1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <h1>LLVM Language Reference Manual</h1>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a>
25 <li><a href="#linkage_private">'<tt>private</tt>' Linkage</a></li>
26 <li><a href="#linkage_linker_private">'<tt>linker_private</tt>' Linkage</a></li>
27 <li><a href="#linkage_linker_private_weak">'<tt>linker_private_weak</tt>' Linkage</a></li>
28 <li><a href="#linkage_linker_private_weak_def_auto">'<tt>linker_private_weak_def_auto</tt>' Linkage</a></li>
29 <li><a href="#linkage_internal">'<tt>internal</tt>' Linkage</a></li>
30 <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
31 <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
32 <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
33 <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
34 <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
35 <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
36 <li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
37 <li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
38 <li><a href="#linkage_external">'<tt>external</tt>' Linkage</a></li>
39 <li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
40 <li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
43 <li><a href="#callingconv">Calling Conventions</a></li>
44 <li><a href="#namedtypes">Named Types</a></li>
45 <li><a href="#globalvars">Global Variables</a></li>
46 <li><a href="#functionstructure">Functions</a></li>
47 <li><a href="#aliasstructure">Aliases</a></li>
48 <li><a href="#namedmetadatastructure">Named Metadata</a></li>
49 <li><a href="#paramattrs">Parameter Attributes</a></li>
50 <li><a href="#fnattrs">Function Attributes</a></li>
51 <li><a href="#gc">Garbage Collector Names</a></li>
52 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
53 <li><a href="#datalayout">Data Layout</a></li>
54 <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
55 <li><a href="#volatile">Volatile Memory Accesses</a></li>
56 <li><a href="#memmodel">Memory Model for Concurrent Operations</a></li>
57 <li><a href="#ordering">Atomic Memory Ordering Constraints</a></li>
60 <li><a href="#typesystem">Type System</a>
62 <li><a href="#t_classifications">Type Classifications</a></li>
63 <li><a href="#t_primitive">Primitive Types</a>
65 <li><a href="#t_integer">Integer Type</a></li>
66 <li><a href="#t_floating">Floating Point Types</a></li>
67 <li><a href="#t_x86mmx">X86mmx Type</a></li>
68 <li><a href="#t_void">Void Type</a></li>
69 <li><a href="#t_label">Label Type</a></li>
70 <li><a href="#t_metadata">Metadata Type</a></li>
73 <li><a href="#t_derived">Derived Types</a>
75 <li><a href="#t_aggregate">Aggregate Types</a>
77 <li><a href="#t_array">Array Type</a></li>
78 <li><a href="#t_struct">Structure Type</a></li>
79 <li><a href="#t_opaque">Opaque Structure Types</a></li>
80 <li><a href="#t_vector">Vector Type</a></li>
83 <li><a href="#t_function">Function Type</a></li>
84 <li><a href="#t_pointer">Pointer Type</a></li>
89 <li><a href="#constants">Constants</a>
91 <li><a href="#simpleconstants">Simple Constants</a></li>
92 <li><a href="#complexconstants">Complex Constants</a></li>
93 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
94 <li><a href="#undefvalues">Undefined Values</a></li>
95 <li><a href="#poisonvalues">Poison Values</a></li>
96 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
97 <li><a href="#constantexprs">Constant Expressions</a></li>
100 <li><a href="#othervalues">Other Values</a>
102 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
103 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a>
105 <li><a href="#tbaa">'<tt>tbaa</tt>' Metadata</a></li>
106 <li><a href="#fpaccuracy">'<tt>fpaccuracy</tt>' Metadata</a></li>
111 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
113 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
114 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
115 Global Variable</a></li>
116 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
117 Global Variable</a></li>
118 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
119 Global Variable</a></li>
122 <li><a href="#instref">Instruction Reference</a>
124 <li><a href="#terminators">Terminator Instructions</a>
126 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
127 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
128 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
129 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
130 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
131 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
132 <li><a href="#i_resume">'<tt>resume</tt>' Instruction</a></li>
133 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
136 <li><a href="#binaryops">Binary Operations</a>
138 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
139 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
140 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
141 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
142 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
143 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
144 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
145 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
146 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
147 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
148 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
149 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
152 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
154 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
155 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
156 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
157 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
158 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
159 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
162 <li><a href="#vectorops">Vector Operations</a>
164 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
165 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
166 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
169 <li><a href="#aggregateops">Aggregate Operations</a>
171 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
172 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
175 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
177 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
178 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
179 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
180 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
181 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
182 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
183 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
186 <li><a href="#convertops">Conversion Operations</a>
188 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
189 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
190 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
191 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
192 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
193 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
194 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
195 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
196 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
197 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
198 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
199 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
202 <li><a href="#otherops">Other Operations</a>
204 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
205 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
206 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
207 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
208 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
209 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
210 <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
215 <li><a href="#intrinsics">Intrinsic Functions</a>
217 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
219 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
220 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
221 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
224 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
226 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
227 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
228 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
231 <li><a href="#int_codegen">Code Generator Intrinsics</a>
233 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
234 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
235 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
236 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
237 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
238 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
239 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
242 <li><a href="#int_libc">Standard C Library Intrinsics</a>
244 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
245 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
246 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
247 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
248 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
249 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
250 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
251 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
252 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
253 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
254 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
257 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
259 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
260 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
261 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
262 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
265 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
267 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
268 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
269 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
270 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
271 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
272 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
275 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
277 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
278 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
281 <li><a href="#int_debugger">Debugger intrinsics</a></li>
282 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
283 <li><a href="#int_trampoline">Trampoline Intrinsics</a>
285 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
286 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
289 <li><a href="#int_memorymarkers">Memory Use Markers</a>
291 <li><a href="#int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a></li>
292 <li><a href="#int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a></li>
293 <li><a href="#int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a></li>
294 <li><a href="#int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a></li>
297 <li><a href="#int_general">General intrinsics</a>
299 <li><a href="#int_var_annotation">
300 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
301 <li><a href="#int_annotation">
302 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
303 <li><a href="#int_trap">
304 '<tt>llvm.trap</tt>' Intrinsic</a></li>
305 <li><a href="#int_stackprotector">
306 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
307 <li><a href="#int_objectsize">
308 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
309 <li><a href="#int_expect">
310 '<tt>llvm.expect</tt>' Intrinsic</a></li>
317 <div class="doc_author">
318 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
319 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
322 <!-- *********************************************************************** -->
323 <h2><a name="abstract">Abstract</a></h2>
324 <!-- *********************************************************************** -->
328 <p>This document is a reference manual for the LLVM assembly language. LLVM is
329 a Static Single Assignment (SSA) based representation that provides type
330 safety, low-level operations, flexibility, and the capability of representing
331 'all' high-level languages cleanly. It is the common code representation
332 used throughout all phases of the LLVM compilation strategy.</p>
336 <!-- *********************************************************************** -->
337 <h2><a name="introduction">Introduction</a></h2>
338 <!-- *********************************************************************** -->
342 <p>The LLVM code representation is designed to be used in three different forms:
343 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
344 for fast loading by a Just-In-Time compiler), and as a human readable
345 assembly language representation. This allows LLVM to provide a powerful
346 intermediate representation for efficient compiler transformations and
347 analysis, while providing a natural means to debug and visualize the
348 transformations. The three different forms of LLVM are all equivalent. This
349 document describes the human readable representation and notation.</p>
351 <p>The LLVM representation aims to be light-weight and low-level while being
352 expressive, typed, and extensible at the same time. It aims to be a
353 "universal IR" of sorts, by being at a low enough level that high-level ideas
354 may be cleanly mapped to it (similar to how microprocessors are "universal
355 IR's", allowing many source languages to be mapped to them). By providing
356 type information, LLVM can be used as the target of optimizations: for
357 example, through pointer analysis, it can be proven that a C automatic
358 variable is never accessed outside of the current function, allowing it to
359 be promoted to a simple SSA value instead of a memory location.</p>
361 <!-- _______________________________________________________________________ -->
363 <a name="wellformed">Well-Formedness</a>
368 <p>It is important to note that this document describes 'well formed' LLVM
369 assembly language. There is a difference between what the parser accepts and
370 what is considered 'well formed'. For example, the following instruction is
371 syntactically okay, but not well formed:</p>
373 <pre class="doc_code">
374 %x = <a href="#i_add">add</a> i32 1, %x
377 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
378 LLVM infrastructure provides a verification pass that may be used to verify
379 that an LLVM module is well formed. This pass is automatically run by the
380 parser after parsing input assembly and by the optimizer before it outputs
381 bitcode. The violations pointed out by the verifier pass indicate bugs in
382 transformation passes or input to the parser.</p>
388 <!-- Describe the typesetting conventions here. -->
390 <!-- *********************************************************************** -->
391 <h2><a name="identifiers">Identifiers</a></h2>
392 <!-- *********************************************************************** -->
396 <p>LLVM identifiers come in two basic types: global and local. Global
397 identifiers (functions, global variables) begin with the <tt>'@'</tt>
398 character. Local identifiers (register names, types) begin with
399 the <tt>'%'</tt> character. Additionally, there are three different formats
400 for identifiers, for different purposes:</p>
403 <li>Named values are represented as a string of characters with their prefix.
404 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
405 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
406 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
407 other characters in their names can be surrounded with quotes. Special
408 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
409 ASCII code for the character in hexadecimal. In this way, any character
410 can be used in a name value, even quotes themselves.</li>
412 <li>Unnamed values are represented as an unsigned numeric value with their
413 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
415 <li>Constants, which are described in a <a href="#constants">section about
416 constants</a>, below.</li>
419 <p>LLVM requires that values start with a prefix for two reasons: Compilers
420 don't need to worry about name clashes with reserved words, and the set of
421 reserved words may be expanded in the future without penalty. Additionally,
422 unnamed identifiers allow a compiler to quickly come up with a temporary
423 variable without having to avoid symbol table conflicts.</p>
425 <p>Reserved words in LLVM are very similar to reserved words in other
426 languages. There are keywords for different opcodes
427 ('<tt><a href="#i_add">add</a></tt>',
428 '<tt><a href="#i_bitcast">bitcast</a></tt>',
429 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
430 ('<tt><a href="#t_void">void</a></tt>',
431 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
432 reserved words cannot conflict with variable names, because none of them
433 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
435 <p>Here is an example of LLVM code to multiply the integer variable
436 '<tt>%X</tt>' by 8:</p>
440 <pre class="doc_code">
441 %result = <a href="#i_mul">mul</a> i32 %X, 8
444 <p>After strength reduction:</p>
446 <pre class="doc_code">
447 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
450 <p>And the hard way:</p>
452 <pre class="doc_code">
453 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
454 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
455 %result = <a href="#i_add">add</a> i32 %1, %1
458 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
459 lexical features of LLVM:</p>
462 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
465 <li>Unnamed temporaries are created when the result of a computation is not
466 assigned to a named value.</li>
468 <li>Unnamed temporaries are numbered sequentially</li>
471 <p>It also shows a convention that we follow in this document. When
472 demonstrating instructions, we will follow an instruction with a comment that
473 defines the type and name of value produced. Comments are shown in italic
478 <!-- *********************************************************************** -->
479 <h2><a name="highlevel">High Level Structure</a></h2>
480 <!-- *********************************************************************** -->
482 <!-- ======================================================================= -->
484 <a name="modulestructure">Module Structure</a>
489 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
490 of the input programs. Each module consists of functions, global variables,
491 and symbol table entries. Modules may be combined together with the LLVM
492 linker, which merges function (and global variable) definitions, resolves
493 forward declarations, and merges symbol table entries. Here is an example of
494 the "hello world" module:</p>
496 <pre class="doc_code">
497 <i>; Declare the string constant as a global constant.</i>
498 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
500 <i>; External declaration of the puts function</i>
501 <a href="#functionstructure">declare</a> i32 @puts(i8*) <i>; i32 (i8*)* </i>
503 <i>; Definition of main function</i>
504 define i32 @main() { <i>; i32()* </i>
505 <i>; Convert [13 x i8]* to i8 *...</i>
506 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8*</i>
508 <i>; Call puts function to write out the string to stdout.</i>
509 <a href="#i_call">call</a> i32 @puts(i8* %cast210) <i>; i32</i>
510 <a href="#i_ret">ret</a> i32 0
513 <i>; Named metadata</i>
514 !1 = metadata !{i32 41}
518 <p>This example is made up of a <a href="#globalvars">global variable</a> named
519 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
520 a <a href="#functionstructure">function definition</a> for
521 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
524 <p>In general, a module is made up of a list of global values, where both
525 functions and global variables are global values. Global values are
526 represented by a pointer to a memory location (in this case, a pointer to an
527 array of char, and a pointer to a function), and have one of the
528 following <a href="#linkage">linkage types</a>.</p>
532 <!-- ======================================================================= -->
534 <a name="linkage">Linkage Types</a>
539 <p>All Global Variables and Functions have one of the following types of
543 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
544 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
545 by objects in the current module. In particular, linking code into a
546 module with an private global value may cause the private to be renamed as
547 necessary to avoid collisions. Because the symbol is private to the
548 module, all references can be updated. This doesn't show up in any symbol
549 table in the object file.</dd>
551 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
552 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
553 assembler and evaluated by the linker. Unlike normal strong symbols, they
554 are removed by the linker from the final linked image (executable or
555 dynamic library).</dd>
557 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
558 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
559 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
560 linker. The symbols are removed by the linker from the final linked image
561 (executable or dynamic library).</dd>
563 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
564 <dd>Similar to "<tt>linker_private_weak</tt>", but it's known that the address
565 of the object is not taken. For instance, functions that had an inline
566 definition, but the compiler decided not to inline it. Note,
567 unlike <tt>linker_private</tt> and <tt>linker_private_weak</tt>,
568 <tt>linker_private_weak_def_auto</tt> may have only <tt>default</tt>
569 visibility. The symbols are removed by the linker from the final linked
570 image (executable or dynamic library).</dd>
572 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
573 <dd>Similar to private, but the value shows as a local symbol
574 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
575 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
577 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
578 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
579 into the object file corresponding to the LLVM module. They exist to
580 allow inlining and other optimizations to take place given knowledge of
581 the definition of the global, which is known to be somewhere outside the
582 module. Globals with <tt>available_externally</tt> linkage are allowed to
583 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
584 This linkage type is only allowed on definitions, not declarations.</dd>
586 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
587 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
588 the same name when linkage occurs. This can be used to implement
589 some forms of inline functions, templates, or other code which must be
590 generated in each translation unit that uses it, but where the body may
591 be overridden with a more definitive definition later. Unreferenced
592 <tt>linkonce</tt> globals are allowed to be discarded. Note that
593 <tt>linkonce</tt> linkage does not actually allow the optimizer to
594 inline the body of this function into callers because it doesn't know if
595 this definition of the function is the definitive definition within the
596 program or whether it will be overridden by a stronger definition.
597 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
600 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
601 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
602 <tt>linkonce</tt> linkage, except that unreferenced globals with
603 <tt>weak</tt> linkage may not be discarded. This is used for globals that
604 are declared "weak" in C source code.</dd>
606 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
607 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
608 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
610 Symbols with "<tt>common</tt>" linkage are merged in the same way as
611 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
612 <tt>common</tt> symbols may not have an explicit section,
613 must have a zero initializer, and may not be marked '<a
614 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
615 have common linkage.</dd>
618 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
619 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
620 pointer to array type. When two global variables with appending linkage
621 are linked together, the two global arrays are appended together. This is
622 the LLVM, typesafe, equivalent of having the system linker append together
623 "sections" with identical names when .o files are linked.</dd>
625 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
626 <dd>The semantics of this linkage follow the ELF object file model: the symbol
627 is weak until linked, if not linked, the symbol becomes null instead of
628 being an undefined reference.</dd>
630 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
631 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
632 <dd>Some languages allow differing globals to be merged, such as two functions
633 with different semantics. Other languages, such as <tt>C++</tt>, ensure
634 that only equivalent globals are ever merged (the "one definition rule"
635 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
636 and <tt>weak_odr</tt> linkage types to indicate that the global will only
637 be merged with equivalent globals. These linkage types are otherwise the
638 same as their non-<tt>odr</tt> versions.</dd>
640 <dt><tt><b><a name="linkage_external">external</a></b></tt></dt>
641 <dd>If none of the above identifiers are used, the global is externally
642 visible, meaning that it participates in linkage and can be used to
643 resolve external symbol references.</dd>
646 <p>The next two types of linkage are targeted for Microsoft Windows platform
647 only. They are designed to support importing (exporting) symbols from (to)
648 DLLs (Dynamic Link Libraries).</p>
651 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
652 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
653 or variable via a global pointer to a pointer that is set up by the DLL
654 exporting the symbol. On Microsoft Windows targets, the pointer name is
655 formed by combining <code>__imp_</code> and the function or variable
658 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
659 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
660 pointer to a pointer in a DLL, so that it can be referenced with the
661 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
662 name is formed by combining <code>__imp_</code> and the function or
666 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
667 another module defined a "<tt>.LC0</tt>" variable and was linked with this
668 one, one of the two would be renamed, preventing a collision. Since
669 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
670 declarations), they are accessible outside of the current module.</p>
672 <p>It is illegal for a function <i>declaration</i> to have any linkage type
673 other than <tt>external</tt>, <tt>dllimport</tt>
674 or <tt>extern_weak</tt>.</p>
676 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
677 or <tt>weak_odr</tt> linkages.</p>
681 <!-- ======================================================================= -->
683 <a name="callingconv">Calling Conventions</a>
688 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
689 and <a href="#i_invoke">invokes</a> can all have an optional calling
690 convention specified for the call. The calling convention of any pair of
691 dynamic caller/callee must match, or the behavior of the program is
692 undefined. The following calling conventions are supported by LLVM, and more
693 may be added in the future:</p>
696 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
697 <dd>This calling convention (the default if no other calling convention is
698 specified) matches the target C calling conventions. This calling
699 convention supports varargs function calls and tolerates some mismatch in
700 the declared prototype and implemented declaration of the function (as
703 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
704 <dd>This calling convention attempts to make calls as fast as possible
705 (e.g. by passing things in registers). This calling convention allows the
706 target to use whatever tricks it wants to produce fast code for the
707 target, without having to conform to an externally specified ABI
708 (Application Binary Interface).
709 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
710 when this or the GHC convention is used.</a> This calling convention
711 does not support varargs and requires the prototype of all callees to
712 exactly match the prototype of the function definition.</dd>
714 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
715 <dd>This calling convention attempts to make code in the caller as efficient
716 as possible under the assumption that the call is not commonly executed.
717 As such, these calls often preserve all registers so that the call does
718 not break any live ranges in the caller side. This calling convention
719 does not support varargs and requires the prototype of all callees to
720 exactly match the prototype of the function definition.</dd>
722 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
723 <dd>This calling convention has been implemented specifically for use by the
724 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
725 It passes everything in registers, going to extremes to achieve this by
726 disabling callee save registers. This calling convention should not be
727 used lightly but only for specific situations such as an alternative to
728 the <em>register pinning</em> performance technique often used when
729 implementing functional programming languages.At the moment only X86
730 supports this convention and it has the following limitations:
732 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
733 floating point types are supported.</li>
734 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
735 6 floating point parameters.</li>
737 This calling convention supports
738 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
739 requires both the caller and callee are using it.
742 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
743 <dd>Any calling convention may be specified by number, allowing
744 target-specific calling conventions to be used. Target specific calling
745 conventions start at 64.</dd>
748 <p>More calling conventions can be added/defined on an as-needed basis, to
749 support Pascal conventions or any other well-known target-independent
754 <!-- ======================================================================= -->
756 <a name="visibility">Visibility Styles</a>
761 <p>All Global Variables and Functions have one of the following visibility
765 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
766 <dd>On targets that use the ELF object file format, default visibility means
767 that the declaration is visible to other modules and, in shared libraries,
768 means that the declared entity may be overridden. On Darwin, default
769 visibility means that the declaration is visible to other modules. Default
770 visibility corresponds to "external linkage" in the language.</dd>
772 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
773 <dd>Two declarations of an object with hidden visibility refer to the same
774 object if they are in the same shared object. Usually, hidden visibility
775 indicates that the symbol will not be placed into the dynamic symbol
776 table, so no other module (executable or shared library) can reference it
779 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
780 <dd>On ELF, protected visibility indicates that the symbol will be placed in
781 the dynamic symbol table, but that references within the defining module
782 will bind to the local symbol. That is, the symbol cannot be overridden by
788 <!-- ======================================================================= -->
790 <a name="namedtypes">Named Types</a>
795 <p>LLVM IR allows you to specify name aliases for certain types. This can make
796 it easier to read the IR and make the IR more condensed (particularly when
797 recursive types are involved). An example of a name specification is:</p>
799 <pre class="doc_code">
800 %mytype = type { %mytype*, i32 }
803 <p>You may give a name to any <a href="#typesystem">type</a> except
804 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
805 is expected with the syntax "%mytype".</p>
807 <p>Note that type names are aliases for the structural type that they indicate,
808 and that you can therefore specify multiple names for the same type. This
809 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
810 uses structural typing, the name is not part of the type. When printing out
811 LLVM IR, the printer will pick <em>one name</em> to render all types of a
812 particular shape. This means that if you have code where two different
813 source types end up having the same LLVM type, that the dumper will sometimes
814 print the "wrong" or unexpected type. This is an important design point and
815 isn't going to change.</p>
819 <!-- ======================================================================= -->
821 <a name="globalvars">Global Variables</a>
826 <p>Global variables define regions of memory allocated at compilation time
827 instead of run-time. Global variables may optionally be initialized, may
828 have an explicit section to be placed in, and may have an optional explicit
829 alignment specified. A variable may be defined as "thread_local", which
830 means that it will not be shared by threads (each thread will have a
831 separated copy of the variable). A variable may be defined as a global
832 "constant," which indicates that the contents of the variable
833 will <b>never</b> be modified (enabling better optimization, allowing the
834 global data to be placed in the read-only section of an executable, etc).
835 Note that variables that need runtime initialization cannot be marked
836 "constant" as there is a store to the variable.</p>
838 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
839 constant, even if the final definition of the global is not. This capability
840 can be used to enable slightly better optimization of the program, but
841 requires the language definition to guarantee that optimizations based on the
842 'constantness' are valid for the translation units that do not include the
845 <p>As SSA values, global variables define pointer values that are in scope
846 (i.e. they dominate) all basic blocks in the program. Global variables
847 always define a pointer to their "content" type because they describe a
848 region of memory, and all memory objects in LLVM are accessed through
851 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
852 that the address is not significant, only the content. Constants marked
853 like this can be merged with other constants if they have the same
854 initializer. Note that a constant with significant address <em>can</em>
855 be merged with a <tt>unnamed_addr</tt> constant, the result being a
856 constant whose address is significant.</p>
858 <p>A global variable may be declared to reside in a target-specific numbered
859 address space. For targets that support them, address spaces may affect how
860 optimizations are performed and/or what target instructions are used to
861 access the variable. The default address space is zero. The address space
862 qualifier must precede any other attributes.</p>
864 <p>LLVM allows an explicit section to be specified for globals. If the target
865 supports it, it will emit globals to the section specified.</p>
867 <p>An explicit alignment may be specified for a global, which must be a power
868 of 2. If not present, or if the alignment is set to zero, the alignment of
869 the global is set by the target to whatever it feels convenient. If an
870 explicit alignment is specified, the global is forced to have exactly that
871 alignment. Targets and optimizers are not allowed to over-align the global
872 if the global has an assigned section. In this case, the extra alignment
873 could be observable: for example, code could assume that the globals are
874 densely packed in their section and try to iterate over them as an array,
875 alignment padding would break this iteration.</p>
877 <p>For example, the following defines a global in a numbered address space with
878 an initializer, section, and alignment:</p>
880 <pre class="doc_code">
881 @G = addrspace(5) constant float 1.0, section "foo", align 4
887 <!-- ======================================================================= -->
889 <a name="functionstructure">Functions</a>
894 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
895 optional <a href="#linkage">linkage type</a>, an optional
896 <a href="#visibility">visibility style</a>, an optional
897 <a href="#callingconv">calling convention</a>,
898 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
899 <a href="#paramattrs">parameter attribute</a> for the return type, a function
900 name, a (possibly empty) argument list (each with optional
901 <a href="#paramattrs">parameter attributes</a>), optional
902 <a href="#fnattrs">function attributes</a>, an optional section, an optional
903 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
904 curly brace, a list of basic blocks, and a closing curly brace.</p>
906 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
907 optional <a href="#linkage">linkage type</a>, an optional
908 <a href="#visibility">visibility style</a>, an optional
909 <a href="#callingconv">calling convention</a>,
910 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
911 <a href="#paramattrs">parameter attribute</a> for the return type, a function
912 name, a possibly empty list of arguments, an optional alignment, and an
913 optional <a href="#gc">garbage collector name</a>.</p>
915 <p>A function definition contains a list of basic blocks, forming the CFG
916 (Control Flow Graph) for the function. Each basic block may optionally start
917 with a label (giving the basic block a symbol table entry), contains a list
918 of instructions, and ends with a <a href="#terminators">terminator</a>
919 instruction (such as a branch or function return).</p>
921 <p>The first basic block in a function is special in two ways: it is immediately
922 executed on entrance to the function, and it is not allowed to have
923 predecessor basic blocks (i.e. there can not be any branches to the entry
924 block of a function). Because the block can have no predecessors, it also
925 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
927 <p>LLVM allows an explicit section to be specified for functions. If the target
928 supports it, it will emit functions to the section specified.</p>
930 <p>An explicit alignment may be specified for a function. If not present, or if
931 the alignment is set to zero, the alignment of the function is set by the
932 target to whatever it feels convenient. If an explicit alignment is
933 specified, the function is forced to have at least that much alignment. All
934 alignments must be a power of 2.</p>
936 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
937 be significant and two identical functions can be merged.</p>
940 <pre class="doc_code">
941 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
942 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
943 <ResultType> @<FunctionName> ([argument list])
944 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
945 [<a href="#gc">gc</a>] { ... }
950 <!-- ======================================================================= -->
952 <a name="aliasstructure">Aliases</a>
957 <p>Aliases act as "second name" for the aliasee value (which can be either
958 function, global variable, another alias or bitcast of global value). Aliases
959 may have an optional <a href="#linkage">linkage type</a>, and an
960 optional <a href="#visibility">visibility style</a>.</p>
963 <pre class="doc_code">
964 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
969 <!-- ======================================================================= -->
971 <a name="namedmetadatastructure">Named Metadata</a>
976 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
977 nodes</a> (but not metadata strings) are the only valid operands for
978 a named metadata.</p>
981 <pre class="doc_code">
982 ; Some unnamed metadata nodes, which are referenced by the named metadata.
983 !0 = metadata !{metadata !"zero"}
984 !1 = metadata !{metadata !"one"}
985 !2 = metadata !{metadata !"two"}
987 !name = !{!0, !1, !2}
992 <!-- ======================================================================= -->
994 <a name="paramattrs">Parameter Attributes</a>
999 <p>The return type and each parameter of a function type may have a set of
1000 <i>parameter attributes</i> associated with them. Parameter attributes are
1001 used to communicate additional information about the result or parameters of
1002 a function. Parameter attributes are considered to be part of the function,
1003 not of the function type, so functions with different parameter attributes
1004 can have the same function type.</p>
1006 <p>Parameter attributes are simple keywords that follow the type specified. If
1007 multiple parameter attributes are needed, they are space separated. For
1010 <pre class="doc_code">
1011 declare i32 @printf(i8* noalias nocapture, ...)
1012 declare i32 @atoi(i8 zeroext)
1013 declare signext i8 @returns_signed_char()
1016 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1017 <tt>readonly</tt>) come immediately after the argument list.</p>
1019 <p>Currently, only the following parameter attributes are defined:</p>
1022 <dt><tt><b>zeroext</b></tt></dt>
1023 <dd>This indicates to the code generator that the parameter or return value
1024 should be zero-extended to the extent required by the target's ABI (which
1025 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1026 parameter) or the callee (for a return value).</dd>
1028 <dt><tt><b>signext</b></tt></dt>
1029 <dd>This indicates to the code generator that the parameter or return value
1030 should be sign-extended to the extent required by the target's ABI (which
1031 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1034 <dt><tt><b>inreg</b></tt></dt>
1035 <dd>This indicates that this parameter or return value should be treated in a
1036 special target-dependent fashion during while emitting code for a function
1037 call or return (usually, by putting it in a register as opposed to memory,
1038 though some targets use it to distinguish between two different kinds of
1039 registers). Use of this attribute is target-specific.</dd>
1041 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1042 <dd><p>This indicates that the pointer parameter should really be passed by
1043 value to the function. The attribute implies that a hidden copy of the
1045 is made between the caller and the callee, so the callee is unable to
1046 modify the value in the callee. This attribute is only valid on LLVM
1047 pointer arguments. It is generally used to pass structs and arrays by
1048 value, but is also valid on pointers to scalars. The copy is considered
1049 to belong to the caller not the callee (for example,
1050 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1051 <tt>byval</tt> parameters). This is not a valid attribute for return
1054 <p>The byval attribute also supports specifying an alignment with
1055 the align attribute. It indicates the alignment of the stack slot to
1056 form and the known alignment of the pointer specified to the call site. If
1057 the alignment is not specified, then the code generator makes a
1058 target-specific assumption.</p></dd>
1060 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1061 <dd>This indicates that the pointer parameter specifies the address of a
1062 structure that is the return value of the function in the source program.
1063 This pointer must be guaranteed by the caller to be valid: loads and
1064 stores to the structure may be assumed by the callee to not to trap. This
1065 may only be applied to the first parameter. This is not a valid attribute
1066 for return values. </dd>
1068 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1069 <dd>This indicates that pointer values
1070 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1071 value do not alias pointer values which are not <i>based</i> on it,
1072 ignoring certain "irrelevant" dependencies.
1073 For a call to the parent function, dependencies between memory
1074 references from before or after the call and from those during the call
1075 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1076 return value used in that call.
1077 The caller shares the responsibility with the callee for ensuring that
1078 these requirements are met.
1079 For further details, please see the discussion of the NoAlias response in
1080 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1082 Note that this definition of <tt>noalias</tt> is intentionally
1083 similar to the definition of <tt>restrict</tt> in C99 for function
1084 arguments, though it is slightly weaker.
1086 For function return values, C99's <tt>restrict</tt> is not meaningful,
1087 while LLVM's <tt>noalias</tt> is.
1090 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1091 <dd>This indicates that the callee does not make any copies of the pointer
1092 that outlive the callee itself. This is not a valid attribute for return
1095 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1096 <dd>This indicates that the pointer parameter can be excised using the
1097 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1098 attribute for return values.</dd>
1103 <!-- ======================================================================= -->
1105 <a name="gc">Garbage Collector Names</a>
1110 <p>Each function may specify a garbage collector name, which is simply a
1113 <pre class="doc_code">
1114 define void @f() gc "name" { ... }
1117 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1118 collector which will cause the compiler to alter its output in order to
1119 support the named garbage collection algorithm.</p>
1123 <!-- ======================================================================= -->
1125 <a name="fnattrs">Function Attributes</a>
1130 <p>Function attributes are set to communicate additional information about a
1131 function. Function attributes are considered to be part of the function, not
1132 of the function type, so functions with different parameter attributes can
1133 have the same function type.</p>
1135 <p>Function attributes are simple keywords that follow the type specified. If
1136 multiple attributes are needed, they are space separated. For example:</p>
1138 <pre class="doc_code">
1139 define void @f() noinline { ... }
1140 define void @f() alwaysinline { ... }
1141 define void @f() alwaysinline optsize { ... }
1142 define void @f() optsize { ... }
1146 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1147 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1148 the backend should forcibly align the stack pointer. Specify the
1149 desired alignment, which must be a power of two, in parentheses.
1151 <dt><tt><b>alwaysinline</b></tt></dt>
1152 <dd>This attribute indicates that the inliner should attempt to inline this
1153 function into callers whenever possible, ignoring any active inlining size
1154 threshold for this caller.</dd>
1156 <dt><tt><b>nonlazybind</b></tt></dt>
1157 <dd>This attribute suppresses lazy symbol binding for the function. This
1158 may make calls to the function faster, at the cost of extra program
1159 startup time if the function is not called during program startup.</dd>
1161 <dt><tt><b>inlinehint</b></tt></dt>
1162 <dd>This attribute indicates that the source code contained a hint that inlining
1163 this function is desirable (such as the "inline" keyword in C/C++). It
1164 is just a hint; it imposes no requirements on the inliner.</dd>
1166 <dt><tt><b>naked</b></tt></dt>
1167 <dd>This attribute disables prologue / epilogue emission for the function.
1168 This can have very system-specific consequences.</dd>
1170 <dt><tt><b>noimplicitfloat</b></tt></dt>
1171 <dd>This attributes disables implicit floating point instructions.</dd>
1173 <dt><tt><b>noinline</b></tt></dt>
1174 <dd>This attribute indicates that the inliner should never inline this
1175 function in any situation. This attribute may not be used together with
1176 the <tt>alwaysinline</tt> attribute.</dd>
1178 <dt><tt><b>noredzone</b></tt></dt>
1179 <dd>This attribute indicates that the code generator should not use a red
1180 zone, even if the target-specific ABI normally permits it.</dd>
1182 <dt><tt><b>noreturn</b></tt></dt>
1183 <dd>This function attribute indicates that the function never returns
1184 normally. This produces undefined behavior at runtime if the function
1185 ever does dynamically return.</dd>
1187 <dt><tt><b>nounwind</b></tt></dt>
1188 <dd>This function attribute indicates that the function never returns with an
1189 unwind or exceptional control flow. If the function does unwind, its
1190 runtime behavior is undefined.</dd>
1192 <dt><tt><b>optsize</b></tt></dt>
1193 <dd>This attribute suggests that optimization passes and code generator passes
1194 make choices that keep the code size of this function low, and otherwise
1195 do optimizations specifically to reduce code size.</dd>
1197 <dt><tt><b>readnone</b></tt></dt>
1198 <dd>This attribute indicates that the function computes its result (or decides
1199 to unwind an exception) based strictly on its arguments, without
1200 dereferencing any pointer arguments or otherwise accessing any mutable
1201 state (e.g. memory, control registers, etc) visible to caller functions.
1202 It does not write through any pointer arguments
1203 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1204 changes any state visible to callers. This means that it cannot unwind
1205 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1206 could use the <tt>unwind</tt> instruction.</dd>
1208 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1209 <dd>This attribute indicates that the function does not write through any
1210 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1211 arguments) or otherwise modify any state (e.g. memory, control registers,
1212 etc) visible to caller functions. It may dereference pointer arguments
1213 and read state that may be set in the caller. A readonly function always
1214 returns the same value (or unwinds an exception identically) when called
1215 with the same set of arguments and global state. It cannot unwind an
1216 exception by calling the <tt>C++</tt> exception throwing methods, but may
1217 use the <tt>unwind</tt> instruction.</dd>
1219 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1220 <dd>This attribute indicates that this function can return twice. The
1221 C <code>setjmp</code> is an example of such a function. The compiler
1222 disables some optimizations (like tail calls) in the caller of these
1225 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1226 <dd>This attribute indicates that the function should emit a stack smashing
1227 protector. It is in the form of a "canary"—a random value placed on
1228 the stack before the local variables that's checked upon return from the
1229 function to see if it has been overwritten. A heuristic is used to
1230 determine if a function needs stack protectors or not.<br>
1232 If a function that has an <tt>ssp</tt> attribute is inlined into a
1233 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1234 function will have an <tt>ssp</tt> attribute.</dd>
1236 <dt><tt><b>sspreq</b></tt></dt>
1237 <dd>This attribute indicates that the function should <em>always</em> emit a
1238 stack smashing protector. This overrides
1239 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1241 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1242 function that doesn't have an <tt>sspreq</tt> attribute or which has
1243 an <tt>ssp</tt> attribute, then the resulting function will have
1244 an <tt>sspreq</tt> attribute.</dd>
1246 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1247 <dd>This attribute indicates that the ABI being targeted requires that
1248 an unwind table entry be produce for this function even if we can
1249 show that no exceptions passes by it. This is normally the case for
1250 the ELF x86-64 abi, but it can be disabled for some compilation
1256 <!-- ======================================================================= -->
1258 <a name="moduleasm">Module-Level Inline Assembly</a>
1263 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1264 the GCC "file scope inline asm" blocks. These blocks are internally
1265 concatenated by LLVM and treated as a single unit, but may be separated in
1266 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1268 <pre class="doc_code">
1269 module asm "inline asm code goes here"
1270 module asm "more can go here"
1273 <p>The strings can contain any character by escaping non-printable characters.
1274 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1277 <p>The inline asm code is simply printed to the machine code .s file when
1278 assembly code is generated.</p>
1282 <!-- ======================================================================= -->
1284 <a name="datalayout">Data Layout</a>
1289 <p>A module may specify a target specific data layout string that specifies how
1290 data is to be laid out in memory. The syntax for the data layout is
1293 <pre class="doc_code">
1294 target datalayout = "<i>layout specification</i>"
1297 <p>The <i>layout specification</i> consists of a list of specifications
1298 separated by the minus sign character ('-'). Each specification starts with
1299 a letter and may include other information after the letter to define some
1300 aspect of the data layout. The specifications accepted are as follows:</p>
1304 <dd>Specifies that the target lays out data in big-endian form. That is, the
1305 bits with the most significance have the lowest address location.</dd>
1308 <dd>Specifies that the target lays out data in little-endian form. That is,
1309 the bits with the least significance have the lowest address
1312 <dt><tt>S<i>size</i></tt></dt>
1313 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1314 of stack variables is limited to the natural stack alignment to avoid
1315 dynamic stack realignment. The stack alignment must be a multiple of
1316 8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1317 which does not prevent any alignment promotions.</dd>
1319 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1320 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1321 <i>preferred</i> alignments. All sizes are in bits. Specifying
1322 the <i>pref</i> alignment is optional. If omitted, the
1323 preceding <tt>:</tt> should be omitted too.</dd>
1325 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1326 <dd>This specifies the alignment for an integer type of a given bit
1327 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1329 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1330 <dd>This specifies the alignment for a vector type of a given bit
1333 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1334 <dd>This specifies the alignment for a floating point type of a given bit
1335 <i>size</i>. Only values of <i>size</i> that are supported by the target
1336 will work. 32 (float) and 64 (double) are supported on all targets;
1337 80 or 128 (different flavors of long double) are also supported on some
1340 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1341 <dd>This specifies the alignment for an aggregate type of a given bit
1344 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1345 <dd>This specifies the alignment for a stack object of a given bit
1348 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1349 <dd>This specifies a set of native integer widths for the target CPU
1350 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1351 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1352 this set are considered to support most general arithmetic
1353 operations efficiently.</dd>
1356 <p>When constructing the data layout for a given target, LLVM starts with a
1357 default set of specifications which are then (possibly) overridden by the
1358 specifications in the <tt>datalayout</tt> keyword. The default specifications
1359 are given in this list:</p>
1362 <li><tt>E</tt> - big endian</li>
1363 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1364 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1365 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1366 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1367 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1368 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1369 alignment of 64-bits</li>
1370 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1371 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1372 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1373 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1374 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1375 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1378 <p>When LLVM is determining the alignment for a given type, it uses the
1379 following rules:</p>
1382 <li>If the type sought is an exact match for one of the specifications, that
1383 specification is used.</li>
1385 <li>If no match is found, and the type sought is an integer type, then the
1386 smallest integer type that is larger than the bitwidth of the sought type
1387 is used. If none of the specifications are larger than the bitwidth then
1388 the the largest integer type is used. For example, given the default
1389 specifications above, the i7 type will use the alignment of i8 (next
1390 largest) while both i65 and i256 will use the alignment of i64 (largest
1393 <li>If no match is found, and the type sought is a vector type, then the
1394 largest vector type that is smaller than the sought vector type will be
1395 used as a fall back. This happens because <128 x double> can be
1396 implemented in terms of 64 <2 x double>, for example.</li>
1399 <p>The function of the data layout string may not be what you expect. Notably,
1400 this is not a specification from the frontend of what alignment the code
1401 generator should use.</p>
1403 <p>Instead, if specified, the target data layout is required to match what the
1404 ultimate <em>code generator</em> expects. This string is used by the
1405 mid-level optimizers to
1406 improve code, and this only works if it matches what the ultimate code
1407 generator uses. If you would like to generate IR that does not embed this
1408 target-specific detail into the IR, then you don't have to specify the
1409 string. This will disable some optimizations that require precise layout
1410 information, but this also prevents those optimizations from introducing
1411 target specificity into the IR.</p>
1417 <!-- ======================================================================= -->
1419 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1424 <p>Any memory access must be done through a pointer value associated
1425 with an address range of the memory access, otherwise the behavior
1426 is undefined. Pointer values are associated with address ranges
1427 according to the following rules:</p>
1430 <li>A pointer value is associated with the addresses associated with
1431 any value it is <i>based</i> on.
1432 <li>An address of a global variable is associated with the address
1433 range of the variable's storage.</li>
1434 <li>The result value of an allocation instruction is associated with
1435 the address range of the allocated storage.</li>
1436 <li>A null pointer in the default address-space is associated with
1438 <li>An integer constant other than zero or a pointer value returned
1439 from a function not defined within LLVM may be associated with address
1440 ranges allocated through mechanisms other than those provided by
1441 LLVM. Such ranges shall not overlap with any ranges of addresses
1442 allocated by mechanisms provided by LLVM.</li>
1445 <p>A pointer value is <i>based</i> on another pointer value according
1446 to the following rules:</p>
1449 <li>A pointer value formed from a
1450 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1451 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1452 <li>The result value of a
1453 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1454 of the <tt>bitcast</tt>.</li>
1455 <li>A pointer value formed by an
1456 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1457 pointer values that contribute (directly or indirectly) to the
1458 computation of the pointer's value.</li>
1459 <li>The "<i>based</i> on" relationship is transitive.</li>
1462 <p>Note that this definition of <i>"based"</i> is intentionally
1463 similar to the definition of <i>"based"</i> in C99, though it is
1464 slightly weaker.</p>
1466 <p>LLVM IR does not associate types with memory. The result type of a
1467 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1468 alignment of the memory from which to load, as well as the
1469 interpretation of the value. The first operand type of a
1470 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1471 and alignment of the store.</p>
1473 <p>Consequently, type-based alias analysis, aka TBAA, aka
1474 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1475 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1476 additional information which specialized optimization passes may use
1477 to implement type-based alias analysis.</p>
1481 <!-- ======================================================================= -->
1483 <a name="volatile">Volatile Memory Accesses</a>
1488 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1489 href="#i_store"><tt>store</tt></a>s, and <a
1490 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1491 The optimizers must not change the number of volatile operations or change their
1492 order of execution relative to other volatile operations. The optimizers
1493 <i>may</i> change the order of volatile operations relative to non-volatile
1494 operations. This is not Java's "volatile" and has no cross-thread
1495 synchronization behavior.</p>
1499 <!-- ======================================================================= -->
1501 <a name="memmodel">Memory Model for Concurrent Operations</a>
1506 <p>The LLVM IR does not define any way to start parallel threads of execution
1507 or to register signal handlers. Nonetheless, there are platform-specific
1508 ways to create them, and we define LLVM IR's behavior in their presence. This
1509 model is inspired by the C++0x memory model.</p>
1511 <p>For a more informal introduction to this model, see the
1512 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1514 <p>We define a <i>happens-before</i> partial order as the least partial order
1517 <li>Is a superset of single-thread program order, and</li>
1518 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1519 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1520 by platform-specific techniques, like pthread locks, thread
1521 creation, thread joining, etc., and by atomic instructions.
1522 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1526 <p>Note that program order does not introduce <i>happens-before</i> edges
1527 between a thread and signals executing inside that thread.</p>
1529 <p>Every (defined) read operation (load instructions, memcpy, atomic
1530 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1531 (defined) write operations (store instructions, atomic
1532 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1533 initialized globals are considered to have a write of the initializer which is
1534 atomic and happens before any other read or write of the memory in question.
1535 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1536 any write to the same byte, except:</p>
1539 <li>If <var>write<sub>1</sub></var> happens before
1540 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1541 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1542 does not see <var>write<sub>1</sub></var>.
1543 <li>If <var>R<sub>byte</sub></var> happens before
1544 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1545 see <var>write<sub>3</sub></var>.
1548 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1550 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1551 is supposed to give guarantees which can support
1552 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1553 addresses which do not behave like normal memory. It does not generally
1554 provide cross-thread synchronization.)
1555 <li>Otherwise, if there is no write to the same byte that happens before
1556 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1557 <tt>undef</tt> for that byte.
1558 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1559 <var>R<sub>byte</sub></var> returns the value written by that
1561 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1562 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1563 values written. See the <a href="#ordering">Atomic Memory Ordering
1564 Constraints</a> section for additional constraints on how the choice
1566 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1569 <p><var>R</var> returns the value composed of the series of bytes it read.
1570 This implies that some bytes within the value may be <tt>undef</tt>
1571 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1572 defines the semantics of the operation; it doesn't mean that targets will
1573 emit more than one instruction to read the series of bytes.</p>
1575 <p>Note that in cases where none of the atomic intrinsics are used, this model
1576 places only one restriction on IR transformations on top of what is required
1577 for single-threaded execution: introducing a store to a byte which might not
1578 otherwise be stored is not allowed in general. (Specifically, in the case
1579 where another thread might write to and read from an address, introducing a
1580 store can change a load that may see exactly one write into a load that may
1581 see multiple writes.)</p>
1583 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1584 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1585 none of the backends currently in the tree fall into this category; however,
1586 there might be targets which care. If there are, we want a paragraph
1589 Targets may specify that stores narrower than a certain width are not
1590 available; on such a target, for the purposes of this model, treat any
1591 non-atomic write with an alignment or width less than the minimum width
1592 as if it writes to the relevant surrounding bytes.
1597 <!-- ======================================================================= -->
1599 <a name="ordering">Atomic Memory Ordering Constraints</a>
1604 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1605 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1606 <a href="#i_fence"><code>fence</code></a>,
1607 <a href="#i_load"><code>atomic load</code></a>, and
1608 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1609 that determines which other atomic instructions on the same address they
1610 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1611 but are somewhat more colloquial. If these descriptions aren't precise enough,
1612 check those specs (see spec references in the
1613 <a href="Atomic.html#introduction">atomics guide</a>).
1614 <a href="#i_fence"><code>fence</code></a> instructions
1615 treat these orderings somewhat differently since they don't take an address.
1616 See that instruction's documentation for details.</p>
1618 <p>For a simpler introduction to the ordering constraints, see the
1619 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1622 <dt><code>unordered</code></dt>
1623 <dd>The set of values that can be read is governed by the happens-before
1624 partial order. A value cannot be read unless some operation wrote it.
1625 This is intended to provide a guarantee strong enough to model Java's
1626 non-volatile shared variables. This ordering cannot be specified for
1627 read-modify-write operations; it is not strong enough to make them atomic
1628 in any interesting way.</dd>
1629 <dt><code>monotonic</code></dt>
1630 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1631 total order for modifications by <code>monotonic</code> operations on each
1632 address. All modification orders must be compatible with the happens-before
1633 order. There is no guarantee that the modification orders can be combined to
1634 a global total order for the whole program (and this often will not be
1635 possible). The read in an atomic read-modify-write operation
1636 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1637 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1638 reads the value in the modification order immediately before the value it
1639 writes. If one atomic read happens before another atomic read of the same
1640 address, the later read must see the same value or a later value in the
1641 address's modification order. This disallows reordering of
1642 <code>monotonic</code> (or stronger) operations on the same address. If an
1643 address is written <code>monotonic</code>ally by one thread, and other threads
1644 <code>monotonic</code>ally read that address repeatedly, the other threads must
1645 eventually see the write. This corresponds to the C++0x/C1x
1646 <code>memory_order_relaxed</code>.</dd>
1647 <dt><code>acquire</code></dt>
1648 <dd>In addition to the guarantees of <code>monotonic</code>,
1649 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1650 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1651 <dt><code>release</code></dt>
1652 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1653 writes a value which is subsequently read by an <code>acquire</code> operation,
1654 it <i>synchronizes-with</i> that operation. (This isn't a complete
1655 description; see the C++0x definition of a release sequence.) This corresponds
1656 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1657 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1658 <code>acquire</code> and <code>release</code> operation on its address.
1659 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1660 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1661 <dd>In addition to the guarantees of <code>acq_rel</code>
1662 (<code>acquire</code> for an operation which only reads, <code>release</code>
1663 for an operation which only writes), there is a global total order on all
1664 sequentially-consistent operations on all addresses, which is consistent with
1665 the <i>happens-before</i> partial order and with the modification orders of
1666 all the affected addresses. Each sequentially-consistent read sees the last
1667 preceding write to the same address in this global order. This corresponds
1668 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1671 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1672 it only <i>synchronizes with</i> or participates in modification and seq_cst
1673 total orderings with other operations running in the same thread (for example,
1674 in signal handlers).</p>
1680 <!-- *********************************************************************** -->
1681 <h2><a name="typesystem">Type System</a></h2>
1682 <!-- *********************************************************************** -->
1686 <p>The LLVM type system is one of the most important features of the
1687 intermediate representation. Being typed enables a number of optimizations
1688 to be performed on the intermediate representation directly, without having
1689 to do extra analyses on the side before the transformation. A strong type
1690 system makes it easier to read the generated code and enables novel analyses
1691 and transformations that are not feasible to perform on normal three address
1692 code representations.</p>
1694 <!-- ======================================================================= -->
1696 <a name="t_classifications">Type Classifications</a>
1701 <p>The types fall into a few useful classifications:</p>
1703 <table border="1" cellspacing="0" cellpadding="4">
1705 <tr><th>Classification</th><th>Types</th></tr>
1707 <td><a href="#t_integer">integer</a></td>
1708 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1711 <td><a href="#t_floating">floating point</a></td>
1712 <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1715 <td><a name="t_firstclass">first class</a></td>
1716 <td><a href="#t_integer">integer</a>,
1717 <a href="#t_floating">floating point</a>,
1718 <a href="#t_pointer">pointer</a>,
1719 <a href="#t_vector">vector</a>,
1720 <a href="#t_struct">structure</a>,
1721 <a href="#t_array">array</a>,
1722 <a href="#t_label">label</a>,
1723 <a href="#t_metadata">metadata</a>.
1727 <td><a href="#t_primitive">primitive</a></td>
1728 <td><a href="#t_label">label</a>,
1729 <a href="#t_void">void</a>,
1730 <a href="#t_integer">integer</a>,
1731 <a href="#t_floating">floating point</a>,
1732 <a href="#t_x86mmx">x86mmx</a>,
1733 <a href="#t_metadata">metadata</a>.</td>
1736 <td><a href="#t_derived">derived</a></td>
1737 <td><a href="#t_array">array</a>,
1738 <a href="#t_function">function</a>,
1739 <a href="#t_pointer">pointer</a>,
1740 <a href="#t_struct">structure</a>,
1741 <a href="#t_vector">vector</a>,
1742 <a href="#t_opaque">opaque</a>.
1748 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1749 important. Values of these types are the only ones which can be produced by
1754 <!-- ======================================================================= -->
1756 <a name="t_primitive">Primitive Types</a>
1761 <p>The primitive types are the fundamental building blocks of the LLVM
1764 <!-- _______________________________________________________________________ -->
1766 <a name="t_integer">Integer Type</a>
1772 <p>The integer type is a very simple type that simply specifies an arbitrary
1773 bit width for the integer type desired. Any bit width from 1 bit to
1774 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1781 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1785 <table class="layout">
1787 <td class="left"><tt>i1</tt></td>
1788 <td class="left">a single-bit integer.</td>
1791 <td class="left"><tt>i32</tt></td>
1792 <td class="left">a 32-bit integer.</td>
1795 <td class="left"><tt>i1942652</tt></td>
1796 <td class="left">a really big integer of over 1 million bits.</td>
1802 <!-- _______________________________________________________________________ -->
1804 <a name="t_floating">Floating Point Types</a>
1811 <tr><th>Type</th><th>Description</th></tr>
1812 <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr>
1813 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1814 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1815 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1816 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1817 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1823 <!-- _______________________________________________________________________ -->
1825 <a name="t_x86mmx">X86mmx Type</a>
1831 <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>
1840 <!-- _______________________________________________________________________ -->
1842 <a name="t_void">Void Type</a>
1848 <p>The void type does not represent any value and has no size.</p>
1857 <!-- _______________________________________________________________________ -->
1859 <a name="t_label">Label Type</a>
1865 <p>The label type represents code labels.</p>
1874 <!-- _______________________________________________________________________ -->
1876 <a name="t_metadata">Metadata Type</a>
1882 <p>The metadata type represents embedded metadata. No derived types may be
1883 created from metadata except for <a href="#t_function">function</a>
1895 <!-- ======================================================================= -->
1897 <a name="t_derived">Derived Types</a>
1902 <p>The real power in LLVM comes from the derived types in the system. This is
1903 what allows a programmer to represent arrays, functions, pointers, and other
1904 useful types. Each of these types contain one or more element types which
1905 may be a primitive type, or another derived type. For example, it is
1906 possible to have a two dimensional array, using an array as the element type
1907 of another array.</p>
1909 <!-- _______________________________________________________________________ -->
1911 <a name="t_aggregate">Aggregate Types</a>
1916 <p>Aggregate Types are a subset of derived types that can contain multiple
1917 member types. <a href="#t_array">Arrays</a> and
1918 <a href="#t_struct">structs</a> are aggregate types.
1919 <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p>
1923 <!-- _______________________________________________________________________ -->
1925 <a name="t_array">Array Type</a>
1931 <p>The array type is a very simple derived type that arranges elements
1932 sequentially in memory. The array type requires a size (number of elements)
1933 and an underlying data type.</p>
1937 [<# elements> x <elementtype>]
1940 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1941 be any type with a size.</p>
1944 <table class="layout">
1946 <td class="left"><tt>[40 x i32]</tt></td>
1947 <td class="left">Array of 40 32-bit integer values.</td>
1950 <td class="left"><tt>[41 x i32]</tt></td>
1951 <td class="left">Array of 41 32-bit integer values.</td>
1954 <td class="left"><tt>[4 x i8]</tt></td>
1955 <td class="left">Array of 4 8-bit integer values.</td>
1958 <p>Here are some examples of multidimensional arrays:</p>
1959 <table class="layout">
1961 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1962 <td class="left">3x4 array of 32-bit integer values.</td>
1965 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1966 <td class="left">12x10 array of single precision floating point values.</td>
1969 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1970 <td class="left">2x3x4 array of 16-bit integer values.</td>
1974 <p>There is no restriction on indexing beyond the end of the array implied by
1975 a static type (though there are restrictions on indexing beyond the bounds
1976 of an allocated object in some cases). This means that single-dimension
1977 'variable sized array' addressing can be implemented in LLVM with a zero
1978 length array type. An implementation of 'pascal style arrays' in LLVM could
1979 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1983 <!-- _______________________________________________________________________ -->
1985 <a name="t_function">Function Type</a>
1991 <p>The function type can be thought of as a function signature. It consists of
1992 a return type and a list of formal parameter types. The return type of a
1993 function type is a first class type or a void type.</p>
1997 <returntype> (<parameter list>)
2000 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
2001 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
2002 which indicates that the function takes a variable number of arguments.
2003 Variable argument functions can access their arguments with
2004 the <a href="#int_varargs">variable argument handling intrinsic</a>
2005 functions. '<tt><returntype></tt>' is any type except
2006 <a href="#t_label">label</a>.</p>
2009 <table class="layout">
2011 <td class="left"><tt>i32 (i32)</tt></td>
2012 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2014 </tr><tr class="layout">
2015 <td class="left"><tt>float (i16, i32 *) *
2017 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2018 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2019 returning <tt>float</tt>.
2021 </tr><tr class="layout">
2022 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2023 <td class="left">A vararg function that takes at least one
2024 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2025 which returns an integer. This is the signature for <tt>printf</tt> in
2028 </tr><tr class="layout">
2029 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2030 <td class="left">A function taking an <tt>i32</tt>, returning a
2031 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2038 <!-- _______________________________________________________________________ -->
2040 <a name="t_struct">Structure Type</a>
2046 <p>The structure type is used to represent a collection of data members together
2047 in memory. The elements of a structure may be any type that has a size.</p>
2049 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2050 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2051 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2052 Structures in registers are accessed using the
2053 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2054 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2056 <p>Structures may optionally be "packed" structures, which indicate that the
2057 alignment of the struct is one byte, and that there is no padding between
2058 the elements. In non-packed structs, padding between field types is inserted
2059 as defined by the TargetData string in the module, which is required to match
2060 what the underlying code generator expects.</p>
2062 <p>Structures can either be "literal" or "identified". A literal structure is
2063 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2064 types are always defined at the top level with a name. Literal types are
2065 uniqued by their contents and can never be recursive or opaque since there is
2066 no way to write one. Identified types can be recursive, can be opaqued, and are
2072 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2073 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2077 <table class="layout">
2079 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2080 <td class="left">A triple of three <tt>i32</tt> values</td>
2083 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2084 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2085 second element is a <a href="#t_pointer">pointer</a> to a
2086 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2087 an <tt>i32</tt>.</td>
2090 <td class="left"><tt><{ i8, i32 }></tt></td>
2091 <td class="left">A packed struct known to be 5 bytes in size.</td>
2097 <!-- _______________________________________________________________________ -->
2099 <a name="t_opaque">Opaque Structure Types</a>
2105 <p>Opaque structure types are used to represent named structure types that do
2106 not have a body specified. This corresponds (for example) to the C notion of
2107 a forward declared structure.</p>
2116 <table class="layout">
2118 <td class="left"><tt>opaque</tt></td>
2119 <td class="left">An opaque type.</td>
2127 <!-- _______________________________________________________________________ -->
2129 <a name="t_pointer">Pointer Type</a>
2135 <p>The pointer type is used to specify memory locations.
2136 Pointers are commonly used to reference objects in memory.</p>
2138 <p>Pointer types may have an optional address space attribute defining the
2139 numbered address space where the pointed-to object resides. The default
2140 address space is number zero. The semantics of non-zero address
2141 spaces are target-specific.</p>
2143 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2144 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2152 <table class="layout">
2154 <td class="left"><tt>[4 x i32]*</tt></td>
2155 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2156 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2159 <td class="left"><tt>i32 (i32*) *</tt></td>
2160 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2161 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2165 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2166 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2167 that resides in address space #5.</td>
2173 <!-- _______________________________________________________________________ -->
2175 <a name="t_vector">Vector Type</a>
2181 <p>A vector type is a simple derived type that represents a vector of elements.
2182 Vector types are used when multiple primitive data are operated in parallel
2183 using a single instruction (SIMD). A vector type requires a size (number of
2184 elements) and an underlying primitive data type. Vector types are considered
2185 <a href="#t_firstclass">first class</a>.</p>
2189 < <# elements> x <elementtype> >
2192 <p>The number of elements is a constant integer value larger than 0; elementtype
2193 may be any integer or floating point type, or a pointer to these types.
2194 Vectors of size zero are not allowed. </p>
2197 <table class="layout">
2199 <td class="left"><tt><4 x i32></tt></td>
2200 <td class="left">Vector of 4 32-bit integer values.</td>
2203 <td class="left"><tt><8 x float></tt></td>
2204 <td class="left">Vector of 8 32-bit floating-point values.</td>
2207 <td class="left"><tt><2 x i64></tt></td>
2208 <td class="left">Vector of 2 64-bit integer values.</td>
2211 <td class="left"><tt><4 x i64*></tt></td>
2212 <td class="left">Vector of 4 pointers to 64-bit integer values.</td>
2222 <!-- *********************************************************************** -->
2223 <h2><a name="constants">Constants</a></h2>
2224 <!-- *********************************************************************** -->
2228 <p>LLVM has several different basic types of constants. This section describes
2229 them all and their syntax.</p>
2231 <!-- ======================================================================= -->
2233 <a name="simpleconstants">Simple Constants</a>
2239 <dt><b>Boolean constants</b></dt>
2240 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2241 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2243 <dt><b>Integer constants</b></dt>
2244 <dd>Standard integers (such as '4') are constants of
2245 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2246 with integer types.</dd>
2248 <dt><b>Floating point constants</b></dt>
2249 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2250 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2251 notation (see below). The assembler requires the exact decimal value of a
2252 floating-point constant. For example, the assembler accepts 1.25 but
2253 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2254 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2256 <dt><b>Null pointer constants</b></dt>
2257 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2258 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2261 <p>The one non-intuitive notation for constants is the hexadecimal form of
2262 floating point constants. For example, the form '<tt>double
2263 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2264 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2265 constants are required (and the only time that they are generated by the
2266 disassembler) is when a floating point constant must be emitted but it cannot
2267 be represented as a decimal floating point number in a reasonable number of
2268 digits. For example, NaN's, infinities, and other special values are
2269 represented in their IEEE hexadecimal format so that assembly and disassembly
2270 do not cause any bits to change in the constants.</p>
2272 <p>When using the hexadecimal form, constants of types half, float, and double are
2273 represented using the 16-digit form shown above (which matches the IEEE754
2274 representation for double); half and float values must, however, be exactly
2275 representable as IEE754 half and single precision, respectively.
2276 Hexadecimal format is always used
2277 for long double, and there are three forms of long double. The 80-bit format
2278 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2279 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2280 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2281 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2282 currently supported target uses this format. Long doubles will only work if
2283 they match the long double format on your target. All hexadecimal formats
2284 are big-endian (sign bit at the left).</p>
2286 <p>There are no constants of type x86mmx.</p>
2289 <!-- ======================================================================= -->
2291 <a name="aggregateconstants"></a> <!-- old anchor -->
2292 <a name="complexconstants">Complex Constants</a>
2297 <p>Complex constants are a (potentially recursive) combination of simple
2298 constants and smaller complex constants.</p>
2301 <dt><b>Structure constants</b></dt>
2302 <dd>Structure constants are represented with notation similar to structure
2303 type definitions (a comma separated list of elements, surrounded by braces
2304 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2305 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2306 Structure constants must have <a href="#t_struct">structure type</a>, and
2307 the number and types of elements must match those specified by the
2310 <dt><b>Array constants</b></dt>
2311 <dd>Array constants are represented with notation similar to array type
2312 definitions (a comma separated list of elements, surrounded by square
2313 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2314 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2315 the number and types of elements must match those specified by the
2318 <dt><b>Vector constants</b></dt>
2319 <dd>Vector constants are represented with notation similar to vector type
2320 definitions (a comma separated list of elements, surrounded by
2321 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2322 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2323 have <a href="#t_vector">vector type</a>, and the number and types of
2324 elements must match those specified by the type.</dd>
2326 <dt><b>Zero initialization</b></dt>
2327 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2328 value to zero of <em>any</em> type, including scalar and
2329 <a href="#t_aggregate">aggregate</a> types.
2330 This is often used to avoid having to print large zero initializers
2331 (e.g. for large arrays) and is always exactly equivalent to using explicit
2332 zero initializers.</dd>
2334 <dt><b>Metadata node</b></dt>
2335 <dd>A metadata node is a structure-like constant with
2336 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2337 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2338 be interpreted as part of the instruction stream, metadata is a place to
2339 attach additional information such as debug info.</dd>
2344 <!-- ======================================================================= -->
2346 <a name="globalconstants">Global Variable and Function Addresses</a>
2351 <p>The addresses of <a href="#globalvars">global variables</a>
2352 and <a href="#functionstructure">functions</a> are always implicitly valid
2353 (link-time) constants. These constants are explicitly referenced when
2354 the <a href="#identifiers">identifier for the global</a> is used and always
2355 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2356 legal LLVM file:</p>
2358 <pre class="doc_code">
2361 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2366 <!-- ======================================================================= -->
2368 <a name="undefvalues">Undefined Values</a>
2373 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2374 indicates that the user of the value may receive an unspecified bit-pattern.
2375 Undefined values may be of any type (other than '<tt>label</tt>'
2376 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2378 <p>Undefined values are useful because they indicate to the compiler that the
2379 program is well defined no matter what value is used. This gives the
2380 compiler more freedom to optimize. Here are some examples of (potentially
2381 surprising) transformations that are valid (in pseudo IR):</p>
2384 <pre class="doc_code">
2394 <p>This is safe because all of the output bits are affected by the undef bits.
2395 Any output bit can have a zero or one depending on the input bits.</p>
2397 <pre class="doc_code">
2408 <p>These logical operations have bits that are not always affected by the input.
2409 For example, if <tt>%X</tt> has a zero bit, then the output of the
2410 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2411 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2412 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2413 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2414 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2415 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2416 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2418 <pre class="doc_code">
2419 %A = select undef, %X, %Y
2420 %B = select undef, 42, %Y
2421 %C = select %X, %Y, undef
2432 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2433 branch) conditions can go <em>either way</em>, but they have to come from one
2434 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2435 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2436 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2437 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2438 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2441 <pre class="doc_code">
2442 %A = xor undef, undef
2460 <p>This example points out that two '<tt>undef</tt>' operands are not
2461 necessarily the same. This can be surprising to people (and also matches C
2462 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2463 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2464 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2465 its value over its "live range". This is true because the variable doesn't
2466 actually <em>have a live range</em>. Instead, the value is logically read
2467 from arbitrary registers that happen to be around when needed, so the value
2468 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2469 need to have the same semantics or the core LLVM "replace all uses with"
2470 concept would not hold.</p>
2472 <pre class="doc_code">
2480 <p>These examples show the crucial difference between an <em>undefined
2481 value</em> and <em>undefined behavior</em>. An undefined value (like
2482 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2483 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2484 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2485 defined on SNaN's. However, in the second example, we can make a more
2486 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2487 arbitrary value, we are allowed to assume that it could be zero. Since a
2488 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2489 the operation does not execute at all. This allows us to delete the divide and
2490 all code after it. Because the undefined operation "can't happen", the
2491 optimizer can assume that it occurs in dead code.</p>
2493 <pre class="doc_code">
2494 a: store undef -> %X
2495 b: store %X -> undef
2501 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2502 undefined value can be assumed to not have any effect; we can assume that the
2503 value is overwritten with bits that happen to match what was already there.
2504 However, a store <em>to</em> an undefined location could clobber arbitrary
2505 memory, therefore, it has undefined behavior.</p>
2509 <!-- ======================================================================= -->
2511 <a name="poisonvalues">Poison Values</a>
2516 <p>Poison values are similar to <a href="#undefvalues">undef values</a>, however
2517 they also represent the fact that an instruction or constant expression which
2518 cannot evoke side effects has nevertheless detected a condition which results
2519 in undefined behavior.</p>
2521 <p>There is currently no way of representing a poison value in the IR; they
2522 only exist when produced by operations such as
2523 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2525 <p>Poison value behavior is defined in terms of value <i>dependence</i>:</p>
2528 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2529 their operands.</li>
2531 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2532 to their dynamic predecessor basic block.</li>
2534 <li>Function arguments depend on the corresponding actual argument values in
2535 the dynamic callers of their functions.</li>
2537 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2538 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2539 control back to them.</li>
2541 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2542 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_unwind"><tt>unwind</tt></a>,
2543 or exception-throwing call instructions that dynamically transfer control
2546 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2547 referenced memory addresses, following the order in the IR
2548 (including loads and stores implied by intrinsics such as
2549 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2551 <!-- TODO: In the case of multiple threads, this only applies if the store
2552 "happens-before" the load or store. -->
2554 <!-- TODO: floating-point exception state -->
2556 <li>An instruction with externally visible side effects depends on the most
2557 recent preceding instruction with externally visible side effects, following
2558 the order in the IR. (This includes
2559 <a href="#volatile">volatile operations</a>.)</li>
2561 <li>An instruction <i>control-depends</i> on a
2562 <a href="#terminators">terminator instruction</a>
2563 if the terminator instruction has multiple successors and the instruction
2564 is always executed when control transfers to one of the successors, and
2565 may not be executed when control is transferred to another.</li>
2567 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2568 instruction if the set of instructions it otherwise depends on would be
2569 different if the terminator had transferred control to a different
2572 <li>Dependence is transitive.</li>
2576 <p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>,
2577 with the additional affect that any instruction which has a <i>dependence</i>
2578 on a poison value has undefined behavior.</p>
2580 <p>Here are some examples:</p>
2582 <pre class="doc_code">
2584 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2585 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2586 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2587 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2589 store i32 %poison, i32* @g ; Poison value stored to memory.
2590 %poison2 = load i32* @g ; Poison value loaded back from memory.
2592 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2594 %narrowaddr = bitcast i32* @g to i16*
2595 %wideaddr = bitcast i32* @g to i64*
2596 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2597 %poison4 = load i64* %wideaddr ; Returns a poison value.
2599 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2600 br i1 %cmp, label %true, label %end ; Branch to either destination.
2603 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2604 ; it has undefined behavior.
2608 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2609 ; Both edges into this PHI are
2610 ; control-dependent on %cmp, so this
2611 ; always results in a poison value.
2613 store volatile i32 0, i32* @g ; This would depend on the store in %true
2614 ; if %cmp is true, or the store in %entry
2615 ; otherwise, so this is undefined behavior.
2617 br i1 %cmp, label %second_true, label %second_end
2618 ; The same branch again, but this time the
2619 ; true block doesn't have side effects.
2626 store volatile i32 0, i32* @g ; This time, the instruction always depends
2627 ; on the store in %end. Also, it is
2628 ; control-equivalent to %end, so this is
2629 ; well-defined (ignoring earlier undefined
2630 ; behavior in this example).
2635 <!-- ======================================================================= -->
2637 <a name="blockaddress">Addresses of Basic Blocks</a>
2642 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2644 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2645 basic block in the specified function, and always has an i8* type. Taking
2646 the address of the entry block is illegal.</p>
2648 <p>This value only has defined behavior when used as an operand to the
2649 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2650 comparisons against null. Pointer equality tests between labels addresses
2651 results in undefined behavior — though, again, comparison against null
2652 is ok, and no label is equal to the null pointer. This may be passed around
2653 as an opaque pointer sized value as long as the bits are not inspected. This
2654 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2655 long as the original value is reconstituted before the <tt>indirectbr</tt>
2658 <p>Finally, some targets may provide defined semantics when using the value as
2659 the operand to an inline assembly, but that is target specific.</p>
2664 <!-- ======================================================================= -->
2666 <a name="constantexprs">Constant Expressions</a>
2671 <p>Constant expressions are used to allow expressions involving other constants
2672 to be used as constants. Constant expressions may be of
2673 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2674 operation that does not have side effects (e.g. load and call are not
2675 supported). The following is the syntax for constant expressions:</p>
2678 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2679 <dd>Truncate a constant to another type. The bit size of CST must be larger
2680 than the bit size of TYPE. Both types must be integers.</dd>
2682 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2683 <dd>Zero extend a constant to another type. The bit size of CST must be
2684 smaller than the bit size of TYPE. Both types must be integers.</dd>
2686 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2687 <dd>Sign extend a constant to another type. The bit size of CST must be
2688 smaller than the bit size of TYPE. Both types must be integers.</dd>
2690 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2691 <dd>Truncate a floating point constant to another floating point type. The
2692 size of CST must be larger than the size of TYPE. Both types must be
2693 floating point.</dd>
2695 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2696 <dd>Floating point extend a constant to another type. The size of CST must be
2697 smaller or equal to the size of TYPE. Both types must be floating
2700 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2701 <dd>Convert a floating point constant to the corresponding unsigned integer
2702 constant. TYPE must be a scalar or vector integer type. CST must be of
2703 scalar or vector floating point type. Both CST and TYPE must be scalars,
2704 or vectors of the same number of elements. If the value won't fit in the
2705 integer type, the results are undefined.</dd>
2707 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2708 <dd>Convert a floating point constant to the corresponding signed integer
2709 constant. TYPE must be a scalar or vector integer type. CST must be of
2710 scalar or vector floating point type. Both CST and TYPE must be scalars,
2711 or vectors of the same number of elements. If the value won't fit in the
2712 integer type, the results are undefined.</dd>
2714 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2715 <dd>Convert an unsigned integer constant to the corresponding floating point
2716 constant. TYPE must be a scalar or vector floating point type. CST must be
2717 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2718 vectors of the same number of elements. If the value won't fit in the
2719 floating point type, the results are undefined.</dd>
2721 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2722 <dd>Convert a signed integer constant to the corresponding floating point
2723 constant. TYPE must be a scalar or vector floating point type. CST must be
2724 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2725 vectors of the same number of elements. If the value won't fit in the
2726 floating point type, the results are undefined.</dd>
2728 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2729 <dd>Convert a pointer typed constant to the corresponding integer constant
2730 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2731 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2732 make it fit in <tt>TYPE</tt>.</dd>
2734 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2735 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2736 type. CST must be of integer type. The CST value is zero extended,
2737 truncated, or unchanged to make it fit in a pointer size. This one is
2738 <i>really</i> dangerous!</dd>
2740 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2741 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2742 are the same as those for the <a href="#i_bitcast">bitcast
2743 instruction</a>.</dd>
2745 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2746 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2747 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2748 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2749 instruction, the index list may have zero or more indexes, which are
2750 required to make sense for the type of "CSTPTR".</dd>
2752 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2753 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2755 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2756 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2758 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2759 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2761 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2762 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2765 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2766 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2769 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2770 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2773 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2774 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2775 constants. The index list is interpreted in a similar manner as indices in
2776 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2777 index value must be specified.</dd>
2779 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2780 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2781 constants. The index list is interpreted in a similar manner as indices in
2782 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2783 index value must be specified.</dd>
2785 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2786 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2787 be any of the <a href="#binaryops">binary</a>
2788 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2789 on operands are the same as those for the corresponding instruction
2790 (e.g. no bitwise operations on floating point values are allowed).</dd>
2797 <!-- *********************************************************************** -->
2798 <h2><a name="othervalues">Other Values</a></h2>
2799 <!-- *********************************************************************** -->
2801 <!-- ======================================================================= -->
2803 <a name="inlineasm">Inline Assembler Expressions</a>
2808 <p>LLVM supports inline assembler expressions (as opposed
2809 to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of
2810 a special value. This value represents the inline assembler as a string
2811 (containing the instructions to emit), a list of operand constraints (stored
2812 as a string), a flag that indicates whether or not the inline asm
2813 expression has side effects, and a flag indicating whether the function
2814 containing the asm needs to align its stack conservatively. An example
2815 inline assembler expression is:</p>
2817 <pre class="doc_code">
2818 i32 (i32) asm "bswap $0", "=r,r"
2821 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2822 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2825 <pre class="doc_code">
2826 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2829 <p>Inline asms with side effects not visible in the constraint list must be
2830 marked as having side effects. This is done through the use of the
2831 '<tt>sideeffect</tt>' keyword, like so:</p>
2833 <pre class="doc_code">
2834 call void asm sideeffect "eieio", ""()
2837 <p>In some cases inline asms will contain code that will not work unless the
2838 stack is aligned in some way, such as calls or SSE instructions on x86,
2839 yet will not contain code that does that alignment within the asm.
2840 The compiler should make conservative assumptions about what the asm might
2841 contain and should generate its usual stack alignment code in the prologue
2842 if the '<tt>alignstack</tt>' keyword is present:</p>
2844 <pre class="doc_code">
2845 call void asm alignstack "eieio", ""()
2848 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2852 <p>TODO: The format of the asm and constraints string still need to be
2853 documented here. Constraints on what can be done (e.g. duplication, moving,
2854 etc need to be documented). This is probably best done by reference to
2855 another document that covers inline asm from a holistic perspective.</p>
2858 <!-- _______________________________________________________________________ -->
2860 <a name="inlineasm_md">Inline Asm Metadata</a>
2865 <p>The call instructions that wrap inline asm nodes may have a
2866 "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant
2867 integers. If present, the code generator will use the integer as the
2868 location cookie value when report errors through the <tt>LLVMContext</tt>
2869 error reporting mechanisms. This allows a front-end to correlate backend
2870 errors that occur with inline asm back to the source code that produced it.
2873 <pre class="doc_code">
2874 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2876 !42 = !{ i32 1234567 }
2879 <p>It is up to the front-end to make sense of the magic numbers it places in the
2880 IR. If the MDNode contains multiple constants, the code generator will use
2881 the one that corresponds to the line of the asm that the error occurs on.</p>
2887 <!-- ======================================================================= -->
2889 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2894 <p>LLVM IR allows metadata to be attached to instructions in the program that
2895 can convey extra information about the code to the optimizers and code
2896 generator. One example application of metadata is source-level debug
2897 information. There are two metadata primitives: strings and nodes. All
2898 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2899 preceding exclamation point ('<tt>!</tt>').</p>
2901 <p>A metadata string is a string surrounded by double quotes. It can contain
2902 any character by escaping non-printable characters with "<tt>\xx</tt>" where
2903 "<tt>xx</tt>" is the two digit hex code. For example:
2904 "<tt>!"test\00"</tt>".</p>
2906 <p>Metadata nodes are represented with notation similar to structure constants
2907 (a comma separated list of elements, surrounded by braces and preceded by an
2908 exclamation point). Metadata nodes can have any values as their operand. For
2911 <div class="doc_code">
2913 !{ metadata !"test\00", i32 10}
2917 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2918 metadata nodes, which can be looked up in the module symbol table. For
2921 <div class="doc_code">
2923 !foo = metadata !{!4, !3}
2927 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2928 function is using two metadata arguments:</p>
2930 <div class="doc_code">
2932 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2936 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2937 attached to the <tt>add</tt> instruction using the <tt>!dbg</tt>
2940 <div class="doc_code">
2942 %indvar.next = add i64 %indvar, 1, !dbg !21
2946 <p>More information about specific metadata nodes recognized by the optimizers
2947 and code generator is found below.</p>
2949 <!-- _______________________________________________________________________ -->
2951 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
2956 <p>In LLVM IR, memory does not have types, so LLVM's own type system is not
2957 suitable for doing TBAA. Instead, metadata is added to the IR to describe
2958 a type system of a higher level language. This can be used to implement
2959 typical C/C++ TBAA, but it can also be used to implement custom alias
2960 analysis behavior for other languages.</p>
2962 <p>The current metadata format is very simple. TBAA metadata nodes have up to
2963 three fields, e.g.:</p>
2965 <div class="doc_code">
2967 !0 = metadata !{ metadata !"an example type tree" }
2968 !1 = metadata !{ metadata !"int", metadata !0 }
2969 !2 = metadata !{ metadata !"float", metadata !0 }
2970 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2974 <p>The first field is an identity field. It can be any value, usually
2975 a metadata string, which uniquely identifies the type. The most important
2976 name in the tree is the name of the root node. Two trees with
2977 different root node names are entirely disjoint, even if they
2978 have leaves with common names.</p>
2980 <p>The second field identifies the type's parent node in the tree, or
2981 is null or omitted for a root node. A type is considered to alias
2982 all of its descendants and all of its ancestors in the tree. Also,
2983 a type is considered to alias all types in other trees, so that
2984 bitcode produced from multiple front-ends is handled conservatively.</p>
2986 <p>If the third field is present, it's an integer which if equal to 1
2987 indicates that the type is "constant" (meaning
2988 <tt>pointsToConstantMemory</tt> should return true; see
2989 <a href="AliasAnalysis.html#OtherItfs">other useful
2990 <tt>AliasAnalysis</tt> methods</a>).</p>
2994 <!-- _______________________________________________________________________ -->
2996 <a name="fpaccuracy">'<tt>fpaccuracy</tt>' Metadata</a>
3001 <p><tt>fpaccuracy</tt> metadata may be attached to any instruction of floating
3002 point type. It expresses the maximum relative error of the result of
3003 that instruction, in ULPs. ULP is defined as follows:</p>
3007 <p>If <tt>x</tt> is a real number that lies between two finite consecutive
3008 floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one
3009 of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the
3010 distance between the two non-equal finite floating-point numbers nearest
3011 <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p>
3015 <p>The maximum relative error may be any rational number. The metadata node
3016 shall consist of a pair of unsigned integers respectively representing
3017 the numerator and denominator. For example, 2.5 ULP:</p>
3019 <div class="doc_code">
3021 !0 = metadata !{ i32 5, i32 2 }
3031 <!-- *********************************************************************** -->
3033 <a name="intrinsic_globals">Intrinsic Global Variables</a>
3035 <!-- *********************************************************************** -->
3037 <p>LLVM has a number of "magic" global variables that contain data that affect
3038 code generation or other IR semantics. These are documented here. All globals
3039 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
3040 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
3043 <!-- ======================================================================= -->
3045 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
3050 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
3051 href="#linkage_appending">appending linkage</a>. This array contains a list of
3052 pointers to global variables and functions which may optionally have a pointer
3053 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
3055 <div class="doc_code">
3060 @llvm.used = appending global [2 x i8*] [
3062 i8* bitcast (i32* @Y to i8*)
3063 ], section "llvm.metadata"
3067 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
3068 compiler, assembler, and linker are required to treat the symbol as if there
3069 is a reference to the global that it cannot see. For example, if a variable
3070 has internal linkage and no references other than that from
3071 the <tt>@llvm.used</tt> list, it cannot be deleted. This is commonly used to
3072 represent references from inline asms and other things the compiler cannot
3073 "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p>
3075 <p>On some targets, the code generator must emit a directive to the assembler or
3076 object file to prevent the assembler and linker from molesting the
3081 <!-- ======================================================================= -->
3083 <a name="intg_compiler_used">
3084 The '<tt>llvm.compiler.used</tt>' Global Variable
3090 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
3091 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
3092 touching the symbol. On targets that support it, this allows an intelligent
3093 linker to optimize references to the symbol without being impeded as it would
3094 be by <tt>@llvm.used</tt>.</p>
3096 <p>This is a rare construct that should only be used in rare circumstances, and
3097 should not be exposed to source languages.</p>
3101 <!-- ======================================================================= -->
3103 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
3108 <div class="doc_code">
3110 %0 = type { i32, void ()* }
3111 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3115 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor
3116 functions and associated priorities. The functions referenced by this array
3117 will be called in ascending order of priority (i.e. lowest first) when the
3118 module is loaded. The order of functions with the same priority is not
3123 <!-- ======================================================================= -->
3125 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3130 <div class="doc_code">
3132 %0 = type { i32, void ()* }
3133 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3137 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions
3138 and associated priorities. The functions referenced by this array will be
3139 called in descending order of priority (i.e. highest first) when the module
3140 is loaded. The order of functions with the same priority is not defined.</p>
3146 <!-- *********************************************************************** -->
3147 <h2><a name="instref">Instruction Reference</a></h2>
3148 <!-- *********************************************************************** -->
3152 <p>The LLVM instruction set consists of several different classifications of
3153 instructions: <a href="#terminators">terminator
3154 instructions</a>, <a href="#binaryops">binary instructions</a>,
3155 <a href="#bitwiseops">bitwise binary instructions</a>,
3156 <a href="#memoryops">memory instructions</a>, and
3157 <a href="#otherops">other instructions</a>.</p>
3159 <!-- ======================================================================= -->
3161 <a name="terminators">Terminator Instructions</a>
3166 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3167 in a program ends with a "Terminator" instruction, which indicates which
3168 block should be executed after the current block is finished. These
3169 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3170 control flow, not values (the one exception being the
3171 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3173 <p>The terminator instructions are:
3174 '<a href="#i_ret"><tt>ret</tt></a>',
3175 '<a href="#i_br"><tt>br</tt></a>',
3176 '<a href="#i_switch"><tt>switch</tt></a>',
3177 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3178 '<a href="#i_invoke"><tt>invoke</tt></a>',
3179 '<a href="#i_unwind"><tt>unwind</tt></a>',
3180 '<a href="#i_resume"><tt>resume</tt></a>', and
3181 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3183 <!-- _______________________________________________________________________ -->
3185 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3192 ret <type> <value> <i>; Return a value from a non-void function</i>
3193 ret void <i>; Return from void function</i>
3197 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3198 a value) from a function back to the caller.</p>
3200 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3201 value and then causes control flow, and one that just causes control flow to
3205 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3206 return value. The type of the return value must be a
3207 '<a href="#t_firstclass">first class</a>' type.</p>
3209 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3210 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3211 value or a return value with a type that does not match its type, or if it
3212 has a void return type and contains a '<tt>ret</tt>' instruction with a
3216 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3217 the calling function's context. If the caller is a
3218 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3219 instruction after the call. If the caller was an
3220 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3221 the beginning of the "normal" destination block. If the instruction returns
3222 a value, that value shall set the call or invoke instruction's return
3227 ret i32 5 <i>; Return an integer value of 5</i>
3228 ret void <i>; Return from a void function</i>
3229 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3233 <!-- _______________________________________________________________________ -->
3235 <a name="i_br">'<tt>br</tt>' Instruction</a>
3242 br i1 <cond>, label <iftrue>, label <iffalse>
3243 br label <dest> <i>; Unconditional branch</i>
3247 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3248 different basic block in the current function. There are two forms of this
3249 instruction, corresponding to a conditional branch and an unconditional
3253 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3254 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3255 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3259 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3260 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3261 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3262 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3267 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3268 br i1 %cond, label %IfEqual, label %IfUnequal
3270 <a href="#i_ret">ret</a> i32 1
3272 <a href="#i_ret">ret</a> i32 0
3277 <!-- _______________________________________________________________________ -->
3279 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3286 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3290 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3291 several different places. It is a generalization of the '<tt>br</tt>'
3292 instruction, allowing a branch to occur to one of many possible
3296 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3297 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3298 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3299 The table is not allowed to contain duplicate constant entries.</p>
3302 <p>The <tt>switch</tt> instruction specifies a table of values and
3303 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3304 is searched for the given value. If the value is found, control flow is
3305 transferred to the corresponding destination; otherwise, control flow is
3306 transferred to the default destination.</p>
3308 <h5>Implementation:</h5>
3309 <p>Depending on properties of the target machine and the particular
3310 <tt>switch</tt> instruction, this instruction may be code generated in
3311 different ways. For example, it could be generated as a series of chained
3312 conditional branches or with a lookup table.</p>
3316 <i>; Emulate a conditional br instruction</i>
3317 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3318 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3320 <i>; Emulate an unconditional br instruction</i>
3321 switch i32 0, label %dest [ ]
3323 <i>; Implement a jump table:</i>
3324 switch i32 %val, label %otherwise [ i32 0, label %onzero
3326 i32 2, label %ontwo ]
3332 <!-- _______________________________________________________________________ -->
3334 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3341 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3346 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3347 within the current function, whose address is specified by
3348 "<tt>address</tt>". Address must be derived from a <a
3349 href="#blockaddress">blockaddress</a> constant.</p>
3353 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3354 rest of the arguments indicate the full set of possible destinations that the
3355 address may point to. Blocks are allowed to occur multiple times in the
3356 destination list, though this isn't particularly useful.</p>
3358 <p>This destination list is required so that dataflow analysis has an accurate
3359 understanding of the CFG.</p>
3363 <p>Control transfers to the block specified in the address argument. All
3364 possible destination blocks must be listed in the label list, otherwise this
3365 instruction has undefined behavior. This implies that jumps to labels
3366 defined in other functions have undefined behavior as well.</p>
3368 <h5>Implementation:</h5>
3370 <p>This is typically implemented with a jump through a register.</p>
3374 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3380 <!-- _______________________________________________________________________ -->
3382 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3389 <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>]
3390 to label <normal label> unwind label <exception label>
3394 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3395 function, with the possibility of control flow transfer to either the
3396 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3397 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3398 control flow will return to the "normal" label. If the callee (or any
3399 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
3400 instruction, control is interrupted and continued at the dynamically nearest
3401 "exception" label.</p>
3403 <p>The '<tt>exception</tt>' label is a
3404 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3405 exception. As such, '<tt>exception</tt>' label is required to have the
3406 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3407 the information about the behavior of the program after unwinding
3408 happens, as its first non-PHI instruction. The restrictions on the
3409 "<tt>landingpad</tt>" instruction's tightly couples it to the
3410 "<tt>invoke</tt>" instruction, so that the important information contained
3411 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3415 <p>This instruction requires several arguments:</p>
3418 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3419 convention</a> the call should use. If none is specified, the call
3420 defaults to using C calling conventions.</li>
3422 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3423 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3424 '<tt>inreg</tt>' attributes are valid here.</li>
3426 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3427 function value being invoked. In most cases, this is a direct function
3428 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3429 off an arbitrary pointer to function value.</li>
3431 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3432 function to be invoked. </li>
3434 <li>'<tt>function args</tt>': argument list whose types match the function
3435 signature argument types and parameter attributes. All arguments must be
3436 of <a href="#t_firstclass">first class</a> type. If the function
3437 signature indicates the function accepts a variable number of arguments,
3438 the extra arguments can be specified.</li>
3440 <li>'<tt>normal label</tt>': the label reached when the called function
3441 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3443 <li>'<tt>exception label</tt>': the label reached when a callee returns with
3444 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
3446 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3447 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3448 '<tt>readnone</tt>' attributes are valid here.</li>
3452 <p>This instruction is designed to operate as a standard
3453 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3454 primary difference is that it establishes an association with a label, which
3455 is used by the runtime library to unwind the stack.</p>
3457 <p>This instruction is used in languages with destructors to ensure that proper
3458 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3459 exception. Additionally, this is important for implementation of
3460 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3462 <p>For the purposes of the SSA form, the definition of the value returned by the
3463 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3464 block to the "normal" label. If the callee unwinds then no return value is
3467 <p>Note that the code generator does not yet completely support unwind, and
3468 that the invoke/unwind semantics are likely to change in future versions.</p>
3472 %retval = invoke i32 @Test(i32 15) to label %Continue
3473 unwind label %TestCleanup <i>; {i32}:retval set</i>
3474 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3475 unwind label %TestCleanup <i>; {i32}:retval set</i>
3480 <!-- _______________________________________________________________________ -->
3483 <a name="i_unwind">'<tt>unwind</tt>' Instruction</a>
3494 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
3495 at the first callee in the dynamic call stack which used
3496 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
3497 This is primarily used to implement exception handling.</p>
3500 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
3501 immediately halt. The dynamic call stack is then searched for the
3502 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
3503 Once found, execution continues at the "exceptional" destination block
3504 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
3505 instruction in the dynamic call chain, undefined behavior results.</p>
3507 <p>Note that the code generator does not yet completely support unwind, and
3508 that the invoke/unwind semantics are likely to change in future versions.</p>
3512 <!-- _______________________________________________________________________ -->
3515 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3522 resume <type> <value>
3526 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3530 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3531 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3535 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3536 (in-flight) exception whose unwinding was interrupted with
3537 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3541 resume { i8*, i32 } %exn
3546 <!-- _______________________________________________________________________ -->
3549 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3560 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3561 instruction is used to inform the optimizer that a particular portion of the
3562 code is not reachable. This can be used to indicate that the code after a
3563 no-return function cannot be reached, and other facts.</p>
3566 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3572 <!-- ======================================================================= -->
3574 <a name="binaryops">Binary Operations</a>
3579 <p>Binary operators are used to do most of the computation in a program. They
3580 require two operands of the same type, execute an operation on them, and
3581 produce a single value. The operands might represent multiple data, as is
3582 the case with the <a href="#t_vector">vector</a> data type. The result value
3583 has the same type as its operands.</p>
3585 <p>There are several different binary operators:</p>
3587 <!-- _______________________________________________________________________ -->
3589 <a name="i_add">'<tt>add</tt>' Instruction</a>
3596 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3597 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3598 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3599 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3603 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3606 <p>The two arguments to the '<tt>add</tt>' instruction must
3607 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3608 integer values. Both arguments must have identical types.</p>
3611 <p>The value produced is the integer sum of the two operands.</p>
3613 <p>If the sum has unsigned overflow, the result returned is the mathematical
3614 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3616 <p>Because LLVM integers use a two's complement representation, this instruction
3617 is appropriate for both signed and unsigned integers.</p>
3619 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3620 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3621 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3622 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3623 respectively, occurs.</p>
3627 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3632 <!-- _______________________________________________________________________ -->
3634 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3641 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3645 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3648 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3649 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3650 floating point values. Both arguments must have identical types.</p>
3653 <p>The value produced is the floating point sum of the two operands.</p>
3657 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3662 <!-- _______________________________________________________________________ -->
3664 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3671 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3672 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3673 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3674 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3678 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3681 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3682 '<tt>neg</tt>' instruction present in most other intermediate
3683 representations.</p>
3686 <p>The two arguments to the '<tt>sub</tt>' instruction must
3687 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3688 integer values. Both arguments must have identical types.</p>
3691 <p>The value produced is the integer difference of the two operands.</p>
3693 <p>If the difference has unsigned overflow, the result returned is the
3694 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3697 <p>Because LLVM integers use a two's complement representation, this instruction
3698 is appropriate for both signed and unsigned integers.</p>
3700 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3701 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3702 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3703 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3704 respectively, occurs.</p>
3708 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3709 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3714 <!-- _______________________________________________________________________ -->
3716 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3723 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3727 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3730 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3731 '<tt>fneg</tt>' instruction present in most other intermediate
3732 representations.</p>
3735 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3736 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3737 floating point values. Both arguments must have identical types.</p>
3740 <p>The value produced is the floating point difference of the two operands.</p>
3744 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3745 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3750 <!-- _______________________________________________________________________ -->
3752 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3759 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3760 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3761 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3762 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3766 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3769 <p>The two arguments to the '<tt>mul</tt>' instruction must
3770 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3771 integer values. Both arguments must have identical types.</p>
3774 <p>The value produced is the integer product of the two operands.</p>
3776 <p>If the result of the multiplication has unsigned overflow, the result
3777 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3778 width of the result.</p>
3780 <p>Because LLVM integers use a two's complement representation, and the result
3781 is the same width as the operands, this instruction returns the correct
3782 result for both signed and unsigned integers. If a full product
3783 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3784 be sign-extended or zero-extended as appropriate to the width of the full
3787 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3788 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3789 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3790 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3791 respectively, occurs.</p>
3795 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3800 <!-- _______________________________________________________________________ -->
3802 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3809 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3813 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3816 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3817 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3818 floating point values. Both arguments must have identical types.</p>
3821 <p>The value produced is the floating point product of the two operands.</p>
3825 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3830 <!-- _______________________________________________________________________ -->
3832 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
3839 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3840 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3844 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3847 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3848 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3849 values. Both arguments must have identical types.</p>
3852 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3854 <p>Note that unsigned integer division and signed integer division are distinct
3855 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3857 <p>Division by zero leads to undefined behavior.</p>
3859 <p>If the <tt>exact</tt> keyword is present, the result value of the
3860 <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a
3861 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
3866 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3871 <!-- _______________________________________________________________________ -->
3873 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
3880 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3881 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3885 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3888 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3889 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3890 values. Both arguments must have identical types.</p>
3893 <p>The value produced is the signed integer quotient of the two operands rounded
3896 <p>Note that signed integer division and unsigned integer division are distinct
3897 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3899 <p>Division by zero leads to undefined behavior. Overflow also leads to
3900 undefined behavior; this is a rare case, but can occur, for example, by doing
3901 a 32-bit division of -2147483648 by -1.</p>
3903 <p>If the <tt>exact</tt> keyword is present, the result value of the
3904 <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would
3909 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3914 <!-- _______________________________________________________________________ -->
3916 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
3923 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3927 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3930 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3931 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3932 floating point values. Both arguments must have identical types.</p>
3935 <p>The value produced is the floating point quotient of the two operands.</p>
3939 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3944 <!-- _______________________________________________________________________ -->
3946 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3953 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3957 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3958 division of its two arguments.</p>
3961 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3962 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3963 values. Both arguments must have identical types.</p>
3966 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3967 This instruction always performs an unsigned division to get the
3970 <p>Note that unsigned integer remainder and signed integer remainder are
3971 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3973 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3977 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3982 <!-- _______________________________________________________________________ -->
3984 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3991 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3995 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3996 division of its two operands. This instruction can also take
3997 <a href="#t_vector">vector</a> versions of the values in which case the
3998 elements must be integers.</p>
4001 <p>The two arguments to the '<tt>srem</tt>' instruction must be
4002 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4003 values. Both arguments must have identical types.</p>
4006 <p>This instruction returns the <i>remainder</i> of a division (where the result
4007 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
4008 <i>modulo</i> operator (where the result is either zero or has the same sign
4009 as the divisor, <tt>op2</tt>) of a value.
4010 For more information about the difference,
4011 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
4012 Math Forum</a>. For a table of how this is implemented in various languages,
4013 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
4014 Wikipedia: modulo operation</a>.</p>
4016 <p>Note that signed integer remainder and unsigned integer remainder are
4017 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
4019 <p>Taking the remainder of a division by zero leads to undefined behavior.
4020 Overflow also leads to undefined behavior; this is a rare case, but can
4021 occur, for example, by taking the remainder of a 32-bit division of
4022 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
4023 lets srem be implemented using instructions that return both the result of
4024 the division and the remainder.)</p>
4028 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4033 <!-- _______________________________________________________________________ -->
4035 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
4042 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4046 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
4047 its two operands.</p>
4050 <p>The two arguments to the '<tt>frem</tt>' instruction must be
4051 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4052 floating point values. Both arguments must have identical types.</p>
4055 <p>This instruction returns the <i>remainder</i> of a division. The remainder
4056 has the same sign as the dividend.</p>
4060 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
4067 <!-- ======================================================================= -->
4069 <a name="bitwiseops">Bitwise Binary Operations</a>
4074 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
4075 program. They are generally very efficient instructions and can commonly be
4076 strength reduced from other instructions. They require two operands of the
4077 same type, execute an operation on them, and produce a single value. The
4078 resulting value is the same type as its operands.</p>
4080 <!-- _______________________________________________________________________ -->
4082 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
4089 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4090 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4091 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4092 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4096 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
4097 a specified number of bits.</p>
4100 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
4101 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4102 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4105 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
4106 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
4107 is (statically or dynamically) negative or equal to or larger than the number
4108 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4109 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4110 shift amount in <tt>op2</tt>.</p>
4112 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
4113 <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits. If
4114 the <tt>nsw</tt> keyword is present, then the shift produces a
4115 <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree
4116 with the resultant sign bit. As such, NUW/NSW have the same semantics as
4117 they would if the shift were expressed as a mul instruction with the same
4118 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
4122 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
4123 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
4124 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
4125 <result> = shl i32 1, 32 <i>; undefined</i>
4126 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4131 <!-- _______________________________________________________________________ -->
4133 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4140 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4141 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4145 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4146 operand shifted to the right a specified number of bits with zero fill.</p>
4149 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4150 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4151 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4154 <p>This instruction always performs a logical shift right operation. The most
4155 significant bits of the result will be filled with zero bits after the shift.
4156 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4157 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4158 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4159 shift amount in <tt>op2</tt>.</p>
4161 <p>If the <tt>exact</tt> keyword is present, the result value of the
4162 <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4163 shifted out are non-zero.</p>
4168 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4169 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4170 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4171 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4172 <result> = lshr i32 1, 32 <i>; undefined</i>
4173 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4178 <!-- _______________________________________________________________________ -->
4180 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4187 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4188 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4192 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4193 operand shifted to the right a specified number of bits with sign
4197 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4198 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4199 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4202 <p>This instruction always performs an arithmetic shift right operation, The
4203 most significant bits of the result will be filled with the sign bit
4204 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4205 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4206 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4207 the corresponding shift amount in <tt>op2</tt>.</p>
4209 <p>If the <tt>exact</tt> keyword is present, the result value of the
4210 <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4211 shifted out are non-zero.</p>
4215 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4216 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4217 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4218 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4219 <result> = ashr i32 1, 32 <i>; undefined</i>
4220 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4225 <!-- _______________________________________________________________________ -->
4227 <a name="i_and">'<tt>and</tt>' Instruction</a>
4234 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4238 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4242 <p>The two arguments to the '<tt>and</tt>' instruction must be
4243 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4244 values. Both arguments must have identical types.</p>
4247 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4249 <table border="1" cellspacing="0" cellpadding="4">
4281 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4282 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4283 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4286 <!-- _______________________________________________________________________ -->
4288 <a name="i_or">'<tt>or</tt>' Instruction</a>
4295 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4299 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4303 <p>The two arguments to the '<tt>or</tt>' instruction must be
4304 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4305 values. Both arguments must have identical types.</p>
4308 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4310 <table border="1" cellspacing="0" cellpadding="4">
4342 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4343 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4344 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4349 <!-- _______________________________________________________________________ -->
4351 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4358 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4362 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4363 its two operands. The <tt>xor</tt> is used to implement the "one's
4364 complement" operation, which is the "~" operator in C.</p>
4367 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4368 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4369 values. Both arguments must have identical types.</p>
4372 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4374 <table border="1" cellspacing="0" cellpadding="4">
4406 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4407 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4408 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4409 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4416 <!-- ======================================================================= -->
4418 <a name="vectorops">Vector Operations</a>
4423 <p>LLVM supports several instructions to represent vector operations in a
4424 target-independent manner. These instructions cover the element-access and
4425 vector-specific operations needed to process vectors effectively. While LLVM
4426 does directly support these vector operations, many sophisticated algorithms
4427 will want to use target-specific intrinsics to take full advantage of a
4428 specific target.</p>
4430 <!-- _______________________________________________________________________ -->
4432 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4439 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4443 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4444 from a vector at a specified index.</p>
4448 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4449 of <a href="#t_vector">vector</a> type. The second operand is an index
4450 indicating the position from which to extract the element. The index may be
4454 <p>The result is a scalar of the same type as the element type of
4455 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4456 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4457 results are undefined.</p>
4461 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4466 <!-- _______________________________________________________________________ -->
4468 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4475 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4479 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4480 vector at a specified index.</p>
4483 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4484 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4485 whose type must equal the element type of the first operand. The third
4486 operand is an index indicating the position at which to insert the value.
4487 The index may be a variable.</p>
4490 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4491 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4492 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4493 results are undefined.</p>
4497 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4502 <!-- _______________________________________________________________________ -->
4504 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4511 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4515 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4516 from two input vectors, returning a vector with the same element type as the
4517 input and length that is the same as the shuffle mask.</p>
4520 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4521 with types that match each other. The third argument is a shuffle mask whose
4522 element type is always 'i32'. The result of the instruction is a vector
4523 whose length is the same as the shuffle mask and whose element type is the
4524 same as the element type of the first two operands.</p>
4526 <p>The shuffle mask operand is required to be a constant vector with either
4527 constant integer or undef values.</p>
4530 <p>The elements of the two input vectors are numbered from left to right across
4531 both of the vectors. The shuffle mask operand specifies, for each element of
4532 the result vector, which element of the two input vectors the result element
4533 gets. The element selector may be undef (meaning "don't care") and the
4534 second operand may be undef if performing a shuffle from only one vector.</p>
4538 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4539 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4540 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4541 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4542 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4543 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4544 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4545 <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>
4552 <!-- ======================================================================= -->
4554 <a name="aggregateops">Aggregate Operations</a>
4559 <p>LLVM supports several instructions for working with
4560 <a href="#t_aggregate">aggregate</a> values.</p>
4562 <!-- _______________________________________________________________________ -->
4564 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4571 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4575 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4576 from an <a href="#t_aggregate">aggregate</a> value.</p>
4579 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4580 of <a href="#t_struct">struct</a> or
4581 <a href="#t_array">array</a> type. The operands are constant indices to
4582 specify which value to extract in a similar manner as indices in a
4583 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4584 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4586 <li>Since the value being indexed is not a pointer, the first index is
4587 omitted and assumed to be zero.</li>
4588 <li>At least one index must be specified.</li>
4589 <li>Not only struct indices but also array indices must be in
4594 <p>The result is the value at the position in the aggregate specified by the
4599 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4604 <!-- _______________________________________________________________________ -->
4606 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4613 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4617 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4618 in an <a href="#t_aggregate">aggregate</a> value.</p>
4621 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4622 of <a href="#t_struct">struct</a> or
4623 <a href="#t_array">array</a> type. The second operand is a first-class
4624 value to insert. The following operands are constant indices indicating
4625 the position at which to insert the value in a similar manner as indices in a
4626 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4627 value to insert must have the same type as the value identified by the
4631 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4632 that of <tt>val</tt> except that the value at the position specified by the
4633 indices is that of <tt>elt</tt>.</p>
4637 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4638 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4639 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4646 <!-- ======================================================================= -->
4648 <a name="memoryops">Memory Access and Addressing Operations</a>
4653 <p>A key design point of an SSA-based representation is how it represents
4654 memory. In LLVM, no memory locations are in SSA form, which makes things
4655 very simple. This section describes how to read, write, and allocate
4658 <!-- _______________________________________________________________________ -->
4660 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4667 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4671 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4672 currently executing function, to be automatically released when this function
4673 returns to its caller. The object is always allocated in the generic address
4674 space (address space zero).</p>
4677 <p>The '<tt>alloca</tt>' instruction
4678 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4679 runtime stack, returning a pointer of the appropriate type to the program.
4680 If "NumElements" is specified, it is the number of elements allocated,
4681 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4682 specified, the value result of the allocation is guaranteed to be aligned to
4683 at least that boundary. If not specified, or if zero, the target can choose
4684 to align the allocation on any convenient boundary compatible with the
4687 <p>'<tt>type</tt>' may be any sized type.</p>
4690 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4691 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4692 memory is automatically released when the function returns. The
4693 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4694 variables that must have an address available. When the function returns
4695 (either with the <tt><a href="#i_ret">ret</a></tt>
4696 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
4697 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4701 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4702 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4703 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4704 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4709 <!-- _______________________________________________________________________ -->
4711 <a name="i_load">'<tt>load</tt>' Instruction</a>
4718 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4719 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4720 !<index> = !{ i32 1 }
4724 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4727 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4728 from which to load. The pointer must point to
4729 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4730 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4731 number or order of execution of this <tt>load</tt> with other <a
4732 href="#volatile">volatile operations</a>.</p>
4734 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
4735 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4736 argument. The <code>release</code> and <code>acq_rel</code> orderings are
4737 not valid on <code>load</code> instructions. Atomic loads produce <a
4738 href="#memorymodel">defined</a> results when they may see multiple atomic
4739 stores. The type of the pointee must be an integer type whose bit width
4740 is a power of two greater than or equal to eight and less than or equal
4741 to a target-specific size limit. <code>align</code> must be explicitly
4742 specified on atomic loads, and the load has undefined behavior if the
4743 alignment is not set to a value which is at least the size in bytes of
4744 the pointee. <code>!nontemporal</code> does not have any defined semantics
4745 for atomic loads.</p>
4747 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4748 operation (that is, the alignment of the memory address). A value of 0 or an
4749 omitted <tt>align</tt> argument means that the operation has the preferential
4750 alignment for the target. It is the responsibility of the code emitter to
4751 ensure that the alignment information is correct. Overestimating the
4752 alignment results in undefined behavior. Underestimating the alignment may
4753 produce less efficient code. An alignment of 1 is always safe.</p>
4755 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4756 metatadata name <index> corresponding to a metadata node with
4757 one <tt>i32</tt> entry of value 1. The existence of
4758 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4759 and code generator that this load is not expected to be reused in the cache.
4760 The code generator may select special instructions to save cache bandwidth,
4761 such as the <tt>MOVNT</tt> instruction on x86.</p>
4764 <p>The location of memory pointed to is loaded. If the value being loaded is of
4765 scalar type then the number of bytes read does not exceed the minimum number
4766 of bytes needed to hold all bits of the type. For example, loading an
4767 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4768 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4769 is undefined if the value was not originally written using a store of the
4774 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4775 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4776 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4781 <!-- _______________________________________________________________________ -->
4783 <a name="i_store">'<tt>store</tt>' Instruction</a>
4790 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4791 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
4795 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4798 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4799 and an address at which to store it. The type of the
4800 '<tt><pointer></tt>' operand must be a pointer to
4801 the <a href="#t_firstclass">first class</a> type of the
4802 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
4803 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
4804 order of execution of this <tt>store</tt> with other <a
4805 href="#volatile">volatile operations</a>.</p>
4807 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
4808 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4809 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
4810 valid on <code>store</code> instructions. Atomic loads produce <a
4811 href="#memorymodel">defined</a> results when they may see multiple atomic
4812 stores. The type of the pointee must be an integer type whose bit width
4813 is a power of two greater than or equal to eight and less than or equal
4814 to a target-specific size limit. <code>align</code> must be explicitly
4815 specified on atomic stores, and the store has undefined behavior if the
4816 alignment is not set to a value which is at least the size in bytes of
4817 the pointee. <code>!nontemporal</code> does not have any defined semantics
4818 for atomic stores.</p>
4820 <p>The optional constant "align" argument specifies the alignment of the
4821 operation (that is, the alignment of the memory address). A value of 0 or an
4822 omitted "align" argument means that the operation has the preferential
4823 alignment for the target. It is the responsibility of the code emitter to
4824 ensure that the alignment information is correct. Overestimating the
4825 alignment results in an undefined behavior. Underestimating the alignment may
4826 produce less efficient code. An alignment of 1 is always safe.</p>
4828 <p>The optional !nontemporal metadata must reference a single metatadata
4829 name <index> corresponding to a metadata node with one i32 entry of
4830 value 1. The existence of the !nontemporal metatadata on the
4831 instruction tells the optimizer and code generator that this load is
4832 not expected to be reused in the cache. The code generator may
4833 select special instructions to save cache bandwidth, such as the
4834 MOVNT instruction on x86.</p>
4838 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4839 location specified by the '<tt><pointer></tt>' operand. If
4840 '<tt><value></tt>' is of scalar type then the number of bytes written
4841 does not exceed the minimum number of bytes needed to hold all bits of the
4842 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4843 writing a value of a type like <tt>i20</tt> with a size that is not an
4844 integral number of bytes, it is unspecified what happens to the extra bits
4845 that do not belong to the type, but they will typically be overwritten.</p>
4849 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4850 store i32 3, i32* %ptr <i>; yields {void}</i>
4851 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4856 <!-- _______________________________________________________________________ -->
4858 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
4865 fence [singlethread] <ordering> <i>; yields {void}</i>
4869 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
4870 between operations.</p>
4872 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
4873 href="#ordering">ordering</a> argument which defines what
4874 <i>synchronizes-with</i> edges they add. They can only be given
4875 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
4876 <code>seq_cst</code> orderings.</p>
4879 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
4880 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
4881 <code>acquire</code> ordering semantics if and only if there exist atomic
4882 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
4883 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
4884 <var>X</var> modifies <var>M</var> (either directly or through some side effect
4885 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
4886 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
4887 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
4888 than an explicit <code>fence</code>, one (but not both) of the atomic operations
4889 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
4890 <code>acquire</code> (resp.) ordering constraint and still
4891 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
4892 <i>happens-before</i> edge.</p>
4894 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
4895 having both <code>acquire</code> and <code>release</code> semantics specified
4896 above, participates in the global program order of other <code>seq_cst</code>
4897 operations and/or fences.</p>
4899 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
4900 specifies that the fence only synchronizes with other fences in the same
4901 thread. (This is useful for interacting with signal handlers.)</p>
4905 fence acquire <i>; yields {void}</i>
4906 fence singlethread seq_cst <i>; yields {void}</i>
4911 <!-- _______________________________________________________________________ -->
4913 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
4920 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
4924 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
4925 It loads a value in memory and compares it to a given value. If they are
4926 equal, it stores a new value into the memory.</p>
4929 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
4930 address to operate on, a value to compare to the value currently be at that
4931 address, and a new value to place at that address if the compared values are
4932 equal. The type of '<var><cmp></var>' must be an integer type whose
4933 bit width is a power of two greater than or equal to eight and less than
4934 or equal to a target-specific size limit. '<var><cmp></var>' and
4935 '<var><new></var>' must have the same type, and the type of
4936 '<var><pointer></var>' must be a pointer to that type. If the
4937 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
4938 optimizer is not allowed to modify the number or order of execution
4939 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
4942 <!-- FIXME: Extend allowed types. -->
4944 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
4945 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
4947 <p>The optional "<code>singlethread</code>" argument declares that the
4948 <code>cmpxchg</code> is only atomic with respect to code (usually signal
4949 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
4950 cmpxchg is atomic with respect to all other code in the system.</p>
4952 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
4953 the size in memory of the operand.
4956 <p>The contents of memory at the location specified by the
4957 '<tt><pointer></tt>' operand is read and compared to
4958 '<tt><cmp></tt>'; if the read value is the equal,
4959 '<tt><new></tt>' is written. The original value at the location
4962 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
4963 purpose of identifying <a href="#release_sequence">release sequences</a>. A
4964 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
4965 parameter determined by dropping any <code>release</code> part of the
4966 <code>cmpxchg</code>'s ordering.</p>
4969 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
4970 optimization work on ARM.)
4972 FIXME: Is a weaker ordering constraint on failure helpful in practice?
4978 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
4979 <a href="#i_br">br</a> label %loop
4982 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
4983 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
4984 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
4985 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
4986 <a href="#i_br">br</a> i1 %success, label %done, label %loop
4994 <!-- _______________________________________________________________________ -->
4996 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
5003 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
5007 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
5010 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
5011 operation to apply, an address whose value to modify, an argument to the
5012 operation. The operation must be one of the following keywords:</p>
5027 <p>The type of '<var><value></var>' must be an integer type whose
5028 bit width is a power of two greater than or equal to eight and less than
5029 or equal to a target-specific size limit. The type of the
5030 '<code><pointer></code>' operand must be a pointer to that type.
5031 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
5032 optimizer is not allowed to modify the number or order of execution of this
5033 <code>atomicrmw</code> with other <a href="#volatile">volatile
5036 <!-- FIXME: Extend allowed types. -->
5039 <p>The contents of memory at the location specified by the
5040 '<tt><pointer></tt>' operand are atomically read, modified, and written
5041 back. The original value at the location is returned. The modification is
5042 specified by the <var>operation</var> argument:</p>
5045 <li>xchg: <code>*ptr = val</code></li>
5046 <li>add: <code>*ptr = *ptr + val</code></li>
5047 <li>sub: <code>*ptr = *ptr - val</code></li>
5048 <li>and: <code>*ptr = *ptr & val</code></li>
5049 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
5050 <li>or: <code>*ptr = *ptr | val</code></li>
5051 <li>xor: <code>*ptr = *ptr ^ val</code></li>
5052 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
5053 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
5054 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
5055 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
5060 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
5065 <!-- _______________________________________________________________________ -->
5067 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
5074 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5075 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5076 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5080 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
5081 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
5082 It performs address calculation only and does not access memory.</p>
5085 <p>The first argument is always a pointer or a vector of pointers,
5086 and forms the basis of the
5087 calculation. The remaining arguments are indices that indicate which of the
5088 elements of the aggregate object are indexed. The interpretation of each
5089 index is dependent on the type being indexed into. The first index always
5090 indexes the pointer value given as the first argument, the second index
5091 indexes a value of the type pointed to (not necessarily the value directly
5092 pointed to, since the first index can be non-zero), etc. The first type
5093 indexed into must be a pointer value, subsequent types can be arrays,
5094 vectors, and structs. Note that subsequent types being indexed into
5095 can never be pointers, since that would require loading the pointer before
5096 continuing calculation.</p>
5098 <p>The type of each index argument depends on the type it is indexing into.
5099 When indexing into a (optionally packed) structure, only <tt>i32</tt>
5100 integer <b>constants</b> are allowed. When indexing into an array, pointer
5101 or vector, integers of any width are allowed, and they are not required to be
5102 constant. These integers are treated as signed values where relevant.</p>
5104 <p>For example, let's consider a C code fragment and how it gets compiled to
5107 <pre class="doc_code">
5119 int *foo(struct ST *s) {
5120 return &s[1].Z.B[5][13];
5124 <p>The LLVM code generated by Clang is:</p>
5126 <pre class="doc_code">
5127 %struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 }
5128 %struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT }
5130 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5132 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5138 <p>In the example above, the first index is indexing into the
5139 '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a
5140 '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a
5141 structure. The second index indexes into the third element of the structure,
5142 yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>'
5143 type, another structure. The third index indexes into the second element of
5144 the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The
5145 two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>'
5146 type. The '<tt>getelementptr</tt>' instruction returns a pointer to this
5147 element, thus computing a value of '<tt>i32*</tt>' type.</p>
5149 <p>Note that it is perfectly legal to index partially through a structure,
5150 returning a pointer to an inner element. Because of this, the LLVM code for
5151 the given testcase is equivalent to:</p>
5153 <pre class="doc_code">
5154 define i32* @foo(%struct.ST* %s) {
5155 %t1 = getelementptr %struct.ST* %s, i32 1 <i>; yields %struct.ST*:%t1</i>
5156 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 <i>; yields %struct.RT*:%t2</i>
5157 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5158 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5159 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5164 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5165 <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the
5166 base pointer is not an <i>in bounds</i> address of an allocated object,
5167 or if any of the addresses that would be formed by successive addition of
5168 the offsets implied by the indices to the base address with infinitely
5169 precise signed arithmetic are not an <i>in bounds</i> address of that
5170 allocated object. The <i>in bounds</i> addresses for an allocated object
5171 are all the addresses that point into the object, plus the address one
5173 In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword
5174 applies to each of the computations element-wise. </p>
5176 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5177 the base address with silently-wrapping two's complement arithmetic. If the
5178 offsets have a different width from the pointer, they are sign-extended or
5179 truncated to the width of the pointer. The result value of the
5180 <tt>getelementptr</tt> may be outside the object pointed to by the base
5181 pointer. The result value may not necessarily be used to access memory
5182 though, even if it happens to point into allocated storage. See the
5183 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5186 <p>The getelementptr instruction is often confusing. For some more insight into
5187 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5191 <i>; yields [12 x i8]*:aptr</i>
5192 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5193 <i>; yields i8*:vptr</i>
5194 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5195 <i>; yields i8*:eptr</i>
5196 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5197 <i>; yields i32*:iptr</i>
5198 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5201 <p>In cases where the pointer argument is a vector of pointers, only a
5202 single index may be used, and the number of vector elements has to be
5203 the same. For example: </p>
5204 <pre class="doc_code">
5205 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5212 <!-- ======================================================================= -->
5214 <a name="convertops">Conversion Operations</a>
5219 <p>The instructions in this category are the conversion instructions (casting)
5220 which all take a single operand and a type. They perform various bit
5221 conversions on the operand.</p>
5223 <!-- _______________________________________________________________________ -->
5225 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5232 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5236 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5237 type <tt>ty2</tt>.</p>
5240 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5241 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5242 of the same number of integers.
5243 The bit size of the <tt>value</tt> must be larger than
5244 the bit size of the destination type, <tt>ty2</tt>.
5245 Equal sized types are not allowed.</p>
5248 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5249 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5250 source size must be larger than the destination size, <tt>trunc</tt> cannot
5251 be a <i>no-op cast</i>. It will always truncate bits.</p>
5255 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5256 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5257 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5258 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5263 <!-- _______________________________________________________________________ -->
5265 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5272 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5276 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5281 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5282 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5283 of the same number of integers.
5284 The bit size of the <tt>value</tt> must be smaller than
5285 the bit size of the destination type,
5289 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5290 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5292 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5296 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5297 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5298 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5303 <!-- _______________________________________________________________________ -->
5305 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5312 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5316 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5319 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5320 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5321 of the same number of integers.
5322 The bit size of the <tt>value</tt> must be smaller than
5323 the bit size of the destination type,
5327 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5328 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5329 of the type <tt>ty2</tt>.</p>
5331 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5335 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5336 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5337 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5342 <!-- _______________________________________________________________________ -->
5344 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5351 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5355 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5359 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5360 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5361 to cast it to. The size of <tt>value</tt> must be larger than the size of
5362 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5363 <i>no-op cast</i>.</p>
5366 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5367 <a href="#t_floating">floating point</a> type to a smaller
5368 <a href="#t_floating">floating point</a> type. If the value cannot fit
5369 within the destination type, <tt>ty2</tt>, then the results are
5374 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5375 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5380 <!-- _______________________________________________________________________ -->
5382 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5389 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5393 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5394 floating point value.</p>
5397 <p>The '<tt>fpext</tt>' instruction takes a
5398 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5399 a <a href="#t_floating">floating point</a> type to cast it to. The source
5400 type must be smaller than the destination type.</p>
5403 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5404 <a href="#t_floating">floating point</a> type to a larger
5405 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5406 used to make a <i>no-op cast</i> because it always changes bits. Use
5407 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5411 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5412 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5417 <!-- _______________________________________________________________________ -->
5419 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5426 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5430 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5431 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5434 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5435 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5436 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5437 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5438 vector integer type with the same number of elements as <tt>ty</tt></p>
5441 <p>The '<tt>fptoui</tt>' instruction converts its
5442 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5443 towards zero) unsigned integer value. If the value cannot fit
5444 in <tt>ty2</tt>, the results are undefined.</p>
5448 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5449 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5450 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5455 <!-- _______________________________________________________________________ -->
5457 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5464 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5468 <p>The '<tt>fptosi</tt>' instruction converts
5469 <a href="#t_floating">floating point</a> <tt>value</tt> to
5470 type <tt>ty2</tt>.</p>
5473 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5474 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5475 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5476 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5477 vector integer type with the same number of elements as <tt>ty</tt></p>
5480 <p>The '<tt>fptosi</tt>' instruction converts its
5481 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5482 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5483 the results are undefined.</p>
5487 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5488 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5489 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5494 <!-- _______________________________________________________________________ -->
5496 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5503 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5507 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5508 integer and converts that value to the <tt>ty2</tt> type.</p>
5511 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5512 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5513 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5514 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5515 floating point type with the same number of elements as <tt>ty</tt></p>
5518 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5519 integer quantity and converts it to the corresponding floating point
5520 value. If the value cannot fit in the floating point value, the results are
5525 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5526 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5531 <!-- _______________________________________________________________________ -->
5533 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5540 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5544 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5545 and converts that value to the <tt>ty2</tt> type.</p>
5548 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5549 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5550 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5551 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5552 floating point type with the same number of elements as <tt>ty</tt></p>
5555 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5556 quantity and converts it to the corresponding floating point value. If the
5557 value cannot fit in the floating point value, the results are undefined.</p>
5561 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5562 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5567 <!-- _______________________________________________________________________ -->
5569 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5576 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5580 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of
5581 pointers <tt>value</tt> to
5582 the integer (or vector of integers) type <tt>ty2</tt>.</p>
5585 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5586 must be a a value of type <a href="#t_pointer">pointer</a> or a vector of
5587 pointers, and a type to cast it to
5588 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector
5589 of integers type.</p>
5592 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5593 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5594 truncating or zero extending that value to the size of the integer type. If
5595 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5596 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5597 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5602 %X = ptrtoint i32* %P to i8 <i>; yields truncation on 32-bit architecture</i>
5603 %Y = ptrtoint i32* %P to i64 <i>; yields zero extension on 32-bit architecture</i>
5604 %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>
5609 <!-- _______________________________________________________________________ -->
5611 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5618 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5622 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5623 pointer type, <tt>ty2</tt>.</p>
5626 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5627 value to cast, and a type to cast it to, which must be a
5628 <a href="#t_pointer">pointer</a> type.</p>
5631 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5632 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5633 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5634 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5635 than the size of a pointer then a zero extension is done. If they are the
5636 same size, nothing is done (<i>no-op cast</i>).</p>
5640 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5641 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5642 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5643 %Z = inttoptr <4 x i32> %G to <4 x i8*><i>; yields truncation of vector G to four pointers</i>
5648 <!-- _______________________________________________________________________ -->
5650 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5657 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5661 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5662 <tt>ty2</tt> without changing any bits.</p>
5665 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5666 non-aggregate first class value, and a type to cast it to, which must also be
5667 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5668 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5669 identical. If the source type is a pointer, the destination type must also be
5670 a pointer. This instruction supports bitwise conversion of vectors to
5671 integers and to vectors of other types (as long as they have the same
5675 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5676 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5677 this conversion. The conversion is done as if the <tt>value</tt> had been
5678 stored to memory and read back as type <tt>ty2</tt>.
5679 Pointer (or vector of pointers) types may only be converted to other pointer
5680 (or vector of pointers) types with this instruction. To convert
5681 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5682 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5686 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
5687 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
5688 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
5689 %Z = bitcast <2 x i32*> %V to <2 x i64*> <i>; yields <2 x i64*></i>
5696 <!-- ======================================================================= -->
5698 <a name="otherops">Other Operations</a>
5703 <p>The instructions in this category are the "miscellaneous" instructions, which
5704 defy better classification.</p>
5706 <!-- _______________________________________________________________________ -->
5708 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5715 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5719 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5720 boolean values based on comparison of its two integer, integer vector,
5721 pointer, or pointer vector operands.</p>
5724 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
5725 the condition code indicating the kind of comparison to perform. It is not a
5726 value, just a keyword. The possible condition code are:</p>
5729 <li><tt>eq</tt>: equal</li>
5730 <li><tt>ne</tt>: not equal </li>
5731 <li><tt>ugt</tt>: unsigned greater than</li>
5732 <li><tt>uge</tt>: unsigned greater or equal</li>
5733 <li><tt>ult</tt>: unsigned less than</li>
5734 <li><tt>ule</tt>: unsigned less or equal</li>
5735 <li><tt>sgt</tt>: signed greater than</li>
5736 <li><tt>sge</tt>: signed greater or equal</li>
5737 <li><tt>slt</tt>: signed less than</li>
5738 <li><tt>sle</tt>: signed less or equal</li>
5741 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
5742 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
5743 typed. They must also be identical types.</p>
5746 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
5747 condition code given as <tt>cond</tt>. The comparison performed always yields
5748 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
5749 result, as follows:</p>
5752 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
5753 <tt>false</tt> otherwise. No sign interpretation is necessary or
5756 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
5757 <tt>false</tt> otherwise. No sign interpretation is necessary or
5760 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
5761 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5763 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
5764 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5765 to <tt>op2</tt>.</li>
5767 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
5768 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5770 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
5771 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5773 <li><tt>sgt</tt>: interprets the operands as signed values and yields
5774 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5776 <li><tt>sge</tt>: interprets the operands as signed values and yields
5777 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5778 to <tt>op2</tt>.</li>
5780 <li><tt>slt</tt>: interprets the operands as signed values and yields
5781 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5783 <li><tt>sle</tt>: interprets the operands as signed values and yields
5784 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5787 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
5788 values are compared as if they were integers.</p>
5790 <p>If the operands are integer vectors, then they are compared element by
5791 element. The result is an <tt>i1</tt> vector with the same number of elements
5792 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
5796 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
5797 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
5798 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
5799 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
5800 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
5801 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
5804 <p>Note that the code generator does not yet support vector types with
5805 the <tt>icmp</tt> instruction.</p>
5809 <!-- _______________________________________________________________________ -->
5811 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
5818 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5822 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
5823 values based on comparison of its operands.</p>
5825 <p>If the operands are floating point scalars, then the result type is a boolean
5826 (<a href="#t_integer"><tt>i1</tt></a>).</p>
5828 <p>If the operands are floating point vectors, then the result type is a vector
5829 of boolean with the same number of elements as the operands being
5833 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
5834 the condition code indicating the kind of comparison to perform. It is not a
5835 value, just a keyword. The possible condition code are:</p>
5838 <li><tt>false</tt>: no comparison, always returns false</li>
5839 <li><tt>oeq</tt>: ordered and equal</li>
5840 <li><tt>ogt</tt>: ordered and greater than </li>
5841 <li><tt>oge</tt>: ordered and greater than or equal</li>
5842 <li><tt>olt</tt>: ordered and less than </li>
5843 <li><tt>ole</tt>: ordered and less than or equal</li>
5844 <li><tt>one</tt>: ordered and not equal</li>
5845 <li><tt>ord</tt>: ordered (no nans)</li>
5846 <li><tt>ueq</tt>: unordered or equal</li>
5847 <li><tt>ugt</tt>: unordered or greater than </li>
5848 <li><tt>uge</tt>: unordered or greater than or equal</li>
5849 <li><tt>ult</tt>: unordered or less than </li>
5850 <li><tt>ule</tt>: unordered or less than or equal</li>
5851 <li><tt>une</tt>: unordered or not equal</li>
5852 <li><tt>uno</tt>: unordered (either nans)</li>
5853 <li><tt>true</tt>: no comparison, always returns true</li>
5856 <p><i>Ordered</i> means that neither operand is a QNAN while
5857 <i>unordered</i> means that either operand may be a QNAN.</p>
5859 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
5860 a <a href="#t_floating">floating point</a> type or
5861 a <a href="#t_vector">vector</a> of floating point type. They must have
5862 identical types.</p>
5865 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
5866 according to the condition code given as <tt>cond</tt>. If the operands are
5867 vectors, then the vectors are compared element by element. Each comparison
5868 performed always yields an <a href="#t_integer">i1</a> result, as
5872 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
5874 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5875 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5877 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5878 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5880 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5881 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5883 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5884 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5886 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5887 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5889 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5890 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5892 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
5894 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
5895 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5897 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
5898 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5900 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
5901 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5903 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
5904 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5906 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
5907 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5909 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
5910 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5912 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
5914 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
5919 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
5920 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
5921 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
5922 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
5925 <p>Note that the code generator does not yet support vector types with
5926 the <tt>fcmp</tt> instruction.</p>
5930 <!-- _______________________________________________________________________ -->
5932 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
5939 <result> = phi <ty> [ <val0>, <label0>], ...
5943 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
5944 SSA graph representing the function.</p>
5947 <p>The type of the incoming values is specified with the first type field. After
5948 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
5949 one pair for each predecessor basic block of the current block. Only values
5950 of <a href="#t_firstclass">first class</a> type may be used as the value
5951 arguments to the PHI node. Only labels may be used as the label
5954 <p>There must be no non-phi instructions between the start of a basic block and
5955 the PHI instructions: i.e. PHI instructions must be first in a basic
5958 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5959 occur on the edge from the corresponding predecessor block to the current
5960 block (but after any definition of an '<tt>invoke</tt>' instruction's return
5961 value on the same edge).</p>
5964 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
5965 specified by the pair corresponding to the predecessor basic block that
5966 executed just prior to the current block.</p>
5970 Loop: ; Infinite loop that counts from 0 on up...
5971 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5972 %nextindvar = add i32 %indvar, 1
5978 <!-- _______________________________________________________________________ -->
5980 <a name="i_select">'<tt>select</tt>' Instruction</a>
5987 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5989 <i>selty</i> is either i1 or {<N x i1>}
5993 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
5994 condition, without branching.</p>
5998 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
5999 values indicating the condition, and two values of the
6000 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
6001 vectors and the condition is a scalar, then entire vectors are selected, not
6002 individual elements.</p>
6005 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
6006 first value argument; otherwise, it returns the second value argument.</p>
6008 <p>If the condition is a vector of i1, then the value arguments must be vectors
6009 of the same size, and the selection is done element by element.</p>
6013 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
6018 <!-- _______________________________________________________________________ -->
6020 <a name="i_call">'<tt>call</tt>' Instruction</a>
6027 <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>]
6031 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
6034 <p>This instruction requires several arguments:</p>
6037 <li>The optional "tail" marker indicates that the callee function does not
6038 access any allocas or varargs in the caller. Note that calls may be
6039 marked "tail" even if they do not occur before
6040 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
6041 present, the function call is eligible for tail call optimization,
6042 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
6043 optimized into a jump</a>. The code generator may optimize calls marked
6044 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
6045 sibling call optimization</a> when the caller and callee have
6046 matching signatures, or 2) forced tail call optimization when the
6047 following extra requirements are met:
6049 <li>Caller and callee both have the calling
6050 convention <tt>fastcc</tt>.</li>
6051 <li>The call is in tail position (ret immediately follows call and ret
6052 uses value of call or is void).</li>
6053 <li>Option <tt>-tailcallopt</tt> is enabled,
6054 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
6055 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
6056 constraints are met.</a></li>
6060 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
6061 convention</a> the call should use. If none is specified, the call
6062 defaults to using C calling conventions. The calling convention of the
6063 call must match the calling convention of the target function, or else the
6064 behavior is undefined.</li>
6066 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
6067 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
6068 '<tt>inreg</tt>' attributes are valid here.</li>
6070 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
6071 type of the return value. Functions that return no value are marked
6072 <tt><a href="#t_void">void</a></tt>.</li>
6074 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
6075 being invoked. The argument types must match the types implied by this
6076 signature. This type can be omitted if the function is not varargs and if
6077 the function type does not return a pointer to a function.</li>
6079 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
6080 be invoked. In most cases, this is a direct function invocation, but
6081 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
6082 to function value.</li>
6084 <li>'<tt>function args</tt>': argument list whose types match the function
6085 signature argument types and parameter attributes. All arguments must be
6086 of <a href="#t_firstclass">first class</a> type. If the function
6087 signature indicates the function accepts a variable number of arguments,
6088 the extra arguments can be specified.</li>
6090 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
6091 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
6092 '<tt>readnone</tt>' attributes are valid here.</li>
6096 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
6097 a specified function, with its incoming arguments bound to the specified
6098 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
6099 function, control flow continues with the instruction after the function
6100 call, and the return value of the function is bound to the result
6105 %retval = call i32 @test(i32 %argc)
6106 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
6107 %X = tail call i32 @foo() <i>; yields i32</i>
6108 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
6109 call void %foo(i8 97 signext)
6111 %struct.A = type { i32, i8 }
6112 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
6113 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
6114 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
6115 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
6116 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
6119 <p>llvm treats calls to some functions with names and arguments that match the
6120 standard C99 library as being the C99 library functions, and may perform
6121 optimizations or generate code for them under that assumption. This is
6122 something we'd like to change in the future to provide better support for
6123 freestanding environments and non-C-based languages.</p>
6127 <!-- _______________________________________________________________________ -->
6129 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
6136 <resultval> = va_arg <va_list*> <arglist>, <argty>
6140 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6141 the "variable argument" area of a function call. It is used to implement the
6142 <tt>va_arg</tt> macro in C.</p>
6145 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6146 argument. It returns a value of the specified argument type and increments
6147 the <tt>va_list</tt> to point to the next argument. The actual type
6148 of <tt>va_list</tt> is target specific.</p>
6151 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6152 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6153 to the next argument. For more information, see the variable argument
6154 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6156 <p>It is legal for this instruction to be called in a function which does not
6157 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6160 <p><tt>va_arg</tt> is an LLVM instruction instead of
6161 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6165 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6167 <p>Note that the code generator does not yet fully support va_arg on many
6168 targets. Also, it does not currently support va_arg with aggregate types on
6173 <!-- _______________________________________________________________________ -->
6175 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6182 <resultval> = landingpad <somety> personality <type> <pers_fn> <clause>+
6183 <resultval> = landingpad <somety> personality <type> <pers_fn> cleanup <clause>*
6185 <clause> := catch <type> <value>
6186 <clause> := filter <array constant type> <array constant>
6190 <p>The '<tt>landingpad</tt>' instruction is used by
6191 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6192 system</a> to specify that a basic block is a landing pad — one where
6193 the exception lands, and corresponds to the code found in the
6194 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6195 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6196 re-entry to the function. The <tt>resultval</tt> has the
6197 type <tt>somety</tt>.</p>
6200 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6201 function associated with the unwinding mechanism. The optional
6202 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6204 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6205 or <tt>filter</tt> — and contains the global variable representing the
6206 "type" that may be caught or filtered respectively. Unlike the
6207 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6208 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6209 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6210 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6213 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6214 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6215 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6216 calling conventions, how the personality function results are represented in
6217 LLVM IR is target specific.</p>
6219 <p>The clauses are applied in order from top to bottom. If two
6220 <tt>landingpad</tt> instructions are merged together through inlining, the
6221 clauses from the calling function are appended to the list of clauses.</p>
6223 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6226 <li>A landing pad block is a basic block which is the unwind destination of an
6227 '<tt>invoke</tt>' instruction.</li>
6228 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6229 first non-PHI instruction.</li>
6230 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6232 <li>A basic block that is not a landing pad block may not include a
6233 '<tt>landingpad</tt>' instruction.</li>
6234 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6235 personality function.</li>
6240 ;; A landing pad which can catch an integer.
6241 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6243 ;; A landing pad that is a cleanup.
6244 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6246 ;; A landing pad which can catch an integer and can only throw a double.
6247 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6249 filter [1 x i8**] [@_ZTId]
6258 <!-- *********************************************************************** -->
6259 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6260 <!-- *********************************************************************** -->
6264 <p>LLVM supports the notion of an "intrinsic function". These functions have
6265 well known names and semantics and are required to follow certain
6266 restrictions. Overall, these intrinsics represent an extension mechanism for
6267 the LLVM language that does not require changing all of the transformations
6268 in LLVM when adding to the language (or the bitcode reader/writer, the
6269 parser, etc...).</p>
6271 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6272 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6273 begin with this prefix. Intrinsic functions must always be external
6274 functions: you cannot define the body of intrinsic functions. Intrinsic
6275 functions may only be used in call or invoke instructions: it is illegal to
6276 take the address of an intrinsic function. Additionally, because intrinsic
6277 functions are part of the LLVM language, it is required if any are added that
6278 they be documented here.</p>
6280 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6281 family of functions that perform the same operation but on different data
6282 types. Because LLVM can represent over 8 million different integer types,
6283 overloading is used commonly to allow an intrinsic function to operate on any
6284 integer type. One or more of the argument types or the result type can be
6285 overloaded to accept any integer type. Argument types may also be defined as
6286 exactly matching a previous argument's type or the result type. This allows
6287 an intrinsic function which accepts multiple arguments, but needs all of them
6288 to be of the same type, to only be overloaded with respect to a single
6289 argument or the result.</p>
6291 <p>Overloaded intrinsics will have the names of its overloaded argument types
6292 encoded into its function name, each preceded by a period. Only those types
6293 which are overloaded result in a name suffix. Arguments whose type is matched
6294 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6295 can take an integer of any width and returns an integer of exactly the same
6296 integer width. This leads to a family of functions such as
6297 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6298 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6299 suffix is required. Because the argument's type is matched against the return
6300 type, it does not require its own name suffix.</p>
6302 <p>To learn how to add an intrinsic function, please see the
6303 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6305 <!-- ======================================================================= -->
6307 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6312 <p>Variable argument support is defined in LLVM with
6313 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6314 intrinsic functions. These functions are related to the similarly named
6315 macros defined in the <tt><stdarg.h></tt> header file.</p>
6317 <p>All of these functions operate on arguments that use a target-specific value
6318 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6319 not define what this type is, so all transformations should be prepared to
6320 handle these functions regardless of the type used.</p>
6322 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6323 instruction and the variable argument handling intrinsic functions are
6326 <pre class="doc_code">
6327 define i32 @test(i32 %X, ...) {
6328 ; Initialize variable argument processing
6330 %ap2 = bitcast i8** %ap to i8*
6331 call void @llvm.va_start(i8* %ap2)
6333 ; Read a single integer argument
6334 %tmp = va_arg i8** %ap, i32
6336 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6338 %aq2 = bitcast i8** %aq to i8*
6339 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6340 call void @llvm.va_end(i8* %aq2)
6342 ; Stop processing of arguments.
6343 call void @llvm.va_end(i8* %ap2)
6347 declare void @llvm.va_start(i8*)
6348 declare void @llvm.va_copy(i8*, i8*)
6349 declare void @llvm.va_end(i8*)
6352 <!-- _______________________________________________________________________ -->
6354 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6362 declare void %llvm.va_start(i8* <arglist>)
6366 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6367 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6370 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6373 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6374 macro available in C. In a target-dependent way, it initializes
6375 the <tt>va_list</tt> element to which the argument points, so that the next
6376 call to <tt>va_arg</tt> will produce the first variable argument passed to
6377 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6378 need to know the last argument of the function as the compiler can figure
6383 <!-- _______________________________________________________________________ -->
6385 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6392 declare void @llvm.va_end(i8* <arglist>)
6396 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6397 which has been initialized previously
6398 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6399 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6402 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6405 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6406 macro available in C. In a target-dependent way, it destroys
6407 the <tt>va_list</tt> element to which the argument points. Calls
6408 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6409 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6410 with calls to <tt>llvm.va_end</tt>.</p>
6414 <!-- _______________________________________________________________________ -->
6416 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6423 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6427 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6428 from the source argument list to the destination argument list.</p>
6431 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6432 The second argument is a pointer to a <tt>va_list</tt> element to copy
6436 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6437 macro available in C. In a target-dependent way, it copies the
6438 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6439 element. This intrinsic is necessary because
6440 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6441 arbitrarily complex and require, for example, memory allocation.</p>
6447 <!-- ======================================================================= -->
6449 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6454 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6455 Collection</a> (GC) requires the implementation and generation of these
6456 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6457 roots on the stack</a>, as well as garbage collector implementations that
6458 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6459 barriers. Front-ends for type-safe garbage collected languages should generate
6460 these intrinsics to make use of the LLVM garbage collectors. For more details,
6461 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6464 <p>The garbage collection intrinsics only operate on objects in the generic
6465 address space (address space zero).</p>
6467 <!-- _______________________________________________________________________ -->
6469 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6476 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6480 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6481 the code generator, and allows some metadata to be associated with it.</p>
6484 <p>The first argument specifies the address of a stack object that contains the
6485 root pointer. The second pointer (which must be either a constant or a
6486 global value address) contains the meta-data to be associated with the
6490 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6491 location. At compile-time, the code generator generates information to allow
6492 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6493 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6498 <!-- _______________________________________________________________________ -->
6500 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6507 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6511 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6512 locations, allowing garbage collector implementations that require read
6516 <p>The second argument is the address to read from, which should be an address
6517 allocated from the garbage collector. The first object is a pointer to the
6518 start of the referenced object, if needed by the language runtime (otherwise
6522 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6523 instruction, but may be replaced with substantially more complex code by the
6524 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6525 may only be used in a function which <a href="#gc">specifies a GC
6530 <!-- _______________________________________________________________________ -->
6532 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6539 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6543 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6544 locations, allowing garbage collector implementations that require write
6545 barriers (such as generational or reference counting collectors).</p>
6548 <p>The first argument is the reference to store, the second is the start of the
6549 object to store it to, and the third is the address of the field of Obj to
6550 store to. If the runtime does not require a pointer to the object, Obj may
6554 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6555 instruction, but may be replaced with substantially more complex code by the
6556 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6557 may only be used in a function which <a href="#gc">specifies a GC
6564 <!-- ======================================================================= -->
6566 <a name="int_codegen">Code Generator Intrinsics</a>
6571 <p>These intrinsics are provided by LLVM to expose special features that may
6572 only be implemented with code generator support.</p>
6574 <!-- _______________________________________________________________________ -->
6576 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6583 declare i8 *@llvm.returnaddress(i32 <level>)
6587 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6588 target-specific value indicating the return address of the current function
6589 or one of its callers.</p>
6592 <p>The argument to this intrinsic indicates which function to return the address
6593 for. Zero indicates the calling function, one indicates its caller, etc.
6594 The argument is <b>required</b> to be a constant integer value.</p>
6597 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6598 indicating the return address of the specified call frame, or zero if it
6599 cannot be identified. The value returned by this intrinsic is likely to be
6600 incorrect or 0 for arguments other than zero, so it should only be used for
6601 debugging purposes.</p>
6603 <p>Note that calling this intrinsic does not prevent function inlining or other
6604 aggressive transformations, so the value returned may not be that of the
6605 obvious source-language caller.</p>
6609 <!-- _______________________________________________________________________ -->
6611 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6618 declare i8* @llvm.frameaddress(i32 <level>)
6622 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6623 target-specific frame pointer value for the specified stack frame.</p>
6626 <p>The argument to this intrinsic indicates which function to return the frame
6627 pointer for. Zero indicates the calling function, one indicates its caller,
6628 etc. The argument is <b>required</b> to be a constant integer value.</p>
6631 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6632 indicating the frame address of the specified call frame, or zero if it
6633 cannot be identified. The value returned by this intrinsic is likely to be
6634 incorrect or 0 for arguments other than zero, so it should only be used for
6635 debugging purposes.</p>
6637 <p>Note that calling this intrinsic does not prevent function inlining or other
6638 aggressive transformations, so the value returned may not be that of the
6639 obvious source-language caller.</p>
6643 <!-- _______________________________________________________________________ -->
6645 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6652 declare i8* @llvm.stacksave()
6656 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6657 of the function stack, for use
6658 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6659 useful for implementing language features like scoped automatic variable
6660 sized arrays in C99.</p>
6663 <p>This intrinsic returns a opaque pointer value that can be passed
6664 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6665 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6666 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6667 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6668 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6669 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6673 <!-- _______________________________________________________________________ -->
6675 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6682 declare void @llvm.stackrestore(i8* %ptr)
6686 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6687 the function stack to the state it was in when the
6688 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6689 executed. This is useful for implementing language features like scoped
6690 automatic variable sized arrays in C99.</p>
6693 <p>See the description
6694 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6698 <!-- _______________________________________________________________________ -->
6700 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6707 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6711 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6712 insert a prefetch instruction if supported; otherwise, it is a noop.
6713 Prefetches have no effect on the behavior of the program but can change its
6714 performance characteristics.</p>
6717 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6718 specifier determining if the fetch should be for a read (0) or write (1),
6719 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
6720 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6721 specifies whether the prefetch is performed on the data (1) or instruction (0)
6722 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
6723 must be constant integers.</p>
6726 <p>This intrinsic does not modify the behavior of the program. In particular,
6727 prefetches cannot trap and do not produce a value. On targets that support
6728 this intrinsic, the prefetch can provide hints to the processor cache for
6729 better performance.</p>
6733 <!-- _______________________________________________________________________ -->
6735 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
6742 declare void @llvm.pcmarker(i32 <id>)
6746 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
6747 Counter (PC) in a region of code to simulators and other tools. The method
6748 is target specific, but it is expected that the marker will use exported
6749 symbols to transmit the PC of the marker. The marker makes no guarantees
6750 that it will remain with any specific instruction after optimizations. It is
6751 possible that the presence of a marker will inhibit optimizations. The
6752 intended use is to be inserted after optimizations to allow correlations of
6753 simulation runs.</p>
6756 <p><tt>id</tt> is a numerical id identifying the marker.</p>
6759 <p>This intrinsic does not modify the behavior of the program. Backends that do
6760 not support this intrinsic may ignore it.</p>
6764 <!-- _______________________________________________________________________ -->
6766 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
6773 declare i64 @llvm.readcyclecounter()
6777 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
6778 counter register (or similar low latency, high accuracy clocks) on those
6779 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6780 should map to RPCC. As the backing counters overflow quickly (on the order
6781 of 9 seconds on alpha), this should only be used for small timings.</p>
6784 <p>When directly supported, reading the cycle counter should not modify any
6785 memory. Implementations are allowed to either return a application specific
6786 value or a system wide value. On backends without support, this is lowered
6787 to a constant 0.</p>
6793 <!-- ======================================================================= -->
6795 <a name="int_libc">Standard C Library Intrinsics</a>
6800 <p>LLVM provides intrinsics for a few important standard C library functions.
6801 These intrinsics allow source-language front-ends to pass information about
6802 the alignment of the pointer arguments to the code generator, providing
6803 opportunity for more efficient code generation.</p>
6805 <!-- _______________________________________________________________________ -->
6807 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
6813 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
6814 integer bit width and for different address spaces. Not all targets support
6815 all bit widths however.</p>
6818 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6819 i32 <len>, i32 <align>, i1 <isvolatile>)
6820 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6821 i64 <len>, i32 <align>, i1 <isvolatile>)
6825 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6826 source location to the destination location.</p>
6828 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
6829 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6830 and the pointers can be in specified address spaces.</p>
6834 <p>The first argument is a pointer to the destination, the second is a pointer
6835 to the source. The third argument is an integer argument specifying the
6836 number of bytes to copy, the fourth argument is the alignment of the
6837 source and destination locations, and the fifth is a boolean indicating a
6838 volatile access.</p>
6840 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6841 then the caller guarantees that both the source and destination pointers are
6842 aligned to that boundary.</p>
6844 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6845 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
6846 The detailed access behavior is not very cleanly specified and it is unwise
6847 to depend on it.</p>
6851 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6852 source location to the destination location, which are not allowed to
6853 overlap. It copies "len" bytes of memory over. If the argument is known to
6854 be aligned to some boundary, this can be specified as the fourth argument,
6855 otherwise it should be set to 0 or 1.</p>
6859 <!-- _______________________________________________________________________ -->
6861 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
6867 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
6868 width and for different address space. Not all targets support all bit
6872 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6873 i32 <len>, i32 <align>, i1 <isvolatile>)
6874 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6875 i64 <len>, i32 <align>, i1 <isvolatile>)
6879 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
6880 source location to the destination location. It is similar to the
6881 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
6884 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
6885 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6886 and the pointers can be in specified address spaces.</p>
6890 <p>The first argument is a pointer to the destination, the second is a pointer
6891 to the source. The third argument is an integer argument specifying the
6892 number of bytes to copy, the fourth argument is the alignment of the
6893 source and destination locations, and the fifth is a boolean indicating a
6894 volatile access.</p>
6896 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6897 then the caller guarantees that the source and destination pointers are
6898 aligned to that boundary.</p>
6900 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6901 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
6902 The detailed access behavior is not very cleanly specified and it is unwise
6903 to depend on it.</p>
6907 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
6908 source location to the destination location, which may overlap. It copies
6909 "len" bytes of memory over. If the argument is known to be aligned to some
6910 boundary, this can be specified as the fourth argument, otherwise it should
6911 be set to 0 or 1.</p>
6915 <!-- _______________________________________________________________________ -->
6917 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
6923 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6924 width and for different address spaces. However, not all targets support all
6928 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6929 i32 <len>, i32 <align>, i1 <isvolatile>)
6930 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6931 i64 <len>, i32 <align>, i1 <isvolatile>)
6935 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
6936 particular byte value.</p>
6938 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
6939 intrinsic does not return a value and takes extra alignment/volatile
6940 arguments. Also, the destination can be in an arbitrary address space.</p>
6943 <p>The first argument is a pointer to the destination to fill, the second is the
6944 byte value with which to fill it, the third argument is an integer argument
6945 specifying the number of bytes to fill, and the fourth argument is the known
6946 alignment of the destination location.</p>
6948 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6949 then the caller guarantees that the destination pointer is aligned to that
6952 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6953 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
6954 The detailed access behavior is not very cleanly specified and it is unwise
6955 to depend on it.</p>
6958 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
6959 at the destination location. If the argument is known to be aligned to some
6960 boundary, this can be specified as the fourth argument, otherwise it should
6961 be set to 0 or 1.</p>
6965 <!-- _______________________________________________________________________ -->
6967 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
6973 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
6974 floating point or vector of floating point type. Not all targets support all
6978 declare float @llvm.sqrt.f32(float %Val)
6979 declare double @llvm.sqrt.f64(double %Val)
6980 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6981 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6982 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6986 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
6987 returning the same value as the libm '<tt>sqrt</tt>' functions would.
6988 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
6989 behavior for negative numbers other than -0.0 (which allows for better
6990 optimization, because there is no need to worry about errno being
6991 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
6994 <p>The argument and return value are floating point numbers of the same
6998 <p>This function returns the sqrt of the specified operand if it is a
6999 nonnegative floating point number.</p>
7003 <!-- _______________________________________________________________________ -->
7005 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
7011 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
7012 floating point or vector of floating point type. Not all targets support all
7016 declare float @llvm.powi.f32(float %Val, i32 %power)
7017 declare double @llvm.powi.f64(double %Val, i32 %power)
7018 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7019 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7020 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7024 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
7025 specified (positive or negative) power. The order of evaluation of
7026 multiplications is not defined. When a vector of floating point type is
7027 used, the second argument remains a scalar integer value.</p>
7030 <p>The second argument is an integer power, and the first is a value to raise to
7034 <p>This function returns the first value raised to the second power with an
7035 unspecified sequence of rounding operations.</p>
7039 <!-- _______________________________________________________________________ -->
7041 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
7047 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
7048 floating point or vector of floating point type. Not all targets support all
7052 declare float @llvm.sin.f32(float %Val)
7053 declare double @llvm.sin.f64(double %Val)
7054 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7055 declare fp128 @llvm.sin.f128(fp128 %Val)
7056 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7060 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
7063 <p>The argument and return value are floating point numbers of the same
7067 <p>This function returns the sine of the specified operand, returning the same
7068 values as the libm <tt>sin</tt> functions would, and handles error conditions
7069 in the same way.</p>
7073 <!-- _______________________________________________________________________ -->
7075 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
7081 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
7082 floating point or vector of floating point type. Not all targets support all
7086 declare float @llvm.cos.f32(float %Val)
7087 declare double @llvm.cos.f64(double %Val)
7088 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7089 declare fp128 @llvm.cos.f128(fp128 %Val)
7090 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7094 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
7097 <p>The argument and return value are floating point numbers of the same
7101 <p>This function returns the cosine of the specified operand, returning the same
7102 values as the libm <tt>cos</tt> functions would, and handles error conditions
7103 in the same way.</p>
7107 <!-- _______________________________________________________________________ -->
7109 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
7115 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
7116 floating point or vector of floating point type. Not all targets support all
7120 declare float @llvm.pow.f32(float %Val, float %Power)
7121 declare double @llvm.pow.f64(double %Val, double %Power)
7122 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7123 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7124 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7128 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
7129 specified (positive or negative) power.</p>
7132 <p>The second argument is a floating point power, and the first is a value to
7133 raise to that power.</p>
7136 <p>This function returns the first value raised to the second power, returning
7137 the same values as the libm <tt>pow</tt> functions would, and handles error
7138 conditions in the same way.</p>
7142 <!-- _______________________________________________________________________ -->
7144 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7150 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7151 floating point or vector of floating point type. Not all targets support all
7155 declare float @llvm.exp.f32(float %Val)
7156 declare double @llvm.exp.f64(double %Val)
7157 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7158 declare fp128 @llvm.exp.f128(fp128 %Val)
7159 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7163 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7166 <p>The argument and return value are floating point numbers of the same
7170 <p>This function returns the same values as the libm <tt>exp</tt> functions
7171 would, and handles error conditions in the same way.</p>
7175 <!-- _______________________________________________________________________ -->
7177 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7183 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7184 floating point or vector of floating point type. Not all targets support all
7188 declare float @llvm.log.f32(float %Val)
7189 declare double @llvm.log.f64(double %Val)
7190 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7191 declare fp128 @llvm.log.f128(fp128 %Val)
7192 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7196 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7199 <p>The argument and return value are floating point numbers of the same
7203 <p>This function returns the same values as the libm <tt>log</tt> functions
7204 would, and handles error conditions in the same way.</p>
7208 <!-- _______________________________________________________________________ -->
7210 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7216 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7217 floating point or vector of floating point type. Not all targets support all
7221 declare float @llvm.fma.f32(float %a, float %b, float %c)
7222 declare double @llvm.fma.f64(double %a, double %b, double %c)
7223 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7224 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7225 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7229 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7233 <p>The argument and return value are floating point numbers of the same
7237 <p>This function returns the same values as the libm <tt>fma</tt> functions
7244 <!-- ======================================================================= -->
7246 <a name="int_manip">Bit Manipulation Intrinsics</a>
7251 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7252 These allow efficient code generation for some algorithms.</p>
7254 <!-- _______________________________________________________________________ -->
7256 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7262 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7263 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7266 declare i16 @llvm.bswap.i16(i16 <id>)
7267 declare i32 @llvm.bswap.i32(i32 <id>)
7268 declare i64 @llvm.bswap.i64(i64 <id>)
7272 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7273 values with an even number of bytes (positive multiple of 16 bits). These
7274 are useful for performing operations on data that is not in the target's
7275 native byte order.</p>
7278 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7279 and low byte of the input i16 swapped. Similarly,
7280 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7281 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7282 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7283 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7284 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7285 more, respectively).</p>
7289 <!-- _______________________________________________________________________ -->
7291 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7297 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7298 width, or on any vector with integer elements. Not all targets support all
7299 bit widths or vector types, however.</p>
7302 declare i8 @llvm.ctpop.i8(i8 <src>)
7303 declare i16 @llvm.ctpop.i16(i16 <src>)
7304 declare i32 @llvm.ctpop.i32(i32 <src>)
7305 declare i64 @llvm.ctpop.i64(i64 <src>)
7306 declare i256 @llvm.ctpop.i256(i256 <src>)
7307 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7311 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7315 <p>The only argument is the value to be counted. The argument may be of any
7316 integer type, or a vector with integer elements.
7317 The return type must match the argument type.</p>
7320 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7321 element of a vector.</p>
7325 <!-- _______________________________________________________________________ -->
7327 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7333 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7334 integer bit width, or any vector whose elements are integers. Not all
7335 targets support all bit widths or vector types, however.</p>
7338 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7339 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7340 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7341 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7342 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7343 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7347 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7348 leading zeros in a variable.</p>
7351 <p>The first argument is the value to be counted. This argument may be of any
7352 integer type, or a vectory with integer element type. The return type
7353 must match the first argument type.</p>
7355 <p>The second argument must be a constant and is a flag to indicate whether the
7356 intrinsic should ensure that a zero as the first argument produces a defined
7357 result. Historically some architectures did not provide a defined result for
7358 zero values as efficiently, and many algorithms are now predicated on
7359 avoiding zero-value inputs.</p>
7362 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7363 zeros in a variable, or within each element of the vector.
7364 If <tt>src == 0</tt> then the result is the size in bits of the type of
7365 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7366 For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7370 <!-- _______________________________________________________________________ -->
7372 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7378 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7379 integer bit width, or any vector of integer elements. Not all targets
7380 support all bit widths or vector types, however.</p>
7383 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7384 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7385 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7386 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7387 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7388 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7392 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7396 <p>The first argument is the value to be counted. This argument may be of any
7397 integer type, or a vectory with integer element type. The return type
7398 must match the first argument type.</p>
7400 <p>The second argument must be a constant and is a flag to indicate whether the
7401 intrinsic should ensure that a zero as the first argument produces a defined
7402 result. Historically some architectures did not provide a defined result for
7403 zero values as efficiently, and many algorithms are now predicated on
7404 avoiding zero-value inputs.</p>
7407 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7408 zeros in a variable, or within each element of a vector.
7409 If <tt>src == 0</tt> then the result is the size in bits of the type of
7410 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7411 For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7417 <!-- ======================================================================= -->
7419 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7424 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7426 <!-- _______________________________________________________________________ -->
7428 <a name="int_sadd_overflow">
7429 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7436 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7437 on any integer bit width.</p>
7440 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7441 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7442 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7446 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7447 a signed addition of the two arguments, and indicate whether an overflow
7448 occurred during the signed summation.</p>
7451 <p>The arguments (%a and %b) and the first element of the result structure may
7452 be of integer types of any bit width, but they must have the same bit
7453 width. The second element of the result structure must be of
7454 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7455 undergo signed addition.</p>
7458 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7459 a signed addition of the two variables. They return a structure — the
7460 first element of which is the signed summation, and the second element of
7461 which is a bit specifying if the signed summation resulted in an
7466 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7467 %sum = extractvalue {i32, i1} %res, 0
7468 %obit = extractvalue {i32, i1} %res, 1
7469 br i1 %obit, label %overflow, label %normal
7474 <!-- _______________________________________________________________________ -->
7476 <a name="int_uadd_overflow">
7477 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7484 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7485 on any integer bit width.</p>
7488 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7489 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7490 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7494 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7495 an unsigned addition of the two arguments, and indicate whether a carry
7496 occurred during the unsigned summation.</p>
7499 <p>The arguments (%a and %b) and the first element of the result structure may
7500 be of integer types of any bit width, but they must have the same bit
7501 width. The second element of the result structure must be of
7502 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7503 undergo unsigned addition.</p>
7506 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7507 an unsigned addition of the two arguments. They return a structure —
7508 the first element of which is the sum, and the second element of which is a
7509 bit specifying if the unsigned summation resulted in a carry.</p>
7513 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7514 %sum = extractvalue {i32, i1} %res, 0
7515 %obit = extractvalue {i32, i1} %res, 1
7516 br i1 %obit, label %carry, label %normal
7521 <!-- _______________________________________________________________________ -->
7523 <a name="int_ssub_overflow">
7524 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7531 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7532 on any integer bit width.</p>
7535 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7536 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7537 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7541 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7542 a signed subtraction of the two arguments, and indicate whether an overflow
7543 occurred during the signed subtraction.</p>
7546 <p>The arguments (%a and %b) and the first element of the result structure may
7547 be of integer types of any bit width, but they must have the same bit
7548 width. The second element of the result structure must be of
7549 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7550 undergo signed subtraction.</p>
7553 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7554 a signed subtraction of the two arguments. They return a structure —
7555 the first element of which is the subtraction, and the second element of
7556 which is a bit specifying if the signed subtraction resulted in an
7561 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7562 %sum = extractvalue {i32, i1} %res, 0
7563 %obit = extractvalue {i32, i1} %res, 1
7564 br i1 %obit, label %overflow, label %normal
7569 <!-- _______________________________________________________________________ -->
7571 <a name="int_usub_overflow">
7572 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7579 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7580 on any integer bit width.</p>
7583 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7584 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7585 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7589 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7590 an unsigned subtraction of the two arguments, and indicate whether an
7591 overflow occurred during the unsigned subtraction.</p>
7594 <p>The arguments (%a and %b) and the first element of the result structure may
7595 be of integer types of any bit width, but they must have the same bit
7596 width. The second element of the result structure must be of
7597 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7598 undergo unsigned subtraction.</p>
7601 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7602 an unsigned subtraction of the two arguments. They return a structure —
7603 the first element of which is the subtraction, and the second element of
7604 which is a bit specifying if the unsigned subtraction resulted in an
7609 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7610 %sum = extractvalue {i32, i1} %res, 0
7611 %obit = extractvalue {i32, i1} %res, 1
7612 br i1 %obit, label %overflow, label %normal
7617 <!-- _______________________________________________________________________ -->
7619 <a name="int_smul_overflow">
7620 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7627 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7628 on any integer bit width.</p>
7631 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7632 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7633 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7638 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7639 a signed multiplication of the two arguments, and indicate whether an
7640 overflow occurred during the signed multiplication.</p>
7643 <p>The arguments (%a and %b) and the first element of the result structure may
7644 be of integer types of any bit width, but they must have the same bit
7645 width. The second element of the result structure must be of
7646 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7647 undergo signed multiplication.</p>
7650 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7651 a signed multiplication of the two arguments. They return a structure —
7652 the first element of which is the multiplication, and the second element of
7653 which is a bit specifying if the signed multiplication resulted in an
7658 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7659 %sum = extractvalue {i32, i1} %res, 0
7660 %obit = extractvalue {i32, i1} %res, 1
7661 br i1 %obit, label %overflow, label %normal
7666 <!-- _______________________________________________________________________ -->
7668 <a name="int_umul_overflow">
7669 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
7676 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
7677 on any integer bit width.</p>
7680 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7681 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7682 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7686 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7687 a unsigned multiplication of the two arguments, and indicate whether an
7688 overflow occurred during the unsigned multiplication.</p>
7691 <p>The arguments (%a and %b) and the first element of the result structure may
7692 be of integer types of any bit width, but they must have the same bit
7693 width. The second element of the result structure must be of
7694 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7695 undergo unsigned multiplication.</p>
7698 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7699 an unsigned multiplication of the two arguments. They return a structure
7700 — the first element of which is the multiplication, and the second
7701 element of which is a bit specifying if the unsigned multiplication resulted
7706 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7707 %sum = extractvalue {i32, i1} %res, 0
7708 %obit = extractvalue {i32, i1} %res, 1
7709 br i1 %obit, label %overflow, label %normal
7716 <!-- ======================================================================= -->
7718 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
7723 <p>Half precision floating point is a storage-only format. This means that it is
7724 a dense encoding (in memory) but does not support computation in the
7727 <p>This means that code must first load the half-precision floating point
7728 value as an i16, then convert it to float with <a
7729 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
7730 Computation can then be performed on the float value (including extending to
7731 double etc). To store the value back to memory, it is first converted to
7732 float if needed, then converted to i16 with
7733 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
7734 storing as an i16 value.</p>
7736 <!-- _______________________________________________________________________ -->
7738 <a name="int_convert_to_fp16">
7739 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
7747 declare i16 @llvm.convert.to.fp16(f32 %a)
7751 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7752 a conversion from single precision floating point format to half precision
7753 floating point format.</p>
7756 <p>The intrinsic function contains single argument - the value to be
7760 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7761 a conversion from single precision floating point format to half precision
7762 floating point format. The return value is an <tt>i16</tt> which
7763 contains the converted number.</p>
7767 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7768 store i16 %res, i16* @x, align 2
7773 <!-- _______________________________________________________________________ -->
7775 <a name="int_convert_from_fp16">
7776 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
7784 declare f32 @llvm.convert.from.fp16(i16 %a)
7788 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
7789 a conversion from half precision floating point format to single precision
7790 floating point format.</p>
7793 <p>The intrinsic function contains single argument - the value to be
7797 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
7798 conversion from half single precision floating point format to single
7799 precision floating point format. The input half-float value is represented by
7800 an <tt>i16</tt> value.</p>
7804 %a = load i16* @x, align 2
7805 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7812 <!-- ======================================================================= -->
7814 <a name="int_debugger">Debugger Intrinsics</a>
7819 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
7820 prefix), are described in
7821 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
7822 Level Debugging</a> document.</p>
7826 <!-- ======================================================================= -->
7828 <a name="int_eh">Exception Handling Intrinsics</a>
7833 <p>The LLVM exception handling intrinsics (which all start with
7834 <tt>llvm.eh.</tt> prefix), are described in
7835 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
7836 Handling</a> document.</p>
7840 <!-- ======================================================================= -->
7842 <a name="int_trampoline">Trampoline Intrinsics</a>
7847 <p>These intrinsics make it possible to excise one parameter, marked with
7848 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
7849 The result is a callable
7850 function pointer lacking the nest parameter - the caller does not need to
7851 provide a value for it. Instead, the value to use is stored in advance in a
7852 "trampoline", a block of memory usually allocated on the stack, which also
7853 contains code to splice the nest value into the argument list. This is used
7854 to implement the GCC nested function address extension.</p>
7856 <p>For example, if the function is
7857 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
7858 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
7861 <pre class="doc_code">
7862 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
7863 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
7864 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
7865 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
7866 %fp = bitcast i8* %p to i32 (i32, i32)*
7869 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
7870 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
7872 <!-- _______________________________________________________________________ -->
7875 '<tt>llvm.init.trampoline</tt>' Intrinsic
7883 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
7887 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
7888 turning it into a trampoline.</p>
7891 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
7892 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
7893 sufficiently aligned block of memory; this memory is written to by the
7894 intrinsic. Note that the size and the alignment are target-specific - LLVM
7895 currently provides no portable way of determining them, so a front-end that
7896 generates this intrinsic needs to have some target-specific knowledge.
7897 The <tt>func</tt> argument must hold a function bitcast to
7898 an <tt>i8*</tt>.</p>
7901 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
7902 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
7903 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
7904 which can be <a href="#int_trampoline">bitcast (to a new function) and
7905 called</a>. The new function's signature is the same as that of
7906 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
7907 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
7908 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
7909 with the same argument list, but with <tt>nval</tt> used for the missing
7910 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
7911 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
7912 to the returned function pointer is undefined.</p>
7915 <!-- _______________________________________________________________________ -->
7918 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
7926 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
7930 <p>This performs any required machine-specific adjustment to the address of a
7931 trampoline (passed as <tt>tramp</tt>).</p>
7934 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
7935 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
7939 <p>On some architectures the address of the code to be executed needs to be
7940 different to the address where the trampoline is actually stored. This
7941 intrinsic returns the executable address corresponding to <tt>tramp</tt>
7942 after performing the required machine specific adjustments.
7943 The pointer returned can then be <a href="#int_trampoline"> bitcast and
7951 <!-- ======================================================================= -->
7953 <a name="int_memorymarkers">Memory Use Markers</a>
7958 <p>This class of intrinsics exists to information about the lifetime of memory
7959 objects and ranges where variables are immutable.</p>
7961 <!-- _______________________________________________________________________ -->
7963 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
7970 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
7974 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
7975 object's lifetime.</p>
7978 <p>The first argument is a constant integer representing the size of the
7979 object, or -1 if it is variable sized. The second argument is a pointer to
7983 <p>This intrinsic indicates that before this point in the code, the value of the
7984 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7985 never be used and has an undefined value. A load from the pointer that
7986 precedes this intrinsic can be replaced with
7987 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
7991 <!-- _______________________________________________________________________ -->
7993 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8000 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8004 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8005 object's lifetime.</p>
8008 <p>The first argument is a constant integer representing the size of the
8009 object, or -1 if it is variable sized. The second argument is a pointer to
8013 <p>This intrinsic indicates that after this point in the code, the value of the
8014 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8015 never be used and has an undefined value. Any stores into the memory object
8016 following this intrinsic may be removed as dead.
8020 <!-- _______________________________________________________________________ -->
8022 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8029 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8033 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8034 a memory object will not change.</p>
8037 <p>The first argument is a constant integer representing the size of the
8038 object, or -1 if it is variable sized. The second argument is a pointer to
8042 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8043 the return value, the referenced memory location is constant and
8048 <!-- _______________________________________________________________________ -->
8050 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8057 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8061 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8062 a memory object are mutable.</p>
8065 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8066 The second argument is a constant integer representing the size of the
8067 object, or -1 if it is variable sized and the third argument is a pointer
8071 <p>This intrinsic indicates that the memory is mutable again.</p>
8077 <!-- ======================================================================= -->
8079 <a name="int_general">General Intrinsics</a>
8084 <p>This class of intrinsics is designed to be generic and has no specific
8087 <!-- _______________________________________________________________________ -->
8089 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8096 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8100 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8103 <p>The first argument is a pointer to a value, the second is a pointer to a
8104 global string, the third is a pointer to a global string which is the source
8105 file name, and the last argument is the line number.</p>
8108 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8109 This can be useful for special purpose optimizations that want to look for
8110 these annotations. These have no other defined use; they are ignored by code
8111 generation and optimization.</p>
8115 <!-- _______________________________________________________________________ -->
8117 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8123 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8124 any integer bit width.</p>
8127 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8128 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8129 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8130 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8131 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8135 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8138 <p>The first argument is an integer value (result of some expression), the
8139 second is a pointer to a global string, the third is a pointer to a global
8140 string which is the source file name, and the last argument is the line
8141 number. It returns the value of the first argument.</p>
8144 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8145 arbitrary strings. This can be useful for special purpose optimizations that
8146 want to look for these annotations. These have no other defined use; they
8147 are ignored by code generation and optimization.</p>
8151 <!-- _______________________________________________________________________ -->
8153 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8160 declare void @llvm.trap()
8164 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8170 <p>This intrinsics is lowered to the target dependent trap instruction. If the
8171 target does not have a trap instruction, this intrinsic will be lowered to
8172 the call of the <tt>abort()</tt> function.</p>
8176 <!-- _______________________________________________________________________ -->
8178 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8185 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8189 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8190 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8191 ensure that it is placed on the stack before local variables.</p>
8194 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8195 arguments. The first argument is the value loaded from the stack
8196 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8197 that has enough space to hold the value of the guard.</p>
8200 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8201 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8202 stack. This is to ensure that if a local variable on the stack is
8203 overwritten, it will destroy the value of the guard. When the function exits,
8204 the guard on the stack is checked against the original guard. If they are
8205 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8210 <!-- _______________________________________________________________________ -->
8212 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8219 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <type>)
8220 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <type>)
8224 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8225 the optimizers to determine at compile time whether a) an operation (like
8226 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8227 runtime check for overflow isn't necessary. An object in this context means
8228 an allocation of a specific class, structure, array, or other object.</p>
8231 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8232 argument is a pointer to or into the <tt>object</tt>. The second argument
8233 is a boolean 0 or 1. This argument determines whether you want the
8234 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
8235 1, variables are not allowed.</p>
8238 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
8239 representing the size of the object concerned, or <tt>i32/i64 -1 or 0</tt>,
8240 depending on the <tt>type</tt> argument, if the size cannot be determined at
8244 <!-- _______________________________________________________________________ -->
8246 <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a>
8253 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8254 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8258 <p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the
8259 most probable) value of <tt>val</tt>, which can be used by optimizers.</p>
8262 <p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first
8263 argument is a value. The second argument is an expected value, this needs to
8264 be a constant value, variables are not allowed.</p>
8267 <p>This intrinsic is lowered to the <tt>val</tt>.</p>
8273 <!-- *********************************************************************** -->
8276 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8277 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8278 <a href="http://validator.w3.org/check/referer"><img
8279 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8281 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8282 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8283 Last modified: $Date$