanisms other than those provided by 1488 LLVM. Such ranges shall not overlap with any ranges of addresses 1489 allocated by mechanisms provided by LLVM.</li> 1490 </ul> 1491 1492 <p>A pointer value is <i>based</i> on another pointer value according 1493 to the following rules:</p> 1494 1495 <ul> 1496 <li>A pointer value formed from a 1497 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation 1498 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li> 1499 <li>The result value of a 1500 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand 1501 of the <tt>bitcast</tt>.</li> 1502 <li>A pointer value formed by an 1503 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all 1504 pointer values that contribute (directly or indirectly) to the 1505 computation of the pointer's value.</li> 1506 <li>The "<i>based</i> on" relationship is transitive.</li> 1507 </ul> 1508 1509 <p>Note that this definition of <i>"based"</i> is intentionally 1510 similar to the definition of <i>"based"</i> in C99, though it is 1511 slightly weaker.</p> 1512 1513 <p>LLVM IR does not associate types with memory. The result type of a 1514 <tt><a href="#i_load">load</a></tt> merely indicates the size and 1515 alignment of the memory from which to load, as well as the 1516 interpretation of the value. The first operand type of a 1517 <tt><a href="#i_store">store</a></tt> similarly only indicates the size 1518 and alignment of the store.</p> 1519 1520 <p>Consequently, type-based alias analysis, aka TBAA, aka 1521 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned 1522 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode 1523 additional information which specialized optimization passes may use 1524 to implement type-based alias analysis.</p> 1525 1526 </div> 1527 1528 <!-- ======================================================================= --> 1529 <h3> 1530 <a name="volatile">Volatile Memory Accesses</a> 1531 </h3> 1532 1533 <div> 1534 1535 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a 1536 href="#i_store"><tt>store</tt></a>s, and <a 1537 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>. 1538 The optimizers must not change the number of volatile operations or change their 1539 order of execution relative to other volatile operations. The optimizers 1540 <i>may</i> change the order of volatile operations relative to non-volatile 1541 operations. This is not Java's "volatile" and has no cross-thread 1542 synchronization behavior.</p> 1543 1544 </div> 1545 1546 <!-- ======================================================================= --> 1547 <h3> 1548 <a name="memmodel">Memory Model for Concurrent Operations</a> 1549 </h3> 1550 1551 <div> 1552 1553 <p>The LLVM IR does not define any way to start parallel threads of execution 1554 or to register signal handlers. Nonetheless, there are platform-specific 1555 ways to create them, and we define LLVM IR's behavior in their presence. This 1556 model is inspired by the C++0x memory model.</p> 1557 1558 <p>For a more informal introduction to this model, see the 1559 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>. 1560 1561 <p>We define a <i>happens-before</i> partial order as the least partial order 1562 that</p> 1563 <ul> 1564 <li>Is a superset of single-thread program order, and</li> 1565 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from 1566 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced 1567 by platform-specific techniques, like pthread locks, thread 1568 creation, thread joining, etc., and by atomic instructions. 1569 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>). 1570 </li> 1571 </ul> 1572 1573 <p>Note that program order does not introduce <i>happens-before</i> edges 1574 between a thread and signals executing inside that thread.</p> 1575 1576 <p>Every (defined) read operation (load instructions, memcpy, atomic 1577 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by 1578 (defined) write operations (store instructions, atomic 1579 stores/read-modify-writes, memcpy, etc.). For the purposes of this section, 1580 initialized globals are considered to have a write of the initializer which is 1581 atomic and happens before any other read or write of the memory in question. 1582 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see 1583 any write to the same byte, except:</p> 1584 1585 <ul> 1586 <li>If <var>write<sub>1</sub></var> happens before 1587 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens 1588 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var> 1589 does not see <var>write<sub>1</sub></var>. 1590 <li>If <var>R<sub>byte</sub></var> happens before 1591 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not 1592 see <var>write<sub>3</sub></var>. 1593 </ul> 1594 1595 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows: 1596 <ul> 1597 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile 1598 is supposed to give guarantees which can support 1599 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to 1600 addresses which do not behave like normal memory. It does not generally 1601 provide cross-thread synchronization.) 1602 <li>Otherwise, if there is no write to the same byte that happens before 1603 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns 1604 <tt>undef</tt> for that byte. 1605 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write, 1606 <var>R<sub>byte</sub></var> returns the value written by that 1607 write.</li> 1608 <li>Otherwise, if <var>R</var> is atomic, and all the writes 1609 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the 1610 values written. See the <a href="#ordering">Atomic Memory Ordering 1611 Constraints</a> section for additional constraints on how the choice 1612 is made. 1613 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li> 1614 </ul> 1615 1616 <p><var>R</var> returns the value composed of the series of bytes it read. 1617 This implies that some bytes within the value may be <tt>undef</tt> 1618 <b>without</b> the entire value being <tt>undef</tt>. Note that this only 1619 defines the semantics of the operation; it doesn't mean that targets will 1620 emit more than one instruction to read the series of bytes.</p> 1621 1622 <p>Note that in cases where none of the atomic intrinsics are used, this model 1623 places only one restriction on IR transformations on top of what is required 1624 for single-threaded execution: introducing a store to a byte which might not 1625 otherwise be stored is not allowed in general. (Specifically, in the case 1626 where another thread might write to and read from an address, introducing a 1627 store can change a load that may see exactly one write into a load that may 1628 see multiple writes.)</p> 1629 1630 <!-- FIXME: This model assumes all targets where concurrency is relevant have 1631 a byte-size store which doesn't affect adjacent bytes. As far as I can tell, 1632 none of the backends currently in the tree fall into this category; however, 1633 there might be targets which care. If there are, we want a paragraph 1634 like the following: 1635 1636 Targets may specify that stores narrower than a certain width are not 1637 available; on such a target, for the purposes of this model, treat any 1638 non-atomic write with an alignment or width less than the minimum width 1639 as if it writes to the relevant surrounding bytes. 1640 --> 1641 1642 </div> 1643 1644 <!-- ======================================================================= --> 1645 <h3> 1646 <a name="ordering">Atomic Memory Ordering Constraints</a> 1647 </h3> 1648 1649 <div> 1650 1651 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>, 1652 <a href="#i_atomicrmw"><code>atomicrmw</code></a>, 1653 <a href="#i_fence"><code>fence</code></a>, 1654 <a href="#i_load"><code>atomic load</code></a>, and 1655 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter 1656 that determines which other atomic instructions on the same address they 1657 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x, 1658 but are somewhat more colloquial. If these descriptions aren't precise enough, 1659 check those specs (see spec references in the 1660 <a href="Atomics.html#introduction">atomics guide</a>). 1661 <a href="#i_fence"><code>fence</code></a> instructions 1662 treat these orderings somewhat differently since they don't take an address. 1663 See that instruction's documentation for details.</p> 1664 1665 <p>For a simpler introduction to the ordering constraints, see the 1666 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p> 1667 1668 <dl> 1669 <dt><code>unordered</code></dt> 1670 <dd>The set of values that can be read is governed by the happens-before 1671 partial order. A value cannot be read unless some operation wrote it. 1672 This is intended to provide a guarantee strong enough to model Java's 1673 non-volatile shared variables. This ordering cannot be specified for 1674 read-modify-write operations; it is not strong enough to make them atomic 1675 in any interesting way.</dd> 1676 <dt><code>monotonic</code></dt> 1677 <dd>In addition to the guarantees of <code>unordered</code>, there is a single 1678 total order for modifications by <code>monotonic</code> operations on each 1679 address. All modification orders must be compatible with the happens-before 1680 order. There is no guarantee that the modification orders can be combined to 1681 a global total order for the whole program (and this often will not be 1682 possible). The read in an atomic read-modify-write operation 1683 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and 1684 <a href="#i_atomicrmw"><code>atomicrmw</code></a>) 1685 reads the value in the modification order immediately before the value it 1686 writes. If one atomic read happens before another atomic read of the same 1687 address, the later read must see the same value or a later value in the 1688 address's modification order. This disallows reordering of 1689 <code>monotonic</code> (or stronger) operations on the same address. If an 1690 address is written <code>monotonic</code>ally by one thread, and other threads 1691 <code>monotonic</code>ally read that address repeatedly, the other threads must 1692 eventually see the write. This corresponds to the C++0x/C1x 1693 <code>memory_order_relaxed</code>.</dd> 1694 <dt><code>acquire</code></dt> 1695 <dd>In addition to the guarantees of <code>monotonic</code>, 1696 a <i>synchronizes-with</i> edge may be formed with a <code>release</code> 1697 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd> 1698 <dt><code>release</code></dt> 1699 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation 1700 writes a value which is subsequently read by an <code>acquire</code> operation, 1701 it <i>synchronizes-with</i> that operation. (This isn't a complete 1702 description; see the C++0x definition of a release sequence.) This corresponds 1703 to the C++0x/C1x <code>memory_order_release</code>.</dd> 1704 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an 1705 <code>acquire</code> and <code>release</code> operation on its address. 1706 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd> 1707 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd> 1708 <dd>In addition to the guarantees of <code>acq_rel</code> 1709 (<code>acquire</code> for an operation which only reads, <code>release</code> 1710 for an operation which only writes), there is a global total order on all 1711 sequentially-consistent operations on all addresses, which is consistent with 1712 the <i>happens-before</i> partial order and with the modification orders of 1713 all the affected addresses. Each sequentially-consistent read sees the last 1714 preceding write to the same address in this global order. This corresponds 1715 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd> 1716 </dl> 1717 1718 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>, 1719 it only <i>synchronizes with</i> or participates in modification and seq_cst 1720 total orderings with other operations running in the same thread (for example, 1721 in signal handlers).</p> 1722 1723 </div> 1724 1725 </div> 1726 1727 <!-- *********************************************************************** --> 1728 <h2><a name="typesystem">Type System</a></h2> 1729 <!-- *********************************************************************** --> 1730 1731 <div> 1732 1733 <p>The LLVM type system is one of the most important features of the 1734 intermediate representation. Being typed enables a number of optimizations 1735 to be performed on the intermediate representation directly, without having 1736 to do extra analyses on the side before the transformation. A strong type 1737 system makes it easier to read the generated code and enables novel analyses 1738 and transformations that are not feasible to perform on normal three address 1739 code representations.</p> 1740 1741 <!-- ======================================================================= --> 1742 <h3> 1743 <a name="t_classifications">Type Classifications</a> 1744 </h3> 1745 1746 <div> 1747 1748 <p>The types fall into a few useful classifications:</p> 1749 1750 <table border="1" cellspacing="0" cellpadding="4"> 1751 <tbody> 1752 <tr><th>Classification</th><th>Types</th></tr> 1753 <tr> 1754 <td><a href="#t_integer">integer</a></td> 1755 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td> 1756 </tr> 1757 <tr> 1758 <td><a href="#t_floating">floating point</a></td> 1759 <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td> 1760 </tr> 1761 <tr> 1762 <td><a name="t_firstclass">first class</a></td> 1763 <td><a href="#t_integer">integer</a>, 1764 <a href="#t_floating">floating point</a>, 1765 <a href="#t_pointer">pointer</a>, 1766 <a href="#t_vector">vector</a>, 1767 <a href="#t_struct">structure</a>, 1768 <a href="#t_array">array</a>, 1769 <a href="#t_label">label</a>, 1770 <a href="#t_metadata">metadata</a>. 1771 </td> 1772 </tr> 1773 <tr> 1774 <td><a href="#t_primitive">primitive</a></td> 1775 <td><a href="#t_label">label</a>, 1776 <a href="#t_void">void</a>, 1777 <a href="#t_integer">integer</a>, 1778 <a href="#t_floating">floating point</a>, 1779 <a href="#t_x86mmx">x86mmx</a>, 1780 <a href="#t_metadata">metadata</a>.</td> 1781 </tr> 1782 <tr> 1783 <td><a href="#t_derived">derived</a></td> 1784 <td><a href="#t_array">array</a>, 1785 <a href="#t_function">function</a>, 1786 <a href="#t_pointer">pointer</a>, 1787 <a href="#t_struct">structure</a>, 1788 <a href="#t_vector">vector</a>, 1789 <a href="#t_opaque">opaque</a>. 1790 </td> 1791 </tr> 1792 </tbody> 1793 </table> 1794 1795 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most 1796 important. Values of these types are the only ones which can be produced by 1797 instructions.</p> 1798 1799 </div> 1800 1801 <!-- ======================================================================= --> 1802 <h3> 1803 <a name="t_primitive">Primitive Types</a> 1804 </h3> 1805 1806 <div> 1807 1808 <p>The primitive types are the fundamental building blocks of the LLVM 1809 system.</p> 1810 1811 <!-- _______________________________________________________________________ --> 1812 <h4> 1813 <a name="t_integer">Integer Type</a> 1814 </h4> 1815 1816 <div> 1817 1818 <h5>Overview:</h5> 1819 <p>The integer type is a very simple type that simply specifies an arbitrary 1820 bit width for the integer type desired. Any bit width from 1 bit to 1821 2<sup>23</sup>-1 (about 8 million) can be specified.</p> 1822 1823 <h5>Syntax:</h5> 1824 <pre> 1825 iN 1826 </pre> 1827 1828 <p>The number of bits the integer will occupy is specified by the <tt>N</tt> 1829 value.</p> 1830 1831 <h5>Examples:</h5> 1832 <table class="layout"> 1833 <tr class="layout"> 1834 <td class="left"><tt>i1</tt></td> 1835 <td class="left">a single-bit integer.</td> 1836 </tr> 1837 <tr class="layout"> 1838 <td class="left"><tt>i32</tt></td> 1839 <td class="left">a 32-bit integer.</td> 1840 </tr> 1841 <tr class="layout"> 1842 <td class="left"><tt>i1942652</tt></td> 1843 <td class="left">a really big integer of over 1 million bits.</td> 1844 </tr> 1845 </table> 1846 1847 </div> 1848 1849 <!-- _______________________________________________________________________ --> 1850 <h4> 1851 <a name="t_floating">Floating Point Types</a> 1852 </h4> 1853 1854 <div> 1855 1856 <table> 1857 <tbody> 1858 <tr><th>Type</th><th>Description</th></tr> 1859 <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr> 1860 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr> 1861 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr> 1862 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr> 1863 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr> 1864 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr> 1865 </tbody> 1866 </table> 1867 1868 </div> 1869 1870 <!-- _______________________________________________________________________ --> 1871 <h4> 1872 <a name="t_x86mmx">X86mmx Type</a> 1873 </h4> 1874 1875 <div> 1876 1877 <h5>Overview:</h5> 1878 <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> 1879 1880 <h5>Syntax:</h5> 1881 <pre> 1882 x86mmx 1883 </pre> 1884 1885 </div> 1886 1887 <!-- _______________________________________________________________________ --> 1888 <h4> 1889 <a name="t_void">Void Type</a> 1890 </h4> 1891 1892 <div> 1893 1894 <h5>Overview:</h5> 1895 <p>The void type does not represent any value and has no size.</p> 1896 1897 <h5>Syntax:</h5> 1898 <pre> 1899 void 1900 </pre> 1901 1902 </div> 1903 1904 <!-- _______________________________________________________________________ --> 1905 <h4> 1906 <a name="t_label">Label Type</a> 1907 </h4> 1908 1909 <div> 1910 1911 <h5>Overview:</h5> 1912 <p>The label type represents code labels.</p> 1913 1914 <h5>Syntax:</h5> 1915 <pre> 1916 label 1917 </pre> 1918 1919 </div> 1920 1921 <!-- _______________________________________________________________________ --> 1922 <h4> 1923 <a name="t_metadata">Metadata Type</a> 1924 </h4> 1925 1926 <div> 1927 1928 <h5>Overview:</h5> 1929 <p>The metadata type represents embedded metadata. No derived types may be 1930 created from metadata except for <a href="#t_function">function</a> 1931 arguments. 1932 1933 <h5>Syntax:</h5> 1934 <pre> 1935 metadata 1936 </pre> 1937 1938 </div> 1939 1940 </div> 1941 1942 <!-- ======================================================================= --> 1943 <h3> 1944 <a name="t_derived">Derived Types</a> 1945 </h3> 1946 1947 <div> 1948 1949 <p>The real power in LLVM comes from the derived types in the system. This is 1950 what allows a programmer to represent arrays, functions, pointers, and other 1951 useful types. Each of these types contain one or more element types which 1952 may be a primitive type, or another derived type. For example, it is 1953 possible to have a two dimensional array, using an array as the element type 1954 of another array.</p> 1955 1956 <!-- _______________________________________________________________________ --> 1957 <h4> 1958 <a name="t_aggregate">Aggregate Types</a> 1959 </h4> 1960 1961 <div> 1962 1963 <p>Aggregate Types are a subset of derived types that can contain multiple 1964 member types. <a href="#t_array">Arrays</a> and 1965 <a href="#t_struct">structs</a> are aggregate types. 1966 <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p> 1967 1968 </div> 1969 1970 <!-- _______________________________________________________________________ --> 1971 <h4> 1972 <a name="t_array">Array Type</a> 1973 </h4> 1974 1975 <div> 1976 1977 <h5>Overview:</h5> 1978 <p>The array type is a very simple derived type that arranges elements 1979 sequentially in memory. The array type requires a size (number of elements) 1980 and an underlying data type.</p> 1981 1982 <h5>Syntax:</h5> 1983 <pre> 1984 [<# elements> x <elementtype>] 1985 </pre> 1986 1987 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may 1988 be any type with a size.</p> 1989 1990 <h5>Examples:</h5> 1991 <table class="layout"> 1992 <tr class="layout"> 1993 <td class="left"><tt>[40 x i32]</tt></td> 1994 <td class="left">Array of 40 32-bit integer values.</td> 1995 </tr> 1996 <tr class="layout"> 1997 <td class="left"><tt>[41 x i32]</tt></td> 1998 <td class="left">Array of 41 32-bit integer values.</td> 1999 </tr> 2000 <tr class="layout"> 2001 <td class="left"><tt>[4 x i8]</tt></td> 2002 <td class="left">Array of 4 8-bit integer values.</td> 2003 </tr> 2004 </table> 2005 <p>Here are some examples of multidimensional arrays:</p> 2006 <table class="layout"> 2007 <tr class="layout"> 2008 <td class="left"><tt>[3 x [4 x i32]]</tt></td> 2009 <td class="left">3x4 array of 32-bit integer values.</td> 2010 </tr> 2011 <tr class="layout"> 2012 <td class="left"><tt>[12 x [10 x float]]</tt></td> 2013 <td class="left">12x10 array of single precision floating point values.</td> 2014 </tr> 2015 <tr class="layout"> 2016 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td> 2017 <td class="left">2x3x4 array of 16-bit integer values.</td> 2018 </tr> 2019 </table> 2020 2021 <p>There is no restriction on indexing beyond the end of the array implied by 2022 a static type (though there are restrictions on indexing beyond the bounds 2023 of an allocated object in some cases). This means that single-dimension 2024 'variable sized array' addressing can be implemented in LLVM with a zero 2025 length array type. An implementation of 'pascal style arrays' in LLVM could 2026 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p> 2027 2028 </div> 2029 2030 <!-- _______________________________________________________________________ --> 2031 <h4> 2032 <a name="t_function">Function Type</a> 2033 </h4> 2034 2035 <div> 2036 2037 <h5>Overview:</h5> 2038 <p>The function type can be thought of as a function signature. It consists of 2039 a return type and a list of formal parameter types. The return type of a 2040 function type is a first class type or a void type.</p> 2041 2042 <h5>Syntax:</h5> 2043 <pre> 2044 <returntype> (<parameter list>) 2045 </pre> 2046 2047 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type 2048 specifiers. Optionally, the parameter list may include a type <tt>...</tt>, 2049 which indicates that the function takes a variable number of arguments. 2050 Variable argument functions can access their arguments with 2051 the <a href="#int_varargs">variable argument handling intrinsic</a> 2052 functions. '<tt><returntype></tt>' is any type except 2053 <a href="#t_label">label</a>.</p> 2054 2055 <h5>Examples:</h5> 2056 <table class="layout"> 2057 <tr class="layout"> 2058 <td class="left"><tt>i32 (i32)</tt></td> 2059 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt> 2060 </td> 2061 </tr><tr class="layout"> 2062 <td class="left"><tt>float (i16, i32 *) * 2063 </tt></td> 2064 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes 2065 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>, 2066 returning <tt>float</tt>. 2067 </td> 2068 </tr><tr class="layout"> 2069 <td class="left"><tt>i32 (i8*, ...)</tt></td> 2070 <td class="left">A vararg function that takes at least one 2071 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C), 2072 which returns an integer. This is the signature for <tt>printf</tt> in 2073 LLVM. 2074 </td> 2075 </tr><tr class="layout"> 2076 <td class="left"><tt>{i32, i32} (i32)</tt></td> 2077 <td class="left">A function taking an <tt>i32</tt>, returning a 2078 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values 2079 </td> 2080 </tr> 2081 </table> 2082 2083 </div> 2084 2085 <!-- _______________________________________________________________________ --> 2086 <h4> 2087 <a name="t_struct">Structure Type</a> 2088 </h4> 2089 2090 <div> 2091 2092 <h5>Overview:</h5> 2093 <p>The structure type is used to represent a collection of data members together 2094 in memory. The elements of a structure may be any type that has a size.</p> 2095 2096 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>' 2097 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field 2098 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction. 2099 Structures in registers are accessed using the 2100 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and 2101 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p> 2102 2103 <p>Structures may optionally be "packed" structures, which indicate that the 2104 alignment of the struct is one byte, and that there is no padding between 2105 the elements. In non-packed structs, padding between field types is inserted 2106 as defined by the TargetData string in the module, which is required to match 2107 what the underlying code generator expects.</p> 2108 2109 <p>Structures can either be "literal" or "identified". A literal structure is 2110 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified 2111 types are always defined at the top level with a name. Literal types are 2112 uniqued by their contents and can never be recursive or opaque since there is 2113 no way to write one. Identified types can be recursive, can be opaqued, and are 2114 never uniqued. 2115 </p> 2116 2117 <h5>Syntax:</h5> 2118 <pre> 2119 %T1 = type { <type list> } <i>; Identified normal struct type</i> 2120 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i> 2121 </pre> 2122 2123 <h5>Examples:</h5> 2124 <table class="layout"> 2125 <tr class="layout"> 2126 <td class="left"><tt>{ i32, i32, i32 }</tt></td> 2127 <td class="left">A triple of three <tt>i32</tt> values</td> 2128 </tr> 2129 <tr class="layout"> 2130 <td class="left"><tt>{ float, i32 (i32) * }</tt></td> 2131 <td class="left">A pair, where the first element is a <tt>float</tt> and the 2132 second element is a <a href="#t_pointer">pointer</a> to a 2133 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning 2134 an <tt>i32</tt>.</td> 2135 </tr> 2136 <tr class="layout"> 2137 <td class="left"><tt><{ i8, i32 }></tt></td> 2138 <td class="left">A packed struct known to be 5 bytes in size.</td> 2139 </tr> 2140 </table> 2141 2142 </div> 2143 2144 <!-- _______________________________________________________________________ --> 2145 <h4> 2146 <a name="t_opaque">Opaque Structure Types</a> 2147 </h4> 2148 2149 <div> 2150 2151 <h5>Overview:</h5> 2152 <p>Opaque structure types are used to represent named structure types that do 2153 not have a body specified. This corresponds (for example) to the C notion of 2154 a forward declared structure.</p> 2155 2156 <h5>Syntax:</h5> 2157 <pre> 2158 %X = type opaque 2159 %52 = type opaque 2160 </pre> 2161 2162 <h5>Examples:</h5> 2163 <table class="layout"> 2164 <tr class="layout"> 2165 <td class="left"><tt>opaque</tt></td> 2166 <td class="left">An opaque type.</td> 2167 </tr> 2168 </table> 2169 2170 </div> 2171 2172 2173 2174 <!-- _______________________________________________________________________ --> 2175 <h4> 2176 <a name="t_pointer">Pointer Type</a> 2177 </h4> 2178 2179 <div> 2180 2181 <h5>Overview:</h5> 2182 <p>The pointer type is used to specify memory locations. 2183 Pointers are commonly used to reference objects in memory.</p> 2184 2185 <p>Pointer types may have an optional address space attribute defining the 2186 numbered address space where the pointed-to object resides. The default 2187 address space is number zero. The semantics of non-zero address 2188 spaces are target-specific.</p> 2189 2190 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it 2191 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p> 2192 2193 <h5>Syntax:</h5> 2194 <pre> 2195 <type> * 2196 </pre> 2197 2198 <h5>Examples:</h5> 2199 <table class="layout"> 2200 <tr class="layout"> 2201 <td class="left"><tt>[4 x i32]*</tt></td> 2202 <td class="left">A <a href="#t_pointer">pointer</a> to <a 2203 href="#t_array">array</a> of four <tt>i32</tt> values.</td> 2204 </tr> 2205 <tr class="layout"> 2206 <td class="left"><tt>i32 (i32*) *</tt></td> 2207 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a 2208 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an 2209 <tt>i32</tt>.</td> 2210 </tr> 2211 <tr class="layout"> 2212 <td class="left"><tt>i32 addrspace(5)*</tt></td> 2213 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value 2214 that resides in address space #5.</td> 2215 </tr> 2216 </table> 2217 2218 </div> 2219 2220 <!-- _______________________________________________________________________ --> 2221 <h4> 2222 <a name="t_vector">Vector Type</a> 2223 </h4> 2224 2225 <div> 2226 2227 <h5>Overview:</h5> 2228 <p>A vector type is a simple derived type that represents a vector of elements. 2229 Vector types are used when multiple primitive data are operated in parallel 2230 using a single instruction (SIMD). A vector type requires a size (number of 2231 elements) and an underlying primitive data type. Vector types are considered 2232 <a href="#t_firstclass">first class</a>.</p> 2233 2234 <h5>Syntax:</h5> 2235 <pre> 2236 < <# elements> x <elementtype> > 2237 </pre> 2238 2239 <p>The number of elements is a constant integer value larger than 0; elementtype 2240 may be any integer or floating point type, or a pointer to these types. 2241 Vectors of size zero are not allowed. </p> 2242 2243 <h5>Examples:</h5> 2244 <table class="layout"> 2245 <tr class="layout"> 2246 <td class="left"><tt><4 x i32></tt></td> 2247 <td class="left">Vector of 4 32-bit integer values.</td> 2248 </tr> 2249 <tr class="layout"> 2250 <td class="left"><tt><8 x float></tt></td> 2251 <td class="left">Vector of 8 32-bit floating-point values.</td> 2252 </tr> 2253 <tr class="layout"> 2254 <td class="left"><tt><2 x i64></tt></td> 2255 <td class="left">Vector of 2 64-bit integer values.</td> 2256 </tr> 2257 <tr class="layout"> 2258 <td class="left"><tt><4 x i64*></tt></td> 2259 <td class="left">Vector of 4 pointers to 64-bit integer values.</td> 2260 </tr> 2261 </table> 2262 2263 </div> 2264 2265 </div> 2266 2267 </div> 2268 2269 <!-- *********************************************************************** --> 2270 <h2><a name="constants">Constants</a></h2> 2271 <!-- *********************************************************************** --> 2272 2273 <div> 2274 2275 <p>LLVM has several different basic types of constants. This section describes 2276 them all and their syntax.</p> 2277 2278 <!-- ======================================================================= --> 2279 <h3> 2280 <a name="simpleconstants">Simple Constants</a> 2281 </h3> 2282 2283 <div> 2284 2285 <dl> 2286 <dt><b>Boolean constants</b></dt> 2287 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid 2288 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd> 2289 2290 <dt><b>Integer constants</b></dt> 2291 <dd>Standard integers (such as '4') are constants of 2292 the <a href="#t_integer">integer</a> type. Negative numbers may be used 2293 with integer types.</dd> 2294 2295 <dt><b>Floating point constants</b></dt> 2296 <dd>Floating point constants use standard decimal notation (e.g. 123.421), 2297 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal 2298 notation (see below). The assembler requires the exact decimal value of a 2299 floating-point constant. For example, the assembler accepts 1.25 but 2300 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point 2301 constants must have a <a href="#t_floating">floating point</a> type. </dd> 2302 2303 <dt><b>Null pointer constants</b></dt> 2304 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant 2305 and must be of <a href="#t_pointer">pointer type</a>.</dd> 2306 </dl> 2307 2308 <p>The one non-intuitive notation for constants is the hexadecimal form of 2309 floating point constants. For example, the form '<tt>double 2310 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) 2311 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point 2312 constants are required (and the only time that they are generated by the 2313 disassembler) is when a floating point constant must be emitted but it cannot 2314 be represented as a decimal floating point number in a reasonable number of 2315 digits. For example, NaN's, infinities, and other special values are 2316 represented in their IEEE hexadecimal format so that assembly and disassembly 2317 do not cause any bits to change in the constants.</p> 2318 2319 <p>When using the hexadecimal form, constants of types half, float, and double are 2320 represented using the 16-digit form shown above (which matches the IEEE754 2321 representation for double); half and float values must, however, be exactly 2322 representable as IEE754 half and single precision, respectively. 2323 Hexadecimal format is always used 2324 for long double, and there are three forms of long double. The 80-bit format 2325 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits. 2326 The 128-bit format used by PowerPC (two adjacent doubles) is represented 2327 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format 2328 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no 2329 currently supported target uses this format. Long doubles will only work if 2330 they match the long double format on your target. The IEEE 16-bit format 2331 (half precision) is represented by <tt>0xH</tt> followed by 4 hexadecimal 2332 digits. All hexadecimal formats are big-endian (sign bit at the left).</p> 2333 2334 <p>There are no constants of type x86mmx.</p> 2335 </div> 2336 2337 <!-- ======================================================================= --> 2338 <h3> 2339 <a name="aggregateconstants"></a> <!-- old anchor --> 2340 <a name="complexconstants">Complex Constants</a> 2341 </h3> 2342 2343 <div> 2344 2345 <p>Complex constants are a (potentially recursive) combination of simple 2346 constants and smaller complex constants.</p> 2347 2348 <dl> 2349 <dt><b>Structure constants</b></dt> 2350 <dd>Structure constants are represented with notation similar to structure 2351 type definitions (a comma separated list of elements, surrounded by braces 2352 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>", 2353 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". 2354 Structure constants must have <a href="#t_struct">structure type</a>, and 2355 the number and types of elements must match those specified by the 2356 type.</dd> 2357 2358 <dt><b>Array constants</b></dt> 2359 <dd>Array constants are represented with notation similar to array type 2360 definitions (a comma separated list of elements, surrounded by square 2361 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 2362 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and 2363 the number and types of elements must match those specified by the 2364 type.</dd> 2365 2366 <dt><b>Vector constants</b></dt> 2367 <dd>Vector constants are represented with notation similar to vector type 2368 definitions (a comma separated list of elements, surrounded by 2369 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 2370 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must 2371 have <a href="#t_vector">vector type</a>, and the number and types of 2372 elements must match those specified by the type.</dd> 2373 2374 <dt><b>Zero initialization</b></dt> 2375 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a 2376 value to zero of <em>any</em> type, including scalar and 2377 <a href="#t_aggregate">aggregate</a> types. 2378 This is often used to avoid having to print large zero initializers 2379 (e.g. for large arrays) and is always exactly equivalent to using explicit 2380 zero initializers.</dd> 2381 2382 <dt><b>Metadata node</b></dt> 2383 <dd>A metadata node is a structure-like constant with 2384 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{ 2385 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to 2386 be interpreted as part of the instruction stream, metadata is a place to 2387 attach additional information such as debug info.</dd> 2388 </dl> 2389 2390 </div> 2391 2392 <!-- ======================================================================= --> 2393 <h3> 2394 <a name="globalconstants">Global Variable and Function Addresses</a> 2395 </h3> 2396 2397 <div> 2398 2399 <p>The addresses of <a href="#globalvars">global variables</a> 2400 and <a href="#functionstructure">functions</a> are always implicitly valid 2401 (link-time) constants. These constants are explicitly referenced when 2402 the <a href="#identifiers">identifier for the global</a> is used and always 2403 have <a href="#t_pointer">pointer</a> type. For example, the following is a 2404 legal LLVM file:</p> 2405 2406 <pre class="doc_code"> 2407 @X = global i32 17 2408 @Y = global i32 42 2409 @Z = global [2 x i32*] [ i32* @X, i32* @Y ] 2410 </pre> 2411 2412 </div> 2413 2414 <!-- ======================================================================= --> 2415 <h3> 2416 <a name="undefvalues">Undefined Values</a> 2417 </h3> 2418 2419 <div> 2420 2421 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and 2422 indicates that the user of the value may receive an unspecified bit-pattern. 2423 Undefined values may be of any type (other than '<tt>label</tt>' 2424 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p> 2425 2426 <p>Undefined values are useful because they indicate to the compiler that the 2427 program is well defined no matter what value is used. This gives the 2428 compiler more freedom to optimize. Here are some examples of (potentially 2429 surprising) transformations that are valid (in pseudo IR):</p> 2430 2431 2432 <pre class="doc_code"> 2433 %A = add %X, undef 2434 %B = sub %X, undef 2435 %C = xor %X, undef 2436 Safe: 2437 %A = undef 2438 %B = undef 2439 %C = undef 2440 </pre> 2441 2442 <p>This is safe because all of the output bits are affected by the undef bits. 2443 Any output bit can have a zero or one depending on the input bits.</p> 2444 2445 <pre class="doc_code"> 2446 %A = or %X, undef 2447 %B = and %X, undef 2448 Safe: 2449 %A = -1 2450 %B = 0 2451 Unsafe: 2452 %A = undef 2453 %B = undef 2454 </pre> 2455 2456 <p>These logical operations have bits that are not always affected by the input. 2457 For example, if <tt>%X</tt> has a zero bit, then the output of the 2458 '<tt>and</tt>' operation will always be a zero for that bit, no matter what 2459 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to 2460 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'. 2461 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be 2462 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that 2463 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be 2464 set, allowing the '<tt>or</tt>' to be folded to -1.</p> 2465 2466 <pre class="doc_code"> 2467 %A = select undef, %X, %Y 2468 %B = select undef, 42, %Y 2469 %C = select %X, %Y, undef 2470 Safe: 2471 %A = %X (or %Y) 2472 %B = 42 (or %Y) 2473 %C = %Y 2474 Unsafe: 2475 %A = undef 2476 %B = undef 2477 %C = undef 2478 </pre> 2479 2480 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional 2481 branch) conditions can go <em>either way</em>, but they have to come from one 2482 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and 2483 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would 2484 have to have a cleared low bit. However, in the <tt>%C</tt> example, the 2485 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the 2486 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be 2487 eliminated.</p> 2488 2489 <pre class="doc_code"> 2490 %A = xor undef, undef 2491 2492 %B = undef 2493 %C = xor %B, %B 2494 2495 %D = undef 2496 %E = icmp lt %D, 4 2497 %F = icmp gte %D, 4 2498 2499 Safe: 2500 %A = undef 2501 %B = undef 2502 %C = undef 2503 %D = undef 2504 %E = undef 2505 %F = undef 2506 </pre> 2507 2508 <p>This example points out that two '<tt>undef</tt>' operands are not 2509 necessarily the same. This can be surprising to people (and also matches C 2510 semantics) where they assume that "<tt>X^X</tt>" is always zero, even 2511 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the 2512 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change 2513 its value over its "live range". This is true because the variable doesn't 2514 actually <em>have a live range</em>. Instead, the value is logically read 2515 from arbitrary registers that happen to be around when needed, so the value 2516 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt> 2517 need to have the same semantics or the core LLVM "replace all uses with" 2518 concept would not hold.</p> 2519 2520 <pre class="doc_code"> 2521 %A = fdiv undef, %X 2522 %B = fdiv %X, undef 2523 Safe: 2524 %A = undef 2525 b: unreachable 2526 </pre> 2527 2528 <p>These examples show the crucial difference between an <em>undefined 2529 value</em> and <em>undefined behavior</em>. An undefined value (like 2530 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that 2531 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because 2532 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently) 2533 defined on SNaN's. However, in the second example, we can make a more 2534 aggressive assumption: because the <tt>undef</tt> is allowed to be an 2535 arbitrary value, we are allowed to assume that it could be zero. Since a 2536 divide by zero has <em>undefined behavior</em>, we are allowed to assume that 2537 the operation does not execute at all. This allows us to delete the divide and 2538 all code after it. Because the undefined operation "can't happen", the 2539 optimizer can assume that it occurs in dead code.</p> 2540 2541 <pre class="doc_code"> 2542 a: store undef -> %X 2543 b: store %X -> undef 2544 Safe: 2545 a: <deleted> 2546 b: unreachable 2547 </pre> 2548 2549 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an 2550 undefined value can be assumed to not have any effect; we can assume that the 2551 value is overwritten with bits that happen to match what was already there. 2552 However, a store <em>to</em> an undefined location could clobber arbitrary 2553 memory, therefore, it has undefined behavior.</p> 2554 2555 </div> 2556 2557 <!-- ======================================================================= --> 2558 <h3> 2559 <a name="poisonvalues">Poison Values</a> 2560 </h3> 2561 2562 <div> 2563 2564 <p>Poison values are similar to <a href="#undefvalues">undef values</a>, however 2565 they also represent the fact that an instruction or constant expression which 2566 cannot evoke side effects has nevertheless detected a condition which results 2567 in undefined behavior.</p> 2568 2569 <p>There is currently no way of representing a poison value in the IR; they 2570 only exist when produced by operations such as 2571 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p> 2572 2573 <p>Poison value behavior is defined in terms of value <i>dependence</i>:</p> 2574 2575 <ul> 2576 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on 2577 their operands.</li> 2578 2579 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding 2580 to their dynamic predecessor basic block.</li> 2581 2582 <li>Function arguments depend on the corresponding actual argument values in 2583 the dynamic callers of their functions.</li> 2584 2585 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the 2586 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer 2587 control back to them.</li> 2588 2589 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the 2590 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>, 2591 or exception-throwing call instructions that dynamically transfer control 2592 back to them.</li> 2593 2594 <li>Non-volatile loads and stores depend on the most recent stores to all of the 2595 referenced memory addresses, following the order in the IR 2596 (including loads and stores implied by intrinsics such as 2597 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li> 2598 2599 <!-- TODO: In the case of multiple threads, this only applies if the store 2600 "happens-before" the load or store. --> 2601 2602 <!-- TODO: floating-point exception state --> 2603 2604 <li>An instruction with externally visible side effects depends on the most 2605 recent preceding instruction with externally visible side effects, following 2606 the order in the IR. (This includes 2607 <a href="#volatile">volatile operations</a>.)</li> 2608 2609 <li>An instruction <i>control-depends</i> on a 2610 <a href="#terminators">terminator instruction</a> 2611 if the terminator instruction has multiple successors and the instruction 2612 is always executed when control transfers to one of the successors, and 2613 may not be executed when control is transferred to another.</li> 2614 2615 <li>Additionally, an instruction also <i>control-depends</i> on a terminator 2616 instruction if the set of instructions it otherwise depends on would be 2617 different if the terminator had transferred control to a different 2618 successor.</li> 2619 2620 <li>Dependence is transitive.</li> 2621 2622 </ul> 2623 2624 <p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>, 2625 with the additional affect that any instruction which has a <i>dependence</i> 2626 on a poison value has undefined behavior.</p> 2627 2628 <p>Here are some examples:</p> 2629 2630 <pre class="doc_code"> 2631 entry: 2632 %poison = sub nuw i32 0, 1 ; Results in a poison value. 2633 %still_poison = and i32 %poison, 0 ; 0, but also poison. 2634 %poison_yet_again = getelementptr i32* @h, i32 %still_poison 2635 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned 2636 2637 store i32 %poison, i32* @g ; Poison value stored to memory. 2638 %poison2 = load i32* @g ; Poison value loaded back from memory. 2639 2640 store volatile i32 %poison, i32* @g ; External observation; undefined behavior. 2641 2642 %narrowaddr = bitcast i32* @g to i16* 2643 %wideaddr = bitcast i32* @g to i64* 2644 %poison3 = load i16* %narrowaddr ; Returns a poison value. 2645 %poison4 = load i64* %wideaddr ; Returns a poison value. 2646 2647 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value. 2648 br i1 %cmp, label %true, label %end ; Branch to either destination. 2649 2650 true: 2651 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so 2652 ; it has undefined behavior. 2653 br label %end 2654 2655 end: 2656 %p = phi i32 [ 0, %entry ], [ 1, %true ] 2657 ; Both edges into this PHI are 2658 ; control-dependent on %cmp, so this 2659 ; always results in a poison value. 2660 2661 store volatile i32 0, i32* @g ; This would depend on the store in %true 2662 ; if %cmp is true, or the store in %entry 2663 ; otherwise, so this is undefined behavior. 2664 2665 br i1 %cmp, label %second_true, label %second_end 2666 ; The same branch again, but this time the 2667 ; true block doesn't have side effects. 2668 2669 second_true: 2670 ; No side effects! 2671 ret void 2672 2673 second_end: 2674 store volatile i32 0, i32* @g ; This time, the instruction always depends 2675 ; on the store in %end. Also, it is 2676 ; control-equivalent to %end, so this is 2677 ; well-defined (ignoring earlier undefined 2678 ; behavior in this example). 2679 </pre> 2680 2681 </div> 2682 2683 <!-- ======================================================================= --> 2684 <h3> 2685 <a name="blockaddress">Addresses of Basic Blocks</a> 2686 </h3> 2687 2688 <div> 2689 2690 <p><b><tt>blockaddress(@function, %block)</tt></b></p> 2691 2692 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified 2693 basic block in the specified function, and always has an i8* type. Taking 2694 the address of the entry block is illegal.</p> 2695 2696 <p>This value only has defined behavior when used as an operand to the 2697 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for 2698 comparisons against null. Pointer equality tests between labels addresses 2699 results in undefined behavior — though, again, comparison against null 2700 is ok, and no label is equal to the null pointer. This may be passed around 2701 as an opaque pointer sized value as long as the bits are not inspected. This 2702 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so 2703 long as the original value is reconstituted before the <tt>indirectbr</tt> 2704 instruction.</p> 2705 2706 <p>Finally, some targets may provide defined semantics when using the value as 2707 the operand to an inline assembly, but that is target specific.</p> 2708 2709 </div> 2710 2711 2712 <!-- ======================================================================= --> 2713 <h3> 2714 <a name="constantexprs">Constant Expressions</a> 2715 </h3> 2716 2717 <div> 2718 2719 <p>Constant expressions are used to allow expressions involving other constants 2720 to be used as constants. Constant expressions may be of 2721 any <a href="#t_firstclass">first class</a> type and may involve any LLVM 2722 operation that does not have side effects (e.g. load and call are not 2723 supported). The following is the syntax for constant expressions:</p> 2724 2725 <dl> 2726 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt> 2727 <dd>Truncate a constant to another type. The bit size of CST must be larger 2728 than the bit size of TYPE. Both types must be integers.</dd> 2729 2730 <dt><b><tt>zext (CST to TYPE)</tt></b></dt> 2731 <dd>Zero extend a constant to another type. The bit size of CST must be 2732 smaller than the bit size of TYPE. Both types must be integers.</dd> 2733 2734 <dt><b><tt>sext (CST to TYPE)</tt></b></dt> 2735 <dd>Sign extend a constant to another type. The bit size of CST must be 2736 smaller than the bit size of TYPE. Both types must be integers.</dd> 2737 2738 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt> 2739 <dd>Truncate a floating point constant to another floating point type. The 2740 size of CST must be larger than the size of TYPE. Both types must be 2741 floating point.</dd> 2742 2743 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt> 2744 <dd>Floating point extend a constant to another type. The size of CST must be 2745 smaller or equal to the size of TYPE. Both types must be floating 2746 point.</dd> 2747 2748 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt> 2749 <dd>Convert a floating point constant to the corresponding unsigned integer 2750 constant. TYPE must be a scalar or vector integer type. CST must be of 2751 scalar or vector floating point type. Both CST and TYPE must be scalars, 2752 or vectors of the same number of elements. If the value won't fit in the 2753 integer type, the results are undefined.</dd> 2754 2755 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt> 2756 <dd>Convert a floating point constant to the corresponding signed integer 2757 constant. TYPE must be a scalar or vector integer type. CST must be of 2758 scalar or vector floating point type. Both CST and TYPE must be scalars, 2759 or vectors of the same number of elements. If the value won't fit in the 2760 integer type, the results are undefined.</dd> 2761 2762 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt> 2763 <dd>Convert an unsigned integer constant to the corresponding floating point 2764 constant. TYPE must be a scalar or vector floating point type. CST must be 2765 of scalar or vector integer type. Both CST and TYPE must be scalars, or 2766 vectors of the same number of elements. If the value won't fit in the 2767 floating point type, the results are undefined.</dd> 2768 2769 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt> 2770 <dd>Convert a signed integer constant to the corresponding floating point 2771 constant. TYPE must be a scalar or vector floating point type. CST must be 2772 of scalar or vector integer type. Both CST and TYPE must be scalars, or 2773 vectors of the same number of elements. If the value won't fit in the 2774 floating point type, the results are undefined.</dd> 2775 2776 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt> 2777 <dd>Convert a pointer typed constant to the corresponding integer constant 2778 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer 2779 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to 2780 make it fit in <tt>TYPE</tt>.</dd> 2781 2782 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt> 2783 <dd>Convert an integer constant to a pointer constant. TYPE must be a pointer 2784 type. CST must be of integer type. The CST value is zero extended, 2785 truncated, or unchanged to make it fit in a pointer size. This one is 2786 <i>really</i> dangerous!</dd> 2787 2788 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt> 2789 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands 2790 are the same as those for the <a href="#i_bitcast">bitcast 2791 instruction</a>.</dd> 2792 2793 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt> 2794 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt> 2795 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on 2796 constants. As with the <a href="#i_getelementptr">getelementptr</a> 2797 instruction, the index list may have zero or more indexes, which are 2798 required to make sense for the type of "CSTPTR".</dd> 2799 2800 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt> 2801 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd> 2802 2803 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt> 2804 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd> 2805 2806 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt> 2807 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd> 2808 2809 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt> 2810 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on 2811 constants.</dd> 2812 2813 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt> 2814 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on 2815 constants.</dd> 2816 2817 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt> 2818 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on 2819 constants.</dd> 2820 2821 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt> 2822 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on 2823 constants. The index list is interpreted in a similar manner as indices in 2824 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one 2825 index value must be specified.</dd> 2826 2827 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt> 2828 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on 2829 constants. The index list is interpreted in a similar manner as indices in 2830 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one 2831 index value must be specified.</dd> 2832 2833 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt> 2834 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may 2835 be any of the <a href="#binaryops">binary</a> 2836 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints 2837 on operands are the same as those for the corresponding instruction 2838 (e.g. no bitwise operations on floating point values are allowed).</dd> 2839 </dl> 2840 2841 </div> 2842 2843 </div> 2844 2845 <!-- *********************************************************************** --> 2846 <h2><a name="othervalues">Other Values</a></h2> 2847 <!-- *********************************************************************** --> 2848 <div> 2849 <!-- ======================================================================= --> 2850 <h3> 2851 <a name="inlineasm">Inline Assembler Expressions</a> 2852 </h3> 2853 2854 <div> 2855 2856 <p>LLVM supports inline assembler expressions (as opposed 2857 to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of 2858 a special value. This value represents the inline assembler as a string 2859 (containing the instructions to emit), a list of operand constraints (stored 2860 as a string), a flag that indicates whether or not the inline asm 2861 expression has side effects, and a flag indicating whether the function 2862 containing the asm needs to align its stack conservatively. An example 2863 inline assembler expression is:</p> 2864 2865 <pre class="doc_code"> 2866 i32 (i32) asm "bswap $0", "=r,r" 2867 </pre> 2868 2869 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of 2870 a <a href="#i_call"><tt>call</tt></a> or an 2871 <a href="#i_invoke"><tt>invoke</tt></a> instruction. 2872 Thus, typically we have:</p> 2873 2874 <pre class="doc_code"> 2875 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y) 2876 </pre> 2877 2878 <p>Inline asms with side effects not visible in the constraint list must be 2879 marked as having side effects. This is done through the use of the 2880 '<tt>sideeffect</tt>' keyword, like so:</p> 2881 2882 <pre class="doc_code"> 2883 call void asm sideeffect "eieio", ""() 2884 </pre> 2885 2886 <p>In some cases inline asms will contain code that will not work unless the 2887 stack is aligned in some way, such as calls or SSE instructions on x86, 2888 yet will not contain code that does that alignment within the asm. 2889 The compiler should make conservative assumptions about what the asm might 2890 contain and should generate its usual stack alignment code in the prologue 2891 if the '<tt>alignstack</tt>' keyword is present:</p> 2892 2893 <pre class="doc_code"> 2894 call void asm alignstack "eieio", ""() 2895 </pre> 2896 2897 <p>Inline asms also support using non-standard assembly dialects. The assumed 2898 dialect is ATT. When the '<tt>inteldialect</tt>' keyword is present, the 2899 inline asm is using the Intel dialect. Currently, ATT and Intel are the 2900 only supported dialects. An example is:</p> 2901 2902 <pre class="doc_code"> 2903 call void asm inteldialect "eieio", ""() 2904 </pre> 2905 2906 <p>If multiple keywords appear the '<tt>sideeffect</tt>' keyword must come 2907 first, the '<tt>alignstack</tt>' keyword second and the 2908 '<tt>inteldialect</tt>' keyword last.</p> 2909 2910 <!-- 2911 <p>TODO: The format of the asm and constraints string still need to be 2912 documented here. Constraints on what can be done (e.g. duplication, moving, 2913 etc need to be documented). This is probably best done by reference to 2914 another document that covers inline asm from a holistic perspective.</p> 2915 --> 2916 2917 <!-- _______________________________________________________________________ --> 2918 <h4> 2919 <a name="inlineasm_md">Inline Asm Metadata</a> 2920 </h4> 2921 2922 <div> 2923 2924 <p>The call instructions that wrap inline asm nodes may have a 2925 "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant 2926 integers. If present, the code generator will use the integer as the 2927 location cookie value when report errors through the <tt>LLVMContext</tt> 2928 error reporting mechanisms. This allows a front-end to correlate backend 2929 errors that occur with inline asm back to the source code that produced it. 2930 For example:</p> 2931 2932 <pre class="doc_code"> 2933 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b> 2934 ... 2935 !42 = !{ i32 1234567 } 2936 </pre> 2937 2938 <p>It is up to the front-end to make sense of the magic numbers it places in the 2939 IR. If the MDNode contains multiple constants, the code generator will use 2940 the one that corresponds to the line of the asm that the error occurs on.</p> 2941 2942 </div> 2943 2944 </div> 2945 2946 <!-- ======================================================================= --> 2947 <h3> 2948 <a name="metadata">Metadata Nodes and Metadata Strings</a> 2949 </h3> 2950 2951 <div> 2952 2953 <p>LLVM IR allows metadata to be attached to instructions in the program that 2954 can convey extra information about the code to the optimizers and code 2955 generator. One example application of metadata is source-level debug 2956 information. There are two metadata primitives: strings and nodes. All 2957 metadata has the <tt>metadata</tt> type and is identified in syntax by a 2958 preceding exclamation point ('<tt>!</tt>').</p> 2959 2960 <p>A metadata string is a string surrounded by double quotes. It can contain 2961 any character by escaping non-printable characters with "<tt>\xx</tt>" where 2962 "<tt>xx</tt>" is the two digit hex code. For example: 2963 "<tt>!"test\00"</tt>".</p> 2964 2965 <p>Metadata nodes are represented with notation similar to structure constants 2966 (a comma separated list of elements, surrounded by braces and preceded by an 2967 exclamation point). Metadata nodes can have any values as their operand. For 2968 example:</p> 2969 2970 <div class="doc_code"> 2971 <pre> 2972 !{ metadata !"test\00", i32 10} 2973 </pre> 2974 </div> 2975 2976 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of 2977 metadata nodes, which can be looked up in the module symbol table. For 2978 example:</p> 2979 2980 <div class="doc_code"> 2981 <pre> 2982 !foo = metadata !{!4, !3} 2983 </pre> 2984 </div> 2985 2986 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt> 2987 function is using two metadata arguments:</p> 2988 2989 <div class="doc_code"> 2990 <pre> 2991 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25) 2992 </pre> 2993 </div> 2994 2995 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is 2996 attached to the <tt>add</tt> instruction using the <tt>!dbg</tt> 2997 identifier:</p> 2998 2999 <div class="doc_code"> 3000 <pre> 3001 %indvar.next = add i64 %indvar, 1, !dbg !21 3002 </pre> 3003 </div> 3004 3005 <p>More information about specific metadata nodes recognized by the optimizers 3006 and code generator is found below.</p> 3007 3008 <!-- _______________________________________________________________________ --> 3009 <h4> 3010 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a> 3011 </h4> 3012 3013 <div> 3014 3015 <p>In LLVM IR, memory does not have types, so LLVM's own type system is not 3016 suitable for doing TBAA. Instead, metadata is added to the IR to describe 3017 a type system of a higher level language. This can be used to implement 3018 typical C/C++ TBAA, but it can also be used to implement custom alias 3019 analysis behavior for other languages.</p> 3020 3021 <p>The current metadata format is very simple. TBAA metadata nodes have up to 3022 three fields, e.g.:</p> 3023 3024 <div class="doc_code"> 3025 <pre> 3026 !0 = metadata !{ metadata !"an example type tree" } 3027 !1 = metadata !{ metadata !"int", metadata !0 } 3028 !2 = metadata !{ metadata !"float", metadata !0 } 3029 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 } 3030 </pre> 3031 </div> 3032 3033 <p>The first field is an identity field. It can be any value, usually 3034 a metadata string, which uniquely identifies the type. The most important 3035 name in the tree is the name of the root node. Two trees with 3036 different root node names are entirely disjoint, even if they 3037 have leaves with common names.</p> 3038 3039 <p>The second field identifies the type's parent node in the tree, or 3040 is null or omitted for a root node. A type is considered to alias 3041 all of its descendants and all of its ancestors in the tree. Also, 3042 a type is considered to alias all types in other trees, so that 3043 bitcode produced from multiple front-ends is handled conservatively.</p> 3044 3045 <p>If the third field is present, it's an integer which if equal to 1 3046 indicates that the type is "constant" (meaning 3047 <tt>pointsToConstantMemory</tt> should return true; see 3048 <a href="AliasAnalysis.html#OtherItfs">other useful 3049 <tt>AliasAnalysis</tt> methods</a>).</p> 3050 3051 </div> 3052 3053 <!-- _______________________________________________________________________ --> 3054 <h4> 3055 <a name="fpmath">'<tt>fpmath</tt>' Metadata</a> 3056 </h4> 3057 3058 <div> 3059 3060 <p><tt>fpmath</tt> metadata may be attached to any instruction of floating point 3061 type. It can be used to express the maximum acceptable error in the result of 3062 that instruction, in ULPs, thus potentially allowing the compiler to use a 3063 more efficient but less accurate method of computing it. ULP is defined as 3064 follows:</p> 3065 3066 <blockquote> 3067 3068 <p>If <tt>x</tt> is a real number that lies between two finite consecutive 3069 floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one 3070 of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the 3071 distance between the two non-equal finite floating-point numbers nearest 3072 <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p> 3073 3074 </blockquote> 3075 3076 <p>The metadata node shall consist of a single positive floating point number 3077 representing the maximum relative error, for example:</p> 3078 3079 <div class="doc_code"> 3080 <pre> 3081 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs 3082 </pre> 3083 </div> 3084 3085 </div> 3086 3087 <!-- _______________________________________________________________________ --> 3088 <h4> 3089 <a name="range">'<tt>range</tt>' Metadata</a> 3090 </h4> 3091 3092 <div> 3093 <p><tt>range</tt> metadata may be attached only to loads of integer types. It 3094 expresses the possible ranges the loaded value is in. The ranges are 3095 represented with a flattened list of integers. The loaded value is known to 3096 be in the union of the ranges defined by each consecutive pair. Each pair 3097 has the following properties:</p> 3098 <ul> 3099 <li>The type must match the type loaded by the instruction.</li> 3100 <li>The pair <tt>a,b</tt> represents the range <tt>[a,b)</tt>.</li> 3101 <li>Both <tt>a</tt> and <tt>b</tt> are constants.</li> 3102 <li>The range is allowed to wrap.</li> 3103 <li>The range should not represent the full or empty set. That is, 3104 <tt>a!=b</tt>. </li> 3105 </ul> 3106 <p> In addition, the pairs must be in signed order of the lower bound and 3107 they must be non-contiguous.</p> 3108 3109 <p>Examples:</p> 3110 <div class="doc_code"> 3111 <pre> 3112 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1 3113 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1 3114 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5 3115 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5 3116 ... 3117 !0 = metadata !{ i8 0, i8 2 } 3118 !1 = metadata !{ i8 255, i8 2 } 3119 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 } 3120 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 } 3121 </pre> 3122 </div> 3123 </div> 3124 </div> 3125 3126 </div> 3127 3128 <!-- *********************************************************************** --> 3129 <h2> 3130 <a name="module_flags">Module Flags Metadata</a> 3131 </h2> 3132 <!-- *********************************************************************** --> 3133 3134 <div> 3135 3136 <p>Information about the module as a whole is difficult to convey to LLVM's 3137 subsystems. The LLVM IR isn't sufficient to transmit this 3138 information. The <tt>llvm.module.flags</tt> named metadata exists in order to 3139 facilitate this. These flags are in the form of key / value pairs — 3140 much like a dictionary — making it easy for any subsystem who cares 3141 about a flag to look it up.</p> 3142 3143 <p>The <tt>llvm.module.flags</tt> metadata contains a list of metadata 3144 triplets. Each triplet has the following form:</p> 3145 3146 <ul> 3147 <li>The first element is a <i>behavior</i> flag, which specifies the behavior 3148 when two (or more) modules are merged together, and it encounters two (or 3149 more) metadata with the same ID. The supported behaviors are described 3150 below.</li> 3151 3152 <li>The second element is a metadata string that is a unique ID for the 3153 metadata. How each ID is interpreted is documented below.</li> 3154 3155 <li>The third element is the value of the flag.</li> 3156 </ul> 3157 3158 <p>When two (or more) modules are merged together, the resulting 3159 <tt>llvm.module.flags</tt> metadata is the union of the 3160 modules' <tt>llvm.module.flags</tt> metadata. The only exception being a flag 3161 with the <i>Override</i> behavior, which may override another flag's value 3162 (see below).</p> 3163 3164 <p>The following behaviors are supported:</p> 3165 3166 <table border="1" cellspacing="0" cellpadding="4"> 3167 <tbody> 3168 <tr> 3169 <th>Value</th> 3170 <th>Behavior</th> 3171 </tr> 3172 <tr> 3173 <td>1</td> 3174 <td align="left"> 3175 <dl> 3176 <dt><b>Error</b></dt> 3177 <dd>Emits an error if two values disagree. It is an error to have an ID 3178 with both an Error and a Warning behavior.</dd> 3179 </dl> 3180 </td> 3181 </tr> 3182 <tr> 3183 <td>2</td> 3184 <td align="left"> 3185 <dl> 3186 <dt><b>Warning</b></dt> 3187 <dd>Emits a warning if two values disagree.</dd> 3188 </dl> 3189 </td> 3190 </tr> 3191 <tr> 3192 <td>3</td> 3193 <td align="left"> 3194 <dl> 3195 <dt><b>Require</b></dt> 3196 <dd>Emits an error when the specified value is not present or doesn't 3197 have the specified value. It is an error for two (or more) 3198 <tt>llvm.module.flags</tt> with the same ID to have the Require 3199 behavior but different values. There may be multiple Require flags 3200 per ID.</dd> 3201 </dl> 3202 </td> 3203 </tr> 3204 <tr> 3205 <td>4</td> 3206 <td align="left"> 3207 <dl> 3208 <dt><b>Override</b></dt> 3209 <dd>Uses the specified value if the two values disagree. It is an 3210 error for two (or more) <tt>llvm.module.flags</tt> with the same 3211 ID to have the Override behavior but different values.</dd> 3212 </dl> 3213 </td> 3214 </tr> 3215 </tbody> 3216 </table> 3217 3218 <p>An example of module flags:</p> 3219 3220 <pre class="doc_code"> 3221 !0 = metadata !{ i32 1, metadata !"foo", i32 1 } 3222 !1 = metadata !{ i32 4, metadata !"bar", i32 37 } 3223 !2 = metadata !{ i32 2, metadata !"qux", i32 42 } 3224 !3 = metadata !{ i32 3, metadata !"qux", 3225 metadata !{ 3226 metadata !"foo", i32 1 3227 } 3228 } 3229 !llvm.module.flags = !{ !0, !1, !2, !3 } 3230 </pre> 3231 3232 <ul> 3233 <li><p>Metadata <tt>!0</tt> has the ID <tt>!"foo"</tt> and the value '1'. The 3234 behavior if two or more <tt>!"foo"</tt> flags are seen is to emit an 3235 error if their values are not equal.</p></li> 3236 3237 <li><p>Metadata <tt>!1</tt> has the ID <tt>!"bar"</tt> and the value '37'. The 3238 behavior if two or more <tt>!"bar"</tt> flags are seen is to use the 3239 value '37' if their values are not equal.</p></li> 3240 3241 <li><p>Metadata <tt>!2</tt> has the ID <tt>!"qux"</tt> and the value '42'. The 3242 behavior if two or more <tt>!"qux"</tt> flags are seen is to emit a 3243 warning if their values are not equal.</p></li> 3244 3245 <li><p>Metadata <tt>!3</tt> has the ID <tt>!"qux"</tt> and the value:</p> 3246 3247 <pre class="doc_code"> 3248 metadata !{ metadata !"foo", i32 1 } 3249 </pre> 3250 3251 <p>The behavior is to emit an error if the <tt>llvm.module.flags</tt> does 3252 not contain a flag with the ID <tt>!"foo"</tt> that has the value 3253 '1'. If two or more <tt>!"qux"</tt> flags exist, then they must have 3254 the same value or an error will be issued.</p></li> 3255 </ul> 3256 3257 3258 <!-- ======================================================================= --> 3259 <h3> 3260 <a name="objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a> 3261 </h3> 3262 3263 <div> 3264 3265 <p>On the Mach-O platform, Objective-C stores metadata about garbage collection 3266 in a special section called "image info". The metadata consists of a version 3267 number and a bitmask specifying what types of garbage collection are 3268 supported (if any) by the file. If two or more modules are linked together 3269 their garbage collection metadata needs to be merged rather than appended 3270 together.</p> 3271 3272 <p>The Objective-C garbage collection module flags metadata consists of the 3273 following key-value pairs:</p> 3274 3275 <table border="1" cellspacing="0" cellpadding="4"> 3276 <col width="30%"> 3277 <tbody> 3278 <tr> 3279 <th>Key</th> 3280 <th>Value</th> 3281 </tr> 3282 <tr> 3283 <td><tt>Objective-C Version</tt></td> 3284 <td align="left"><b>[Required]</b> — The Objective-C ABI 3285 version. Valid values are 1 and 2.</td> 3286 </tr> 3287 <tr> 3288 <td><tt>Objective-C Image Info Version</tt></td> 3289 <td align="left"><b>[Required]</b> — The version of the image info 3290 section. Currently always 0.</td> 3291 </tr> 3292 <tr> 3293 <td><tt>Objective-C Image Info Section</tt></td> 3294 <td align="left"><b>[Required]</b> — The section to place the 3295 metadata. Valid values are <tt>"__OBJC, __image_info, regular"</tt> for 3296 Objective-C ABI version 1, and <tt>"__DATA,__objc_imageinfo, regular, 3297 no_dead_strip"</tt> for Objective-C ABI version 2.</td> 3298 </tr> 3299 <tr> 3300 <td><tt>Objective-C Garbage Collection</tt></td> 3301 <td align="left"><b>[Required]</b> — Specifies whether garbage 3302 collection is supported or not. Valid values are 0, for no garbage 3303 collection, and 2, for garbage collection supported.</td> 3304 </tr> 3305 <tr> 3306 <td><tt>Objective-C GC Only</tt></td> 3307 <td align="left"><b>[Optional]</b> — Specifies that only garbage 3308 collection is supported. If present, its value must be 6. This flag 3309 requires that the <tt>Objective-C Garbage Collection</tt> flag have the 3310 value 2.</td> 3311 </tr> 3312 </tbody> 3313 </table> 3314 3315 <p>Some important flag interactions:</p> 3316 3317 <ul> 3318 <li>If a module with <tt>Objective-C Garbage Collection</tt> set to 0 is 3319 merged with a module with <tt>Objective-C Garbage Collection</tt> set to 3320 2, then the resulting module has the <tt>Objective-C Garbage 3321 Collection</tt> flag set to 0.</li> 3322 3323 <li>A module with <tt>Objective-C Garbage Collection</tt> set to 0 cannot be 3324 merged with a module with <tt>Objective-C GC Only</tt> set to 6.</li> 3325 </ul> 3326 3327 </div> 3328 3329 </div> 3330 3331 <!-- *********************************************************************** --> 3332 <h2> 3333 <a name="intrinsic_globals">Intrinsic Global Variables</a> 3334 </h2> 3335 <!-- *********************************************************************** --> 3336 <div> 3337 <p>LLVM has a number of "magic" global variables that contain data that affect 3338 code generation or other IR semantics. These are documented here. All globals 3339 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This 3340 section and all globals that start with "<tt>llvm.</tt>" are reserved for use 3341 by LLVM.</p> 3342 3343 <!-- ======================================================================= --> 3344 <h3> 3345 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a> 3346 </h3> 3347 3348 <div> 3349 3350 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a 3351 href="#linkage_appending">appending linkage</a>. This array contains a list of 3352 pointers to global variables and functions which may optionally have a pointer 3353 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p> 3354 3355 <div class="doc_code"> 3356 <pre> 3357 @X = global i8 4 3358 @Y = global i32 123 3359 3360 @llvm.used = appending global [2 x i8*] [ 3361 i8* @X, 3362 i8* bitcast (i32* @Y to i8*) 3363 ], section "llvm.metadata" 3364 </pre> 3365 </div> 3366 3367 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the 3368 compiler, assembler, and linker are required to treat the symbol as if there 3369 is a reference to the global that it cannot see. For example, if a variable 3370 has internal linkage and no references other than that from 3371 the <tt>@llvm.used</tt> list, it cannot be deleted. This is commonly used to 3372 represent references from inline asms and other things the compiler cannot 3373 "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p> 3374 3375 <p>On some targets, the code generator must emit a directive to the assembler or 3376 object file to prevent the assembler and linker from molesting the 3377 symbol.</p> 3378 3379 </div> 3380 3381 <!-- ======================================================================= --> 3382 <h3> 3383 <a name="intg_compiler_used"> 3384 The '<tt>llvm.compiler.used</tt>' Global Variable 3385 </a> 3386 </h3> 3387 3388 <div> 3389 3390 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the 3391 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from 3392 touching the symbol. On targets that support it, this allows an intelligent 3393 linker to optimize references to the symbol without being impeded as it would 3394 be by <tt>@llvm.used</tt>.</p> 3395 3396 <p>This is a rare construct that should only be used in rare circumstances, and 3397 should not be exposed to source languages.</p> 3398 3399 </div> 3400 3401 <!-- ======================================================================= --> 3402 <h3> 3403 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a> 3404 </h3> 3405 3406 <div> 3407 3408 <div class="doc_code"> 3409 <pre> 3410 %0 = type { i32, void ()* } 3411 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }] 3412 </pre> 3413 </div> 3414 3415 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor 3416 functions and associated priorities. The functions referenced by this array 3417 will be called in ascending order of priority (i.e. lowest first) when the 3418 module is loaded. The order of functions with the same priority is not 3419 defined.</p> 3420 3421 </div> 3422 3423 <!-- ======================================================================= --> 3424 <h3> 3425 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a> 3426 </h3> 3427 3428 <div> 3429 3430 <div class="doc_code"> 3431 <pre> 3432 %0 = type { i32, void ()* } 3433 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }] 3434 </pre> 3435 </div> 3436 3437 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions 3438 and associated priorities. The functions referenced by this array will be 3439 called in descending order of priority (i.e. highest first) when the module 3440 is loaded. The order of functions with the same priority is not defined.</p> 3441 3442 </div> 3443 3444 </div> 3445 3446 <!-- *********************************************************************** --> 3447 <h2><a name="instref">Instruction Reference</a></h2> 3448 <!-- *********************************************************************** --> 3449 3450 <div> 3451 3452 <p>The LLVM instruction set consists of several different classifications of 3453 instructions: <a href="#terminators">terminator 3454 instructions</a>, <a href="#binaryops">binary instructions</a>, 3455 <a href="#bitwiseops">bitwise binary instructions</a>, 3456 <a href="#memoryops">memory instructions</a>, and 3457 <a href="#otherops">other instructions</a>.</p> 3458 3459 <!-- ======================================================================= --> 3460 <h3> 3461 <a name="terminators">Terminator Instructions</a> 3462 </h3> 3463 3464 <div> 3465 3466 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block 3467 in a program ends with a "Terminator" instruction, which indicates which 3468 block should be executed after the current block is finished. These 3469 terminator instructions typically yield a '<tt>void</tt>' value: they produce 3470 control flow, not values (the one exception being the 3471 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p> 3472 3473 <p>The terminator instructions are: 3474 '<a href="#i_ret"><tt>ret</tt></a>', 3475 '<a href="#i_br"><tt>br</tt></a>', 3476 '<a href="#i_switch"><tt>switch</tt></a>', 3477 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>', 3478 '<a href="#i_invoke"><tt>invoke</tt></a>', 3479 '<a href="#i_resume"><tt>resume</tt></a>', and 3480 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p> 3481 3482 <!-- _______________________________________________________________________ --> 3483 <h4> 3484 <a name="i_ret">'<tt>ret</tt>' Instruction</a> 3485 </h4> 3486 3487 <div> 3488 3489 <h5>Syntax:</h5> 3490 <pre> 3491 ret <type> <value> <i>; Return a value from a non-void function</i> 3492 ret void <i>; Return from void function</i> 3493 </pre> 3494 3495 <h5>Overview:</h5> 3496 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally 3497 a value) from a function back to the caller.</p> 3498 3499 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a 3500 value and then causes control flow, and one that just causes control flow to 3501 occur.</p> 3502 3503 <h5>Arguments:</h5> 3504 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the 3505 return value. The type of the return value must be a 3506 '<a href="#t_firstclass">first class</a>' type.</p> 3507 3508 <p>A function is not <a href="#wellformed">well formed</a> if it it has a 3509 non-void return type and contains a '<tt>ret</tt>' instruction with no return 3510 value or a return value with a type that does not match its type, or if it 3511 has a void return type and contains a '<tt>ret</tt>' instruction with a 3512 return value.</p> 3513 3514 <h5>Semantics:</h5> 3515 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to 3516 the calling function's context. If the caller is a 3517 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the 3518 instruction after the call. If the caller was an 3519 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at 3520 the beginning of the "normal" destination block. If the instruction returns 3521 a value, that value shall set the call or invoke instruction's return 3522 value.</p> 3523 3524 <h5>Example:</h5> 3525 <pre> 3526 ret i32 5 <i>; Return an integer value of 5</i> 3527 ret void <i>; Return from a void function</i> 3528 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i> 3529 </pre> 3530 3531 </div> 3532 <!-- _______________________________________________________________________ --> 3533 <h4> 3534 <a name="i_br">'<tt>br</tt>' Instruction</a> 3535 </h4> 3536 3537 <div> 3538 3539 <h5>Syntax:</h5> 3540 <pre> 3541 br i1 <cond>, label <iftrue>, label <iffalse> 3542 br label <dest> <i>; Unconditional branch</i> 3543 </pre> 3544 3545 <h5>Overview:</h5> 3546 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a 3547 different basic block in the current function. There are two forms of this 3548 instruction, corresponding to a conditional branch and an unconditional 3549 branch.</p> 3550 3551 <h5>Arguments:</h5> 3552 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single 3553 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form 3554 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a 3555 target.</p> 3556 3557 <h5>Semantics:</h5> 3558 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>' 3559 argument is evaluated. If the value is <tt>true</tt>, control flows to the 3560 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>, 3561 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p> 3562 3563 <h5>Example:</h5> 3564 <pre> 3565 Test: 3566 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b 3567 br i1 %cond, label %IfEqual, label %IfUnequal 3568 IfEqual: 3569 <a href="#i_ret">ret</a> i32 1 3570 IfUnequal: 3571 <a href="#i_ret">ret</a> i32 0 3572 </pre> 3573 3574 </div> 3575 3576 <!-- _______________________________________________________________________ --> 3577 <h4> 3578 <a name="i_switch">'<tt>switch</tt>' Instruction</a> 3579 </h4> 3580 3581 <div> 3582 3583 <h5>Syntax:</h5> 3584 <pre> 3585 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ] 3586 </pre> 3587 3588 <h5>Overview:</h5> 3589 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of 3590 several different places. It is a generalization of the '<tt>br</tt>' 3591 instruction, allowing a branch to occur to one of many possible 3592 destinations.</p> 3593 3594 <h5>Arguments:</h5> 3595 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer 3596 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, 3597 and an array of pairs of comparison value constants and '<tt>label</tt>'s. 3598 The table is not allowed to contain duplicate constant entries.</p> 3599 3600 <h5>Semantics:</h5> 3601 <p>The <tt>switch</tt> instruction specifies a table of values and 3602 destinations. When the '<tt>switch</tt>' instruction is executed, this table 3603 is searched for the given value. If the value is found, control flow is 3604 transferred to the corresponding destination; otherwise, control flow is 3605 transferred to the default destination.</p> 3606 3607 <h5>Implementation:</h5> 3608 <p>Depending on properties of the target machine and the particular 3609 <tt>switch</tt> instruction, this instruction may be code generated in 3610 different ways. For example, it could be generated as a series of chained 3611 conditional branches or with a lookup table.</p> 3612 3613 <h5>Example:</h5> 3614 <pre> 3615 <i>; Emulate a conditional br instruction</i> 3616 %Val = <a href="#i_zext">zext</a> i1 %value to i32 3617 switch i32 %Val, label %truedest [ i32 0, label %falsedest ] 3618 3619 <i>; Emulate an unconditional br instruction</i> 3620 switch i32 0, label %dest [ ] 3621 3622 <i>; Implement a jump table:</i> 3623 switch i32 %val, label %otherwise [ i32 0, label %onzero 3624 i32 1, label %onone 3625 i32 2, label %ontwo ] 3626 </pre> 3627 3628 </div> 3629 3630 3631 <!-- _______________________________________________________________________ --> 3632 <h4> 3633 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a> 3634 </h4> 3635 3636 <div> 3637 3638 <h5>Syntax:</h5> 3639 <pre> 3640 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ] 3641 </pre> 3642 3643 <h5>Overview:</h5> 3644 3645 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label 3646 within the current function, whose address is specified by 3647 "<tt>address</tt>". Address must be derived from a <a 3648 href="#blockaddress">blockaddress</a> constant.</p> 3649 3650 <h5>Arguments:</h5> 3651 3652 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The 3653 rest of the arguments indicate the full set of possible destinations that the 3654 address may point to. Blocks are allowed to occur multiple times in the 3655 destination list, though this isn't particularly useful.</p> 3656 3657 <p>This destination list is required so that dataflow analysis has an accurate 3658 understanding of the CFG.</p> 3659 3660 <h5>Semantics:</h5> 3661 3662 <p>Control transfers to the block specified in the address argument. All 3663 possible destination blocks must be listed in the label list, otherwise this 3664 instruction has undefined behavior. This implies that jumps to labels 3665 defined in other functions have undefined behavior as well.</p> 3666 3667 <h5>Implementation:</h5> 3668 3669 <p>This is typically implemented with a jump through a register.</p> 3670 3671 <h5>Example:</h5> 3672 <pre> 3673 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ] 3674 </pre> 3675 3676 </div> 3677 3678 3679 <!-- _______________________________________________________________________ --> 3680 <h4> 3681 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a> 3682 </h4> 3683 3684 <div> 3685 3686 <h5>Syntax:</h5> 3687 <pre> 3688 <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>] 3689 to label <normal label> unwind label <exception label> 3690 </pre> 3691 3692 <h5>Overview:</h5> 3693 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified 3694 function, with the possibility of control flow transfer to either the 3695 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee 3696 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction, 3697 control flow will return to the "normal" label. If the callee (or any 3698 indirect callees) returns via the "<a href="#i_resume"><tt>resume</tt></a>" 3699 instruction or other exception handling mechanism, control is interrupted and 3700 continued at the dynamically nearest "exception" label.</p> 3701 3702 <p>The '<tt>exception</tt>' label is a 3703 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the 3704 exception. As such, '<tt>exception</tt>' label is required to have the 3705 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains 3706 the information about the behavior of the program after unwinding 3707 happens, as its first non-PHI instruction. The restrictions on the 3708 "<tt>landingpad</tt>" instruction's tightly couples it to the 3709 "<tt>invoke</tt>" instruction, so that the important information contained 3710 within the "<tt>landingpad</tt>" instruction can't be lost through normal 3711 code motion.</p> 3712 3713 <h5>Arguments:</h5> 3714 <p>This instruction requires several arguments:</p> 3715 3716 <ol> 3717 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling 3718 convention</a> the call should use. If none is specified, the call 3719 defaults to using C calling conventions.</li> 3720 3721 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for 3722 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and 3723 '<tt>inreg</tt>' attributes are valid here.</li> 3724 3725 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to 3726 function value being invoked. In most cases, this is a direct function 3727 invocation, but indirect <tt>invoke</tt>s are just as possible, branching 3728 off an arbitrary pointer to function value.</li> 3729 3730 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a 3731 function to be invoked. </li> 3732 3733 <li>'<tt>function args</tt>': argument list whose types match the function 3734 signature argument types and parameter attributes. All arguments must be 3735 of <a href="#t_firstclass">first class</a> type. If the function 3736 signature indicates the function accepts a variable number of arguments, 3737 the extra arguments can be specified.</li> 3738 3739 <li>'<tt>normal label</tt>': the label reached when the called function 3740 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li> 3741 3742 <li>'<tt>exception label</tt>': the label reached when a callee returns via 3743 the <a href="#i_resume"><tt>resume</tt></a> instruction or other exception 3744 handling mechanism.</li> 3745 3746 <li>The optional <a href="#fnattrs">function attributes</a> list. Only 3747 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and 3748 '<tt>readnone</tt>' attributes are valid here.</li> 3749 </ol> 3750 3751 <h5>Semantics:</h5> 3752 <p>This instruction is designed to operate as a standard 3753 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The 3754 primary difference is that it establishes an association with a label, which 3755 is used by the runtime library to unwind the stack.</p> 3756 3757 <p>This instruction is used in languages with destructors to ensure that proper 3758 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown 3759 exception. Additionally, this is important for implementation of 3760 '<tt>catch</tt>' clauses in high-level languages that support them.</p> 3761 3762 <p>For the purposes of the SSA form, the definition of the value returned by the 3763 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current 3764 block to the "normal" label. If the callee unwinds then no return value is 3765 available.</p> 3766 3767 <h5>Example:</h5> 3768 <pre> 3769 %retval = invoke i32 @Test(i32 15) to label %Continue 3770 unwind label %TestCleanup <i>; {i32}:retval set</i> 3771 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue 3772 unwind label %TestCleanup <i>; {i32}:retval set</i> 3773 </pre> 3774 3775 </div> 3776 3777 <!-- _______________________________________________________________________ --> 3778 3779 <h4> 3780 <a name="i_resume">'<tt>resume</tt>' Instruction</a> 3781 </h4> 3782 3783 <div> 3784 3785 <h5>Syntax:</h5> 3786 <pre> 3787 resume <type> <value> 3788 </pre> 3789 3790 <h5>Overview:</h5> 3791 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no 3792 successors.</p> 3793 3794 <h5>Arguments:</h5> 3795 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the 3796 same type as the result of any '<tt>landingpad</tt>' instruction in the same 3797 function.</p> 3798 3799 <h5>Semantics:</h5> 3800 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing 3801 (in-flight) exception whose unwinding was interrupted with 3802 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p> 3803 3804 <h5>Example:</h5> 3805 <pre> 3806 resume { i8*, i32 } %exn 3807 </pre> 3808 3809 </div> 3810 3811 <!-- _______________________________________________________________________ --> 3812 3813 <h4> 3814 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a> 3815 </h4> 3816 3817 <div> 3818 3819 <h5>Syntax:</h5> 3820 <pre> 3821 unreachable 3822 </pre> 3823 3824 <h5>Overview:</h5> 3825 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This 3826 instruction is used to inform the optimizer that a particular portion of the 3827 code is not reachable. This can be used to indicate that the code after a 3828 no-return function cannot be reached, and other facts.</p> 3829 3830 <h5>Semantics:</h5> 3831 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p> 3832 3833 </div> 3834 3835 </div> 3836 3837 <!-- ======================================================================= --> 3838 <h3> 3839 <a name="binaryops">Binary Operations</a> 3840 </h3> 3841 3842 <div> 3843 3844 <p>Binary operators are used to do most of the computation in a program. They 3845 require two operands of the same type, execute an operation on them, and 3846 produce a single value. The operands might represent multiple data, as is 3847 the case with the <a href="#t_vector">vector</a> data type. The result value 3848 has the same type as its operands.</p> 3849 3850 <p>There are several different binary operators:</p> 3851 3852 <!-- _______________________________________________________________________ --> 3853 <h4> 3854 <a name="i_add">'<tt>add</tt>' Instruction</a> 3855 </h4> 3856 3857 <div> 3858 3859 <h5>Syntax:</h5> 3860 <pre> 3861 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3862 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3863 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3864 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3865 </pre> 3866 3867 <h5>Overview:</h5> 3868 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p> 3869 3870 <h5>Arguments:</h5> 3871 <p>The two arguments to the '<tt>add</tt>' instruction must 3872 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of 3873 integer values. Both arguments must have identical types.</p> 3874 3875 <h5>Semantics:</h5> 3876 <p>The value produced is the integer sum of the two operands.</p> 3877 3878 <p>If the sum has unsigned overflow, the result returned is the mathematical 3879 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p> 3880 3881 <p>Because LLVM integers use a two's complement representation, this instruction 3882 is appropriate for both signed and unsigned integers.</p> 3883 3884 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap" 3885 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or 3886 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt> 3887 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow, 3888 respectively, occurs.</p> 3889 3890 <h5>Example:</h5> 3891 <pre> 3892 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i> 3893 </pre> 3894 3895 </div> 3896 3897 <!-- _______________________________________________________________________ --> 3898 <h4> 3899 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a> 3900 </h4> 3901 3902 <div> 3903 3904 <h5>Syntax:</h5> 3905 <pre> 3906 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3907 </pre> 3908 3909 <h5>Overview:</h5> 3910 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p> 3911 3912 <h5>Arguments:</h5> 3913 <p>The two arguments to the '<tt>fadd</tt>' instruction must be 3914 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of 3915 floating point values. Both arguments must have identical types.</p> 3916 3917 <h5>Semantics:</h5> 3918 <p>The value produced is the floating point sum of the two operands.</p> 3919 3920 <h5>Example:</h5> 3921 <pre> 3922 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i> 3923 </pre> 3924 3925 </div> 3926 3927 <!-- _______________________________________________________________________ --> 3928 <h4> 3929 <a name="i_sub">'<tt>sub</tt>' Instruction</a> 3930 </h4> 3931 3932 <div> 3933 3934 <h5>Syntax:</h5> 3935 <pre> 3936 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3937 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3938 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3939 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3940 </pre> 3941 3942 <h5>Overview:</h5> 3943 <p>The '<tt>sub</tt>' instruction returns the difference of its two 3944 operands.</p> 3945 3946 <p>Note that the '<tt>sub</tt>' instruction is used to represent the 3947 '<tt>neg</tt>' instruction present in most other intermediate 3948 representations.</p> 3949 3950 <h5>Arguments:</h5> 3951 <p>The two arguments to the '<tt>sub</tt>' instruction must 3952 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of 3953 integer values. Both arguments must have identical types.</p> 3954 3955 <h5>Semantics:</h5> 3956 <p>The value produced is the integer difference of the two operands.</p> 3957 3958 <p>If the difference has unsigned overflow, the result returned is the 3959 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the 3960 result.</p> 3961 3962 <p>Because LLVM integers use a two's complement representation, this instruction 3963 is appropriate for both signed and unsigned integers.</p> 3964 3965 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap" 3966 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or 3967 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt> 3968 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow, 3969 respectively, occurs.</p> 3970 3971 <h5>Example:</h5> 3972 <pre> 3973 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i> 3974 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i> 3975 </pre> 3976 3977 </div> 3978 3979 <!-- _______________________________________________________________________ --> 3980 <h4> 3981 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a> 3982 </h4> 3983 3984 <div> 3985 3986 <h5>Syntax:</h5> 3987 <pre> 3988 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i> 3989 </pre> 3990 3991 <h5>Overview:</h5> 3992 <p>The '<tt>fsub</tt>' instruction returns the difference of its two 3993 operands.</p> 3994 3995 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the 3996 '<tt>fneg</tt>' instruction present in most other intermediate 3997 representations.</p> 3998 3999 <h5>Arguments:</h5> 4000 <p>The two arguments to the '<tt>fsub</tt>' instruction must be 4001 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of 4002 floating point values. Both arguments must have identical types.</p> 4003 4004 <h5>Semantics:</h5> 4005 <p>The value produced is the floating point difference of the two operands.</p> 4006 4007 <h5>Example:</h5> 4008 <pre> 4009 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i> 4010 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i> 4011 </pre> 4012 4013 </div> 4014 4015 <!-- _______________________________________________________________________ --> 4016 <h4> 4017 <a name="i_mul">'<tt>mul</tt>' Instruction</a> 4018 </h4> 4019 4020 <div> 4021 4022 <h5>Syntax:</h5> 4023 <pre> 4024 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4025 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4026 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4027 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4028 </pre> 4029 4030 <h5>Overview:</h5> 4031 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p> 4032 4033 <h5>Arguments:</h5> 4034 <p>The two arguments to the '<tt>mul</tt>' instruction must 4035 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of 4036 integer values. Both arguments must have identical types.</p> 4037 4038 <h5>Semantics:</h5> 4039 <p>The value produced is the integer product of the two operands.</p> 4040 4041 <p>If the result of the multiplication has unsigned overflow, the result 4042 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit 4043 width of the result.</p> 4044 4045 <p>Because LLVM integers use a two's complement representation, and the result 4046 is the same width as the operands, this instruction returns the correct 4047 result for both signed and unsigned integers. If a full product 4048 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should 4049 be sign-extended or zero-extended as appropriate to the width of the full 4050 product.</p> 4051 4052 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap" 4053 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or 4054 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt> 4055 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow, 4056 respectively, occurs.</p> 4057 4058 <h5>Example:</h5> 4059 <pre> 4060 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i> 4061 </pre> 4062 4063 </div> 4064 4065 <!-- _______________________________________________________________________ --> 4066 <h4> 4067 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a> 4068 </h4> 4069 4070 <div> 4071 4072 <h5>Syntax:</h5> 4073 <pre> 4074 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4075 </pre> 4076 4077 <h5>Overview:</h5> 4078 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p> 4079 4080 <h5>Arguments:</h5> 4081 <p>The two arguments to the '<tt>fmul</tt>' instruction must be 4082 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of 4083 floating point values. Both arguments must have identical types.</p> 4084 4085 <h5>Semantics:</h5> 4086 <p>The value produced is the floating point product of the two operands.</p> 4087 4088 <h5>Example:</h5> 4089 <pre> 4090 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i> 4091 </pre> 4092 4093 </div> 4094 4095 <!-- _______________________________________________________________________ --> 4096 <h4> 4097 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a> 4098 </h4> 4099 4100 <div> 4101 4102 <h5>Syntax:</h5> 4103 <pre> 4104 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4105 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4106 </pre> 4107 4108 <h5>Overview:</h5> 4109 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p> 4110 4111 <h5>Arguments:</h5> 4112 <p>The two arguments to the '<tt>udiv</tt>' instruction must be 4113 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4114 values. Both arguments must have identical types.</p> 4115 4116 <h5>Semantics:</h5> 4117 <p>The value produced is the unsigned integer quotient of the two operands.</p> 4118 4119 <p>Note that unsigned integer division and signed integer division are distinct 4120 operations; for signed integer division, use '<tt>sdiv</tt>'.</p> 4121 4122 <p>Division by zero leads to undefined behavior.</p> 4123 4124 <p>If the <tt>exact</tt> keyword is present, the result value of the 4125 <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a 4126 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p> 4127 4128 4129 <h5>Example:</h5> 4130 <pre> 4131 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i> 4132 </pre> 4133 4134 </div> 4135 4136 <!-- _______________________________________________________________________ --> 4137 <h4> 4138 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a> 4139 </h4> 4140 4141 <div> 4142 4143 <h5>Syntax:</h5> 4144 <pre> 4145 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4146 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4147 </pre> 4148 4149 <h5>Overview:</h5> 4150 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p> 4151 4152 <h5>Arguments:</h5> 4153 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be 4154 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4155 values. Both arguments must have identical types.</p> 4156 4157 <h5>Semantics:</h5> 4158 <p>The value produced is the signed integer quotient of the two operands rounded 4159 towards zero.</p> 4160 4161 <p>Note that signed integer division and unsigned integer division are distinct 4162 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p> 4163 4164 <p>Division by zero leads to undefined behavior. Overflow also leads to 4165 undefined behavior; this is a rare case, but can occur, for example, by doing 4166 a 32-bit division of -2147483648 by -1.</p> 4167 4168 <p>If the <tt>exact</tt> keyword is present, the result value of the 4169 <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would 4170 be rounded.</p> 4171 4172 <h5>Example:</h5> 4173 <pre> 4174 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i> 4175 </pre> 4176 4177 </div> 4178 4179 <!-- _______________________________________________________________________ --> 4180 <h4> 4181 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a> 4182 </h4> 4183 4184 <div> 4185 4186 <h5>Syntax:</h5> 4187 <pre> 4188 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4189 </pre> 4190 4191 <h5>Overview:</h5> 4192 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p> 4193 4194 <h5>Arguments:</h5> 4195 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be 4196 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of 4197 floating point values. Both arguments must have identical types.</p> 4198 4199 <h5>Semantics:</h5> 4200 <p>The value produced is the floating point quotient of the two operands.</p> 4201 4202 <h5>Example:</h5> 4203 <pre> 4204 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i> 4205 </pre> 4206 4207 </div> 4208 4209 <!-- _______________________________________________________________________ --> 4210 <h4> 4211 <a name="i_urem">'<tt>urem</tt>' Instruction</a> 4212 </h4> 4213 4214 <div> 4215 4216 <h5>Syntax:</h5> 4217 <pre> 4218 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4219 </pre> 4220 4221 <h5>Overview:</h5> 4222 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned 4223 division of its two arguments.</p> 4224 4225 <h5>Arguments:</h5> 4226 <p>The two arguments to the '<tt>urem</tt>' instruction must be 4227 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4228 values. Both arguments must have identical types.</p> 4229 4230 <h5>Semantics:</h5> 4231 <p>This instruction returns the unsigned integer <i>remainder</i> of a division. 4232 This instruction always performs an unsigned division to get the 4233 remainder.</p> 4234 4235 <p>Note that unsigned integer remainder and signed integer remainder are 4236 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p> 4237 4238 <p>Taking the remainder of a division by zero leads to undefined behavior.</p> 4239 4240 <h5>Example:</h5> 4241 <pre> 4242 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i> 4243 </pre> 4244 4245 </div> 4246 4247 <!-- _______________________________________________________________________ --> 4248 <h4> 4249 <a name="i_srem">'<tt>srem</tt>' Instruction</a> 4250 </h4> 4251 4252 <div> 4253 4254 <h5>Syntax:</h5> 4255 <pre> 4256 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4257 </pre> 4258 4259 <h5>Overview:</h5> 4260 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed 4261 division of its two operands. This instruction can also take 4262 <a href="#t_vector">vector</a> versions of the values in which case the 4263 elements must be integers.</p> 4264 4265 <h5>Arguments:</h5> 4266 <p>The two arguments to the '<tt>srem</tt>' instruction must be 4267 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4268 values. Both arguments must have identical types.</p> 4269 4270 <h5>Semantics:</h5> 4271 <p>This instruction returns the <i>remainder</i> of a division (where the result 4272 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the 4273 <i>modulo</i> operator (where the result is either zero or has the same sign 4274 as the divisor, <tt>op2</tt>) of a value. 4275 For more information about the difference, 4276 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The 4277 Math Forum</a>. For a table of how this is implemented in various languages, 4278 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation"> 4279 Wikipedia: modulo operation</a>.</p> 4280 4281 <p>Note that signed integer remainder and unsigned integer remainder are 4282 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p> 4283 4284 <p>Taking the remainder of a division by zero leads to undefined behavior. 4285 Overflow also leads to undefined behavior; this is a rare case, but can 4286 occur, for example, by taking the remainder of a 32-bit division of 4287 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule 4288 lets srem be implemented using instructions that return both the result of 4289 the division and the remainder.)</p> 4290 4291 <h5>Example:</h5> 4292 <pre> 4293 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i> 4294 </pre> 4295 4296 </div> 4297 4298 <!-- _______________________________________________________________________ --> 4299 <h4> 4300 <a name="i_frem">'<tt>frem</tt>' Instruction</a> 4301 </h4> 4302 4303 <div> 4304 4305 <h5>Syntax:</h5> 4306 <pre> 4307 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4308 </pre> 4309 4310 <h5>Overview:</h5> 4311 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of 4312 its two operands.</p> 4313 4314 <h5>Arguments:</h5> 4315 <p>The two arguments to the '<tt>frem</tt>' instruction must be 4316 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of 4317 floating point values. Both arguments must have identical types.</p> 4318 4319 <h5>Semantics:</h5> 4320 <p>This instruction returns the <i>remainder</i> of a division. The remainder 4321 has the same sign as the dividend.</p> 4322 4323 <h5>Example:</h5> 4324 <pre> 4325 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i> 4326 </pre> 4327 4328 </div> 4329 4330 </div> 4331 4332 <!-- ======================================================================= --> 4333 <h3> 4334 <a name="bitwiseops">Bitwise Binary Operations</a> 4335 </h3> 4336 4337 <div> 4338 4339 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a 4340 program. They are generally very efficient instructions and can commonly be 4341 strength reduced from other instructions. They require two operands of the 4342 same type, execute an operation on them, and produce a single value. The 4343 resulting value is the same type as its operands.</p> 4344 4345 <!-- _______________________________________________________________________ --> 4346 <h4> 4347 <a name="i_shl">'<tt>shl</tt>' Instruction</a> 4348 </h4> 4349 4350 <div> 4351 4352 <h5>Syntax:</h5> 4353 <pre> 4354 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4355 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4356 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4357 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4358 </pre> 4359 4360 <h5>Overview:</h5> 4361 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left 4362 a specified number of bits.</p> 4363 4364 <h5>Arguments:</h5> 4365 <p>Both arguments to the '<tt>shl</tt>' instruction must be the 4366 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of 4367 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p> 4368 4369 <h5>Semantics:</h5> 4370 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 4371 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt> 4372 is (statically or dynamically) negative or equal to or larger than the number 4373 of bits in <tt>op1</tt>, the result is undefined. If the arguments are 4374 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding 4375 shift amount in <tt>op2</tt>.</p> 4376 4377 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a 4378 <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits. If 4379 the <tt>nsw</tt> keyword is present, then the shift produces a 4380 <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree 4381 with the resultant sign bit. As such, NUW/NSW have the same semantics as 4382 they would if the shift were expressed as a mul instruction with the same 4383 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p> 4384 4385 <h5>Example:</h5> 4386 <pre> 4387 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i> 4388 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i> 4389 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i> 4390 <result> = shl i32 1, 32 <i>; undefined</i> 4391 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i> 4392 </pre> 4393 4394 </div> 4395 4396 <!-- _______________________________________________________________________ --> 4397 <h4> 4398 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a> 4399 </h4> 4400 4401 <div> 4402 4403 <h5>Syntax:</h5> 4404 <pre> 4405 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4406 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4407 </pre> 4408 4409 <h5>Overview:</h5> 4410 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first 4411 operand shifted to the right a specified number of bits with zero fill.</p> 4412 4413 <h5>Arguments:</h5> 4414 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same 4415 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4416 type. '<tt>op2</tt>' is treated as an unsigned value.</p> 4417 4418 <h5>Semantics:</h5> 4419 <p>This instruction always performs a logical shift right operation. The most 4420 significant bits of the result will be filled with zero bits after the shift. 4421 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the 4422 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are 4423 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding 4424 shift amount in <tt>op2</tt>.</p> 4425 4426 <p>If the <tt>exact</tt> keyword is present, the result value of the 4427 <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits 4428 shifted out are non-zero.</p> 4429 4430 4431 <h5>Example:</h5> 4432 <pre> 4433 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i> 4434 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i> 4435 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i> 4436 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i> 4437 <result> = lshr i32 1, 32 <i>; undefined</i> 4438 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i> 4439 </pre> 4440 4441 </div> 4442 4443 <!-- _______________________________________________________________________ --> 4444 <h4> 4445 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a> 4446 </h4> 4447 4448 <div> 4449 4450 <h5>Syntax:</h5> 4451 <pre> 4452 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4453 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4454 </pre> 4455 4456 <h5>Overview:</h5> 4457 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first 4458 operand shifted to the right a specified number of bits with sign 4459 extension.</p> 4460 4461 <h5>Arguments:</h5> 4462 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same 4463 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4464 type. '<tt>op2</tt>' is treated as an unsigned value.</p> 4465 4466 <h5>Semantics:</h5> 4467 <p>This instruction always performs an arithmetic shift right operation, The 4468 most significant bits of the result will be filled with the sign bit 4469 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or 4470 larger than the number of bits in <tt>op1</tt>, the result is undefined. If 4471 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by 4472 the corresponding shift amount in <tt>op2</tt>.</p> 4473 4474 <p>If the <tt>exact</tt> keyword is present, the result value of the 4475 <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits 4476 shifted out are non-zero.</p> 4477 4478 <h5>Example:</h5> 4479 <pre> 4480 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i> 4481 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i> 4482 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i> 4483 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i> 4484 <result> = ashr i32 1, 32 <i>; undefined</i> 4485 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i> 4486 </pre> 4487 4488 </div> 4489 4490 <!-- _______________________________________________________________________ --> 4491 <h4> 4492 <a name="i_and">'<tt>and</tt>' Instruction</a> 4493 </h4> 4494 4495 <div> 4496 4497 <h5>Syntax:</h5> 4498 <pre> 4499 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4500 </pre> 4501 4502 <h5>Overview:</h5> 4503 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two 4504 operands.</p> 4505 4506 <h5>Arguments:</h5> 4507 <p>The two arguments to the '<tt>and</tt>' instruction must be 4508 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4509 values. Both arguments must have identical types.</p> 4510 4511 <h5>Semantics:</h5> 4512 <p>The truth table used for the '<tt>and</tt>' instruction is:</p> 4513 4514 <table border="1" cellspacing="0" cellpadding="4"> 4515 <tbody> 4516 <tr> 4517 <th>In0</th> 4518 <th>In1</th> 4519 <th>Out</th> 4520 </tr> 4521 <tr> 4522 <td>0</td> 4523 <td>0</td> 4524 <td>0</td> 4525 </tr> 4526 <tr> 4527 <td>0</td> 4528 <td>1</td> 4529 <td>0</td> 4530 </tr> 4531 <tr> 4532 <td>1</td> 4533 <td>0</td> 4534 <td>0</td> 4535 </tr> 4536 <tr> 4537 <td>1</td> 4538 <td>1</td> 4539 <td>1</td> 4540 </tr> 4541 </tbody> 4542 </table> 4543 4544 <h5>Example:</h5> 4545 <pre> 4546 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i> 4547 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i> 4548 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i> 4549 </pre> 4550 </div> 4551 <!-- _______________________________________________________________________ --> 4552 <h4> 4553 <a name="i_or">'<tt>or</tt>' Instruction</a> 4554 </h4> 4555 4556 <div> 4557 4558 <h5>Syntax:</h5> 4559 <pre> 4560 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4561 </pre> 4562 4563 <h5>Overview:</h5> 4564 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its 4565 two operands.</p> 4566 4567 <h5>Arguments:</h5> 4568 <p>The two arguments to the '<tt>or</tt>' instruction must be 4569 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4570 values. Both arguments must have identical types.</p> 4571 4572 <h5>Semantics:</h5> 4573 <p>The truth table used for the '<tt>or</tt>' instruction is:</p> 4574 4575 <table border="1" cellspacing="0" cellpadding="4"> 4576 <tbody> 4577 <tr> 4578 <th>In0</th> 4579 <th>In1</th> 4580 <th>Out</th> 4581 </tr> 4582 <tr> 4583 <td>0</td> 4584 <td>0</td> 4585 <td>0</td> 4586 </tr> 4587 <tr> 4588 <td>0</td> 4589 <td>1</td> 4590 <td>1</td> 4591 </tr> 4592 <tr> 4593 <td>1</td> 4594 <td>0</td> 4595 <td>1</td> 4596 </tr> 4597 <tr> 4598 <td>1</td> 4599 <td>1</td> 4600 <td>1</td> 4601 </tr> 4602 </tbody> 4603 </table> 4604 4605 <h5>Example:</h5> 4606 <pre> 4607 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i> 4608 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i> 4609 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i> 4610 </pre> 4611 4612 </div> 4613 4614 <!-- _______________________________________________________________________ --> 4615 <h4> 4616 <a name="i_xor">'<tt>xor</tt>' Instruction</a> 4617 </h4> 4618 4619 <div> 4620 4621 <h5>Syntax:</h5> 4622 <pre> 4623 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i> 4624 </pre> 4625 4626 <h5>Overview:</h5> 4627 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of 4628 its two operands. The <tt>xor</tt> is used to implement the "one's 4629 complement" operation, which is the "~" operator in C.</p> 4630 4631 <h5>Arguments:</h5> 4632 <p>The two arguments to the '<tt>xor</tt>' instruction must be 4633 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer 4634 values. Both arguments must have identical types.</p> 4635 4636 <h5>Semantics:</h5> 4637 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p> 4638 4639 <table border="1" cellspacing="0" cellpadding="4"> 4640 <tbody> 4641 <tr> 4642 <th>In0</th> 4643 <th>In1</th> 4644 <th>Out</th> 4645 </tr> 4646 <tr> 4647 <td>0</td> 4648 <td>0</td> 4649 <td>0</td> 4650 </tr> 4651 <tr> 4652 <td>0</td> 4653 <td>1</td> 4654 <td>1</td> 4655 </tr> 4656 <tr> 4657 <td>1</td> 4658 <td>0</td> 4659 <td>1</td> 4660 </tr> 4661 <tr> 4662 <td>1</td> 4663 <td>1</td> 4664 <td>0</td> 4665 </tr> 4666 </tbody> 4667 </table> 4668 4669 <h5>Example:</h5> 4670 <pre> 4671 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i> 4672 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i> 4673 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i> 4674 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i> 4675 </pre> 4676 4677 </div> 4678 4679 </div> 4680 4681 <!-- ======================================================================= --> 4682 <h3> 4683 <a name="vectorops">Vector Operations</a> 4684 </h3> 4685 4686 <div> 4687 4688 <p>LLVM supports several instructions to represent vector operations in a 4689 target-independent manner. These instructions cover the element-access and 4690 vector-specific operations needed to process vectors effectively. While LLVM 4691 does directly support these vector operations, many sophisticated algorithms 4692 will want to use target-specific intrinsics to take full advantage of a 4693 specific target.</p> 4694 4695 <!-- _______________________________________________________________________ --> 4696 <h4> 4697 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a> 4698 </h4> 4699 4700 <div> 4701 4702 <h5>Syntax:</h5> 4703 <pre> 4704 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i> 4705 </pre> 4706 4707 <h5>Overview:</h5> 4708 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element 4709 from a vector at a specified index.</p> 4710 4711 4712 <h5>Arguments:</h5> 4713 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value 4714 of <a href="#t_vector">vector</a> type. The second operand is an index 4715 indicating the position from which to extract the element. The index may be 4716 a variable.</p> 4717 4718 <h5>Semantics:</h5> 4719 <p>The result is a scalar of the same type as the element type of 4720 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of 4721 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the 4722 results are undefined.</p> 4723 4724 <h5>Example:</h5> 4725 <pre> 4726 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i> 4727 </pre> 4728 4729 </div> 4730 4731 <!-- _______________________________________________________________________ --> 4732 <h4> 4733 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a> 4734 </h4> 4735 4736 <div> 4737 4738 <h5>Syntax:</h5> 4739 <pre> 4740 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i> 4741 </pre> 4742 4743 <h5>Overview:</h5> 4744 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a 4745 vector at a specified index.</p> 4746 4747 <h5>Arguments:</h5> 4748 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value 4749 of <a href="#t_vector">vector</a> type. The second operand is a scalar value 4750 whose type must equal the element type of the first operand. The third 4751 operand is an index indicating the position at which to insert the value. 4752 The index may be a variable.</p> 4753 4754 <h5>Semantics:</h5> 4755 <p>The result is a vector of the same type as <tt>val</tt>. Its element values 4756 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the 4757 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the 4758 results are undefined.</p> 4759 4760 <h5>Example:</h5> 4761 <pre> 4762 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i> 4763 </pre> 4764 4765 </div> 4766 4767 <!-- _______________________________________________________________________ --> 4768 <h4> 4769 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a> 4770 </h4> 4771 4772 <div> 4773 4774 <h5>Syntax:</h5> 4775 <pre> 4776 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i> 4777 </pre> 4778 4779 <h5>Overview:</h5> 4780 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements 4781 from two input vectors, returning a vector with the same element type as the 4782 input and length that is the same as the shuffle mask.</p> 4783 4784 <h5>Arguments:</h5> 4785 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors 4786 with the same type. The third argument is a shuffle mask whose 4787 element type is always 'i32'. The result of the instruction is a vector 4788 whose length is the same as the shuffle mask and whose element type is the 4789 same as the element type of the first two operands.</p> 4790 4791 <p>The shuffle mask operand is required to be a constant vector with either 4792 constant integer or undef values.</p> 4793 4794 <h5>Semantics:</h5> 4795 <p>The elements of the two input vectors are numbered from left to right across 4796 both of the vectors. The shuffle mask operand specifies, for each element of 4797 the result vector, which element of the two input vectors the result element 4798 gets. The element selector may be undef (meaning "don't care") and the 4799 second operand may be undef if performing a shuffle from only one vector.</p> 4800 4801 <h5>Example:</h5> 4802 <pre> 4803 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2, 4804 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i> 4805 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef, 4806 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle. 4807 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef, 4808 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> 4809 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2, 4810 <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> 4811 </pre> 4812 4813 </div> 4814 4815 </div> 4816 4817 <!-- ======================================================================= --> 4818 <h3> 4819 <a name="aggregateops">Aggregate Operations</a> 4820 </h3> 4821 4822 <div> 4823 4824 <p>LLVM supports several instructions for working with 4825 <a href="#t_aggregate">aggregate</a> values.</p> 4826 4827 <!-- _______________________________________________________________________ --> 4828 <h4> 4829 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a> 4830 </h4> 4831 4832 <div> 4833 4834 <h5>Syntax:</h5> 4835 <pre> 4836 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}* 4837 </pre> 4838 4839 <h5>Overview:</h5> 4840 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field 4841 from an <a href="#t_aggregate">aggregate</a> value.</p> 4842 4843 <h5>Arguments:</h5> 4844 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value 4845 of <a href="#t_struct">struct</a> or 4846 <a href="#t_array">array</a> type. The operands are constant indices to 4847 specify which value to extract in a similar manner as indices in a 4848 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p> 4849 <p>The major differences to <tt>getelementptr</tt> indexing are:</p> 4850 <ul> 4851 <li>Since the value being indexed is not a pointer, the first index is 4852 omitted and assumed to be zero.</li> 4853 <li>At least one index must be specified.</li> 4854 <li>Not only struct indices but also array indices must be in 4855 bounds.</li> 4856 </ul> 4857 4858 <h5>Semantics:</h5> 4859 <p>The result is the value at the position in the aggregate specified by the 4860 index operands.</p> 4861 4862 <h5>Example:</h5> 4863 <pre> 4864 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i> 4865 </pre> 4866 4867 </div> 4868 4869 <!-- _______________________________________________________________________ --> 4870 <h4> 4871 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a> 4872 </h4> 4873 4874 <div> 4875 4876 <h5>Syntax:</h5> 4877 <pre> 4878 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i> 4879 </pre> 4880 4881 <h5>Overview:</h5> 4882 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field 4883 in an <a href="#t_aggregate">aggregate</a> value.</p> 4884 4885 <h5>Arguments:</h5> 4886 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value 4887 of <a href="#t_struct">struct</a> or 4888 <a href="#t_array">array</a> type. The second operand is a first-class 4889 value to insert. The following operands are constant indices indicating 4890 the position at which to insert the value in a similar manner as indices in a 4891 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The 4892 value to insert must have the same type as the value identified by the 4893 indices.</p> 4894 4895 <h5>Semantics:</h5> 4896 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is 4897 that of <tt>val</tt> except that the value at the position specified by the 4898 indices is that of <tt>elt</tt>.</p> 4899 4900 <h5>Example:</h5> 4901 <pre> 4902 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i> 4903 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i> 4904 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i> 4905 </pre> 4906 4907 </div> 4908 4909 </div> 4910 4911 <!-- ======================================================================= --> 4912 <h3> 4913 <a name="memoryops">Memory Access and Addressing Operations</a> 4914 </h3> 4915 4916 <div> 4917 4918 <p>A key design point of an SSA-based representation is how it represents 4919 memory. In LLVM, no memory locations are in SSA form, which makes things 4920 very simple. This section describes how to read, write, and allocate 4921 memory in LLVM.</p> 4922 4923 <!-- _______________________________________________________________________ --> 4924 <h4> 4925 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a> 4926 </h4> 4927 4928 <div> 4929 4930 <h5>Syntax:</h5> 4931 <pre> 4932 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i> 4933 </pre> 4934 4935 <h5>Overview:</h5> 4936 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the 4937 currently executing function, to be automatically released when this function 4938 returns to its caller. The object is always allocated in the generic address 4939 space (address space zero).</p> 4940 4941 <h5>Arguments:</h5> 4942 <p>The '<tt>alloca</tt>' instruction 4943 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the 4944 runtime stack, returning a pointer of the appropriate type to the program. 4945 If "NumElements" is specified, it is the number of elements allocated, 4946 otherwise "NumElements" is defaulted to be one. If a constant alignment is 4947 specified, the value result of the allocation is guaranteed to be aligned to 4948 at least that boundary. If not specified, or if zero, the target can choose 4949 to align the allocation on any convenient boundary compatible with the 4950 type.</p> 4951 4952 <p>'<tt>type</tt>' may be any sized type.</p> 4953 4954 <h5>Semantics:</h5> 4955 <p>Memory is allocated; a pointer is returned. The operation is undefined if 4956 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d 4957 memory is automatically released when the function returns. The 4958 '<tt>alloca</tt>' instruction is commonly used to represent automatic 4959 variables that must have an address available. When the function returns 4960 (either with the <tt><a href="#i_ret">ret</a></tt> 4961 or <tt><a href="#i_resume">resume</a></tt> instructions), the memory is 4962 reclaimed. Allocating zero bytes is legal, but the result is undefined. 4963 The order in which memory is allocated (ie., which way the stack grows) is 4964 not specified.</p> 4965 4966 <p> 4967 4968 <h5>Example:</h5> 4969 <pre> 4970 %ptr = alloca i32 <i>; yields {i32*}:ptr</i> 4971 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i> 4972 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i> 4973 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i> 4974 </pre> 4975 4976 </div> 4977 4978 <!-- _______________________________________________________________________ --> 4979 <h4> 4980 <a name="i_load">'<tt>load</tt>' Instruction</a> 4981 </h4> 4982 4983 <div> 4984 4985 <h5>Syntax:</h5> 4986 <pre> 4987 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>] 4988 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment> 4989 !<index> = !{ i32 1 } 4990 </pre> 4991 4992 <h5>Overview:</h5> 4993 <p>The '<tt>load</tt>' instruction is used to read from memory.</p> 4994 4995 <h5>Arguments:</h5> 4996 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address 4997 from which to load. The pointer must point to 4998 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is 4999 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the 5000 number or order of execution of this <tt>load</tt> with other <a 5001 href="#volatile">volatile operations</a>.</p> 5002 5003 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra 5004 <a href="#ordering">ordering</a> and optional <code>singlethread</code> 5005 argument. The <code>release</code> and <code>acq_rel</code> orderings are 5006 not valid on <code>load</code> instructions. Atomic loads produce <a 5007 href="#memorymodel">defined</a> results when they may see multiple atomic 5008 stores. The type of the pointee must be an integer type whose bit width 5009 is a power of two greater than or equal to eight and less than or equal 5010 to a target-specific size limit. <code>align</code> must be explicitly 5011 specified on atomic loads, and the load has undefined behavior if the 5012 alignment is not set to a value which is at least the size in bytes of 5013 the pointee. <code>!nontemporal</code> does not have any defined semantics 5014 for atomic loads.</p> 5015 5016 <p>The optional constant <tt>align</tt> argument specifies the alignment of the 5017 operation (that is, the alignment of the memory address). A value of 0 or an 5018 omitted <tt>align</tt> argument means that the operation has the preferential 5019 alignment for the target. It is the responsibility of the code emitter to 5020 ensure that the alignment information is correct. Overestimating the 5021 alignment results in undefined behavior. Underestimating the alignment may 5022 produce less efficient code. An alignment of 1 is always safe.</p> 5023 5024 <p>The optional <tt>!nontemporal</tt> metadata must reference a single 5025 metatadata name <index> corresponding to a metadata node with 5026 one <tt>i32</tt> entry of value 1. The existence of 5027 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer 5028 and code generator that this load is not expected to be reused in the cache. 5029 The code generator may select special instructions to save cache bandwidth, 5030 such as the <tt>MOVNT</tt> instruction on x86.</p> 5031 5032 <p>The optional <tt>!invariant.load</tt> metadata must reference a single 5033 metatadata name <index> corresponding to a metadata node with no 5034 entries. The existence of the <tt>!invariant.load</tt> metatadata on the 5035 instruction tells the optimizer and code generator that this load address 5036 points to memory which does not change value during program execution. 5037 The optimizer may then move this load around, for example, by hoisting it 5038 out of loops using loop invariant code motion.</p> 5039 5040 <h5>Semantics:</h5> 5041 <p>The location of memory pointed to is loaded. If the value being loaded is of 5042 scalar type then the number of bytes read does not exceed the minimum number 5043 of bytes needed to hold all bits of the type. For example, loading an 5044 <tt>i24</tt> reads at most three bytes. When loading a value of a type like 5045 <tt>i20</tt> with a size that is not an integral number of bytes, the result 5046 is undefined if the value was not originally written using a store of the 5047 same type.</p> 5048 5049 <h5>Examples:</h5> 5050 <pre> 5051 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i> 5052 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i> 5053 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i> 5054 </pre> 5055 5056 </div> 5057 5058 <!-- _______________________________________________________________________ --> 5059 <h4> 5060 <a name="i_store">'<tt>store</tt>' Instruction</a> 5061 </h4> 5062 5063 <div> 5064 5065 <h5>Syntax:</h5> 5066 <pre> 5067 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i> 5068 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i> 5069 </pre> 5070 5071 <h5>Overview:</h5> 5072 <p>The '<tt>store</tt>' instruction is used to write to memory.</p> 5073 5074 <h5>Arguments:</h5> 5075 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store 5076 and an address at which to store it. The type of the 5077 '<tt><pointer></tt>' operand must be a pointer to 5078 the <a href="#t_firstclass">first class</a> type of the 5079 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as 5080 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or 5081 order of execution of this <tt>store</tt> with other <a 5082 href="#volatile">volatile operations</a>.</p> 5083 5084 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra 5085 <a href="#ordering">ordering</a> and optional <code>singlethread</code> 5086 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't 5087 valid on <code>store</code> instructions. Atomic loads produce <a 5088 href="#memorymodel">defined</a> results when they may see multiple atomic 5089 stores. The type of the pointee must be an integer type whose bit width 5090 is a power of two greater than or equal to eight and less than or equal 5091 to a target-specific size limit. <code>align</code> must be explicitly 5092 specified on atomic stores, and the store has undefined behavior if the 5093 alignment is not set to a value which is at least the size in bytes of 5094 the pointee. <code>!nontemporal</code> does not have any defined semantics 5095 for atomic stores.</p> 5096 5097 <p>The optional constant "align" argument specifies the alignment of the 5098 operation (that is, the alignment of the memory address). A value of 0 or an 5099 omitted "align" argument means that the operation has the preferential 5100 alignment for the target. It is the responsibility of the code emitter to 5101 ensure that the alignment information is correct. Overestimating the 5102 alignment results in an undefined behavior. Underestimating the alignment may 5103 produce less efficient code. An alignment of 1 is always safe.</p> 5104 5105 <p>The optional !nontemporal metadata must reference a single metatadata 5106 name <index> corresponding to a metadata node with one i32 entry of 5107 value 1. The existence of the !nontemporal metatadata on the 5108 instruction tells the optimizer and code generator that this load is 5109 not expected to be reused in the cache. The code generator may 5110 select special instructions to save cache bandwidth, such as the 5111 MOVNT instruction on x86.</p> 5112 5113 5114 <h5>Semantics:</h5> 5115 <p>The contents of memory are updated to contain '<tt><value></tt>' at the 5116 location specified by the '<tt><pointer></tt>' operand. If 5117 '<tt><value></tt>' is of scalar type then the number of bytes written 5118 does not exceed the minimum number of bytes needed to hold all bits of the 5119 type. For example, storing an <tt>i24</tt> writes at most three bytes. When 5120 writing a value of a type like <tt>i20</tt> with a size that is not an 5121 integral number of bytes, it is unspecified what happens to the extra bits 5122 that do not belong to the type, but they will typically be overwritten.</p> 5123 5124 <h5>Example:</h5> 5125 <pre> 5126 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i> 5127 store i32 3, i32* %ptr <i>; yields {void}</i> 5128 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i> 5129 </pre> 5130 5131 </div> 5132 5133 <!-- _______________________________________________________________________ --> 5134 <h4> 5135 <a name="i_fence">'<tt>fence</tt>' Instruction</a> 5136 </h4> 5137 5138 <div> 5139 5140 <h5>Syntax:</h5> 5141 <pre> 5142 fence [singlethread] <ordering> <i>; yields {void}</i> 5143 </pre> 5144 5145 <h5>Overview:</h5> 5146 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges 5147 between operations.</p> 5148 5149 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a 5150 href="#ordering">ordering</a> argument which defines what 5151 <i>synchronizes-with</i> edges they add. They can only be given 5152 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and 5153 <code>seq_cst</code> orderings.</p> 5154 5155 <h5>Semantics:</h5> 5156 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering 5157 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least) 5158 <code>acquire</code> ordering semantics if and only if there exist atomic 5159 operations <var>X</var> and <var>Y</var>, both operating on some atomic object 5160 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>, 5161 <var>X</var> modifies <var>M</var> (either directly or through some side effect 5162 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before 5163 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a 5164 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather 5165 than an explicit <code>fence</code>, one (but not both) of the atomic operations 5166 <var>X</var> or <var>Y</var> might provide a <code>release</code> or 5167 <code>acquire</code> (resp.) ordering constraint and still 5168 <i>synchronize-with</i> the explicit <code>fence</code> and establish the 5169 <i>happens-before</i> edge.</p> 5170 5171 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to 5172 having both <code>acquire</code> and <code>release</code> semantics specified 5173 above, participates in the global program order of other <code>seq_cst</code> 5174 operations and/or fences.</p> 5175 5176 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument 5177 specifies that the fence only synchronizes with other fences in the same 5178 thread. (This is useful for interacting with signal handlers.)</p> 5179 5180 <h5>Example:</h5> 5181 <pre> 5182 fence acquire <i>; yields {void}</i> 5183 fence singlethread seq_cst <i>; yields {void}</i> 5184 </pre> 5185 5186 </div> 5187 5188 <!-- _______________________________________________________________________ --> 5189 <h4> 5190 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a> 5191 </h4> 5192 5193 <div> 5194 5195 <h5>Syntax:</h5> 5196 <pre> 5197 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i> 5198 </pre> 5199 5200 <h5>Overview:</h5> 5201 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory. 5202 It loads a value in memory and compares it to a given value. If they are 5203 equal, it stores a new value into the memory.</p> 5204 5205 <h5>Arguments:</h5> 5206 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an 5207 address to operate on, a value to compare to the value currently be at that 5208 address, and a new value to place at that address if the compared values are 5209 equal. The type of '<var><cmp></var>' must be an integer type whose 5210 bit width is a power of two greater than or equal to eight and less than 5211 or equal to a target-specific size limit. '<var><cmp></var>' and 5212 '<var><new></var>' must have the same type, and the type of 5213 '<var><pointer></var>' must be a pointer to that type. If the 5214 <code>cmpxchg</code> is marked as <code>volatile</code>, then the 5215 optimizer is not allowed to modify the number or order of execution 5216 of this <code>cmpxchg</code> with other <a href="#volatile">volatile 5217 operations</a>.</p> 5218 5219 <!-- FIXME: Extend allowed types. --> 5220 5221 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this 5222 <code>cmpxchg</code> synchronizes with other atomic operations.</p> 5223 5224 <p>The optional "<code>singlethread</code>" argument declares that the 5225 <code>cmpxchg</code> is only atomic with respect to code (usually signal 5226 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the 5227 cmpxchg is atomic with respect to all other code in the system.</p> 5228 5229 <p>The pointer passed into cmpxchg must have alignment greater than or equal to 5230 the size in memory of the operand. 5231 5232 <h5>Semantics:</h5> 5233 <p>The contents of memory at the location specified by the 5234 '<tt><pointer></tt>' operand is read and compared to 5235 '<tt><cmp></tt>'; if the read value is the equal, 5236 '<tt><new></tt>' is written. The original value at the location 5237 is returned. 5238 5239 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the 5240 purpose of identifying <a href="#release_sequence">release sequences</a>. A 5241 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering 5242 parameter determined by dropping any <code>release</code> part of the 5243 <code>cmpxchg</code>'s ordering.</p> 5244 5245 <!-- 5246 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done 5247 optimization work on ARM.) 5248 5249 FIXME: Is a weaker ordering constraint on failure helpful in practice? 5250 --> 5251 5252 <h5>Example:</h5> 5253 <pre> 5254 entry: 5255 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i> 5256 <a href="#i_br">br</a> label %loop 5257 5258 loop: 5259 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop] 5260 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp 5261 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i> 5262 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old 5263 <a href="#i_br">br</a> i1 %success, label %done, label %loop 5264 5265 done: 5266 ... 5267 </pre> 5268 5269 </div> 5270 5271 <!-- _______________________________________________________________________ --> 5272 <h4> 5273 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a> 5274 </h4> 5275 5276 <div> 5277 5278 <h5>Syntax:</h5> 5279 <pre> 5280 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i> 5281 </pre> 5282 5283 <h5>Overview:</h5> 5284 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p> 5285 5286 <h5>Arguments:</h5> 5287 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an 5288 operation to apply, an address whose value to modify, an argument to the 5289 operation. The operation must be one of the following keywords:</p> 5290 <ul> 5291 <li>xchg</li> 5292 <li>add</li> 5293 <li>sub</li> 5294 <li>and</li> 5295 <li>nand</li> 5296 <li>or</li> 5297 <li>xor</li> 5298 <li>max</li> 5299 <li>min</li> 5300 <li>umax</li> 5301 <li>umin</li> 5302 </ul> 5303 5304 <p>The type of '<var><value></var>' must be an integer type whose 5305 bit width is a power of two greater than or equal to eight and less than 5306 or equal to a target-specific size limit. The type of the 5307 '<code><pointer></code>' operand must be a pointer to that type. 5308 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the 5309 optimizer is not allowed to modify the number or order of execution of this 5310 <code>atomicrmw</code> with other <a href="#volatile">volatile 5311 operations</a>.</p> 5312 5313 <!-- FIXME: Extend allowed types. --> 5314 5315 <h5>Semantics:</h5> 5316 <p>The contents of memory at the location specified by the 5317 '<tt><pointer></tt>' operand are atomically read, modified, and written 5318 back. The original value at the location is returned. The modification is 5319 specified by the <var>operation</var> argument:</p> 5320 5321 <ul> 5322 <li>xchg: <code>*ptr = val</code></li> 5323 <li>add: <code>*ptr = *ptr + val</code></li> 5324 <li>sub: <code>*ptr = *ptr - val</code></li> 5325 <li>and: <code>*ptr = *ptr & val</code></li> 5326 <li>nand: <code>*ptr = ~(*ptr & val)</code></li> 5327 <li>or: <code>*ptr = *ptr | val</code></li> 5328 <li>xor: <code>*ptr = *ptr ^ val</code></li> 5329 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li> 5330 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li> 5331 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li> 5332 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li> 5333 </ul> 5334 5335 <h5>Example:</h5> 5336 <pre> 5337 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i> 5338 </pre> 5339 5340 </div> 5341 5342 <!-- _______________________________________________________________________ --> 5343 <h4> 5344 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a> 5345 </h4> 5346 5347 <div> 5348 5349 <h5>Syntax:</h5> 5350 <pre> 5351 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}* 5352 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}* 5353 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx 5354 </pre> 5355 5356 <h5>Overview:</h5> 5357 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a 5358 subelement of an <a href="#t_aggregate">aggregate</a> data structure. 5359 It performs address calculation only and does not access memory.</p> 5360 5361 <h5>Arguments:</h5> 5362 <p>The first argument is always a pointer or a vector of pointers, 5363 and forms the basis of the 5364 calculation. The remaining arguments are indices that indicate which of the 5365 elements of the aggregate object are indexed. The interpretation of each 5366 index is dependent on the type being indexed into. The first index always 5367 indexes the pointer value given as the first argument, the second index 5368 indexes a value of the type pointed to (not necessarily the value directly 5369 pointed to, since the first index can be non-zero), etc. The first type 5370 indexed into must be a pointer value, subsequent types can be arrays, 5371 vectors, and structs. Note that subsequent types being indexed into 5372 can never be pointers, since that would require loading the pointer before 5373 continuing calculation.</p> 5374 5375 <p>The type of each index argument depends on the type it is indexing into. 5376 When indexing into a (optionally packed) structure, only <tt>i32</tt> 5377 integer <b>constants</b> are allowed. When indexing into an array, pointer 5378 or vector, integers of any width are allowed, and they are not required to be 5379 constant. These integers are treated as signed values where relevant.</p> 5380 5381 <p>For example, let's consider a C code fragment and how it gets compiled to 5382 LLVM:</p> 5383 5384 <pre class="doc_code"> 5385 struct RT { 5386 char A; 5387 int B[10][20]; 5388 char C; 5389 }; 5390 struct ST { 5391 int X; 5392 double Y; 5393 struct RT Z; 5394 }; 5395 5396 int *foo(struct ST *s) { 5397 return &s[1].Z.B[5][13]; 5398 } 5399 </pre> 5400 5401 <p>The LLVM code generated by Clang is:</p> 5402 5403 <pre class="doc_code"> 5404 %struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 } 5405 %struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT } 5406 5407 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp { 5408 entry: 5409 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13 5410 ret i32* %arrayidx 5411 } 5412 </pre> 5413 5414 <h5>Semantics:</h5> 5415 <p>In the example above, the first index is indexing into the 5416 '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a 5417 '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a 5418 structure. The second index indexes into the third element of the structure, 5419 yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>' 5420 type, another structure. The third index indexes into the second element of 5421 the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The 5422 two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>' 5423 type. The '<tt>getelementptr</tt>' instruction returns a pointer to this 5424 element, thus computing a value of '<tt>i32*</tt>' type.</p> 5425 5426 <p>Note that it is perfectly legal to index partially through a structure, 5427 returning a pointer to an inner element. Because of this, the LLVM code for 5428 the given testcase is equivalent to:</p> 5429 5430 <pre class="doc_code"> 5431 define i32* @foo(%struct.ST* %s) { 5432 %t1 = getelementptr %struct.ST* %s, i32 1 <i>; yields %struct.ST*:%t1</i> 5433 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 <i>; yields %struct.RT*:%t2</i> 5434 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i> 5435 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i> 5436 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i> 5437 ret i32* %t5 5438 } 5439 </pre> 5440 5441 <p>If the <tt>inbounds</tt> keyword is present, the result value of the 5442 <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the 5443 base pointer is not an <i>in bounds</i> address of an allocated object, 5444 or if any of the addresses that would be formed by successive addition of 5445 the offsets implied by the indices to the base address with infinitely 5446 precise signed arithmetic are not an <i>in bounds</i> address of that 5447 allocated object. The <i>in bounds</i> addresses for an allocated object 5448 are all the addresses that point into the object, plus the address one 5449 byte past the end. 5450 In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword 5451 applies to each of the computations element-wise. </p> 5452 5453 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to 5454 the base address with silently-wrapping two's complement arithmetic. If the 5455 offsets have a different width from the pointer, they are sign-extended or 5456 truncated to the width of the pointer. The result value of the 5457 <tt>getelementptr</tt> may be outside the object pointed to by the base 5458 pointer. The result value may not necessarily be used to access memory 5459 though, even if it happens to point into allocated storage. See the 5460 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more 5461 information.</p> 5462 5463 <p>The getelementptr instruction is often confusing. For some more insight into 5464 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p> 5465 5466 <h5>Example:</h5> 5467 <pre> 5468 <i>; yields [12 x i8]*:aptr</i> 5469 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1 5470 <i>; yields i8*:vptr</i> 5471 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1 5472 <i>; yields i8*:eptr</i> 5473 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1 5474 <i>; yields i32*:iptr</i> 5475 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0 5476 </pre> 5477 5478 <p>In cases where the pointer argument is a vector of pointers, only a 5479 single index may be used, and the number of vector elements has to be 5480 the same. For example: </p> 5481 <pre class="doc_code"> 5482 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets, 5483 </pre> 5484 5485 </div> 5486 5487 </div> 5488 5489 <!-- ======================================================================= --> 5490 <h3> 5491 <a name="convertops">Conversion Operations</a> 5492 </h3> 5493 5494 <div> 5495 5496 <p>The instructions in this category are the conversion instructions (casting) 5497 which all take a single operand and a type. They perform various bit 5498 conversions on the operand.</p> 5499 5500 <!-- _______________________________________________________________________ --> 5501 <h4> 5502 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a> 5503 </h4> 5504 5505 <div> 5506 5507 <h5>Syntax:</h5> 5508 <pre> 5509 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i> 5510 </pre> 5511 5512 <h5>Overview:</h5> 5513 <p>The '<tt>trunc</tt>' instruction truncates its operand to the 5514 type <tt>ty2</tt>.</p> 5515 5516 <h5>Arguments:</h5> 5517 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to. 5518 Both types must be of <a href="#t_integer">integer</a> types, or vectors 5519 of the same number of integers. 5520 The bit size of the <tt>value</tt> must be larger than 5521 the bit size of the destination type, <tt>ty2</tt>. 5522 Equal sized types are not allowed.</p> 5523 5524 <h5>Semantics:</h5> 5525 <p>The '<tt>trunc</tt>' instruction truncates the high order bits 5526 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the 5527 source size must be larger than the destination size, <tt>trunc</tt> cannot 5528 be a <i>no-op cast</i>. It will always truncate bits.</p> 5529 5530 <h5>Example:</h5> 5531 <pre> 5532 %X = trunc i32 257 to i8 <i>; yields i8:1</i> 5533 %Y = trunc i32 123 to i1 <i>; yields i1:true</i> 5534 %Z = trunc i32 122 to i1 <i>; yields i1:false</i> 5535 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i> 5536 </pre> 5537 5538 </div> 5539 5540 <!-- _______________________________________________________________________ --> 5541 <h4> 5542 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a> 5543 </h4> 5544 5545 <div> 5546 5547 <h5>Syntax:</h5> 5548 <pre> 5549 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i> 5550 </pre> 5551 5552 <h5>Overview:</h5> 5553 <p>The '<tt>zext</tt>' instruction zero extends its operand to type 5554 <tt>ty2</tt>.</p> 5555 5556 5557 <h5>Arguments:</h5> 5558 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to. 5559 Both types must be of <a href="#t_integer">integer</a> types, or vectors 5560 of the same number of integers. 5561 The bit size of the <tt>value</tt> must be smaller than 5562 the bit size of the destination type, 5563 <tt>ty2</tt>.</p> 5564 5565 <h5>Semantics:</h5> 5566 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero 5567 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p> 5568 5569 <p>When zero extending from i1, the result will always be either 0 or 1.</p> 5570 5571 <h5>Example:</h5> 5572 <pre> 5573 %X = zext i32 257 to i64 <i>; yields i64:257</i> 5574 %Y = zext i1 true to i32 <i>; yields i32:1</i> 5575 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i> 5576 </pre> 5577 5578 </div> 5579 5580 <!-- _______________________________________________________________________ --> 5581 <h4> 5582 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a> 5583 </h4> 5584 5585 <div> 5586 5587 <h5>Syntax:</h5> 5588 <pre> 5589 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i> 5590 </pre> 5591 5592 <h5>Overview:</h5> 5593 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p> 5594 5595 <h5>Arguments:</h5> 5596 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to. 5597 Both types must be of <a href="#t_integer">integer</a> types, or vectors 5598 of the same number of integers. 5599 The bit size of the <tt>value</tt> must be smaller than 5600 the bit size of the destination type, 5601 <tt>ty2</tt>.</p> 5602 5603 <h5>Semantics:</h5> 5604 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign 5605 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size 5606 of the type <tt>ty2</tt>.</p> 5607 5608 <p>When sign extending from i1, the extension always results in -1 or 0.</p> 5609 5610 <h5>Example:</h5> 5611 <pre> 5612 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i> 5613 %Y = sext i1 true to i32 <i>; yields i32:-1</i> 5614 %Z = sext <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_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a> 5622 </h4> 5623 5624 <div> 5625 5626 <h5>Syntax:</h5> 5627 <pre> 5628 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i> 5629 </pre> 5630 5631 <h5>Overview:</h5> 5632 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type 5633 <tt>ty2</tt>.</p> 5634 5635 <h5>Arguments:</h5> 5636 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating 5637 point</a> value to cast and a <a href="#t_floating">floating point</a> type 5638 to cast it to. The size of <tt>value</tt> must be larger than the size of 5639 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a 5640 <i>no-op cast</i>.</p> 5641 5642 <h5>Semantics:</h5> 5643 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger 5644 <a href="#t_floating">floating point</a> type to a smaller 5645 <a href="#t_floating">floating point</a> type. If the value cannot fit 5646 within the destination type, <tt>ty2</tt>, then the results are 5647 undefined.</p> 5648 5649 <h5>Example:</h5> 5650 <pre> 5651 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i> 5652 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i> 5653 </pre> 5654 5655 </div> 5656 5657 <!-- _______________________________________________________________________ --> 5658 <h4> 5659 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a> 5660 </h4> 5661 5662 <div> 5663 5664 <h5>Syntax:</h5> 5665 <pre> 5666 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i> 5667 </pre> 5668 5669 <h5>Overview:</h5> 5670 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger 5671 floating point value.</p> 5672 5673 <h5>Arguments:</h5> 5674 <p>The '<tt>fpext</tt>' instruction takes a 5675 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and 5676 a <a href="#t_floating">floating point</a> type to cast it to. The source 5677 type must be smaller than the destination type.</p> 5678 5679 <h5>Semantics:</h5> 5680 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller 5681 <a href="#t_floating">floating point</a> type to a larger 5682 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be 5683 used to make a <i>no-op cast</i> because it always changes bits. Use 5684 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p> 5685 5686 <h5>Example:</h5> 5687 <pre> 5688 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i> 5689 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i> 5690 </pre> 5691 5692 </div> 5693 5694 <!-- _______________________________________________________________________ --> 5695 <h4> 5696 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a> 5697 </h4> 5698 5699 <div> 5700 5701 <h5>Syntax:</h5> 5702 <pre> 5703 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i> 5704 </pre> 5705 5706 <h5>Overview:</h5> 5707 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its 5708 unsigned integer equivalent of type <tt>ty2</tt>.</p> 5709 5710 <h5>Arguments:</h5> 5711 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a 5712 scalar or vector <a href="#t_floating">floating point</a> value, and a type 5713 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> 5714 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a 5715 vector integer type with the same number of elements as <tt>ty</tt></p> 5716 5717 <h5>Semantics:</h5> 5718 <p>The '<tt>fptoui</tt>' instruction converts its 5719 <a href="#t_floating">floating point</a> operand into the nearest (rounding 5720 towards zero) unsigned integer value. If the value cannot fit 5721 in <tt>ty2</tt>, the results are undefined.</p> 5722 5723 <h5>Example:</h5> 5724 <pre> 5725 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i> 5726 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i> 5727 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i> 5728 </pre> 5729 5730 </div> 5731 5732 <!-- _______________________________________________________________________ --> 5733 <h4> 5734 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a> 5735 </h4> 5736 5737 <div> 5738 5739 <h5>Syntax:</h5> 5740 <pre> 5741 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i> 5742 </pre> 5743 5744 <h5>Overview:</h5> 5745 <p>The '<tt>fptosi</tt>' instruction converts 5746 <a href="#t_floating">floating point</a> <tt>value</tt> to 5747 type <tt>ty2</tt>.</p> 5748 5749 <h5>Arguments:</h5> 5750 <p>The '<tt>fptosi</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>fptosi</tt>' instruction converts its 5758 <a href="#t_floating">floating point</a> operand into the nearest (rounding 5759 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>, 5760 the results are undefined.</p> 5761 5762 <h5>Example:</h5> 5763 <pre> 5764 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i> 5765 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i> 5766 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i> 5767 </pre> 5768 5769 </div> 5770 5771 <!-- _______________________________________________________________________ --> 5772 <h4> 5773 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a> 5774 </h4> 5775 5776 <div> 5777 5778 <h5>Syntax:</h5> 5779 <pre> 5780 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i> 5781 </pre> 5782 5783 <h5>Overview:</h5> 5784 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned 5785 integer and converts that value to the <tt>ty2</tt> type.</p> 5786 5787 <h5>Arguments:</h5> 5788 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a 5789 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast 5790 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a> 5791 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector 5792 floating point type with the same number of elements as <tt>ty</tt></p> 5793 5794 <h5>Semantics:</h5> 5795 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned 5796 integer quantity and converts it to the corresponding floating point 5797 value. If the value cannot fit in the floating point value, the results are 5798 undefined.</p> 5799 5800 <h5>Example:</h5> 5801 <pre> 5802 %X = uitofp i32 257 to float <i>; yields float:257.0</i> 5803 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i> 5804 </pre> 5805 5806 </div> 5807 5808 <!-- _______________________________________________________________________ --> 5809 <h4> 5810 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a> 5811 </h4> 5812 5813 <div> 5814 5815 <h5>Syntax:</h5> 5816 <pre> 5817 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i> 5818 </pre> 5819 5820 <h5>Overview:</h5> 5821 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer 5822 and converts that value to the <tt>ty2</tt> type.</p> 5823 5824 <h5>Arguments:</h5> 5825 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a 5826 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast 5827 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a> 5828 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector 5829 floating point type with the same number of elements as <tt>ty</tt></p> 5830 5831 <h5>Semantics:</h5> 5832 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer 5833 quantity and converts it to the corresponding floating point value. If the 5834 value cannot fit in the floating point value, the results are undefined.</p> 5835 5836 <h5>Example:</h5> 5837 <pre> 5838 %X = sitofp i32 257 to float <i>; yields float:257.0</i> 5839 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i> 5840 </pre> 5841 5842 </div> 5843 5844 <!-- _______________________________________________________________________ --> 5845 <h4> 5846 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a> 5847 </h4> 5848 5849 <div> 5850 5851 <h5>Syntax:</h5> 5852 <pre> 5853 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i> 5854 </pre> 5855 5856 <h5>Overview:</h5> 5857 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of 5858 pointers <tt>value</tt> to 5859 the integer (or vector of integers) type <tt>ty2</tt>.</p> 5860 5861 <h5>Arguments:</h5> 5862 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which 5863 must be a a value of type <a href="#t_pointer">pointer</a> or a vector of 5864 pointers, and a type to cast it to 5865 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector 5866 of integers type.</p> 5867 5868 <h5>Semantics:</h5> 5869 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type 5870 <tt>ty2</tt> by interpreting the pointer value as an integer and either 5871 truncating or zero extending that value to the size of the integer type. If 5872 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If 5873 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they 5874 are the same size, then nothing is done (<i>no-op cast</i>) other than a type 5875 change.</p> 5876 5877 <h5>Example:</h5> 5878 <pre> 5879 %X = ptrtoint i32* %P to i8 <i>; yields truncation on 32-bit architecture</i> 5880 %Y = ptrtoint i32* %P to i64 <i>; yields zero extension on 32-bit architecture</i> 5881 %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> 5882 </pre> 5883 5884 </div> 5885 5886 <!-- _______________________________________________________________________ --> 5887 <h4> 5888 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a> 5889 </h4> 5890 5891 <div> 5892 5893 <h5>Syntax:</h5> 5894 <pre> 5895 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i> 5896 </pre> 5897 5898 <h5>Overview:</h5> 5899 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a 5900 pointer type, <tt>ty2</tt>.</p> 5901 5902 <h5>Arguments:</h5> 5903 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a> 5904 value to cast, and a type to cast it to, which must be a 5905 <a href="#t_pointer">pointer</a> type.</p> 5906 5907 <h5>Semantics:</h5> 5908 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type 5909 <tt>ty2</tt> by applying either a zero extension or a truncation depending on 5910 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the 5911 size of a pointer then a truncation is done. If <tt>value</tt> is smaller 5912 than the size of a pointer then a zero extension is done. If they are the 5913 same size, nothing is done (<i>no-op cast</i>).</p> 5914 5915 <h5>Example:</h5> 5916 <pre> 5917 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i> 5918 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i> 5919 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i> 5920 %Z = inttoptr <4 x i32> %G to <4 x i8*><i>; yields truncation of vector G to four pointers</i> 5921 </pre> 5922 5923 </div> 5924 5925 <!-- _______________________________________________________________________ --> 5926 <h4> 5927 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a> 5928 </h4> 5929 5930 <div> 5931 5932 <h5>Syntax:</h5> 5933 <pre> 5934 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i> 5935 </pre> 5936 5937 <h5>Overview:</h5> 5938 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type 5939 <tt>ty2</tt> without changing any bits.</p> 5940 5941 <h5>Arguments:</h5> 5942 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a 5943 non-aggregate first class value, and a type to cast it to, which must also be 5944 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes 5945 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be 5946 identical. If the source type is a pointer, the destination type must also be 5947 a pointer. This instruction supports bitwise conversion of vectors to 5948 integers and to vectors of other types (as long as they have the same 5949 size).</p> 5950 5951 <h5>Semantics:</h5> 5952 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type 5953 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with 5954 this conversion. The conversion is done as if the <tt>value</tt> had been 5955 stored to memory and read back as type <tt>ty2</tt>. 5956 Pointer (or vector of pointers) types may only be converted to other pointer 5957 (or vector of pointers) types with this instruction. To convert 5958 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or 5959 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p> 5960 5961 <h5>Example:</h5> 5962 <pre> 5963 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i> 5964 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i> 5965 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i> 5966 %Z = bitcast <2 x i32*> %V to <2 x i64*> <i>; yields <2 x i64*></i> 5967 </pre> 5968 5969 </div> 5970 5971 </div> 5972 5973 <!-- ======================================================================= --> 5974 <h3> 5975 <a name="otherops">Other Operations</a> 5976 </h3> 5977 5978 <div> 5979 5980 <p>The instructions in this category are the "miscellaneous" instructions, which 5981 defy better classification.</p> 5982 5983 <!-- _______________________________________________________________________ --> 5984 <h4> 5985 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a> 5986 </h4> 5987 5988 <div> 5989 5990 <h5>Syntax:</h5> 5991 <pre> 5992 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i> 5993 </pre> 5994 5995 <h5>Overview:</h5> 5996 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of 5997 boolean values based on comparison of its two integer, integer vector, 5998 pointer, or pointer vector operands.</p> 5999 6000 <h5>Arguments:</h5> 6001 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is 6002 the condition code indicating the kind of comparison to perform. It is not a 6003 value, just a keyword. The possible condition code are:</p> 6004 6005 <ol> 6006 <li><tt>eq</tt>: equal</li> 6007 <li><tt>ne</tt>: not equal </li> 6008 <li><tt>ugt</tt>: unsigned greater than</li> 6009 <li><tt>uge</tt>: unsigned greater or equal</li> 6010 <li><tt>ult</tt>: unsigned less than</li> 6011 <li><tt>ule</tt>: unsigned less or equal</li> 6012 <li><tt>sgt</tt>: signed greater than</li> 6013 <li><tt>sge</tt>: signed greater or equal</li> 6014 <li><tt>slt</tt>: signed less than</li> 6015 <li><tt>sle</tt>: signed less or equal</li> 6016 </ol> 6017 6018 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or 6019 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a> 6020 typed. They must also be identical types.</p> 6021 6022 <h5>Semantics:</h5> 6023 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the 6024 condition code given as <tt>cond</tt>. The comparison performed always yields 6025 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt> 6026 result, as follows:</p> 6027 6028 <ol> 6029 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal, 6030 <tt>false</tt> otherwise. No sign interpretation is necessary or 6031 performed.</li> 6032 6033 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal, 6034 <tt>false</tt> otherwise. No sign interpretation is necessary or 6035 performed.</li> 6036 6037 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields 6038 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li> 6039 6040 <li><tt>uge</tt>: interprets the operands as unsigned values and yields 6041 <tt>true</tt> if <tt>op1</tt> is greater than or equal 6042 to <tt>op2</tt>.</li> 6043 6044 <li><tt>ult</tt>: interprets the operands as unsigned values and yields 6045 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li> 6046 6047 <li><tt>ule</tt>: interprets the operands as unsigned values and yields 6048 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li> 6049 6050 <li><tt>sgt</tt>: interprets the operands as signed values and yields 6051 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li> 6052 6053 <li><tt>sge</tt>: interprets the operands as signed values and yields 6054 <tt>true</tt> if <tt>op1</tt> is greater than or equal 6055 to <tt>op2</tt>.</li> 6056 6057 <li><tt>slt</tt>: interprets the operands as signed values and yields 6058 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li> 6059 6060 <li><tt>sle</tt>: interprets the operands as signed values and yields 6061 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li> 6062 </ol> 6063 6064 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer 6065 values are compared as if they were integers.</p> 6066 6067 <p>If the operands are integer vectors, then they are compared element by 6068 element. The result is an <tt>i1</tt> vector with the same number of elements 6069 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p> 6070 6071 <h5>Example:</h5> 6072 <pre> 6073 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i> 6074 <result> = icmp ne float* %X, %X <i>; yields: result=false</i> 6075 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i> 6076 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i> 6077 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i> 6078 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i> 6079 </pre> 6080 6081 <p>Note that the code generator does not yet support vector types with 6082 the <tt>icmp</tt> instruction.</p> 6083 6084 </div> 6085 6086 <!-- _______________________________________________________________________ --> 6087 <h4> 6088 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a> 6089 </h4> 6090 6091 <div> 6092 6093 <h5>Syntax:</h5> 6094 <pre> 6095 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i> 6096 </pre> 6097 6098 <h5>Overview:</h5> 6099 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean 6100 values based on comparison of its operands.</p> 6101 6102 <p>If the operands are floating point scalars, then the result type is a boolean 6103 (<a href="#t_integer"><tt>i1</tt></a>).</p> 6104 6105 <p>If the operands are floating point vectors, then the result type is a vector 6106 of boolean with the same number of elements as the operands being 6107 compared.</p> 6108 6109 <h5>Arguments:</h5> 6110 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is 6111 the condition code indicating the kind of comparison to perform. It is not a 6112 value, just a keyword. The possible condition code are:</p> 6113 6114 <ol> 6115 <li><tt>false</tt>: no comparison, always returns false</li> 6116 <li><tt>oeq</tt>: ordered and equal</li> 6117 <li><tt>ogt</tt>: ordered and greater than </li> 6118 <li><tt>oge</tt>: ordered and greater than or equal</li> 6119 <li><tt>olt</tt>: ordered and less than </li> 6120 <li><tt>ole</tt>: ordered and less than or equal</li> 6121 <li><tt>one</tt>: ordered and not equal</li> 6122 <li><tt>ord</tt>: ordered (no nans)</li> 6123 <li><tt>ueq</tt>: unordered or equal</li> 6124 <li><tt>ugt</tt>: unordered or greater than </li> 6125 <li><tt>uge</tt>: unordered or greater than or equal</li> 6126 <li><tt>ult</tt>: unordered or less than </li> 6127 <li><tt>ule</tt>: unordered or less than or equal</li> 6128 <li><tt>une</tt>: unordered or not equal</li> 6129 <li><tt>uno</tt>: unordered (either nans)</li> 6130 <li><tt>true</tt>: no comparison, always returns true</li> 6131 </ol> 6132 6133 <p><i>Ordered</i> means that neither operand is a QNAN while 6134 <i>unordered</i> means that either operand may be a QNAN.</p> 6135 6136 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either 6137 a <a href="#t_floating">floating point</a> type or 6138 a <a href="#t_vector">vector</a> of floating point type. They must have 6139 identical types.</p> 6140 6141 <h5>Semantics:</h5> 6142 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt> 6143 according to the condition code given as <tt>cond</tt>. If the operands are 6144 vectors, then the vectors are compared element by element. Each comparison 6145 performed always yields an <a href="#t_integer">i1</a> result, as 6146 follows:</p> 6147 6148 <ol> 6149 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li> 6150 6151 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and 6152 <tt>op1</tt> is equal to <tt>op2</tt>.</li> 6153 6154 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and 6155 <tt>op1</tt> is greater than <tt>op2</tt>.</li> 6156 6157 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and 6158 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li> 6159 6160 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and 6161 <tt>op1</tt> is less than <tt>op2</tt>.</li> 6162 6163 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and 6164 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li> 6165 6166 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and 6167 <tt>op1</tt> is not equal to <tt>op2</tt>.</li> 6168 6169 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li> 6170 6171 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or 6172 <tt>op1</tt> is equal to <tt>op2</tt>.</li> 6173 6174 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or 6175 <tt>op1</tt> is greater than <tt>op2</tt>.</li> 6176 6177 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or 6178 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li> 6179 6180 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or 6181 <tt>op1</tt> is less than <tt>op2</tt>.</li> 6182 6183 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or 6184 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li> 6185 6186 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or 6187 <tt>op1</tt> is not equal to <tt>op2</tt>.</li> 6188 6189 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li> 6190 6191 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li> 6192 </ol> 6193 6194 <h5>Example:</h5> 6195 <pre> 6196 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i> 6197 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i> 6198 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i> 6199 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i> 6200 </pre> 6201 6202 <p>Note that the code generator does not yet support vector types with 6203 the <tt>fcmp</tt> instruction.</p> 6204 6205 </div> 6206 6207 <!-- _______________________________________________________________________ --> 6208 <h4> 6209 <a name="i_phi">'<tt>phi</tt>' Instruction</a> 6210 </h4> 6211 6212 <div> 6213 6214 <h5>Syntax:</h5> 6215 <pre> 6216 <result> = phi <ty> [ <val0>, <label0>], ... 6217 </pre> 6218 6219 <h5>Overview:</h5> 6220 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the 6221 SSA graph representing the function.</p> 6222 6223 <h5>Arguments:</h5> 6224 <p>The type of the incoming values is specified with the first type field. After 6225 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with 6226 one pair for each predecessor basic block of the current block. Only values 6227 of <a href="#t_firstclass">first class</a> type may be used as the value 6228 arguments to the PHI node. Only labels may be used as the label 6229 arguments.</p> 6230 6231 <p>There must be no non-phi instructions between the start of a basic block and 6232 the PHI instructions: i.e. PHI instructions must be first in a basic 6233 block.</p> 6234 6235 <p>For the purposes of the SSA form, the use of each incoming value is deemed to 6236 occur on the edge from the corresponding predecessor block to the current 6237 block (but after any definition of an '<tt>invoke</tt>' instruction's return 6238 value on the same edge).</p> 6239 6240 <h5>Semantics:</h5> 6241 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value 6242 specified by the pair corresponding to the predecessor basic block that 6243 executed just prior to the current block.</p> 6244 6245 <h5>Example:</h5> 6246 <pre> 6247 Loop: ; Infinite loop that counts from 0 on up... 6248 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ] 6249 %nextindvar = add i32 %indvar, 1 6250 br label %Loop 6251 </pre> 6252 6253 </div> 6254 6255 <!-- _______________________________________________________________________ --> 6256 <h4> 6257 <a name="i_select">'<tt>select</tt>' Instruction</a> 6258 </h4> 6259 6260 <div> 6261 6262 <h5>Syntax:</h5> 6263 <pre> 6264 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i> 6265 6266 <i>selty</i> is either i1 or {<N x i1>} 6267 </pre> 6268 6269 <h5>Overview:</h5> 6270 <p>The '<tt>select</tt>' instruction is used to choose one value based on a 6271 condition, without branching.</p> 6272 6273 6274 <h5>Arguments:</h5> 6275 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1' 6276 values indicating the condition, and two values of the 6277 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are 6278 vectors and the condition is a scalar, then entire vectors are selected, not 6279 individual elements.</p> 6280 6281 <h5>Semantics:</h5> 6282 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the 6283 first value argument; otherwise, it returns the second value argument.</p> 6284 6285 <p>If the condition is a vector of i1, then the value arguments must be vectors 6286 of the same size, and the selection is done element by element.</p> 6287 6288 <h5>Example:</h5> 6289 <pre> 6290 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i> 6291 </pre> 6292 6293 </div> 6294 6295 <!-- _______________________________________________________________________ --> 6296 <h4> 6297 <a name="i_call">'<tt>call</tt>' Instruction</a> 6298 </h4> 6299 6300 <div> 6301 6302 <h5>Syntax:</h5> 6303 <pre> 6304 <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>] 6305 </pre> 6306 6307 <h5>Overview:</h5> 6308 <p>The '<tt>call</tt>' instruction represents a simple function call.</p> 6309 6310 <h5>Arguments:</h5> 6311 <p>This instruction requires several arguments:</p> 6312 6313 <ol> 6314 <li>The optional "tail" marker indicates that the callee function does not 6315 access any allocas or varargs in the caller. Note that calls may be 6316 marked "tail" even if they do not occur before 6317 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is 6318 present, the function call is eligible for tail call optimization, 6319 but <a href="CodeGenerator.html#tailcallopt">might not in fact be 6320 optimized into a jump</a>. The code generator may optimize calls marked 6321 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt"> 6322 sibling call optimization</a> when the caller and callee have 6323 matching signatures, or 2) forced tail call optimization when the 6324 following extra requirements are met: 6325 <ul> 6326 <li>Caller and callee both have the calling 6327 convention <tt>fastcc</tt>.</li> 6328 <li>The call is in tail position (ret immediately follows call and ret 6329 uses value of call or is void).</li> 6330 <li>Option <tt>-tailcallopt</tt> is enabled, 6331 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li> 6332 <li><a href="CodeGenerator.html#tailcallopt">Platform specific 6333 constraints are met.</a></li> 6334 </ul> 6335 </li> 6336 6337 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling 6338 convention</a> the call should use. If none is specified, the call 6339 defaults to using C calling conventions. The calling convention of the 6340 call must match the calling convention of the target function, or else the 6341 behavior is undefined.</li> 6342 6343 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for 6344 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and 6345 '<tt>inreg</tt>' attributes are valid here.</li> 6346 6347 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the 6348 type of the return value. Functions that return no value are marked 6349 <tt><a href="#t_void">void</a></tt>.</li> 6350 6351 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value 6352 being invoked. The argument types must match the types implied by this 6353 signature. This type can be omitted if the function is not varargs and if 6354 the function type does not return a pointer to a function.</li> 6355 6356 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to 6357 be invoked. In most cases, this is a direct function invocation, but 6358 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer 6359 to function value.</li> 6360 6361 <li>'<tt>function args</tt>': argument list whose types match the function 6362 signature argument types and parameter attributes. All arguments must be 6363 of <a href="#t_firstclass">first class</a> type. If the function 6364 signature indicates the function accepts a variable number of arguments, 6365 the extra arguments can be specified.</li> 6366 6367 <li>The optional <a href="#fnattrs">function attributes</a> list. Only 6368 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and 6369 '<tt>readnone</tt>' attributes are valid here.</li> 6370 </ol> 6371 6372 <h5>Semantics:</h5> 6373 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to 6374 a specified function, with its incoming arguments bound to the specified 6375 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called 6376 function, control flow continues with the instruction after the function 6377 call, and the return value of the function is bound to the result 6378 argument.</p> 6379 6380 <h5>Example:</h5> 6381 <pre> 6382 %retval = call i32 @test(i32 %argc) 6383 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i> 6384 %X = tail call i32 @foo() <i>; yields i32</i> 6385 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i> 6386 call void %foo(i8 97 signext) 6387 6388 %struct.A = type { i32, i8 } 6389 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i> 6390 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i> 6391 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i> 6392 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i> 6393 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i> 6394 </pre> 6395 6396 <p>llvm treats calls to some functions with names and arguments that match the 6397 standard C99 library as being the C99 library functions, and may perform 6398 optimizations or generate code for them under that assumption. This is 6399 something we'd like to change in the future to provide better support for 6400 freestanding environments and non-C-based languages.</p> 6401 6402 </div> 6403 6404 <!-- _______________________________________________________________________ --> 6405 <h4> 6406 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a> 6407 </h4> 6408 6409 <div> 6410 6411 <h5>Syntax:</h5> 6412 <pre> 6413 <resultval> = va_arg <va_list*> <arglist>, <argty> 6414 </pre> 6415 6416 <h5>Overview:</h5> 6417 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through 6418 the "variable argument" area of a function call. It is used to implement the 6419 <tt>va_arg</tt> macro in C.</p> 6420 6421 <h5>Arguments:</h5> 6422 <p>This instruction takes a <tt>va_list*</tt> value and the type of the 6423 argument. It returns a value of the specified argument type and increments 6424 the <tt>va_list</tt> to point to the next argument. The actual type 6425 of <tt>va_list</tt> is target specific.</p> 6426 6427 <h5>Semantics:</h5> 6428 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type 6429 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point 6430 to the next argument. For more information, see the variable argument 6431 handling <a href="#int_varargs">Intrinsic Functions</a>.</p> 6432 6433 <p>It is legal for this instruction to be called in a function which does not 6434 take a variable number of arguments, for example, the <tt>vfprintf</tt> 6435 function.</p> 6436 6437 <p><tt>va_arg</tt> is an LLVM instruction instead of 6438 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an 6439 argument.</p> 6440 6441 <h5>Example:</h5> 6442 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p> 6443 6444 <p>Note that the code generator does not yet fully support va_arg on many 6445 targets. Also, it does not currently support va_arg with aggregate types on 6446 any target.</p> 6447 6448 </div> 6449 6450 <!-- _______________________________________________________________________ --> 6451 <h4> 6452 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a> 6453 </h4> 6454 6455 <div> 6456 6457 <h5>Syntax:</h5> 6458 <pre> 6459 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+ 6460 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>* 6461 6462 <clause> := catch <type> <value> 6463 <clause> := filter <array constant type> <array constant> 6464 </pre> 6465 6466 <h5>Overview:</h5> 6467 <p>The '<tt>landingpad</tt>' instruction is used by 6468 <a href="ExceptionHandling.html#overview">LLVM's exception handling 6469 system</a> to specify that a basic block is a landing pad — one where 6470 the exception lands, and corresponds to the code found in the 6471 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It 6472 defines values supplied by the personality function (<tt>pers_fn</tt>) upon 6473 re-entry to the function. The <tt>resultval</tt> has the 6474 type <tt>resultty</tt>.</p> 6475 6476 <h5>Arguments:</h5> 6477 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality 6478 function associated with the unwinding mechanism. The optional 6479 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p> 6480 6481 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt> 6482 or <tt>filter</tt> — and contains the global variable representing the 6483 "type" that may be caught or filtered respectively. Unlike the 6484 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as 6485 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot 6486 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em> 6487 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p> 6488 6489 <h5>Semantics:</h5> 6490 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the 6491 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and 6492 therefore the "result type" of the <tt>landingpad</tt> instruction. As with 6493 calling conventions, how the personality function results are represented in 6494 LLVM IR is target specific.</p> 6495 6496 <p>The clauses are applied in order from top to bottom. If two 6497 <tt>landingpad</tt> instructions are merged together through inlining, the 6498 clauses from the calling function are appended to the list of clauses. 6499 When the call stack is being unwound due to an exception being thrown, the 6500 exception is compared against each <tt>clause</tt> in turn. If it doesn't 6501 match any of the clauses, and the <tt>cleanup</tt> flag is not set, then 6502 unwinding continues further up the call stack.</p> 6503 6504 <p>The <tt>landingpad</tt> instruction has several restrictions:</p> 6505 6506 <ul> 6507 <li>A landing pad block is a basic block which is the unwind destination of an 6508 '<tt>invoke</tt>' instruction.</li> 6509 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its 6510 first non-PHI instruction.</li> 6511 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing 6512 pad block.</li> 6513 <li>A basic block that is not a landing pad block may not include a 6514 '<tt>landingpad</tt>' instruction.</li> 6515 <li>All '<tt>landingpad</tt>' instructions in a function must have the same 6516 personality function.</li> 6517 </ul> 6518 6519 <h5>Example:</h5> 6520 <pre> 6521 ;; A landing pad which can catch an integer. 6522 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0 6523 catch i8** @_ZTIi 6524 ;; A landing pad that is a cleanup. 6525 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0 6526 cleanup 6527 ;; A landing pad which can catch an integer and can only throw a double. 6528 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0 6529 catch i8** @_ZTIi 6530 filter [1 x i8**] [@_ZTId] 6531 </pre> 6532 6533 </div> 6534 6535 </div> 6536 6537 </div> 6538 6539 <!-- *********************************************************************** --> 6540 <h2><a name="intrinsics">Intrinsic Functions</a></h2> 6541 <!-- *********************************************************************** --> 6542 6543 <div> 6544 6545 <p>LLVM supports the notion of an "intrinsic function". These functions have 6546 well known names and semantics and are required to follow certain 6547 restrictions. Overall, these intrinsics represent an extension mechanism for 6548 the LLVM language that does not require changing all of the transformations 6549 in LLVM when adding to the language (or the bitcode reader/writer, the 6550 parser, etc...).</p> 6551 6552 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This 6553 prefix is reserved in LLVM for intrinsic names; thus, function names may not 6554 begin with this prefix. Intrinsic functions must always be external 6555 functions: you cannot define the body of intrinsic functions. Intrinsic 6556 functions may only be used in call or invoke instructions: it is illegal to 6557 take the address of an intrinsic function. Additionally, because intrinsic 6558 functions are part of the LLVM language, it is required if any are added that 6559 they be documented here.</p> 6560 6561 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a 6562 family of functions that perform the same operation but on different data 6563 types. Because LLVM can represent over 8 million different integer types, 6564 overloading is used commonly to allow an intrinsic function to operate on any 6565 integer type. One or more of the argument types or the result type can be 6566 overloaded to accept any integer type. Argument types may also be defined as 6567 exactly matching a previous argument's type or the result type. This allows 6568 an intrinsic function which accepts multiple arguments, but needs all of them 6569 to be of the same type, to only be overloaded with respect to a single 6570 argument or the result.</p> 6571 6572 <p>Overloaded intrinsics will have the names of its overloaded argument types 6573 encoded into its function name, each preceded by a period. Only those types 6574 which are overloaded result in a name suffix. Arguments whose type is matched 6575 against another type do not. For example, the <tt>llvm.ctpop</tt> function 6576 can take an integer of any width and returns an integer of exactly the same 6577 integer width. This leads to a family of functions such as 6578 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 6579 %val)</tt>. Only one type, the return type, is overloaded, and only one type 6580 suffix is required. Because the argument's type is matched against the return 6581 type, it does not require its own name suffix.</p> 6582 6583 <p>To learn how to add an intrinsic function, please see the 6584 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p> 6585 6586 <!-- ======================================================================= --> 6587 <h3> 6588 <a name="int_varargs">Variable Argument Handling Intrinsics</a> 6589 </h3> 6590 6591 <div> 6592 6593 <p>Variable argument support is defined in LLVM with 6594 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three 6595 intrinsic functions. These functions are related to the similarly named 6596 macros defined in the <tt><stdarg.h></tt> header file.</p> 6597 6598 <p>All of these functions operate on arguments that use a target-specific value 6599 type "<tt>va_list</tt>". The LLVM assembly language reference manual does 6600 not define what this type is, so all transformations should be prepared to 6601 handle these functions regardless of the type used.</p> 6602 6603 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a> 6604 instruction and the variable argument handling intrinsic functions are 6605 used.</p> 6606 6607 <pre class="doc_code"> 6608 define i32 @test(i32 %X, ...) { 6609 ; Initialize variable argument processing 6610 %ap = alloca i8* 6611 %ap2 = bitcast i8** %ap to i8* 6612 call void @llvm.va_start(i8* %ap2) 6613 6614 ; Read a single integer argument 6615 %tmp = va_arg i8** %ap, i32 6616 6617 ; Demonstrate usage of llvm.va_copy and llvm.va_end 6618 %aq = alloca i8* 6619 %aq2 = bitcast i8** %aq to i8* 6620 call void @llvm.va_copy(i8* %aq2, i8* %ap2) 6621 call void @llvm.va_end(i8* %aq2) 6622 6623 ; Stop processing of arguments. 6624 call void @llvm.va_end(i8* %ap2) 6625 ret i32 %tmp 6626 } 6627 6628 declare void @llvm.va_start(i8*) 6629 declare void @llvm.va_copy(i8*, i8*) 6630 declare void @llvm.va_end(i8*) 6631 </pre> 6632 6633 <!-- _______________________________________________________________________ --> 6634 <h4> 6635 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a> 6636 </h4> 6637 6638 6639 <div> 6640 6641 <h5>Syntax:</h5> 6642 <pre> 6643 declare void %llvm.va_start(i8* <arglist>) 6644 </pre> 6645 6646 <h5>Overview:</h5> 6647 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt> 6648 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p> 6649 6650 <h5>Arguments:</h5> 6651 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p> 6652 6653 <h5>Semantics:</h5> 6654 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt> 6655 macro available in C. In a target-dependent way, it initializes 6656 the <tt>va_list</tt> element to which the argument points, so that the next 6657 call to <tt>va_arg</tt> will produce the first variable argument passed to 6658 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not 6659 need to know the last argument of the function as the compiler can figure 6660 that out.</p> 6661 6662 </div> 6663 6664 <!-- _______________________________________________________________________ --> 6665 <h4> 6666 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a> 6667 </h4> 6668 6669 <div> 6670 6671 <h5>Syntax:</h5> 6672 <pre> 6673 declare void @llvm.va_end(i8* <arglist>) 6674 </pre> 6675 6676 <h5>Overview:</h5> 6677 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>, 6678 which has been initialized previously 6679 with <tt><a href="#int_va_start">llvm.va_start</a></tt> 6680 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p> 6681 6682 <h5>Arguments:</h5> 6683 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p> 6684 6685 <h5>Semantics:</h5> 6686 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt> 6687 macro available in C. In a target-dependent way, it destroys 6688 the <tt>va_list</tt> element to which the argument points. Calls 6689 to <a href="#int_va_start"><tt>llvm.va_start</tt></a> 6690 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly 6691 with calls to <tt>llvm.va_end</tt>.</p> 6692 6693 </div> 6694 6695 <!-- _______________________________________________________________________ --> 6696 <h4> 6697 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a> 6698 </h4> 6699 6700 <div> 6701 6702 <h5>Syntax:</h5> 6703 <pre> 6704 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>) 6705 </pre> 6706 6707 <h5>Overview:</h5> 6708 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position 6709 from the source argument list to the destination argument list.</p> 6710 6711 <h5>Arguments:</h5> 6712 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize. 6713 The second argument is a pointer to a <tt>va_list</tt> element to copy 6714 from.</p> 6715 6716 <h5>Semantics:</h5> 6717 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> 6718 macro available in C. In a target-dependent way, it copies the 6719 source <tt>va_list</tt> element into the destination <tt>va_list</tt> 6720 element. This intrinsic is necessary because 6721 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be 6722 arbitrarily complex and require, for example, memory allocation.</p> 6723 6724 </div> 6725 6726 </div> 6727 6728 <!-- ======================================================================= --> 6729 <h3> 6730 <a name="int_gc">Accurate Garbage Collection Intrinsics</a> 6731 </h3> 6732 6733 <div> 6734 6735 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage 6736 Collection</a> (GC) requires the implementation and generation of these 6737 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC 6738 roots on the stack</a>, as well as garbage collector implementations that 6739 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> 6740 barriers. Front-ends for type-safe garbage collected languages should generate 6741 these intrinsics to make use of the LLVM garbage collectors. For more details, 6742 see <a href="GarbageCollection.html">Accurate Garbage Collection with 6743 LLVM</a>.</p> 6744 6745 <p>The garbage collection intrinsics only operate on objects in the generic 6746 address space (address space zero).</p> 6747 6748 <!-- _______________________________________________________________________ --> 6749 <h4> 6750 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a> 6751 </h4> 6752 6753 <div> 6754 6755 <h5>Syntax:</h5> 6756 <pre> 6757 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata) 6758 </pre> 6759 6760 <h5>Overview:</h5> 6761 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to 6762 the code generator, and allows some metadata to be associated with it.</p> 6763 6764 <h5>Arguments:</h5> 6765 <p>The first argument specifies the address of a stack object that contains the 6766 root pointer. The second pointer (which must be either a constant or a 6767 global value address) contains the meta-data to be associated with the 6768 root.</p> 6769 6770 <h5>Semantics:</h5> 6771 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc" 6772 location. At compile-time, the code generator generates information to allow 6773 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>' 6774 intrinsic may only be used in a function which <a href="#gc">specifies a GC 6775 algorithm</a>.</p> 6776 6777 </div> 6778 6779 <!-- _______________________________________________________________________ --> 6780 <h4> 6781 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a> 6782 </h4> 6783 6784 <div> 6785 6786 <h5>Syntax:</h5> 6787 <pre> 6788 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr) 6789 </pre> 6790 6791 <h5>Overview:</h5> 6792 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap 6793 locations, allowing garbage collector implementations that require read 6794 barriers.</p> 6795 6796 <h5>Arguments:</h5> 6797 <p>The second argument is the address to read from, which should be an address 6798 allocated from the garbage collector. The first object is a pointer to the 6799 start of the referenced object, if needed by the language runtime (otherwise 6800 null).</p> 6801 6802 <h5>Semantics:</h5> 6803 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load 6804 instruction, but may be replaced with substantially more complex code by the 6805 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic 6806 may only be used in a function which <a href="#gc">specifies a GC 6807 algorithm</a>.</p> 6808 6809 </div> 6810 6811 <!-- _______________________________________________________________________ --> 6812 <h4> 6813 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a> 6814 </h4> 6815 6816 <div> 6817 6818 <h5>Syntax:</h5> 6819 <pre> 6820 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2) 6821 </pre> 6822 6823 <h5>Overview:</h5> 6824 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap 6825 locations, allowing garbage collector implementations that require write 6826 barriers (such as generational or reference counting collectors).</p> 6827 6828 <h5>Arguments:</h5> 6829 <p>The first argument is the reference to store, the second is the start of the 6830 object to store it to, and the third is the address of the field of Obj to 6831 store to. If the runtime does not require a pointer to the object, Obj may 6832 be null.</p> 6833 6834 <h5>Semantics:</h5> 6835 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store 6836 instruction, but may be replaced with substantially more complex code by the 6837 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic 6838 may only be used in a function which <a href="#gc">specifies a GC 6839 algorithm</a>.</p> 6840 6841 </div> 6842 6843 </div> 6844 6845 <!-- ======================================================================= --> 6846 <h3> 6847 <a name="int_codegen">Code Generator Intrinsics</a> 6848 </h3> 6849 6850 <div> 6851 6852 <p>These intrinsics are provided by LLVM to expose special features that may 6853 only be implemented with code generator support.</p> 6854 6855 <!-- _______________________________________________________________________ --> 6856 <h4> 6857 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a> 6858 </h4> 6859 6860 <div> 6861 6862 <h5>Syntax:</h5> 6863 <pre> 6864 declare i8 *@llvm.returnaddress(i32 <level>) 6865 </pre> 6866 6867 <h5>Overview:</h5> 6868 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a 6869 target-specific value indicating the return address of the current function 6870 or one of its callers.</p> 6871 6872 <h5>Arguments:</h5> 6873 <p>The argument to this intrinsic indicates which function to return the address 6874 for. Zero indicates the calling function, one indicates its caller, etc. 6875 The argument is <b>required</b> to be a constant integer value.</p> 6876 6877 <h5>Semantics:</h5> 6878 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer 6879 indicating the return address of the specified call frame, or zero if it 6880 cannot be identified. The value returned by this intrinsic is likely to be 6881 incorrect or 0 for arguments other than zero, so it should only be used for 6882 debugging purposes.</p> 6883 6884 <p>Note that calling this intrinsic does not prevent function inlining or other 6885 aggressive transformations, so the value returned may not be that of the 6886 obvious source-language caller.</p> 6887 6888 </div> 6889 6890 <!-- _______________________________________________________________________ --> 6891 <h4> 6892 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a> 6893 </h4> 6894 6895 <div> 6896 6897 <h5>Syntax:</h5> 6898 <pre> 6899 declare i8* @llvm.frameaddress(i32 <level>) 6900 </pre> 6901 6902 <h5>Overview:</h5> 6903 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the 6904 target-specific frame pointer value for the specified stack frame.</p> 6905 6906 <h5>Arguments:</h5> 6907 <p>The argument to this intrinsic indicates which function to return the frame 6908 pointer for. Zero indicates the calling function, one indicates its caller, 6909 etc. The argument is <b>required</b> to be a constant integer value.</p> 6910 6911 <h5>Semantics:</h5> 6912 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer 6913 indicating the frame address of the specified call frame, or zero if it 6914 cannot be identified. The value returned by this intrinsic is likely to be 6915 incorrect or 0 for arguments other than zero, so it should only be used for 6916 debugging purposes.</p> 6917 6918 <p>Note that calling this intrinsic does not prevent function inlining or other 6919 aggressive transformations, so the value returned may not be that of the 6920 obvious source-language caller.</p> 6921 6922 </div> 6923 6924 <!-- _______________________________________________________________________ --> 6925 <h4> 6926 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a> 6927 </h4> 6928 6929 <div> 6930 6931 <h5>Syntax:</h5> 6932 <pre> 6933 declare i8* @llvm.stacksave() 6934 </pre> 6935 6936 <h5>Overview:</h5> 6937 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state 6938 of the function stack, for use 6939 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is 6940 useful for implementing language features like scoped automatic variable 6941 sized arrays in C99.</p> 6942 6943 <h5>Semantics:</h5> 6944 <p>This intrinsic returns a opaque pointer value that can be passed 6945 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When 6946 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved 6947 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack 6948 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. 6949 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the 6950 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p> 6951 6952 </div> 6953 6954 <!-- _______________________________________________________________________ --> 6955 <h4> 6956 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a> 6957 </h4> 6958 6959 <div> 6960 6961 <h5>Syntax:</h5> 6962 <pre> 6963 declare void @llvm.stackrestore(i8* %ptr) 6964 </pre> 6965 6966 <h5>Overview:</h5> 6967 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of 6968 the function stack to the state it was in when the 6969 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic 6970 executed. This is useful for implementing language features like scoped 6971 automatic variable sized arrays in C99.</p> 6972 6973 <h5>Semantics:</h5> 6974 <p>See the description 6975 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p> 6976 6977 </div> 6978 6979 <!-- _______________________________________________________________________ --> 6980 <h4> 6981 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a> 6982 </h4> 6983 6984 <div> 6985 6986 <h5>Syntax:</h5> 6987 <pre> 6988 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>) 6989 </pre> 6990 6991 <h5>Overview:</h5> 6992 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to 6993 insert a prefetch instruction if supported; otherwise, it is a noop. 6994 Prefetches have no effect on the behavior of the program but can change its 6995 performance characteristics.</p> 6996 6997 <h5>Arguments:</h5> 6998 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the 6999 specifier determining if the fetch should be for a read (0) or write (1), 7000 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no 7001 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt> 7002 specifies whether the prefetch is performed on the data (1) or instruction (0) 7003 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments 7004 must be constant integers.</p> 7005 7006 <h5>Semantics:</h5> 7007 <p>This intrinsic does not modify the behavior of the program. In particular, 7008 prefetches cannot trap and do not produce a value. On targets that support 7009 this intrinsic, the prefetch can provide hints to the processor cache for 7010 better performance.</p> 7011 7012 </div> 7013 7014 <!-- _______________________________________________________________________ --> 7015 <h4> 7016 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a> 7017 </h4> 7018 7019 <div> 7020 7021 <h5>Syntax:</h5> 7022 <pre> 7023 declare void @llvm.pcmarker(i32 <id>) 7024 </pre> 7025 7026 <h5>Overview:</h5> 7027 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program 7028 Counter (PC) in a region of code to simulators and other tools. The method 7029 is target specific, but it is expected that the marker will use exported 7030 symbols to transmit the PC of the marker. The marker makes no guarantees 7031 that it will remain with any specific instruction after optimizations. It is 7032 possible that the presence of a marker will inhibit optimizations. The 7033 intended use is to be inserted after optimizations to allow correlations of 7034 simulation runs.</p> 7035 7036 <h5>Arguments:</h5> 7037 <p><tt>id</tt> is a numerical id identifying the marker.</p> 7038 7039 <h5>Semantics:</h5> 7040 <p>This intrinsic does not modify the behavior of the program. Backends that do 7041 not support this intrinsic may ignore it.</p> 7042 7043 </div> 7044 7045 <!-- _______________________________________________________________________ --> 7046 <h4> 7047 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a> 7048 </h4> 7049 7050 <div> 7051 7052 <h5>Syntax:</h5> 7053 <pre> 7054 declare i64 @llvm.readcyclecounter() 7055 </pre> 7056 7057 <h5>Overview:</h5> 7058 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle 7059 counter register (or similar low latency, high accuracy clocks) on those 7060 targets that support it. On X86, it should map to RDTSC. On Alpha, it 7061 should map to RPCC. As the backing counters overflow quickly (on the order 7062 of 9 seconds on alpha), this should only be used for small timings.</p> 7063 7064 <h5>Semantics:</h5> 7065 <p>When directly supported, reading the cycle counter should not modify any 7066 memory. Implementations are allowed to either return a application specific 7067 value or a system wide value. On backends without support, this is lowered 7068 to a constant 0.</p> 7069 7070 </div> 7071 7072 </div> 7073 7074 <!-- ======================================================================= --> 7075 <h3> 7076 <a name="int_libc">Standard C Library Intrinsics</a> 7077 </h3> 7078 7079 <div> 7080 7081 <p>LLVM provides intrinsics for a few important standard C library functions. 7082 These intrinsics allow source-language front-ends to pass information about 7083 the alignment of the pointer arguments to the code generator, providing 7084 opportunity for more efficient code generation.</p> 7085 7086 <!-- _______________________________________________________________________ --> 7087 <h4> 7088 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a> 7089 </h4> 7090 7091 <div> 7092 7093 <h5>Syntax:</h5> 7094 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any 7095 integer bit width and for different address spaces. Not all targets support 7096 all bit widths however.</p> 7097 7098 <pre> 7099 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>, 7100 i32 <len>, i32 <align>, i1 <isvolatile>) 7101 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>, 7102 i64 <len>, i32 <align>, i1 <isvolatile>) 7103 </pre> 7104 7105 <h5>Overview:</h5> 7106 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the 7107 source location to the destination location.</p> 7108 7109 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt> 7110 intrinsics do not return a value, takes extra alignment/isvolatile arguments 7111 and the pointers can be in specified address spaces.</p> 7112 7113 <h5>Arguments:</h5> 7114 7115 <p>The first argument is a pointer to the destination, the second is a pointer 7116 to the source. The third argument is an integer argument specifying the 7117 number of bytes to copy, the fourth argument is the alignment of the 7118 source and destination locations, and the fifth is a boolean indicating a 7119 volatile access.</p> 7120 7121 <p>If the call to this intrinsic has an alignment value that is not 0 or 1, 7122 then the caller guarantees that both the source and destination pointers are 7123 aligned to that boundary.</p> 7124 7125 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the 7126 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>. 7127 The detailed access behavior is not very cleanly specified and it is unwise 7128 to depend on it.</p> 7129 7130 <h5>Semantics:</h5> 7131 7132 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the 7133 source location to the destination location, which are not allowed to 7134 overlap. It copies "len" bytes of memory over. If the argument is known to 7135 be aligned to some boundary, this can be specified as the fourth argument, 7136 otherwise it should be set to 0 or 1.</p> 7137 7138 </div> 7139 7140 <!-- _______________________________________________________________________ --> 7141 <h4> 7142 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a> 7143 </h4> 7144 7145 <div> 7146 7147 <h5>Syntax:</h5> 7148 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit 7149 width and for different address space. Not all targets support all bit 7150 widths however.</p> 7151 7152 <pre> 7153 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>, 7154 i32 <len>, i32 <align>, i1 <isvolatile>) 7155 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>, 7156 i64 <len>, i32 <align>, i1 <isvolatile>) 7157 </pre> 7158 7159 <h5>Overview:</h5> 7160 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the 7161 source location to the destination location. It is similar to the 7162 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to 7163 overlap.</p> 7164 7165 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt> 7166 intrinsics do not return a value, takes extra alignment/isvolatile arguments 7167 and the pointers can be in specified address spaces.</p> 7168 7169 <h5>Arguments:</h5> 7170 7171 <p>The first argument is a pointer to the destination, the second is a pointer 7172 to the source. The third argument is an integer argument specifying the 7173 number of bytes to copy, the fourth argument is the alignment of the 7174 source and destination locations, and the fifth is a boolean indicating a 7175 volatile access.</p> 7176 7177 <p>If the call to this intrinsic has an alignment value that is not 0 or 1, 7178 then the caller guarantees that the source and destination pointers are 7179 aligned to that boundary.</p> 7180 7181 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the 7182 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>. 7183 The detailed access behavior is not very cleanly specified and it is unwise 7184 to depend on it.</p> 7185 7186 <h5>Semantics:</h5> 7187 7188 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the 7189 source location to the destination location, which may overlap. It copies 7190 "len" bytes of memory over. If the argument is known to be aligned to some 7191 boundary, this can be specified as the fourth argument, otherwise it should 7192 be set to 0 or 1.</p> 7193 7194 </div> 7195 7196 <!-- _______________________________________________________________________ --> 7197 <h4> 7198 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a> 7199 </h4> 7200 7201 <div> 7202 7203 <h5>Syntax:</h5> 7204 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit 7205 width and for different address spaces. However, not all targets support all 7206 bit widths.</p> 7207 7208 <pre> 7209 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>, 7210 i32 <len>, i32 <align>, i1 <isvolatile>) 7211 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>, 7212 i64 <len>, i32 <align>, i1 <isvolatile>) 7213 </pre> 7214 7215 <h5>Overview:</h5> 7216 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a 7217 particular byte value.</p> 7218 7219 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt> 7220 intrinsic does not return a value and takes extra alignment/volatile 7221 arguments. Also, the destination can be in an arbitrary address space.</p> 7222 7223 <h5>Arguments:</h5> 7224 <p>The first argument is a pointer to the destination to fill, the second is the 7225 byte value with which to fill it, the third argument is an integer argument 7226 specifying the number of bytes to fill, and the fourth argument is the known 7227 alignment of the destination location.</p> 7228 7229 <p>If the call to this intrinsic has an alignment value that is not 0 or 1, 7230 then the caller guarantees that the destination pointer is aligned to that 7231 boundary.</p> 7232 7233 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the 7234 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>. 7235 The detailed access behavior is not very cleanly specified and it is unwise 7236 to depend on it.</p> 7237 7238 <h5>Semantics:</h5> 7239 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting 7240 at the destination location. If the argument is known to be aligned to some 7241 boundary, this can be specified as the fourth argument, otherwise it should 7242 be set to 0 or 1.</p> 7243 7244 </div> 7245 7246 <!-- _______________________________________________________________________ --> 7247 <h4> 7248 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a> 7249 </h4> 7250 7251 <div> 7252 7253 <h5>Syntax:</h5> 7254 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any 7255 floating point or vector of floating point type. Not all targets support all 7256 types however.</p> 7257 7258 <pre> 7259 declare float @llvm.sqrt.f32(float %Val) 7260 declare double @llvm.sqrt.f64(double %Val) 7261 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val) 7262 declare fp128 @llvm.sqrt.f128(fp128 %Val) 7263 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val) 7264 </pre> 7265 7266 <h5>Overview:</h5> 7267 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand, 7268 returning the same value as the libm '<tt>sqrt</tt>' functions would. 7269 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined 7270 behavior for negative numbers other than -0.0 (which allows for better 7271 optimization, because there is no need to worry about errno being 7272 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p> 7273 7274 <h5>Arguments:</h5> 7275 <p>The argument and return value are floating point numbers of the same 7276 type.</p> 7277 7278 <h5>Semantics:</h5> 7279 <p>This function returns the sqrt of the specified operand if it is a 7280 nonnegative floating point number.</p> 7281 7282 </div> 7283 7284 <!-- _______________________________________________________________________ --> 7285 <h4> 7286 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a> 7287 </h4> 7288 7289 <div> 7290 7291 <h5>Syntax:</h5> 7292 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any 7293 floating point or vector of floating point type. Not all targets support all 7294 types however.</p> 7295 7296 <pre> 7297 declare float @llvm.powi.f32(float %Val, i32 %power) 7298 declare double @llvm.powi.f64(double %Val, i32 %power) 7299 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power) 7300 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power) 7301 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power) 7302 </pre> 7303 7304 <h5>Overview:</h5> 7305 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the 7306 specified (positive or negative) power. The order of evaluation of 7307 multiplications is not defined. When a vector of floating point type is 7308 used, the second argument remains a scalar integer value.</p> 7309 7310 <h5>Arguments:</h5> 7311 <p>The second argument is an integer power, and the first is a value to raise to 7312 that power.</p> 7313 7314 <h5>Semantics:</h5> 7315 <p>This function returns the first value raised to the second power with an 7316 unspecified sequence of rounding operations.</p> 7317 7318 </div> 7319 7320 <!-- _______________________________________________________________________ --> 7321 <h4> 7322 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a> 7323 </h4> 7324 7325 <div> 7326 7327 <h5>Syntax:</h5> 7328 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any 7329 floating point or vector of floating point type. Not all targets support all 7330 types however.</p> 7331 7332 <pre> 7333 declare float @llvm.sin.f32(float %Val) 7334 declare double @llvm.sin.f64(double %Val) 7335 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val) 7336 declare fp128 @llvm.sin.f128(fp128 %Val) 7337 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val) 7338 </pre> 7339 7340 <h5>Overview:</h5> 7341 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p> 7342 7343 <h5>Arguments:</h5> 7344 <p>The argument and return value are floating point numbers of the same 7345 type.</p> 7346 7347 <h5>Semantics:</h5> 7348 <p>This function returns the sine of the specified operand, returning the same 7349 values as the libm <tt>sin</tt> functions would, and handles error conditions 7350 in the same way.</p> 7351 7352 </div> 7353 7354 <!-- _______________________________________________________________________ --> 7355 <h4> 7356 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a> 7357 </h4> 7358 7359 <div> 7360 7361 <h5>Syntax:</h5> 7362 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any 7363 floating point or vector of floating point type. Not all targets support all 7364 types however.</p> 7365 7366 <pre> 7367 declare float @llvm.cos.f32(float %Val) 7368 declare double @llvm.cos.f64(double %Val) 7369 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val) 7370 declare fp128 @llvm.cos.f128(fp128 %Val) 7371 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val) 7372 </pre> 7373 7374 <h5>Overview:</h5> 7375 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p> 7376 7377 <h5>Arguments:</h5> 7378 <p>The argument and return value are floating point numbers of the same 7379 type.</p> 7380 7381 <h5>Semantics:</h5> 7382 <p>This function returns the cosine of the specified operand, returning the same 7383 values as the libm <tt>cos</tt> functions would, and handles error conditions 7384 in the same way.</p> 7385 7386 </div> 7387 7388 <!-- _______________________________________________________________________ --> 7389 <h4> 7390 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a> 7391 </h4> 7392 7393 <div> 7394 7395 <h5>Syntax:</h5> 7396 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any 7397 floating point or vector of floating point type. Not all targets support all 7398 types however.</p> 7399 7400 <pre> 7401 declare float @llvm.pow.f32(float %Val, float %Power) 7402 declare double @llvm.pow.f64(double %Val, double %Power) 7403 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power) 7404 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power) 7405 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power) 7406 </pre> 7407 7408 <h5>Overview:</h5> 7409 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the 7410 specified (positive or negative) power.</p> 7411 7412 <h5>Arguments:</h5> 7413 <p>The second argument is a floating point power, and the first is a value to 7414 raise to that power.</p> 7415 7416 <h5>Semantics:</h5> 7417 <p>This function returns the first value raised to the second power, returning 7418 the same values as the libm <tt>pow</tt> functions would, and handles error 7419 conditions in the same way.</p> 7420 7421 </div> 7422 7423 <!-- _______________________________________________________________________ --> 7424 <h4> 7425 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a> 7426 </h4> 7427 7428 <div> 7429 7430 <h5>Syntax:</h5> 7431 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any 7432 floating point or vector of floating point type. Not all targets support all 7433 types however.</p> 7434 7435 <pre> 7436 declare float @llvm.exp.f32(float %Val) 7437 declare double @llvm.exp.f64(double %Val) 7438 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val) 7439 declare fp128 @llvm.exp.f128(fp128 %Val) 7440 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val) 7441 </pre> 7442 7443 <h5>Overview:</h5> 7444 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p> 7445 7446 <h5>Arguments:</h5> 7447 <p>The argument and return value are floating point numbers of the same 7448 type.</p> 7449 7450 <h5>Semantics:</h5> 7451 <p>This function returns the same values as the libm <tt>exp</tt> functions 7452 would, and handles error conditions in the same way.</p> 7453 7454 </div> 7455 7456 <!-- _______________________________________________________________________ --> 7457 <h4> 7458 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a> 7459 </h4> 7460 7461 <div> 7462 7463 <h5>Syntax:</h5> 7464 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any 7465 floating point or vector of floating point type. Not all targets support all 7466 types however.</p> 7467 7468 <pre> 7469 declare float @llvm.log.f32(float %Val) 7470 declare double @llvm.log.f64(double %Val) 7471 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val) 7472 declare fp128 @llvm.log.f128(fp128 %Val) 7473 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val) 7474 </pre> 7475 7476 <h5>Overview:</h5> 7477 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p> 7478 7479 <h5>Arguments:</h5> 7480 <p>The argument and return value are floating point numbers of the same 7481 type.</p> 7482 7483 <h5>Semantics:</h5> 7484 <p>This function returns the same values as the libm <tt>log</tt> functions 7485 would, and handles error conditions in the same way.</p> 7486 7487 </div> 7488 7489 <!-- _______________________________________________________________________ --> 7490 <h4> 7491 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a> 7492 </h4> 7493 7494 <div> 7495 7496 <h5>Syntax:</h5> 7497 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any 7498 floating point or vector of floating point type. Not all targets support all 7499 types however.</p> 7500 7501 <pre> 7502 declare float @llvm.fma.f32(float %a, float %b, float %c) 7503 declare double @llvm.fma.f64(double %a, double %b, double %c) 7504 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c) 7505 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c) 7506 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c) 7507 </pre> 7508 7509 <h5>Overview:</h5> 7510 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add 7511 operation.</p> 7512 7513 <h5>Arguments:</h5> 7514 <p>The argument and return value are floating point numbers of the same 7515 type.</p> 7516 7517 <h5>Semantics:</h5> 7518 <p>This function returns the same values as the libm <tt>fma</tt> functions 7519 would.</p> 7520 7521 </div> 7522 7523 <!-- _______________________________________________________________________ --> 7524 <h4> 7525 <a name="int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a> 7526 </h4> 7527 7528 <div> 7529 7530 <h5>Syntax:</h5> 7531 <p>This is an overloaded intrinsic. You can use <tt>llvm.fabs</tt> on any 7532 floating point or vector of floating point type. Not all targets support all 7533 types however.</p> 7534 7535 <pre> 7536 declare float @llvm.fabs.f32(float %Val) 7537 declare double @llvm.fabs.f64(double %Val) 7538 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val) 7539 declare fp128 @llvm.fabs.f128(fp128 %Val) 7540 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val) 7541 </pre> 7542 7543 <h5>Overview:</h5> 7544 <p>The '<tt>llvm.fabs.*</tt>' intrinsics return the absolute value of 7545 the operand.</p> 7546 7547 <h5>Arguments:</h5> 7548 <p>The argument and return value are floating point numbers of the same 7549 type.</p> 7550 7551 <h5>Semantics:</h5> 7552 <p>This function returns the same values as the libm <tt>fabs</tt> functions 7553 would, and handles error conditions in the same way.</p> 7554 7555 </div> 7556 7557 <!-- _______________________________________________________________________ --> 7558 <h4> 7559 <a name="int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a> 7560 </h4> 7561 7562 <div> 7563 7564 <h5>Syntax:</h5> 7565 <p>This is an overloaded intrinsic. You can use <tt>llvm.floor</tt> on any 7566 floating point or vector of floating point type. Not all targets support all 7567 types however.</p> 7568 7569 <pre> 7570 declare float @llvm.floor.f32(float %Val) 7571 declare double @llvm.floor.f64(double %Val) 7572 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val) 7573 declare fp128 @llvm.floor.f128(fp128 %Val) 7574 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val) 7575 </pre> 7576 7577 <h5>Overview:</h5> 7578 <p>The '<tt>llvm.floor.*</tt>' intrinsics return the floor of 7579 the operand.</p> 7580 7581 <h5>Arguments:</h5> 7582 <p>The argument and return value are floating point numbers of the same 7583 type.</p> 7584 7585 <h5>Semantics:</h5> 7586 <p>This function returns the same values as the libm <tt>floor</tt> functions 7587 would, and handles error conditions in the same way.</p> 7588 7589 </div> 7590 7591 </div> 7592 7593 <!-- ======================================================================= --> 7594 <h3> 7595 <a name="int_manip">Bit Manipulation Intrinsics</a> 7596 </h3> 7597 7598 <div> 7599 7600 <p>LLVM provides intrinsics for a few important bit manipulation operations. 7601 These allow efficient code generation for some algorithms.</p> 7602 7603 <!-- _______________________________________________________________________ --> 7604 <h4> 7605 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a> 7606 </h4> 7607 7608 <div> 7609 7610 <h5>Syntax:</h5> 7611 <p>This is an overloaded intrinsic function. You can use bswap on any integer 7612 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p> 7613 7614 <pre> 7615 declare i16 @llvm.bswap.i16(i16 <id>) 7616 declare i32 @llvm.bswap.i32(i32 <id>) 7617 declare i64 @llvm.bswap.i64(i64 <id>) 7618 </pre> 7619 7620 <h5>Overview:</h5> 7621 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer 7622 values with an even number of bytes (positive multiple of 16 bits). These 7623 are useful for performing operations on data that is not in the target's 7624 native byte order.</p> 7625 7626 <h5>Semantics:</h5> 7627 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high 7628 and low byte of the input i16 swapped. Similarly, 7629 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four 7630 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1, 7631 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order. 7632 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics 7633 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and 7634 more, respectively).</p> 7635 7636 </div> 7637 7638 <!-- _______________________________________________________________________ --> 7639 <h4> 7640 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a> 7641 </h4> 7642 7643 <div> 7644 7645 <h5>Syntax:</h5> 7646 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit 7647 width, or on any vector with integer elements. Not all targets support all 7648 bit widths or vector types, however.</p> 7649 7650 <pre> 7651 declare i8 @llvm.ctpop.i8(i8 <src>) 7652 declare i16 @llvm.ctpop.i16(i16 <src>) 7653 declare i32 @llvm.ctpop.i32(i32 <src>) 7654 declare i64 @llvm.ctpop.i64(i64 <src>) 7655 declare i256 @llvm.ctpop.i256(i256 <src>) 7656 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>) 7657 </pre> 7658 7659 <h5>Overview:</h5> 7660 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set 7661 in a value.</p> 7662 7663 <h5>Arguments:</h5> 7664 <p>The only argument is the value to be counted. The argument may be of any 7665 integer type, or a vector with integer elements. 7666 The return type must match the argument type.</p> 7667 7668 <h5>Semantics:</h5> 7669 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each 7670 element of a vector.</p> 7671 7672 </div> 7673 7674 <!-- _______________________________________________________________________ --> 7675 <h4> 7676 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a> 7677 </h4> 7678 7679 <div> 7680 7681 <h5>Syntax:</h5> 7682 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any 7683 integer bit width, or any vector whose elements are integers. Not all 7684 targets support all bit widths or vector types, however.</p> 7685 7686 <pre> 7687 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>) 7688 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>) 7689 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>) 7690 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>) 7691 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>) 7692 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>) 7693 </pre> 7694 7695 <h5>Overview:</h5> 7696 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of 7697 leading zeros in a variable.</p> 7698 7699 <h5>Arguments:</h5> 7700 <p>The first argument is the value to be counted. This argument may be of any 7701 integer type, or a vectory with integer element type. The return type 7702 must match the first argument type.</p> 7703 7704 <p>The second argument must be a constant and is a flag to indicate whether the 7705 intrinsic should ensure that a zero as the first argument produces a defined 7706 result. Historically some architectures did not provide a defined result for 7707 zero values as efficiently, and many algorithms are now predicated on 7708 avoiding zero-value inputs.</p> 7709 7710 <h5>Semantics:</h5> 7711 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) 7712 zeros in a variable, or within each element of the vector. 7713 If <tt>src == 0</tt> then the result is the size in bits of the type of 7714 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise. 7715 For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p> 7716 7717 </div> 7718 7719 <!-- _______________________________________________________________________ --> 7720 <h4> 7721 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a> 7722 </h4> 7723 7724 <div> 7725 7726 <h5>Syntax:</h5> 7727 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any 7728 integer bit width, or any vector of integer elements. Not all targets 7729 support all bit widths or vector types, however.</p> 7730 7731 <pre> 7732 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>) 7733 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>) 7734 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>) 7735 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>) 7736 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>) 7737 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>) 7738 </pre> 7739 7740 <h5>Overview:</h5> 7741 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of 7742 trailing zeros.</p> 7743 7744 <h5>Arguments:</h5> 7745 <p>The first argument is the value to be counted. This argument may be of any 7746 integer type, or a vectory with integer element type. The return type 7747 must match the first argument type.</p> 7748 7749 <p>The second argument must be a constant and is a flag to indicate whether the 7750 intrinsic should ensure that a zero as the first argument produces a defined 7751 result. Historically some architectures did not provide a defined result for 7752 zero values as efficiently, and many algorithms are now predicated on 7753 avoiding zero-value inputs.</p> 7754 7755 <h5>Semantics:</h5> 7756 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) 7757 zeros in a variable, or within each element of a vector. 7758 If <tt>src == 0</tt> then the result is the size in bits of the type of 7759 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise. 7760 For example, <tt>llvm.cttz(2) = 1</tt>.</p> 7761 7762 </div> 7763 7764 </div> 7765 7766 <!-- ======================================================================= --> 7767 <h3> 7768 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a> 7769 </h3> 7770 7771 <div> 7772 7773 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p> 7774 7775 <!-- _______________________________________________________________________ --> 7776 <h4> 7777 <a name="int_sadd_overflow"> 7778 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics 7779 </a> 7780 </h4> 7781 7782 <div> 7783 7784 <h5>Syntax:</h5> 7785 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt> 7786 on any integer bit width.</p> 7787 7788 <pre> 7789 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b) 7790 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b) 7791 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b) 7792 </pre> 7793 7794 <h5>Overview:</h5> 7795 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform 7796 a signed addition of the two arguments, and indicate whether an overflow 7797 occurred during the signed summation.</p> 7798 7799 <h5>Arguments:</h5> 7800 <p>The arguments (%a and %b) and the first element of the result structure may 7801 be of integer types of any bit width, but they must have the same bit 7802 width. The second element of the result structure must be of 7803 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will 7804 undergo signed addition.</p> 7805 7806 <h5>Semantics:</h5> 7807 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform 7808 a signed addition of the two variables. They return a structure — the 7809 first element of which is the signed summation, and the second element of 7810 which is a bit specifying if the signed summation resulted in an 7811 overflow.</p> 7812 7813 <h5>Examples:</h5> 7814 <pre> 7815 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b) 7816 %sum = extractvalue {i32, i1} %res, 0 7817 %obit = extractvalue {i32, i1} %res, 1 7818 br i1 %obit, label %overflow, label %normal 7819 </pre> 7820 7821 </div> 7822 7823 <!-- _______________________________________________________________________ --> 7824 <h4> 7825 <a name="int_uadd_overflow"> 7826 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics 7827 </a> 7828 </h4> 7829 7830 <div> 7831 7832 <h5>Syntax:</h5> 7833 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt> 7834 on any integer bit width.</p> 7835 7836 <pre> 7837 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b) 7838 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 7839 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b) 7840 </pre> 7841 7842 <h5>Overview:</h5> 7843 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform 7844 an unsigned addition of the two arguments, and indicate whether a carry 7845 occurred during the unsigned summation.</p> 7846 7847 <h5>Arguments:</h5> 7848 <p>The arguments (%a and %b) and the first element of the result structure may 7849 be of integer types of any bit width, but they must have the same bit 7850 width. The second element of the result structure must be of 7851 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will 7852 undergo unsigned addition.</p> 7853 7854 <h5>Semantics:</h5> 7855 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform 7856 an unsigned addition of the two arguments. They return a structure — 7857 the first element of which is the sum, and the second element of which is a 7858 bit specifying if the unsigned summation resulted in a carry.</p> 7859 7860 <h5>Examples:</h5> 7861 <pre> 7862 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b) 7863 %sum = extractvalue {i32, i1} %res, 0 7864 %obit = extractvalue {i32, i1} %res, 1 7865 br i1 %obit, label %carry, label %normal 7866 </pre> 7867 7868 </div> 7869 7870 <!-- _______________________________________________________________________ --> 7871 <h4> 7872 <a name="int_ssub_overflow"> 7873 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics 7874 </a> 7875 </h4> 7876 7877 <div> 7878 7879 <h5>Syntax:</h5> 7880 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt> 7881 on any integer bit width.</p> 7882 7883 <pre> 7884 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b) 7885 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b) 7886 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b) 7887 </pre> 7888 7889 <h5>Overview:</h5> 7890 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform 7891 a signed subtraction of the two arguments, and indicate whether an overflow 7892 occurred during the signed subtraction.</p> 7893 7894 <h5>Arguments:</h5> 7895 <p>The arguments (%a and %b) and the first element of the result structure may 7896 be of integer types of any bit width, but they must have the same bit 7897 width. The second element of the result structure must be of 7898 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will 7899 undergo signed subtraction.</p> 7900 7901 <h5>Semantics:</h5> 7902 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform 7903 a signed subtraction of the two arguments. They return a structure — 7904 the first element of which is the subtraction, and the second element of 7905 which is a bit specifying if the signed subtraction resulted in an 7906 overflow.</p> 7907 7908 <h5>Examples:</h5> 7909 <pre> 7910 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b) 7911 %sum = extractvalue {i32, i1} %res, 0 7912 %obit = extractvalue {i32, i1} %res, 1 7913 br i1 %obit, label %overflow, label %normal 7914 </pre> 7915 7916 </div> 7917 7918 <!-- _______________________________________________________________________ --> 7919 <h4> 7920 <a name="int_usub_overflow"> 7921 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics 7922 </a> 7923 </h4> 7924 7925 <div> 7926 7927 <h5>Syntax:</h5> 7928 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt> 7929 on any integer bit width.</p> 7930 7931 <pre> 7932 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b) 7933 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b) 7934 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b) 7935 </pre> 7936 7937 <h5>Overview:</h5> 7938 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform 7939 an unsigned subtraction of the two arguments, and indicate whether an 7940 overflow occurred during the unsigned subtraction.</p> 7941 7942 <h5>Arguments:</h5> 7943 <p>The arguments (%a and %b) and the first element of the result structure may 7944 be of integer types of any bit width, but they must have the same bit 7945 width. The second element of the result structure must be of 7946 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will 7947 undergo unsigned subtraction.</p> 7948 7949 <h5>Semantics:</h5> 7950 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform 7951 an unsigned subtraction of the two arguments. They return a structure — 7952 the first element of which is the subtraction, and the second element of 7953 which is a bit specifying if the unsigned subtraction resulted in an 7954 overflow.</p> 7955 7956 <h5>Examples:</h5> 7957 <pre> 7958 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b) 7959 %sum = extractvalue {i32, i1} %res, 0 7960 %obit = extractvalue {i32, i1} %res, 1 7961 br i1 %obit, label %overflow, label %normal 7962 </pre> 7963 7964 </div> 7965 7966 <!-- _______________________________________________________________________ --> 7967 <h4> 7968 <a name="int_smul_overflow"> 7969 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics 7970 </a> 7971 </h4> 7972 7973 <div> 7974 7975 <h5>Syntax:</h5> 7976 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt> 7977 on any integer bit width.</p> 7978 7979 <pre> 7980 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b) 7981 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b) 7982 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b) 7983 </pre> 7984 7985 <h5>Overview:</h5> 7986 7987 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform 7988 a signed multiplication of the two arguments, and indicate whether an 7989 overflow occurred during the signed multiplication.</p> 7990 7991 <h5>Arguments:</h5> 7992 <p>The arguments (%a and %b) and the first element of the result structure may 7993 be of integer types of any bit width, but they must have the same bit 7994 width. The second element of the result structure must be of 7995 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will 7996 undergo signed multiplication.</p> 7997 7998 <h5>Semantics:</h5> 7999 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform 8000 a signed multiplication of the two arguments. They return a structure — 8001 the first element of which is the multiplication, and the second element of 8002 which is a bit specifying if the signed multiplication resulted in an 8003 overflow.</p> 8004 8005 <h5>Examples:</h5> 8006 <pre> 8007 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b) 8008 %sum = extractvalue {i32, i1} %res, 0 8009 %obit = extractvalue {i32, i1} %res, 1 8010 br i1 %obit, label %overflow, label %normal 8011 </pre> 8012 8013 </div> 8014 8015 <!-- _______________________________________________________________________ --> 8016 <h4> 8017 <a name="int_umul_overflow"> 8018 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics 8019 </a> 8020 </h4> 8021 8022 <div> 8023 8024 <h5>Syntax:</h5> 8025 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt> 8026 on any integer bit width.</p> 8027 8028 <pre> 8029 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b) 8030 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b) 8031 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b) 8032 </pre> 8033 8034 <h5>Overview:</h5> 8035 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform 8036 a unsigned multiplication of the two arguments, and indicate whether an 8037 overflow occurred during the unsigned multiplication.</p> 8038 8039 <h5>Arguments:</h5> 8040 <p>The arguments (%a and %b) and the first element of the result structure may 8041 be of integer types of any bit width, but they must have the same bit 8042 width. The second element of the result structure must be of 8043 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will 8044 undergo unsigned multiplication.</p> 8045 8046 <h5>Semantics:</h5> 8047 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform 8048 an unsigned multiplication of the two arguments. They return a structure 8049 — the first element of which is the multiplication, and the second 8050 element of which is a bit specifying if the unsigned multiplication resulted 8051 in an overflow.</p> 8052 8053 <h5>Examples:</h5> 8054 <pre> 8055 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b) 8056 %sum = extractvalue {i32, i1} %res, 0 8057 %obit = extractvalue {i32, i1} %res, 1 8058 br i1 %obit, label %overflow, label %normal 8059 </pre> 8060 8061 </div> 8062 8063 </div> 8064 8065 <!-- ======================================================================= --> 8066 <h3> 8067 <a name="spec_arithmetic">Specialised Arithmetic Intrinsics</a> 8068 </h3> 8069 8070 <!-- _______________________________________________________________________ --> 8071 8072 <h4> 8073 <a name="fmuladd">'<tt>llvm.fmuladd.*</tt>' Intrinsic</a> 8074 </h4> 8075 8076 <div> 8077 8078 <h5>Syntax:</h5> 8079 <pre> 8080 declare float @llvm.fmuladd.f32(float %a, float %b, float %c) 8081 declare double @llvm.fmuladd.f64(double %a, double %b, double %c) 8082 </pre> 8083 8084 <h5>Overview:</h5> 8085 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsic functions represent multiply-add 8086 expressions that can be fused if the code generator determines that the fused 8087 expression would be legal and efficient.</p> 8088 8089 <h5>Arguments:</h5> 8090 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsics each take three arguments: two 8091 multiplicands, a and b, and an addend c.</p> 8092 8093 <h5>Semantics:</h5> 8094 <p>The expression:</p> 8095 <pre> 8096 %0 = call float @llvm.fmuladd.f32(%a, %b, %c) 8097 </pre> 8098 <p>is equivalent to the expression a * b + c, except that rounding will not be 8099 performed between the multiplication and addition steps if the code generator 8100 fuses the operations. Fusion is not guaranteed, even if the target platform 8101 supports it. If a fused multiply-add is required the corresponding llvm.fma.* 8102 intrinsic function should be used instead.</p> 8103 8104 <h5>Examples:</h5> 8105 <pre> 8106 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c 8107 </pre> 8108 8109 </div> 8110 8111 <!-- ======================================================================= --> 8112 <h3> 8113 <a name="int_fp16">Half Precision Floating Point Intrinsics</a> 8114 </h3> 8115 8116 <div> 8117 8118 <p>For most target platforms, half precision floating point is a storage-only 8119 format. This means that it is 8120 a dense encoding (in memory) but does not support computation in the 8121 format.</p> 8122 8123 <p>This means that code must first load the half-precision floating point 8124 value as an i16, then convert it to float with <a 8125 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>. 8126 Computation can then be performed on the float value (including extending to 8127 double etc). To store the value back to memory, it is first converted to 8128 float if needed, then converted to i16 with 8129 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then 8130 storing as an i16 value.</p> 8131 8132 <!-- _______________________________________________________________________ --> 8133 <h4> 8134 <a name="int_convert_to_fp16"> 8135 '<tt>llvm.convert.to.fp16</tt>' Intrinsic 8136 </a> 8137 </h4> 8138 8139 <div> 8140 8141 <h5>Syntax:</h5> 8142 <pre> 8143 declare i16 @llvm.convert.to.fp16(f32 %a) 8144 </pre> 8145 8146 <h5>Overview:</h5> 8147 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs 8148 a conversion from single precision floating point format to half precision 8149 floating point format.</p> 8150 8151 <h5>Arguments:</h5> 8152 <p>The intrinsic function contains single argument - the value to be 8153 converted.</p> 8154 8155 <h5>Semantics:</h5> 8156 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs 8157 a conversion from single precision floating point format to half precision 8158 floating point format. The return value is an <tt>i16</tt> which 8159 contains the converted number.</p> 8160 8161 <h5>Examples:</h5> 8162 <pre> 8163 %res = call i16 @llvm.convert.to.fp16(f32 %a) 8164 store i16 %res, i16* @x, align 2 8165 </pre> 8166 8167 </div> 8168 8169 <!-- _______________________________________________________________________ --> 8170 <h4> 8171 <a name="int_convert_from_fp16"> 8172 '<tt>llvm.convert.from.fp16</tt>' Intrinsic 8173 </a> 8174 </h4> 8175 8176 <div> 8177 8178 <h5>Syntax:</h5> 8179 <pre> 8180 declare f32 @llvm.convert.from.fp16(i16 %a) 8181 </pre> 8182 8183 <h5>Overview:</h5> 8184 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs 8185 a conversion from half precision floating point format to single precision 8186 floating point format.</p> 8187 8188 <h5>Arguments:</h5> 8189 <p>The intrinsic function contains single argument - the value to be 8190 converted.</p> 8191 8192 <h5>Semantics:</h5> 8193 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a 8194 conversion from half single precision floating point format to single 8195 precision floating point format. The input half-float value is represented by 8196 an <tt>i16</tt> value.</p> 8197 8198 <h5>Examples:</h5> 8199 <pre> 8200 %a = load i16* @x, align 2 8201 %res = call f32 @llvm.convert.from.fp16(i16 %a) 8202 </pre> 8203 8204 </div> 8205 8206 </div> 8207 8208 <!-- ======================================================================= --> 8209 <h3> 8210 <a name="int_debugger">Debugger Intrinsics</a> 8211 </h3> 8212 8213 <div> 8214 8215 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> 8216 prefix), are described in 8217 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source 8218 Level Debugging</a> document.</p> 8219 8220 </div> 8221 8222 <!-- ======================================================================= --> 8223 <h3> 8224 <a name="int_eh">Exception Handling Intrinsics</a> 8225 </h3> 8226 8227 <div> 8228 8229 <p>The LLVM exception handling intrinsics (which all start with 8230 <tt>llvm.eh.</tt> prefix), are described in 8231 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception 8232 Handling</a> document.</p> 8233 8234 </div> 8235 8236 <!-- ======================================================================= --> 8237 <h3> 8238 <a name="int_trampoline">Trampoline Intrinsics</a> 8239 </h3> 8240 8241 <div> 8242 8243 <p>These intrinsics make it possible to excise one parameter, marked with 8244 the <a href="#nest"><tt>nest</tt></a> attribute, from a function. 8245 The result is a callable 8246 function pointer lacking the nest parameter - the caller does not need to 8247 provide a value for it. Instead, the value to use is stored in advance in a 8248 "trampoline", a block of memory usually allocated on the stack, which also 8249 contains code to splice the nest value into the argument list. This is used 8250 to implement the GCC nested function address extension.</p> 8251 8252 <p>For example, if the function is 8253 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function 8254 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as 8255 follows:</p> 8256 8257 <pre class="doc_code"> 8258 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86 8259 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0 8260 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval) 8261 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1) 8262 %fp = bitcast i8* %p to i32 (i32, i32)* 8263 </pre> 8264 8265 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent 8266 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p> 8267 8268 <!-- _______________________________________________________________________ --> 8269 <h4> 8270 <a name="int_it"> 8271 '<tt>llvm.init.trampoline</tt>' Intrinsic 8272 </a> 8273 </h4> 8274 8275 <div> 8276 8277 <h5>Syntax:</h5> 8278 <pre> 8279 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>) 8280 </pre> 8281 8282 <h5>Overview:</h5> 8283 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code, 8284 turning it into a trampoline.</p> 8285 8286 <h5>Arguments:</h5> 8287 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all 8288 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and 8289 sufficiently aligned block of memory; this memory is written to by the 8290 intrinsic. Note that the size and the alignment are target-specific - LLVM 8291 currently provides no portable way of determining them, so a front-end that 8292 generates this intrinsic needs to have some target-specific knowledge. 8293 The <tt>func</tt> argument must hold a function bitcast to 8294 an <tt>i8*</tt>.</p> 8295 8296 <h5>Semantics:</h5> 8297 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target 8298 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be 8299 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer 8300 which can be <a href="#int_trampoline">bitcast (to a new function) and 8301 called</a>. The new function's signature is the same as that of 8302 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute 8303 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of 8304 pointer type. Calling the new function is equivalent to calling <tt>func</tt> 8305 with the same argument list, but with <tt>nval</tt> used for the missing 8306 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the 8307 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call 8308 to the returned function pointer is undefined.</p> 8309 </div> 8310 8311 <!-- _______________________________________________________________________ --> 8312 <h4> 8313 <a name="int_at"> 8314 '<tt>llvm.adjust.trampoline</tt>' Intrinsic 8315 </a> 8316 </h4> 8317 8318 <div> 8319 8320 <h5>Syntax:</h5> 8321 <pre> 8322 declare i8* @llvm.adjust.trampoline(i8* <tramp>) 8323 </pre> 8324 8325 <h5>Overview:</h5> 8326 <p>This performs any required machine-specific adjustment to the address of a 8327 trampoline (passed as <tt>tramp</tt>).</p> 8328 8329 <h5>Arguments:</h5> 8330 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code 8331 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt> 8332 </a>.</p> 8333 8334 <h5>Semantics:</h5> 8335 <p>On some architectures the address of the code to be executed needs to be 8336 different to the address where the trampoline is actually stored. This 8337 intrinsic returns the executable address corresponding to <tt>tramp</tt> 8338 after performing the required machine specific adjustments. 8339 The pointer returned can then be <a href="#int_trampoline"> bitcast and 8340 executed</a>. 8341 </p> 8342 8343 </div> 8344 8345 </div> 8346 8347 <!-- ======================================================================= --> 8348 <h3> 8349 <a name="int_memorymarkers">Memory Use Markers</a> 8350 </h3> 8351 8352 <div> 8353 8354 <p>This class of intrinsics exists to information about the lifetime of memory 8355 objects and ranges where variables are immutable.</p> 8356 8357 <!-- _______________________________________________________________________ --> 8358 <h4> 8359 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a> 8360 </h4> 8361 8362 <div> 8363 8364 <h5>Syntax:</h5> 8365 <pre> 8366 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>) 8367 </pre> 8368 8369 <h5>Overview:</h5> 8370 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory 8371 object's lifetime.</p> 8372 8373 <h5>Arguments:</h5> 8374 <p>The first argument is a constant integer representing the size of the 8375 object, or -1 if it is variable sized. The second argument is a pointer to 8376 the object.</p> 8377 8378 <h5>Semantics:</h5> 8379 <p>This intrinsic indicates that before this point in the code, the value of the 8380 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to 8381 never be used and has an undefined value. A load from the pointer that 8382 precedes this intrinsic can be replaced with 8383 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p> 8384 8385 </div> 8386 8387 <!-- _______________________________________________________________________ --> 8388 <h4> 8389 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a> 8390 </h4> 8391 8392 <div> 8393 8394 <h5>Syntax:</h5> 8395 <pre> 8396 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>) 8397 </pre> 8398 8399 <h5>Overview:</h5> 8400 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory 8401 object's lifetime.</p> 8402 8403 <h5>Arguments:</h5> 8404 <p>The first argument is a constant integer representing the size of the 8405 object, or -1 if it is variable sized. The second argument is a pointer to 8406 the object.</p> 8407 8408 <h5>Semantics:</h5> 8409 <p>This intrinsic indicates that after this point in the code, the value of the 8410 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to 8411 never be used and has an undefined value. Any stores into the memory object 8412 following this intrinsic may be removed as dead. 8413 8414 </div> 8415 8416 <!-- _______________________________________________________________________ --> 8417 <h4> 8418 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a> 8419 </h4> 8420 8421 <div> 8422 8423 <h5>Syntax:</h5> 8424 <pre> 8425 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>) 8426 </pre> 8427 8428 <h5>Overview:</h5> 8429 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of 8430 a memory object will not change.</p> 8431 8432 <h5>Arguments:</h5> 8433 <p>The first argument is a constant integer representing the size of the 8434 object, or -1 if it is variable sized. The second argument is a pointer to 8435 the object.</p> 8436 8437 <h5>Semantics:</h5> 8438 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses 8439 the return value, the referenced memory location is constant and 8440 unchanging.</p> 8441 8442 </div> 8443 8444 <!-- _______________________________________________________________________ --> 8445 <h4> 8446 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a> 8447 </h4> 8448 8449 <div> 8450 8451 <h5>Syntax:</h5> 8452 <pre> 8453 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>) 8454 </pre> 8455 8456 <h5>Overview:</h5> 8457 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of 8458 a memory object are mutable.</p> 8459 8460 <h5>Arguments:</h5> 8461 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic. 8462 The second argument is a constant integer representing the size of the 8463 object, or -1 if it is variable sized and the third argument is a pointer 8464 to the object.</p> 8465 8466 <h5>Semantics:</h5> 8467 <p>This intrinsic indicates that the memory is mutable again.</p> 8468 8469 </div> 8470 8471 </div> 8472 8473 <!-- ======================================================================= --> 8474 <h3> 8475 <a name="int_general">General Intrinsics</a> 8476 </h3> 8477 8478 <div> 8479 8480 <p>This class of intrinsics is designed to be generic and has no specific 8481 purpose.</p> 8482 8483 <!-- _______________________________________________________________________ --> 8484 <h4> 8485 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a> 8486 </h4> 8487 8488 <div> 8489 8490 <h5>Syntax:</h5> 8491 <pre> 8492 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>) 8493 </pre> 8494 8495 <h5>Overview:</h5> 8496 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p> 8497 8498 <h5>Arguments:</h5> 8499 <p>The first argument is a pointer to a value, the second is a pointer to a 8500 global string, the third is a pointer to a global string which is the source 8501 file name, and the last argument is the line number.</p> 8502 8503 <h5>Semantics:</h5> 8504 <p>This intrinsic allows annotation of local variables with arbitrary strings. 8505 This can be useful for special purpose optimizations that want to look for 8506 these annotations. These have no other defined use; they are ignored by code 8507 generation and optimization.</p> 8508 8509 </div> 8510 8511 <!-- _______________________________________________________________________ --> 8512 <h4> 8513 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a> 8514 </h4> 8515 8516 <div> 8517 8518 <h5>Syntax:</h5> 8519 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on 8520 any integer bit width.</p> 8521 8522 <pre> 8523 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>) 8524 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>) 8525 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>) 8526 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>) 8527 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>) 8528 </pre> 8529 8530 <h5>Overview:</h5> 8531 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p> 8532 8533 <h5>Arguments:</h5> 8534 <p>The first argument is an integer value (result of some expression), the 8535 second is a pointer to a global string, the third is a pointer to a global 8536 string which is the source file name, and the last argument is the line 8537 number. It returns the value of the first argument.</p> 8538 8539 <h5>Semantics:</h5> 8540 <p>This intrinsic allows annotations to be put on arbitrary expressions with 8541 arbitrary strings. This can be useful for special purpose optimizations that 8542 want to look for these annotations. These have no other defined use; they 8543 are ignored by code generation and optimization.</p> 8544 8545 </div> 8546 8547 <!-- _______________________________________________________________________ --> 8548 <h4> 8549 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a> 8550 </h4> 8551 8552 <div> 8553 8554 <h5>Syntax:</h5> 8555 <pre> 8556 declare void @llvm.trap() noreturn nounwind 8557 </pre> 8558 8559 <h5>Overview:</h5> 8560 <p>The '<tt>llvm.trap</tt>' intrinsic.</p> 8561 8562 <h5>Arguments:</h5> 8563 <p>None.</p> 8564 8565 <h5>Semantics:</h5> 8566 <p>This intrinsic is lowered to the target dependent trap instruction. If the 8567 target does not have a trap instruction, this intrinsic will be lowered to 8568 a call of the <tt>abort()</tt> function.</p> 8569 8570 </div> 8571 8572 <!-- _______________________________________________________________________ --> 8573 <h4> 8574 <a name="int_debugtrap">'<tt>llvm.debugtrap</tt>' Intrinsic</a> 8575 </h4> 8576 8577 <div> 8578 8579 <h5>Syntax:</h5> 8580 <pre> 8581 declare void @llvm.debugtrap() nounwind 8582 </pre> 8583 8584 <h5>Overview:</h5> 8585 <p>The '<tt>llvm.debugtrap</tt>' intrinsic.</p> 8586 8587 <h5>Arguments:</h5> 8588 <p>None.</p> 8589 8590 <h5>Semantics:</h5> 8591 <p>This intrinsic is lowered to code which is intended to cause an execution 8592 trap with the intention of requesting the attention of a debugger.</p> 8593 8594 </div> 8595 8596 <!-- _______________________________________________________________________ --> 8597 <h4> 8598 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a> 8599 </h4> 8600 8601 <div> 8602 8603 <h5>Syntax:</h5> 8604 <pre> 8605 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>) 8606 </pre> 8607 8608 <h5>Overview:</h5> 8609 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and 8610 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to 8611 ensure that it is placed on the stack before local variables.</p> 8612 8613 <h5>Arguments:</h5> 8614 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer 8615 arguments. The first argument is the value loaded from the stack 8616 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt> 8617 that has enough space to hold the value of the guard.</p> 8618 8619 <h5>Semantics:</h5> 8620 <p>This intrinsic causes the prologue/epilogue inserter to force the position of 8621 the <tt>AllocaInst</tt> stack slot to be before local variables on the 8622 stack. This is to ensure that if a local variable on the stack is 8623 overwritten, it will destroy the value of the guard. When the function exits, 8624 the guard on the stack is checked against the original guard. If they are 8625 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt> 8626 function.</p> 8627 8628 </div> 8629 8630 <!-- _______________________________________________________________________ --> 8631 <h4> 8632 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a> 8633 </h4> 8634 8635 <div> 8636 8637 <h5>Syntax:</h5> 8638 <pre> 8639 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>) 8640 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>) 8641 </pre> 8642 8643 <h5>Overview:</h5> 8644 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to 8645 the optimizers to determine at compile time whether a) an operation (like 8646 memcpy) will overflow a buffer that corresponds to an object, or b) that a 8647 runtime check for overflow isn't necessary. An object in this context means 8648 an allocation of a specific class, structure, array, or other object.</p> 8649 8650 <h5>Arguments:</h5> 8651 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first 8652 argument is a pointer to or into the <tt>object</tt>. The second argument 8653 is a boolean and determines whether <tt>llvm.objectsize</tt> returns 0 (if 8654 true) or -1 (if false) when the object size is unknown. 8655 The second argument only accepts constants.</p> 8656 8657 <h5>Semantics:</h5> 8658 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to a constant representing 8659 the size of the object concerned. If the size cannot be determined at compile 8660 time, <tt>llvm.objectsize</tt> returns <tt>i32/i64 -1 or 0</tt> 8661 (depending on the <tt>min</tt> argument).</p> 8662 8663 </div> 8664 <!-- _______________________________________________________________________ --> 8665 <h4> 8666 <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a> 8667 </h4> 8668 8669 <div> 8670 8671 <h5>Syntax:</h5> 8672 <pre> 8673 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>) 8674 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>) 8675 </pre> 8676 8677 <h5>Overview:</h5> 8678 <p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the 8679 most probable) value of <tt>val</tt>, which can be used by optimizers.</p> 8680 8681 <h5>Arguments:</h5> 8682 <p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first 8683 argument is a value. The second argument is an expected value, this needs to 8684 be a constant value, variables are not allowed.</p> 8685 8686 <h5>Semantics:</h5> 8687 <p>This intrinsic is lowered to the <tt>val</tt>.</p> 8688 </div> 8689 8690 <!-- _______________________________________________________________________ --> 8691 <h4> 8692 <a name="int_donothing">'<tt>llvm.donothing</tt>' Intrinsic</a> 8693 </h4> 8694 8695 <div> 8696 8697 <h5>Syntax:</h5> 8698 <pre> 8699 declare void @llvm.donothing() nounwind readnone 8700 </pre> 8701 8702 <h5>Overview:</h5> 8703 <p>The <tt>llvm.donothing</tt> intrinsic doesn't perform any operation. It's the 8704 only intrinsic that can be called with an invoke instruction.</p> 8705 8706 <h5>Arguments:</h5> 8707 <p>None.</p> 8708 8709 <h5>Semantics:</h5> 8710 <p>This intrinsic does nothing, and it's removed by optimizers and ignored by 8711 codegen.</p> 8712 </div> 8713 8714 </div> 8715 8716 </div> 8717 <!-- *********************************************************************** --> 8718 <hr> 8719 <address> 8720 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img 8721 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a> 8722 <a href="http://validator.w3.org/check/referer"><img 8723 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a> 8724 8725 <a href="mailto:sabre (a] nondot.org">Chris Lattner</a><br> 8726 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br> 8727 Last modified: $Date$ 8728 </address> 8729 8730 </body> 8731 </html> 8732
