1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="_static/llvm.css" type="text/css">
15 <h1>LLVM Language Reference Manual</h1>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a>
25 <li><a href="#linkage_private">'<tt>private</tt>' Linkage</a></li>
26 <li><a href="#linkage_linker_private">'<tt>linker_private</tt>' Linkage</a></li>
27 <li><a href="#linkage_linker_private_weak">'<tt>linker_private_weak</tt>' Linkage</a></li>
28 <li><a href="#linkage_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="#fpmath">'<tt>fpmath</tt>' Metadata</a></li>
107 <li><a href="#range">'<tt>range</tt>' Metadata</a></li>
112 <li><a href="#module_flags">Module Flags Metadata</a>
114 <li><a href="#objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a></li>
117 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
119 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
120 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
121 Global Variable</a></li>
122 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
123 Global Variable</a></li>
124 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
125 Global Variable</a></li>
128 <li><a href="#instref">Instruction Reference</a>
130 <li><a href="#terminators">Terminator Instructions</a>
132 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
133 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
134 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
135 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
136 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
137 <li><a href="#i_resume">'<tt>resume</tt>' Instruction</a></li>
138 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
141 <li><a href="#binaryops">Binary Operations</a>
143 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
144 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
145 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
146 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
147 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
148 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
149 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
150 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
151 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
152 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
153 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
154 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
157 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
159 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
160 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
161 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
162 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
163 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
164 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
167 <li><a href="#vectorops">Vector Operations</a>
169 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
170 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
171 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
174 <li><a href="#aggregateops">Aggregate Operations</a>
176 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
177 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
180 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
182 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
183 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
184 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
185 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
186 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
187 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
188 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
191 <li><a href="#convertops">Conversion Operations</a>
193 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
194 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
195 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
196 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
197 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
198 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
199 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
200 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
201 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
202 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
203 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
204 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
207 <li><a href="#otherops">Other Operations</a>
209 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
210 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
211 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
212 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
213 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
214 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
215 <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
220 <li><a href="#intrinsics">Intrinsic Functions</a>
222 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
224 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
225 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
226 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
229 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
231 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
232 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
233 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
236 <li><a href="#int_codegen">Code Generator Intrinsics</a>
238 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
239 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
240 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
241 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
242 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
243 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
244 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
247 <li><a href="#int_libc">Standard C Library Intrinsics</a>
249 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
250 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
251 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
252 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
253 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
254 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
255 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
256 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
257 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
258 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
259 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
262 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
264 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
265 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
266 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
267 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
270 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
272 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
273 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
274 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
275 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
276 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
277 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
280 <li><a href="#spec_arithmetic">Specialised Arithmetic Intrinsics</a>
282 <li><a href="#fmuladd">'<tt>llvm.fmuladd</tt> Intrinsic</a></li>
285 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
287 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
288 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
291 <li><a href="#int_debugger">Debugger intrinsics</a></li>
292 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
293 <li><a href="#int_trampoline">Trampoline Intrinsics</a>
295 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
296 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
299 <li><a href="#int_memorymarkers">Memory Use Markers</a>
301 <li><a href="#int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a></li>
302 <li><a href="#int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a></li>
303 <li><a href="#int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a></li>
304 <li><a href="#int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a></li>
307 <li><a href="#int_general">General intrinsics</a>
309 <li><a href="#int_var_annotation">
310 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
311 <li><a href="#int_annotation">
312 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
313 <li><a href="#int_trap">
314 '<tt>llvm.trap</tt>' Intrinsic</a></li>
315 <li><a href="#int_debugtrap">
316 '<tt>llvm.debugtrap</tt>' Intrinsic</a></li>
317 <li><a href="#int_stackprotector">
318 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
319 <li><a href="#int_objectsize">
320 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
321 <li><a href="#int_expect">
322 '<tt>llvm.expect</tt>' Intrinsic</a></li>
329 <div class="doc_author">
330 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
331 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
334 <!-- *********************************************************************** -->
335 <h2><a name="abstract">Abstract</a></h2>
336 <!-- *********************************************************************** -->
340 <p>This document is a reference manual for the LLVM assembly language. LLVM is
341 a Static Single Assignment (SSA) based representation that provides type
342 safety, low-level operations, flexibility, and the capability of representing
343 'all' high-level languages cleanly. It is the common code representation
344 used throughout all phases of the LLVM compilation strategy.</p>
348 <!-- *********************************************************************** -->
349 <h2><a name="introduction">Introduction</a></h2>
350 <!-- *********************************************************************** -->
354 <p>The LLVM code representation is designed to be used in three different forms:
355 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
356 for fast loading by a Just-In-Time compiler), and as a human readable
357 assembly language representation. This allows LLVM to provide a powerful
358 intermediate representation for efficient compiler transformations and
359 analysis, while providing a natural means to debug and visualize the
360 transformations. The three different forms of LLVM are all equivalent. This
361 document describes the human readable representation and notation.</p>
363 <p>The LLVM representation aims to be light-weight and low-level while being
364 expressive, typed, and extensible at the same time. It aims to be a
365 "universal IR" of sorts, by being at a low enough level that high-level ideas
366 may be cleanly mapped to it (similar to how microprocessors are "universal
367 IR's", allowing many source languages to be mapped to them). By providing
368 type information, LLVM can be used as the target of optimizations: for
369 example, through pointer analysis, it can be proven that a C automatic
370 variable is never accessed outside of the current function, allowing it to
371 be promoted to a simple SSA value instead of a memory location.</p>
373 <!-- _______________________________________________________________________ -->
375 <a name="wellformed">Well-Formedness</a>
380 <p>It is important to note that this document describes 'well formed' LLVM
381 assembly language. There is a difference between what the parser accepts and
382 what is considered 'well formed'. For example, the following instruction is
383 syntactically okay, but not well formed:</p>
385 <pre class="doc_code">
386 %x = <a href="#i_add">add</a> i32 1, %x
389 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
390 LLVM infrastructure provides a verification pass that may be used to verify
391 that an LLVM module is well formed. This pass is automatically run by the
392 parser after parsing input assembly and by the optimizer before it outputs
393 bitcode. The violations pointed out by the verifier pass indicate bugs in
394 transformation passes or input to the parser.</p>
400 <!-- Describe the typesetting conventions here. -->
402 <!-- *********************************************************************** -->
403 <h2><a name="identifiers">Identifiers</a></h2>
404 <!-- *********************************************************************** -->
408 <p>LLVM identifiers come in two basic types: global and local. Global
409 identifiers (functions, global variables) begin with the <tt>'@'</tt>
410 character. Local identifiers (register names, types) begin with
411 the <tt>'%'</tt> character. Additionally, there are three different formats
412 for identifiers, for different purposes:</p>
415 <li>Named values are represented as a string of characters with their prefix.
416 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
417 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
418 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
419 other characters in their names can be surrounded with quotes. Special
420 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
421 ASCII code for the character in hexadecimal. In this way, any character
422 can be used in a name value, even quotes themselves.</li>
424 <li>Unnamed values are represented as an unsigned numeric value with their
425 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
427 <li>Constants, which are described in a <a href="#constants">section about
428 constants</a>, below.</li>
431 <p>LLVM requires that values start with a prefix for two reasons: Compilers
432 don't need to worry about name clashes with reserved words, and the set of
433 reserved words may be expanded in the future without penalty. Additionally,
434 unnamed identifiers allow a compiler to quickly come up with a temporary
435 variable without having to avoid symbol table conflicts.</p>
437 <p>Reserved words in LLVM are very similar to reserved words in other
438 languages. There are keywords for different opcodes
439 ('<tt><a href="#i_add">add</a></tt>',
440 '<tt><a href="#i_bitcast">bitcast</a></tt>',
441 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
442 ('<tt><a href="#t_void">void</a></tt>',
443 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
444 reserved words cannot conflict with variable names, because none of them
445 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
447 <p>Here is an example of LLVM code to multiply the integer variable
448 '<tt>%X</tt>' by 8:</p>
452 <pre class="doc_code">
453 %result = <a href="#i_mul">mul</a> i32 %X, 8
456 <p>After strength reduction:</p>
458 <pre class="doc_code">
459 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
462 <p>And the hard way:</p>
464 <pre class="doc_code">
465 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
466 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
467 %result = <a href="#i_add">add</a> i32 %1, %1
470 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
471 lexical features of LLVM:</p>
474 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
477 <li>Unnamed temporaries are created when the result of a computation is not
478 assigned to a named value.</li>
480 <li>Unnamed temporaries are numbered sequentially</li>
483 <p>It also shows a convention that we follow in this document. When
484 demonstrating instructions, we will follow an instruction with a comment that
485 defines the type and name of value produced. Comments are shown in italic
490 <!-- *********************************************************************** -->
491 <h2><a name="highlevel">High Level Structure</a></h2>
492 <!-- *********************************************************************** -->
494 <!-- ======================================================================= -->
496 <a name="modulestructure">Module Structure</a>
501 <p>LLVM programs are composed of <tt>Module</tt>s, each of which is a
502 translation unit of the input programs. Each module consists of functions,
503 global variables, and symbol table entries. Modules may be combined together
504 with the LLVM linker, which merges function (and global variable)
505 definitions, resolves forward declarations, and merges symbol table
506 entries. Here is an example of the "hello world" module:</p>
508 <pre class="doc_code">
509 <i>; Declare the string constant as a global constant.</i>
510 <a href="#identifiers">@.str</a> = <a href="#linkage_private">private</a> <a href="#globalvars">unnamed_addr</a> <a href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00"
512 <i>; External declaration of the puts function</i>
513 <a href="#functionstructure">declare</a> i32 @puts(i8* <a href="#nocapture">nocapture</a>) <a href="#fnattrs">nounwind</a>
515 <i>; Definition of main function</i>
516 define i32 @main() { <i>; i32()* </i>
517 <i>; Convert [13 x i8]* to i8 *...</i>
518 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.str, i64 0, i64 0
520 <i>; Call puts function to write out the string to stdout.</i>
521 <a href="#i_call">call</a> i32 @puts(i8* %cast210)
522 <a href="#i_ret">ret</a> i32 0
525 <i>; Named metadata</i>
526 !1 = metadata !{i32 42}
530 <p>This example is made up of a <a href="#globalvars">global variable</a> named
531 "<tt>.str</tt>", an external declaration of the "<tt>puts</tt>" function,
532 a <a href="#functionstructure">function definition</a> for
533 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
536 <p>In general, a module is made up of a list of global values (where both
537 functions and global variables are global values). Global values are
538 represented by a pointer to a memory location (in this case, a pointer to an
539 array of char, and a pointer to a function), and have one of the
540 following <a href="#linkage">linkage types</a>.</p>
544 <!-- ======================================================================= -->
546 <a name="linkage">Linkage Types</a>
551 <p>All Global Variables and Functions have one of the following types of
555 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
556 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
557 by objects in the current module. In particular, linking code into a
558 module with an private global value may cause the private to be renamed as
559 necessary to avoid collisions. Because the symbol is private to the
560 module, all references can be updated. This doesn't show up in any symbol
561 table in the object file.</dd>
563 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
564 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
565 assembler and evaluated by the linker. Unlike normal strong symbols, they
566 are removed by the linker from the final linked image (executable or
567 dynamic library).</dd>
569 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
570 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
571 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
572 linker. The symbols are removed by the linker from the final linked image
573 (executable or dynamic library).</dd>
575 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
576 <dd>Similar to "<tt>linker_private_weak</tt>", but it's known that the address
577 of the object is not taken. For instance, functions that had an inline
578 definition, but the compiler decided not to inline it. Note,
579 unlike <tt>linker_private</tt> and <tt>linker_private_weak</tt>,
580 <tt>linker_private_weak_def_auto</tt> may have only <tt>default</tt>
581 visibility. The symbols are removed by the linker from the final linked
582 image (executable or dynamic library).</dd>
584 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
585 <dd>Similar to private, but the value shows as a local symbol
586 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
587 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
589 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
590 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
591 into the object file corresponding to the LLVM module. They exist to
592 allow inlining and other optimizations to take place given knowledge of
593 the definition of the global, which is known to be somewhere outside the
594 module. Globals with <tt>available_externally</tt> linkage are allowed to
595 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
596 This linkage type is only allowed on definitions, not declarations.</dd>
598 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
599 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
600 the same name when linkage occurs. This can be used to implement
601 some forms of inline functions, templates, or other code which must be
602 generated in each translation unit that uses it, but where the body may
603 be overridden with a more definitive definition later. Unreferenced
604 <tt>linkonce</tt> globals are allowed to be discarded. Note that
605 <tt>linkonce</tt> linkage does not actually allow the optimizer to
606 inline the body of this function into callers because it doesn't know if
607 this definition of the function is the definitive definition within the
608 program or whether it will be overridden by a stronger definition.
609 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
612 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
613 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
614 <tt>linkonce</tt> linkage, except that unreferenced globals with
615 <tt>weak</tt> linkage may not be discarded. This is used for globals that
616 are declared "weak" in C source code.</dd>
618 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
619 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
620 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
622 Symbols with "<tt>common</tt>" linkage are merged in the same way as
623 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
624 <tt>common</tt> symbols may not have an explicit section,
625 must have a zero initializer, and may not be marked '<a
626 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
627 have common linkage.</dd>
630 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
631 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
632 pointer to array type. When two global variables with appending linkage
633 are linked together, the two global arrays are appended together. This is
634 the LLVM, typesafe, equivalent of having the system linker append together
635 "sections" with identical names when .o files are linked.</dd>
637 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
638 <dd>The semantics of this linkage follow the ELF object file model: the symbol
639 is weak until linked, if not linked, the symbol becomes null instead of
640 being an undefined reference.</dd>
642 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
643 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
644 <dd>Some languages allow differing globals to be merged, such as two functions
645 with different semantics. Other languages, such as <tt>C++</tt>, ensure
646 that only equivalent globals are ever merged (the "one definition rule"
647 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
648 and <tt>weak_odr</tt> linkage types to indicate that the global will only
649 be merged with equivalent globals. These linkage types are otherwise the
650 same as their non-<tt>odr</tt> versions.</dd>
652 <dt><tt><b><a name="linkage_external">external</a></b></tt></dt>
653 <dd>If none of the above identifiers are used, the global is externally
654 visible, meaning that it participates in linkage and can be used to
655 resolve external symbol references.</dd>
658 <p>The next two types of linkage are targeted for Microsoft Windows platform
659 only. They are designed to support importing (exporting) symbols from (to)
660 DLLs (Dynamic Link Libraries).</p>
663 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
664 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
665 or variable via a global pointer to a pointer that is set up by the DLL
666 exporting the symbol. On Microsoft Windows targets, the pointer name is
667 formed by combining <code>__imp_</code> and the function or variable
670 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
671 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
672 pointer to a pointer in a DLL, so that it can be referenced with the
673 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
674 name is formed by combining <code>__imp_</code> and the function or
678 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
679 another module defined a "<tt>.LC0</tt>" variable and was linked with this
680 one, one of the two would be renamed, preventing a collision. Since
681 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
682 declarations), they are accessible outside of the current module.</p>
684 <p>It is illegal for a function <i>declaration</i> to have any linkage type
685 other than <tt>external</tt>, <tt>dllimport</tt>
686 or <tt>extern_weak</tt>.</p>
688 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
689 or <tt>weak_odr</tt> linkages.</p>
693 <!-- ======================================================================= -->
695 <a name="callingconv">Calling Conventions</a>
700 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
701 and <a href="#i_invoke">invokes</a> can all have an optional calling
702 convention specified for the call. The calling convention of any pair of
703 dynamic caller/callee must match, or the behavior of the program is
704 undefined. The following calling conventions are supported by LLVM, and more
705 may be added in the future:</p>
708 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
709 <dd>This calling convention (the default if no other calling convention is
710 specified) matches the target C calling conventions. This calling
711 convention supports varargs function calls and tolerates some mismatch in
712 the declared prototype and implemented declaration of the function (as
715 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
716 <dd>This calling convention attempts to make calls as fast as possible
717 (e.g. by passing things in registers). This calling convention allows the
718 target to use whatever tricks it wants to produce fast code for the
719 target, without having to conform to an externally specified ABI
720 (Application Binary Interface).
721 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
722 when this or the GHC convention is used.</a> This calling convention
723 does not support varargs and requires the prototype of all callees to
724 exactly match the prototype of the function definition.</dd>
726 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
727 <dd>This calling convention attempts to make code in the caller as efficient
728 as possible under the assumption that the call is not commonly executed.
729 As such, these calls often preserve all registers so that the call does
730 not break any live ranges in the caller side. This calling convention
731 does not support varargs and requires the prototype of all callees to
732 exactly match the prototype of the function definition.</dd>
734 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
735 <dd>This calling convention has been implemented specifically for use by the
736 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
737 It passes everything in registers, going to extremes to achieve this by
738 disabling callee save registers. This calling convention should not be
739 used lightly but only for specific situations such as an alternative to
740 the <em>register pinning</em> performance technique often used when
741 implementing functional programming languages.At the moment only X86
742 supports this convention and it has the following limitations:
744 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
745 floating point types are supported.</li>
746 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
747 6 floating point parameters.</li>
749 This calling convention supports
750 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
751 requires both the caller and callee are using it.
754 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
755 <dd>Any calling convention may be specified by number, allowing
756 target-specific calling conventions to be used. Target specific calling
757 conventions start at 64.</dd>
760 <p>More calling conventions can be added/defined on an as-needed basis, to
761 support Pascal conventions or any other well-known target-independent
766 <!-- ======================================================================= -->
768 <a name="visibility">Visibility Styles</a>
773 <p>All Global Variables and Functions have one of the following visibility
777 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
778 <dd>On targets that use the ELF object file format, default visibility means
779 that the declaration is visible to other modules and, in shared libraries,
780 means that the declared entity may be overridden. On Darwin, default
781 visibility means that the declaration is visible to other modules. Default
782 visibility corresponds to "external linkage" in the language.</dd>
784 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
785 <dd>Two declarations of an object with hidden visibility refer to the same
786 object if they are in the same shared object. Usually, hidden visibility
787 indicates that the symbol will not be placed into the dynamic symbol
788 table, so no other module (executable or shared library) can reference it
791 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
792 <dd>On ELF, protected visibility indicates that the symbol will be placed in
793 the dynamic symbol table, but that references within the defining module
794 will bind to the local symbol. That is, the symbol cannot be overridden by
800 <!-- ======================================================================= -->
802 <a name="namedtypes">Named Types</a>
807 <p>LLVM IR allows you to specify name aliases for certain types. This can make
808 it easier to read the IR and make the IR more condensed (particularly when
809 recursive types are involved). An example of a name specification is:</p>
811 <pre class="doc_code">
812 %mytype = type { %mytype*, i32 }
815 <p>You may give a name to any <a href="#typesystem">type</a> except
816 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
817 is expected with the syntax "%mytype".</p>
819 <p>Note that type names are aliases for the structural type that they indicate,
820 and that you can therefore specify multiple names for the same type. This
821 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
822 uses structural typing, the name is not part of the type. When printing out
823 LLVM IR, the printer will pick <em>one name</em> to render all types of a
824 particular shape. This means that if you have code where two different
825 source types end up having the same LLVM type, that the dumper will sometimes
826 print the "wrong" or unexpected type. This is an important design point and
827 isn't going to change.</p>
831 <!-- ======================================================================= -->
833 <a name="globalvars">Global Variables</a>
838 <p>Global variables define regions of memory allocated at compilation time
839 instead of run-time. Global variables may optionally be initialized, may
840 have an explicit section to be placed in, and may have an optional explicit
841 alignment specified. A variable may be defined as "thread_local", which
842 means that it will not be shared by threads (each thread will have a
843 separated copy of the variable). A variable may be defined as a global
844 "constant," which indicates that the contents of the variable
845 will <b>never</b> be modified (enabling better optimization, allowing the
846 global data to be placed in the read-only section of an executable, etc).
847 Note that variables that need runtime initialization cannot be marked
848 "constant" as there is a store to the variable.</p>
850 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
851 constant, even if the final definition of the global is not. This capability
852 can be used to enable slightly better optimization of the program, but
853 requires the language definition to guarantee that optimizations based on the
854 'constantness' are valid for the translation units that do not include the
857 <p>As SSA values, global variables define pointer values that are in scope
858 (i.e. they dominate) all basic blocks in the program. Global variables
859 always define a pointer to their "content" type because they describe a
860 region of memory, and all memory objects in LLVM are accessed through
863 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
864 that the address is not significant, only the content. Constants marked
865 like this can be merged with other constants if they have the same
866 initializer. Note that a constant with significant address <em>can</em>
867 be merged with a <tt>unnamed_addr</tt> constant, the result being a
868 constant whose address is significant.</p>
870 <p>A global variable may be declared to reside in a target-specific numbered
871 address space. For targets that support them, address spaces may affect how
872 optimizations are performed and/or what target instructions are used to
873 access the variable. The default address space is zero. The address space
874 qualifier must precede any other attributes.</p>
876 <p>LLVM allows an explicit section to be specified for globals. If the target
877 supports it, it will emit globals to the section specified.</p>
879 <p>An explicit alignment may be specified for a global, which must be a power
880 of 2. If not present, or if the alignment is set to zero, the alignment of
881 the global is set by the target to whatever it feels convenient. If an
882 explicit alignment is specified, the global is forced to have exactly that
883 alignment. Targets and optimizers are not allowed to over-align the global
884 if the global has an assigned section. In this case, the extra alignment
885 could be observable: for example, code could assume that the globals are
886 densely packed in their section and try to iterate over them as an array,
887 alignment padding would break this iteration.</p>
889 <p>For example, the following defines a global in a numbered address space with
890 an initializer, section, and alignment:</p>
892 <pre class="doc_code">
893 @G = addrspace(5) constant float 1.0, section "foo", align 4
899 <!-- ======================================================================= -->
901 <a name="functionstructure">Functions</a>
906 <p>LLVM function definitions consist of the "<tt>define</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) argument list (each with optional
913 <a href="#paramattrs">parameter attributes</a>), optional
914 <a href="#fnattrs">function attributes</a>, an optional section, an optional
915 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
916 curly brace, a list of basic blocks, and a closing curly brace.</p>
918 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
919 optional <a href="#linkage">linkage type</a>, an optional
920 <a href="#visibility">visibility style</a>, an optional
921 <a href="#callingconv">calling convention</a>,
922 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
923 <a href="#paramattrs">parameter attribute</a> for the return type, a function
924 name, a possibly empty list of arguments, an optional alignment, and an
925 optional <a href="#gc">garbage collector name</a>.</p>
927 <p>A function definition contains a list of basic blocks, forming the CFG
928 (Control Flow Graph) for the function. Each basic block may optionally start
929 with a label (giving the basic block a symbol table entry), contains a list
930 of instructions, and ends with a <a href="#terminators">terminator</a>
931 instruction (such as a branch or function return).</p>
933 <p>The first basic block in a function is special in two ways: it is immediately
934 executed on entrance to the function, and it is not allowed to have
935 predecessor basic blocks (i.e. there can not be any branches to the entry
936 block of a function). Because the block can have no predecessors, it also
937 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
939 <p>LLVM allows an explicit section to be specified for functions. If the target
940 supports it, it will emit functions to the section specified.</p>
942 <p>An explicit alignment may be specified for a function. If not present, or if
943 the alignment is set to zero, the alignment of the function is set by the
944 target to whatever it feels convenient. If an explicit alignment is
945 specified, the function is forced to have at least that much alignment. All
946 alignments must be a power of 2.</p>
948 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
949 be significant and two identical functions can be merged.</p>
952 <pre class="doc_code">
953 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
954 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
955 <ResultType> @<FunctionName> ([argument list])
956 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
957 [<a href="#gc">gc</a>] { ... }
962 <!-- ======================================================================= -->
964 <a name="aliasstructure">Aliases</a>
969 <p>Aliases act as "second name" for the aliasee value (which can be either
970 function, global variable, another alias or bitcast of global value). Aliases
971 may have an optional <a href="#linkage">linkage type</a>, and an
972 optional <a href="#visibility">visibility style</a>.</p>
975 <pre class="doc_code">
976 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
981 <!-- ======================================================================= -->
983 <a name="namedmetadatastructure">Named Metadata</a>
988 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
989 nodes</a> (but not metadata strings) are the only valid operands for
990 a named metadata.</p>
993 <pre class="doc_code">
994 ; Some unnamed metadata nodes, which are referenced by the named metadata.
995 !0 = metadata !{metadata !"zero"}
996 !1 = metadata !{metadata !"one"}
997 !2 = metadata !{metadata !"two"}
999 !name = !{!0, !1, !2}
1004 <!-- ======================================================================= -->
1006 <a name="paramattrs">Parameter Attributes</a>
1011 <p>The return type and each parameter of a function type may have a set of
1012 <i>parameter attributes</i> associated with them. Parameter attributes are
1013 used to communicate additional information about the result or parameters of
1014 a function. Parameter attributes are considered to be part of the function,
1015 not of the function type, so functions with different parameter attributes
1016 can have the same function type.</p>
1018 <p>Parameter attributes are simple keywords that follow the type specified. If
1019 multiple parameter attributes are needed, they are space separated. For
1022 <pre class="doc_code">
1023 declare i32 @printf(i8* noalias nocapture, ...)
1024 declare i32 @atoi(i8 zeroext)
1025 declare signext i8 @returns_signed_char()
1028 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1029 <tt>readonly</tt>) come immediately after the argument list.</p>
1031 <p>Currently, only the following parameter attributes are defined:</p>
1034 <dt><tt><b>zeroext</b></tt></dt>
1035 <dd>This indicates to the code generator that the parameter or return value
1036 should be zero-extended to the extent required by the target's ABI (which
1037 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1038 parameter) or the callee (for a return value).</dd>
1040 <dt><tt><b>signext</b></tt></dt>
1041 <dd>This indicates to the code generator that the parameter or return value
1042 should be sign-extended to the extent required by the target's ABI (which
1043 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1046 <dt><tt><b>inreg</b></tt></dt>
1047 <dd>This indicates that this parameter or return value should be treated in a
1048 special target-dependent fashion during while emitting code for a function
1049 call or return (usually, by putting it in a register as opposed to memory,
1050 though some targets use it to distinguish between two different kinds of
1051 registers). Use of this attribute is target-specific.</dd>
1053 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1054 <dd><p>This indicates that the pointer parameter should really be passed by
1055 value to the function. The attribute implies that a hidden copy of the
1057 is made between the caller and the callee, so the callee is unable to
1058 modify the value in the caller. This attribute is only valid on LLVM
1059 pointer arguments. It is generally used to pass structs and arrays by
1060 value, but is also valid on pointers to scalars. The copy is considered
1061 to belong to the caller not the callee (for example,
1062 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1063 <tt>byval</tt> parameters). This is not a valid attribute for return
1066 <p>The byval attribute also supports specifying an alignment with
1067 the align attribute. It indicates the alignment of the stack slot to
1068 form and the known alignment of the pointer specified to the call site. If
1069 the alignment is not specified, then the code generator makes a
1070 target-specific assumption.</p></dd>
1072 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1073 <dd>This indicates that the pointer parameter specifies the address of a
1074 structure that is the return value of the function in the source program.
1075 This pointer must be guaranteed by the caller to be valid: loads and
1076 stores to the structure may be assumed by the callee to not to trap. This
1077 may only be applied to the first parameter. This is not a valid attribute
1078 for return values. </dd>
1080 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1081 <dd>This indicates that pointer values
1082 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1083 value do not alias pointer values which are not <i>based</i> on it,
1084 ignoring certain "irrelevant" dependencies.
1085 For a call to the parent function, dependencies between memory
1086 references from before or after the call and from those during the call
1087 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1088 return value used in that call.
1089 The caller shares the responsibility with the callee for ensuring that
1090 these requirements are met.
1091 For further details, please see the discussion of the NoAlias response in
1092 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1094 Note that this definition of <tt>noalias</tt> is intentionally
1095 similar to the definition of <tt>restrict</tt> in C99 for function
1096 arguments, though it is slightly weaker.
1098 For function return values, C99's <tt>restrict</tt> is not meaningful,
1099 while LLVM's <tt>noalias</tt> is.
1102 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1103 <dd>This indicates that the callee does not make any copies of the pointer
1104 that outlive the callee itself. This is not a valid attribute for return
1107 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1108 <dd>This indicates that the pointer parameter can be excised using the
1109 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1110 attribute for return values.</dd>
1115 <!-- ======================================================================= -->
1117 <a name="gc">Garbage Collector Names</a>
1122 <p>Each function may specify a garbage collector name, which is simply a
1125 <pre class="doc_code">
1126 define void @f() gc "name" { ... }
1129 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1130 collector which will cause the compiler to alter its output in order to
1131 support the named garbage collection algorithm.</p>
1135 <!-- ======================================================================= -->
1137 <a name="fnattrs">Function Attributes</a>
1142 <p>Function attributes are set to communicate additional information about a
1143 function. Function attributes are considered to be part of the function, not
1144 of the function type, so functions with different parameter attributes can
1145 have the same function type.</p>
1147 <p>Function attributes are simple keywords that follow the type specified. If
1148 multiple attributes are needed, they are space separated. For example:</p>
1150 <pre class="doc_code">
1151 define void @f() noinline { ... }
1152 define void @f() alwaysinline { ... }
1153 define void @f() alwaysinline optsize { ... }
1154 define void @f() optsize { ... }
1158 <dt><tt><b>address_safety</b></tt></dt>
1159 <dd>This attribute indicates that the address safety analysis
1160 is enabled for this function. </dd>
1162 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1163 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1164 the backend should forcibly align the stack pointer. Specify the
1165 desired alignment, which must be a power of two, in parentheses.
1167 <dt><tt><b>alwaysinline</b></tt></dt>
1168 <dd>This attribute indicates that the inliner should attempt to inline this
1169 function into callers whenever possible, ignoring any active inlining size
1170 threshold for this caller.</dd>
1172 <dt><tt><b>nonlazybind</b></tt></dt>
1173 <dd>This attribute suppresses lazy symbol binding for the function. This
1174 may make calls to the function faster, at the cost of extra program
1175 startup time if the function is not called during program startup.</dd>
1177 <dt><tt><b>inlinehint</b></tt></dt>
1178 <dd>This attribute indicates that the source code contained a hint that inlining
1179 this function is desirable (such as the "inline" keyword in C/C++). It
1180 is just a hint; it imposes no requirements on the inliner.</dd>
1182 <dt><tt><b>naked</b></tt></dt>
1183 <dd>This attribute disables prologue / epilogue emission for the function.
1184 This can have very system-specific consequences.</dd>
1186 <dt><tt><b>noimplicitfloat</b></tt></dt>
1187 <dd>This attributes disables implicit floating point instructions.</dd>
1189 <dt><tt><b>noinline</b></tt></dt>
1190 <dd>This attribute indicates that the inliner should never inline this
1191 function in any situation. This attribute may not be used together with
1192 the <tt>alwaysinline</tt> attribute.</dd>
1194 <dt><tt><b>noredzone</b></tt></dt>
1195 <dd>This attribute indicates that the code generator should not use a red
1196 zone, even if the target-specific ABI normally permits it.</dd>
1198 <dt><tt><b>noreturn</b></tt></dt>
1199 <dd>This function attribute indicates that the function never returns
1200 normally. This produces undefined behavior at runtime if the function
1201 ever does dynamically return.</dd>
1203 <dt><tt><b>nounwind</b></tt></dt>
1204 <dd>This function attribute indicates that the function never returns with an
1205 unwind or exceptional control flow. If the function does unwind, its
1206 runtime behavior is undefined.</dd>
1208 <dt><tt><b>optsize</b></tt></dt>
1209 <dd>This attribute suggests that optimization passes and code generator passes
1210 make choices that keep the code size of this function low, and otherwise
1211 do optimizations specifically to reduce code size.</dd>
1213 <dt><tt><b>readnone</b></tt></dt>
1214 <dd>This attribute indicates that the function computes its result (or decides
1215 to unwind an exception) based strictly on its arguments, without
1216 dereferencing any pointer arguments or otherwise accessing any mutable
1217 state (e.g. memory, control registers, etc) visible to caller functions.
1218 It does not write through any pointer arguments
1219 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1220 changes any state visible to callers. This means that it cannot unwind
1221 exceptions by calling the <tt>C++</tt> exception throwing methods.</dd>
1223 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1224 <dd>This attribute indicates that the function does not write through any
1225 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1226 arguments) or otherwise modify any state (e.g. memory, control registers,
1227 etc) visible to caller functions. It may dereference pointer arguments
1228 and read state that may be set in the caller. A readonly function always
1229 returns the same value (or unwinds an exception identically) when called
1230 with the same set of arguments and global state. It cannot unwind an
1231 exception by calling the <tt>C++</tt> exception throwing methods.</dd>
1233 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1234 <dd>This attribute indicates that this function can return twice. The
1235 C <code>setjmp</code> is an example of such a function. The compiler
1236 disables some optimizations (like tail calls) in the caller of these
1239 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1240 <dd>This attribute indicates that the function should emit a stack smashing
1241 protector. It is in the form of a "canary"—a random value placed on
1242 the stack before the local variables that's checked upon return from the
1243 function to see if it has been overwritten. A heuristic is used to
1244 determine if a function needs stack protectors or not.<br>
1246 If a function that has an <tt>ssp</tt> attribute is inlined into a
1247 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1248 function will have an <tt>ssp</tt> attribute.</dd>
1250 <dt><tt><b>sspreq</b></tt></dt>
1251 <dd>This attribute indicates that the function should <em>always</em> emit a
1252 stack smashing protector. This overrides
1253 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1255 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1256 function that doesn't have an <tt>sspreq</tt> attribute or which has
1257 an <tt>ssp</tt> attribute, then the resulting function will have
1258 an <tt>sspreq</tt> attribute.</dd>
1260 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1261 <dd>This attribute indicates that the ABI being targeted requires that
1262 an unwind table entry be produce for this function even if we can
1263 show that no exceptions passes by it. This is normally the case for
1264 the ELF x86-64 abi, but it can be disabled for some compilation
1270 <!-- ======================================================================= -->
1272 <a name="moduleasm">Module-Level Inline Assembly</a>
1277 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1278 the GCC "file scope inline asm" blocks. These blocks are internally
1279 concatenated by LLVM and treated as a single unit, but may be separated in
1280 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1282 <pre class="doc_code">
1283 module asm "inline asm code goes here"
1284 module asm "more can go here"
1287 <p>The strings can contain any character by escaping non-printable characters.
1288 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1291 <p>The inline asm code is simply printed to the machine code .s file when
1292 assembly code is generated.</p>
1296 <!-- ======================================================================= -->
1298 <a name="datalayout">Data Layout</a>
1303 <p>A module may specify a target specific data layout string that specifies how
1304 data is to be laid out in memory. The syntax for the data layout is
1307 <pre class="doc_code">
1308 target datalayout = "<i>layout specification</i>"
1311 <p>The <i>layout specification</i> consists of a list of specifications
1312 separated by the minus sign character ('-'). Each specification starts with
1313 a letter and may include other information after the letter to define some
1314 aspect of the data layout. The specifications accepted are as follows:</p>
1318 <dd>Specifies that the target lays out data in big-endian form. That is, the
1319 bits with the most significance have the lowest address location.</dd>
1322 <dd>Specifies that the target lays out data in little-endian form. That is,
1323 the bits with the least significance have the lowest address
1326 <dt><tt>S<i>size</i></tt></dt>
1327 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1328 of stack variables is limited to the natural stack alignment to avoid
1329 dynamic stack realignment. The stack alignment must be a multiple of
1330 8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1331 which does not prevent any alignment promotions.</dd>
1333 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1334 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1335 <i>preferred</i> alignments. All sizes are in bits. Specifying
1336 the <i>pref</i> alignment is optional. If omitted, the
1337 preceding <tt>:</tt> should be omitted too.</dd>
1339 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1340 <dd>This specifies the alignment for an integer type of a given bit
1341 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1343 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1344 <dd>This specifies the alignment for a vector type of a given bit
1347 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1348 <dd>This specifies the alignment for a floating point type of a given bit
1349 <i>size</i>. Only values of <i>size</i> that are supported by the target
1350 will work. 32 (float) and 64 (double) are supported on all targets;
1351 80 or 128 (different flavors of long double) are also supported on some
1354 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1355 <dd>This specifies the alignment for an aggregate type of a given bit
1358 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1359 <dd>This specifies the alignment for a stack object of a given bit
1362 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1363 <dd>This specifies a set of native integer widths for the target CPU
1364 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1365 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1366 this set are considered to support most general arithmetic
1367 operations efficiently.</dd>
1370 <p>When constructing the data layout for a given target, LLVM starts with a
1371 default set of specifications which are then (possibly) overridden by the
1372 specifications in the <tt>datalayout</tt> keyword. The default specifications
1373 are given in this list:</p>
1376 <li><tt>E</tt> - big endian</li>
1377 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1378 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1379 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1380 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1381 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1382 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1383 alignment of 64-bits</li>
1384 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1385 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1386 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1387 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1388 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1389 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1392 <p>When LLVM is determining the alignment for a given type, it uses the
1393 following rules:</p>
1396 <li>If the type sought is an exact match for one of the specifications, that
1397 specification is used.</li>
1399 <li>If no match is found, and the type sought is an integer type, then the
1400 smallest integer type that is larger than the bitwidth of the sought type
1401 is used. If none of the specifications are larger than the bitwidth then
1402 the the largest integer type is used. For example, given the default
1403 specifications above, the i7 type will use the alignment of i8 (next
1404 largest) while both i65 and i256 will use the alignment of i64 (largest
1407 <li>If no match is found, and the type sought is a vector type, then the
1408 largest vector type that is smaller than the sought vector type will be
1409 used as a fall back. This happens because <128 x double> can be
1410 implemented in terms of 64 <2 x double>, for example.</li>
1413 <p>The function of the data layout string may not be what you expect. Notably,
1414 this is not a specification from the frontend of what alignment the code
1415 generator should use.</p>
1417 <p>Instead, if specified, the target data layout is required to match what the
1418 ultimate <em>code generator</em> expects. This string is used by the
1419 mid-level optimizers to
1420 improve code, and this only works if it matches what the ultimate code
1421 generator uses. If you would like to generate IR that does not embed this
1422 target-specific detail into the IR, then you don't have to specify the
1423 string. This will disable some optimizations that require precise layout
1424 information, but this also prevents those optimizations from introducing
1425 target specificity into the IR.</p>
1431 <!-- ======================================================================= -->
1433 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1438 <p>Any memory access must be done through a pointer value associated
1439 with an address range of the memory access, otherwise the behavior
1440 is undefined. Pointer values are associated with address ranges
1441 according to the following rules:</p>
1444 <li>A pointer value is associated with the addresses associated with
1445 any value it is <i>based</i> on.
1446 <li>An address of a global variable is associated with the address
1447 range of the variable's storage.</li>
1448 <li>The result value of an allocation instruction is associated with
1449 the address range of the allocated storage.</li>
1450 <li>A null pointer in the default address-space is associated with
1452 <li>An integer constant other than zero or a pointer value returned
1453 from a function not defined within LLVM may be associated with address
1454 ranges allocated through mechanisms other than those provided by
1455 LLVM. Such ranges shall not overlap with any ranges of addresses
1456 allocated by mechanisms provided by LLVM.</li>
1459 <p>A pointer value is <i>based</i> on another pointer value according
1460 to the following rules:</p>
1463 <li>A pointer value formed from a
1464 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1465 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1466 <li>The result value of a
1467 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1468 of the <tt>bitcast</tt>.</li>
1469 <li>A pointer value formed by an
1470 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1471 pointer values that contribute (directly or indirectly) to the
1472 computation of the pointer's value.</li>
1473 <li>The "<i>based</i> on" relationship is transitive.</li>
1476 <p>Note that this definition of <i>"based"</i> is intentionally
1477 similar to the definition of <i>"based"</i> in C99, though it is
1478 slightly weaker.</p>
1480 <p>LLVM IR does not associate types with memory. The result type of a
1481 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1482 alignment of the memory from which to load, as well as the
1483 interpretation of the value. The first operand type of a
1484 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1485 and alignment of the store.</p>
1487 <p>Consequently, type-based alias analysis, aka TBAA, aka
1488 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1489 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1490 additional information which specialized optimization passes may use
1491 to implement type-based alias analysis.</p>
1495 <!-- ======================================================================= -->
1497 <a name="volatile">Volatile Memory Accesses</a>
1502 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1503 href="#i_store"><tt>store</tt></a>s, and <a
1504 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1505 The optimizers must not change the number of volatile operations or change their
1506 order of execution relative to other volatile operations. The optimizers
1507 <i>may</i> change the order of volatile operations relative to non-volatile
1508 operations. This is not Java's "volatile" and has no cross-thread
1509 synchronization behavior.</p>
1513 <!-- ======================================================================= -->
1515 <a name="memmodel">Memory Model for Concurrent Operations</a>
1520 <p>The LLVM IR does not define any way to start parallel threads of execution
1521 or to register signal handlers. Nonetheless, there are platform-specific
1522 ways to create them, and we define LLVM IR's behavior in their presence. This
1523 model is inspired by the C++0x memory model.</p>
1525 <p>For a more informal introduction to this model, see the
1526 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1528 <p>We define a <i>happens-before</i> partial order as the least partial order
1531 <li>Is a superset of single-thread program order, and</li>
1532 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1533 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1534 by platform-specific techniques, like pthread locks, thread
1535 creation, thread joining, etc., and by atomic instructions.
1536 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1540 <p>Note that program order does not introduce <i>happens-before</i> edges
1541 between a thread and signals executing inside that thread.</p>
1543 <p>Every (defined) read operation (load instructions, memcpy, atomic
1544 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1545 (defined) write operations (store instructions, atomic
1546 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1547 initialized globals are considered to have a write of the initializer which is
1548 atomic and happens before any other read or write of the memory in question.
1549 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1550 any write to the same byte, except:</p>
1553 <li>If <var>write<sub>1</sub></var> happens before
1554 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1555 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1556 does not see <var>write<sub>1</sub></var>.
1557 <li>If <var>R<sub>byte</sub></var> happens before
1558 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1559 see <var>write<sub>3</sub></var>.
1562 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1564 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1565 is supposed to give guarantees which can support
1566 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1567 addresses which do not behave like normal memory. It does not generally
1568 provide cross-thread synchronization.)
1569 <li>Otherwise, if there is no write to the same byte that happens before
1570 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1571 <tt>undef</tt> for that byte.
1572 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1573 <var>R<sub>byte</sub></var> returns the value written by that
1575 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1576 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1577 values written. See the <a href="#ordering">Atomic Memory Ordering
1578 Constraints</a> section for additional constraints on how the choice
1580 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1583 <p><var>R</var> returns the value composed of the series of bytes it read.
1584 This implies that some bytes within the value may be <tt>undef</tt>
1585 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1586 defines the semantics of the operation; it doesn't mean that targets will
1587 emit more than one instruction to read the series of bytes.</p>
1589 <p>Note that in cases where none of the atomic intrinsics are used, this model
1590 places only one restriction on IR transformations on top of what is required
1591 for single-threaded execution: introducing a store to a byte which might not
1592 otherwise be stored is not allowed in general. (Specifically, in the case
1593 where another thread might write to and read from an address, introducing a
1594 store can change a load that may see exactly one write into a load that may
1595 see multiple writes.)</p>
1597 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1598 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1599 none of the backends currently in the tree fall into this category; however,
1600 there might be targets which care. If there are, we want a paragraph
1603 Targets may specify that stores narrower than a certain width are not
1604 available; on such a target, for the purposes of this model, treat any
1605 non-atomic write with an alignment or width less than the minimum width
1606 as if it writes to the relevant surrounding bytes.
1611 <!-- ======================================================================= -->
1613 <a name="ordering">Atomic Memory Ordering Constraints</a>
1618 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1619 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1620 <a href="#i_fence"><code>fence</code></a>,
1621 <a href="#i_load"><code>atomic load</code></a>, and
1622 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1623 that determines which other atomic instructions on the same address they
1624 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1625 but are somewhat more colloquial. If these descriptions aren't precise enough,
1626 check those specs (see spec references in the
1627 <a href="Atomics.html#introduction">atomics guide</a>).
1628 <a href="#i_fence"><code>fence</code></a> instructions
1629 treat these orderings somewhat differently since they don't take an address.
1630 See that instruction's documentation for details.</p>
1632 <p>For a simpler introduction to the ordering constraints, see the
1633 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1636 <dt><code>unordered</code></dt>
1637 <dd>The set of values that can be read is governed by the happens-before
1638 partial order. A value cannot be read unless some operation wrote it.
1639 This is intended to provide a guarantee strong enough to model Java's
1640 non-volatile shared variables. This ordering cannot be specified for
1641 read-modify-write operations; it is not strong enough to make them atomic
1642 in any interesting way.</dd>
1643 <dt><code>monotonic</code></dt>
1644 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1645 total order for modifications by <code>monotonic</code> operations on each
1646 address. All modification orders must be compatible with the happens-before
1647 order. There is no guarantee that the modification orders can be combined to
1648 a global total order for the whole program (and this often will not be
1649 possible). The read in an atomic read-modify-write operation
1650 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1651 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1652 reads the value in the modification order immediately before the value it
1653 writes. If one atomic read happens before another atomic read of the same
1654 address, the later read must see the same value or a later value in the
1655 address's modification order. This disallows reordering of
1656 <code>monotonic</code> (or stronger) operations on the same address. If an
1657 address is written <code>monotonic</code>ally by one thread, and other threads
1658 <code>monotonic</code>ally read that address repeatedly, the other threads must
1659 eventually see the write. This corresponds to the C++0x/C1x
1660 <code>memory_order_relaxed</code>.</dd>
1661 <dt><code>acquire</code></dt>
1662 <dd>In addition to the guarantees of <code>monotonic</code>,
1663 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1664 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1665 <dt><code>release</code></dt>
1666 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1667 writes a value which is subsequently read by an <code>acquire</code> operation,
1668 it <i>synchronizes-with</i> that operation. (This isn't a complete
1669 description; see the C++0x definition of a release sequence.) This corresponds
1670 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1671 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1672 <code>acquire</code> and <code>release</code> operation on its address.
1673 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1674 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1675 <dd>In addition to the guarantees of <code>acq_rel</code>
1676 (<code>acquire</code> for an operation which only reads, <code>release</code>
1677 for an operation which only writes), there is a global total order on all
1678 sequentially-consistent operations on all addresses, which is consistent with
1679 the <i>happens-before</i> partial order and with the modification orders of
1680 all the affected addresses. Each sequentially-consistent read sees the last
1681 preceding write to the same address in this global order. This corresponds
1682 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1685 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1686 it only <i>synchronizes with</i> or participates in modification and seq_cst
1687 total orderings with other operations running in the same thread (for example,
1688 in signal handlers).</p>
1694 <!-- *********************************************************************** -->
1695 <h2><a name="typesystem">Type System</a></h2>
1696 <!-- *********************************************************************** -->
1700 <p>The LLVM type system is one of the most important features of the
1701 intermediate representation. Being typed enables a number of optimizations
1702 to be performed on the intermediate representation directly, without having
1703 to do extra analyses on the side before the transformation. A strong type
1704 system makes it easier to read the generated code and enables novel analyses
1705 and transformations that are not feasible to perform on normal three address
1706 code representations.</p>
1708 <!-- ======================================================================= -->
1710 <a name="t_classifications">Type Classifications</a>
1715 <p>The types fall into a few useful classifications:</p>
1717 <table border="1" cellspacing="0" cellpadding="4">
1719 <tr><th>Classification</th><th>Types</th></tr>
1721 <td><a href="#t_integer">integer</a></td>
1722 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1725 <td><a href="#t_floating">floating point</a></td>
1726 <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1729 <td><a name="t_firstclass">first class</a></td>
1730 <td><a href="#t_integer">integer</a>,
1731 <a href="#t_floating">floating point</a>,
1732 <a href="#t_pointer">pointer</a>,
1733 <a href="#t_vector">vector</a>,
1734 <a href="#t_struct">structure</a>,
1735 <a href="#t_array">array</a>,
1736 <a href="#t_label">label</a>,
1737 <a href="#t_metadata">metadata</a>.
1741 <td><a href="#t_primitive">primitive</a></td>
1742 <td><a href="#t_label">label</a>,
1743 <a href="#t_void">void</a>,
1744 <a href="#t_integer">integer</a>,
1745 <a href="#t_floating">floating point</a>,
1746 <a href="#t_x86mmx">x86mmx</a>,
1747 <a href="#t_metadata">metadata</a>.</td>
1750 <td><a href="#t_derived">derived</a></td>
1751 <td><a href="#t_array">array</a>,
1752 <a href="#t_function">function</a>,
1753 <a href="#t_pointer">pointer</a>,
1754 <a href="#t_struct">structure</a>,
1755 <a href="#t_vector">vector</a>,
1756 <a href="#t_opaque">opaque</a>.
1762 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1763 important. Values of these types are the only ones which can be produced by
1768 <!-- ======================================================================= -->
1770 <a name="t_primitive">Primitive Types</a>
1775 <p>The primitive types are the fundamental building blocks of the LLVM
1778 <!-- _______________________________________________________________________ -->
1780 <a name="t_integer">Integer Type</a>
1786 <p>The integer type is a very simple type that simply specifies an arbitrary
1787 bit width for the integer type desired. Any bit width from 1 bit to
1788 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1795 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1799 <table class="layout">
1801 <td class="left"><tt>i1</tt></td>
1802 <td class="left">a single-bit integer.</td>
1805 <td class="left"><tt>i32</tt></td>
1806 <td class="left">a 32-bit integer.</td>
1809 <td class="left"><tt>i1942652</tt></td>
1810 <td class="left">a really big integer of over 1 million bits.</td>
1816 <!-- _______________________________________________________________________ -->
1818 <a name="t_floating">Floating Point Types</a>
1825 <tr><th>Type</th><th>Description</th></tr>
1826 <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr>
1827 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1828 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1829 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1830 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1831 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1837 <!-- _______________________________________________________________________ -->
1839 <a name="t_x86mmx">X86mmx Type</a>
1845 <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>
1854 <!-- _______________________________________________________________________ -->
1856 <a name="t_void">Void Type</a>
1862 <p>The void type does not represent any value and has no size.</p>
1871 <!-- _______________________________________________________________________ -->
1873 <a name="t_label">Label Type</a>
1879 <p>The label type represents code labels.</p>
1888 <!-- _______________________________________________________________________ -->
1890 <a name="t_metadata">Metadata Type</a>
1896 <p>The metadata type represents embedded metadata. No derived types may be
1897 created from metadata except for <a href="#t_function">function</a>
1909 <!-- ======================================================================= -->
1911 <a name="t_derived">Derived Types</a>
1916 <p>The real power in LLVM comes from the derived types in the system. This is
1917 what allows a programmer to represent arrays, functions, pointers, and other
1918 useful types. Each of these types contain one or more element types which
1919 may be a primitive type, or another derived type. For example, it is
1920 possible to have a two dimensional array, using an array as the element type
1921 of another array.</p>
1923 <!-- _______________________________________________________________________ -->
1925 <a name="t_aggregate">Aggregate Types</a>
1930 <p>Aggregate Types are a subset of derived types that can contain multiple
1931 member types. <a href="#t_array">Arrays</a> and
1932 <a href="#t_struct">structs</a> are aggregate types.
1933 <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p>
1937 <!-- _______________________________________________________________________ -->
1939 <a name="t_array">Array Type</a>
1945 <p>The array type is a very simple derived type that arranges elements
1946 sequentially in memory. The array type requires a size (number of elements)
1947 and an underlying data type.</p>
1951 [<# elements> x <elementtype>]
1954 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1955 be any type with a size.</p>
1958 <table class="layout">
1960 <td class="left"><tt>[40 x i32]</tt></td>
1961 <td class="left">Array of 40 32-bit integer values.</td>
1964 <td class="left"><tt>[41 x i32]</tt></td>
1965 <td class="left">Array of 41 32-bit integer values.</td>
1968 <td class="left"><tt>[4 x i8]</tt></td>
1969 <td class="left">Array of 4 8-bit integer values.</td>
1972 <p>Here are some examples of multidimensional arrays:</p>
1973 <table class="layout">
1975 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1976 <td class="left">3x4 array of 32-bit integer values.</td>
1979 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1980 <td class="left">12x10 array of single precision floating point values.</td>
1983 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1984 <td class="left">2x3x4 array of 16-bit integer values.</td>
1988 <p>There is no restriction on indexing beyond the end of the array implied by
1989 a static type (though there are restrictions on indexing beyond the bounds
1990 of an allocated object in some cases). This means that single-dimension
1991 'variable sized array' addressing can be implemented in LLVM with a zero
1992 length array type. An implementation of 'pascal style arrays' in LLVM could
1993 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1997 <!-- _______________________________________________________________________ -->
1999 <a name="t_function">Function Type</a>
2005 <p>The function type can be thought of as a function signature. It consists of
2006 a return type and a list of formal parameter types. The return type of a
2007 function type is a first class type or a void type.</p>
2011 <returntype> (<parameter list>)
2014 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
2015 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
2016 which indicates that the function takes a variable number of arguments.
2017 Variable argument functions can access their arguments with
2018 the <a href="#int_varargs">variable argument handling intrinsic</a>
2019 functions. '<tt><returntype></tt>' is any type except
2020 <a href="#t_label">label</a>.</p>
2023 <table class="layout">
2025 <td class="left"><tt>i32 (i32)</tt></td>
2026 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2028 </tr><tr class="layout">
2029 <td class="left"><tt>float (i16, i32 *) *
2031 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2032 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2033 returning <tt>float</tt>.
2035 </tr><tr class="layout">
2036 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2037 <td class="left">A vararg function that takes at least one
2038 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2039 which returns an integer. This is the signature for <tt>printf</tt> in
2042 </tr><tr class="layout">
2043 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2044 <td class="left">A function taking an <tt>i32</tt>, returning a
2045 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2052 <!-- _______________________________________________________________________ -->
2054 <a name="t_struct">Structure Type</a>
2060 <p>The structure type is used to represent a collection of data members together
2061 in memory. The elements of a structure may be any type that has a size.</p>
2063 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2064 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2065 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2066 Structures in registers are accessed using the
2067 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2068 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2070 <p>Structures may optionally be "packed" structures, which indicate that the
2071 alignment of the struct is one byte, and that there is no padding between
2072 the elements. In non-packed structs, padding between field types is inserted
2073 as defined by the TargetData string in the module, which is required to match
2074 what the underlying code generator expects.</p>
2076 <p>Structures can either be "literal" or "identified". A literal structure is
2077 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2078 types are always defined at the top level with a name. Literal types are
2079 uniqued by their contents and can never be recursive or opaque since there is
2080 no way to write one. Identified types can be recursive, can be opaqued, and are
2086 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2087 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2091 <table class="layout">
2093 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2094 <td class="left">A triple of three <tt>i32</tt> values</td>
2097 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2098 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2099 second element is a <a href="#t_pointer">pointer</a> to a
2100 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2101 an <tt>i32</tt>.</td>
2104 <td class="left"><tt><{ i8, i32 }></tt></td>
2105 <td class="left">A packed struct known to be 5 bytes in size.</td>
2111 <!-- _______________________________________________________________________ -->
2113 <a name="t_opaque">Opaque Structure Types</a>
2119 <p>Opaque structure types are used to represent named structure types that do
2120 not have a body specified. This corresponds (for example) to the C notion of
2121 a forward declared structure.</p>
2130 <table class="layout">
2132 <td class="left"><tt>opaque</tt></td>
2133 <td class="left">An opaque type.</td>
2141 <!-- _______________________________________________________________________ -->
2143 <a name="t_pointer">Pointer Type</a>
2149 <p>The pointer type is used to specify memory locations.
2150 Pointers are commonly used to reference objects in memory.</p>
2152 <p>Pointer types may have an optional address space attribute defining the
2153 numbered address space where the pointed-to object resides. The default
2154 address space is number zero. The semantics of non-zero address
2155 spaces are target-specific.</p>
2157 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2158 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2166 <table class="layout">
2168 <td class="left"><tt>[4 x i32]*</tt></td>
2169 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2170 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2173 <td class="left"><tt>i32 (i32*) *</tt></td>
2174 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2175 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2179 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2180 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2181 that resides in address space #5.</td>
2187 <!-- _______________________________________________________________________ -->
2189 <a name="t_vector">Vector Type</a>
2195 <p>A vector type is a simple derived type that represents a vector of elements.
2196 Vector types are used when multiple primitive data are operated in parallel
2197 using a single instruction (SIMD). A vector type requires a size (number of
2198 elements) and an underlying primitive data type. Vector types are considered
2199 <a href="#t_firstclass">first class</a>.</p>
2203 < <# elements> x <elementtype> >
2206 <p>The number of elements is a constant integer value larger than 0; elementtype
2207 may be any integer or floating point type, or a pointer to these types.
2208 Vectors of size zero are not allowed. </p>
2211 <table class="layout">
2213 <td class="left"><tt><4 x i32></tt></td>
2214 <td class="left">Vector of 4 32-bit integer values.</td>
2217 <td class="left"><tt><8 x float></tt></td>
2218 <td class="left">Vector of 8 32-bit floating-point values.</td>
2221 <td class="left"><tt><2 x i64></tt></td>
2222 <td class="left">Vector of 2 64-bit integer values.</td>
2225 <td class="left"><tt><4 x i64*></tt></td>
2226 <td class="left">Vector of 4 pointers to 64-bit integer values.</td>
2236 <!-- *********************************************************************** -->
2237 <h2><a name="constants">Constants</a></h2>
2238 <!-- *********************************************************************** -->
2242 <p>LLVM has several different basic types of constants. This section describes
2243 them all and their syntax.</p>
2245 <!-- ======================================================================= -->
2247 <a name="simpleconstants">Simple Constants</a>
2253 <dt><b>Boolean constants</b></dt>
2254 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2255 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2257 <dt><b>Integer constants</b></dt>
2258 <dd>Standard integers (such as '4') are constants of
2259 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2260 with integer types.</dd>
2262 <dt><b>Floating point constants</b></dt>
2263 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2264 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2265 notation (see below). The assembler requires the exact decimal value of a
2266 floating-point constant. For example, the assembler accepts 1.25 but
2267 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2268 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2270 <dt><b>Null pointer constants</b></dt>
2271 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2272 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2275 <p>The one non-intuitive notation for constants is the hexadecimal form of
2276 floating point constants. For example, the form '<tt>double
2277 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2278 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2279 constants are required (and the only time that they are generated by the
2280 disassembler) is when a floating point constant must be emitted but it cannot
2281 be represented as a decimal floating point number in a reasonable number of
2282 digits. For example, NaN's, infinities, and other special values are
2283 represented in their IEEE hexadecimal format so that assembly and disassembly
2284 do not cause any bits to change in the constants.</p>
2286 <p>When using the hexadecimal form, constants of types half, float, and double are
2287 represented using the 16-digit form shown above (which matches the IEEE754
2288 representation for double); half and float values must, however, be exactly
2289 representable as IEE754 half and single precision, respectively.
2290 Hexadecimal format is always used
2291 for long double, and there are three forms of long double. The 80-bit format
2292 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2293 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2294 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2295 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2296 currently supported target uses this format. Long doubles will only work if
2297 they match the long double format on your target. The IEEE 16-bit format
2298 (half precision) is represented by <tt>0xH</tt> followed by 4 hexadecimal
2299 digits. All hexadecimal formats are big-endian (sign bit at the left).</p>
2301 <p>There are no constants of type x86mmx.</p>
2304 <!-- ======================================================================= -->
2306 <a name="aggregateconstants"></a> <!-- old anchor -->
2307 <a name="complexconstants">Complex Constants</a>
2312 <p>Complex constants are a (potentially recursive) combination of simple
2313 constants and smaller complex constants.</p>
2316 <dt><b>Structure constants</b></dt>
2317 <dd>Structure constants are represented with notation similar to structure
2318 type definitions (a comma separated list of elements, surrounded by braces
2319 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2320 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2321 Structure constants must have <a href="#t_struct">structure type</a>, and
2322 the number and types of elements must match those specified by the
2325 <dt><b>Array constants</b></dt>
2326 <dd>Array constants are represented with notation similar to array type
2327 definitions (a comma separated list of elements, surrounded by square
2328 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2329 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2330 the number and types of elements must match those specified by the
2333 <dt><b>Vector constants</b></dt>
2334 <dd>Vector constants are represented with notation similar to vector type
2335 definitions (a comma separated list of elements, surrounded by
2336 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2337 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2338 have <a href="#t_vector">vector type</a>, and the number and types of
2339 elements must match those specified by the type.</dd>
2341 <dt><b>Zero initialization</b></dt>
2342 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2343 value to zero of <em>any</em> type, including scalar and
2344 <a href="#t_aggregate">aggregate</a> types.
2345 This is often used to avoid having to print large zero initializers
2346 (e.g. for large arrays) and is always exactly equivalent to using explicit
2347 zero initializers.</dd>
2349 <dt><b>Metadata node</b></dt>
2350 <dd>A metadata node is a structure-like constant with
2351 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2352 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2353 be interpreted as part of the instruction stream, metadata is a place to
2354 attach additional information such as debug info.</dd>
2359 <!-- ======================================================================= -->
2361 <a name="globalconstants">Global Variable and Function Addresses</a>
2366 <p>The addresses of <a href="#globalvars">global variables</a>
2367 and <a href="#functionstructure">functions</a> are always implicitly valid
2368 (link-time) constants. These constants are explicitly referenced when
2369 the <a href="#identifiers">identifier for the global</a> is used and always
2370 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2371 legal LLVM file:</p>
2373 <pre class="doc_code">
2376 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2381 <!-- ======================================================================= -->
2383 <a name="undefvalues">Undefined Values</a>
2388 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2389 indicates that the user of the value may receive an unspecified bit-pattern.
2390 Undefined values may be of any type (other than '<tt>label</tt>'
2391 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2393 <p>Undefined values are useful because they indicate to the compiler that the
2394 program is well defined no matter what value is used. This gives the
2395 compiler more freedom to optimize. Here are some examples of (potentially
2396 surprising) transformations that are valid (in pseudo IR):</p>
2399 <pre class="doc_code">
2409 <p>This is safe because all of the output bits are affected by the undef bits.
2410 Any output bit can have a zero or one depending on the input bits.</p>
2412 <pre class="doc_code">
2423 <p>These logical operations have bits that are not always affected by the input.
2424 For example, if <tt>%X</tt> has a zero bit, then the output of the
2425 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2426 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2427 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2428 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2429 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2430 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2431 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2433 <pre class="doc_code">
2434 %A = select undef, %X, %Y
2435 %B = select undef, 42, %Y
2436 %C = select %X, %Y, undef
2447 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2448 branch) conditions can go <em>either way</em>, but they have to come from one
2449 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2450 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2451 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2452 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2453 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2456 <pre class="doc_code">
2457 %A = xor undef, undef
2475 <p>This example points out that two '<tt>undef</tt>' operands are not
2476 necessarily the same. This can be surprising to people (and also matches C
2477 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2478 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2479 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2480 its value over its "live range". This is true because the variable doesn't
2481 actually <em>have a live range</em>. Instead, the value is logically read
2482 from arbitrary registers that happen to be around when needed, so the value
2483 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2484 need to have the same semantics or the core LLVM "replace all uses with"
2485 concept would not hold.</p>
2487 <pre class="doc_code">
2495 <p>These examples show the crucial difference between an <em>undefined
2496 value</em> and <em>undefined behavior</em>. An undefined value (like
2497 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2498 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2499 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2500 defined on SNaN's. However, in the second example, we can make a more
2501 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2502 arbitrary value, we are allowed to assume that it could be zero. Since a
2503 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2504 the operation does not execute at all. This allows us to delete the divide and
2505 all code after it. Because the undefined operation "can't happen", the
2506 optimizer can assume that it occurs in dead code.</p>
2508 <pre class="doc_code">
2509 a: store undef -> %X
2510 b: store %X -> undef
2516 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2517 undefined value can be assumed to not have any effect; we can assume that the
2518 value is overwritten with bits that happen to match what was already there.
2519 However, a store <em>to</em> an undefined location could clobber arbitrary
2520 memory, therefore, it has undefined behavior.</p>
2524 <!-- ======================================================================= -->
2526 <a name="poisonvalues">Poison Values</a>
2531 <p>Poison values are similar to <a href="#undefvalues">undef values</a>, however
2532 they also represent the fact that an instruction or constant expression which
2533 cannot evoke side effects has nevertheless detected a condition which results
2534 in undefined behavior.</p>
2536 <p>There is currently no way of representing a poison value in the IR; they
2537 only exist when produced by operations such as
2538 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2540 <p>Poison value behavior is defined in terms of value <i>dependence</i>:</p>
2543 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2544 their operands.</li>
2546 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2547 to their dynamic predecessor basic block.</li>
2549 <li>Function arguments depend on the corresponding actual argument values in
2550 the dynamic callers of their functions.</li>
2552 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2553 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2554 control back to them.</li>
2556 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2557 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>,
2558 or exception-throwing call instructions that dynamically transfer control
2561 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2562 referenced memory addresses, following the order in the IR
2563 (including loads and stores implied by intrinsics such as
2564 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2566 <!-- TODO: In the case of multiple threads, this only applies if the store
2567 "happens-before" the load or store. -->
2569 <!-- TODO: floating-point exception state -->
2571 <li>An instruction with externally visible side effects depends on the most
2572 recent preceding instruction with externally visible side effects, following
2573 the order in the IR. (This includes
2574 <a href="#volatile">volatile operations</a>.)</li>
2576 <li>An instruction <i>control-depends</i> on a
2577 <a href="#terminators">terminator instruction</a>
2578 if the terminator instruction has multiple successors and the instruction
2579 is always executed when control transfers to one of the successors, and
2580 may not be executed when control is transferred to another.</li>
2582 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2583 instruction if the set of instructions it otherwise depends on would be
2584 different if the terminator had transferred control to a different
2587 <li>Dependence is transitive.</li>
2591 <p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>,
2592 with the additional affect that any instruction which has a <i>dependence</i>
2593 on a poison value has undefined behavior.</p>
2595 <p>Here are some examples:</p>
2597 <pre class="doc_code">
2599 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2600 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2601 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2602 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2604 store i32 %poison, i32* @g ; Poison value stored to memory.
2605 %poison2 = load i32* @g ; Poison value loaded back from memory.
2607 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2609 %narrowaddr = bitcast i32* @g to i16*
2610 %wideaddr = bitcast i32* @g to i64*
2611 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2612 %poison4 = load i64* %wideaddr ; Returns a poison value.
2614 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2615 br i1 %cmp, label %true, label %end ; Branch to either destination.
2618 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2619 ; it has undefined behavior.
2623 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2624 ; Both edges into this PHI are
2625 ; control-dependent on %cmp, so this
2626 ; always results in a poison value.
2628 store volatile i32 0, i32* @g ; This would depend on the store in %true
2629 ; if %cmp is true, or the store in %entry
2630 ; otherwise, so this is undefined behavior.
2632 br i1 %cmp, label %second_true, label %second_end
2633 ; The same branch again, but this time the
2634 ; true block doesn't have side effects.
2641 store volatile i32 0, i32* @g ; This time, the instruction always depends
2642 ; on the store in %end. Also, it is
2643 ; control-equivalent to %end, so this is
2644 ; well-defined (ignoring earlier undefined
2645 ; behavior in this example).
2650 <!-- ======================================================================= -->
2652 <a name="blockaddress">Addresses of Basic Blocks</a>
2657 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2659 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2660 basic block in the specified function, and always has an i8* type. Taking
2661 the address of the entry block is illegal.</p>
2663 <p>This value only has defined behavior when used as an operand to the
2664 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2665 comparisons against null. Pointer equality tests between labels addresses
2666 results in undefined behavior — though, again, comparison against null
2667 is ok, and no label is equal to the null pointer. This may be passed around
2668 as an opaque pointer sized value as long as the bits are not inspected. This
2669 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2670 long as the original value is reconstituted before the <tt>indirectbr</tt>
2673 <p>Finally, some targets may provide defined semantics when using the value as
2674 the operand to an inline assembly, but that is target specific.</p>
2679 <!-- ======================================================================= -->
2681 <a name="constantexprs">Constant Expressions</a>
2686 <p>Constant expressions are used to allow expressions involving other constants
2687 to be used as constants. Constant expressions may be of
2688 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2689 operation that does not have side effects (e.g. load and call are not
2690 supported). The following is the syntax for constant expressions:</p>
2693 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2694 <dd>Truncate a constant to another type. The bit size of CST must be larger
2695 than the bit size of TYPE. Both types must be integers.</dd>
2697 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2698 <dd>Zero extend a constant to another type. The bit size of CST must be
2699 smaller than the bit size of TYPE. Both types must be integers.</dd>
2701 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2702 <dd>Sign extend a constant to another type. The bit size of CST must be
2703 smaller than the bit size of TYPE. Both types must be integers.</dd>
2705 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2706 <dd>Truncate a floating point constant to another floating point type. The
2707 size of CST must be larger than the size of TYPE. Both types must be
2708 floating point.</dd>
2710 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2711 <dd>Floating point extend a constant to another type. The size of CST must be
2712 smaller or equal to the size of TYPE. Both types must be floating
2715 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2716 <dd>Convert a floating point constant to the corresponding unsigned integer
2717 constant. TYPE must be a scalar or vector integer type. CST must be of
2718 scalar or vector floating point type. Both CST and TYPE must be scalars,
2719 or vectors of the same number of elements. If the value won't fit in the
2720 integer type, the results are undefined.</dd>
2722 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2723 <dd>Convert a floating point constant to the corresponding signed integer
2724 constant. TYPE must be a scalar or vector integer type. CST must be of
2725 scalar or vector floating point type. Both CST and TYPE must be scalars,
2726 or vectors of the same number of elements. If the value won't fit in the
2727 integer type, the results are undefined.</dd>
2729 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2730 <dd>Convert an unsigned integer constant to the corresponding floating point
2731 constant. TYPE must be a scalar or vector floating point type. CST must be
2732 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2733 vectors of the same number of elements. If the value won't fit in the
2734 floating point type, the results are undefined.</dd>
2736 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2737 <dd>Convert a signed integer constant to the corresponding floating point
2738 constant. TYPE must be a scalar or vector floating point type. CST must be
2739 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2740 vectors of the same number of elements. If the value won't fit in the
2741 floating point type, the results are undefined.</dd>
2743 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2744 <dd>Convert a pointer typed constant to the corresponding integer constant
2745 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2746 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2747 make it fit in <tt>TYPE</tt>.</dd>
2749 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2750 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2751 type. CST must be of integer type. The CST value is zero extended,
2752 truncated, or unchanged to make it fit in a pointer size. This one is
2753 <i>really</i> dangerous!</dd>
2755 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2756 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2757 are the same as those for the <a href="#i_bitcast">bitcast
2758 instruction</a>.</dd>
2760 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2761 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2762 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2763 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2764 instruction, the index list may have zero or more indexes, which are
2765 required to make sense for the type of "CSTPTR".</dd>
2767 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2768 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2770 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2771 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2773 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2774 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2776 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2777 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2780 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2781 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2784 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2785 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2788 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2789 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2790 constants. The index list is interpreted in a similar manner as indices in
2791 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2792 index value must be specified.</dd>
2794 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2795 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2796 constants. The index list is interpreted in a similar manner as indices in
2797 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2798 index value must be specified.</dd>
2800 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2801 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2802 be any of the <a href="#binaryops">binary</a>
2803 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2804 on operands are the same as those for the corresponding instruction
2805 (e.g. no bitwise operations on floating point values are allowed).</dd>
2812 <!-- *********************************************************************** -->
2813 <h2><a name="othervalues">Other Values</a></h2>
2814 <!-- *********************************************************************** -->
2816 <!-- ======================================================================= -->
2818 <a name="inlineasm">Inline Assembler Expressions</a>
2823 <p>LLVM supports inline assembler expressions (as opposed
2824 to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of
2825 a special value. This value represents the inline assembler as a string
2826 (containing the instructions to emit), a list of operand constraints (stored
2827 as a string), a flag that indicates whether or not the inline asm
2828 expression has side effects, and a flag indicating whether the function
2829 containing the asm needs to align its stack conservatively. An example
2830 inline assembler expression is:</p>
2832 <pre class="doc_code">
2833 i32 (i32) asm "bswap $0", "=r,r"
2836 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2837 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2840 <pre class="doc_code">
2841 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2844 <p>Inline asms with side effects not visible in the constraint list must be
2845 marked as having side effects. This is done through the use of the
2846 '<tt>sideeffect</tt>' keyword, like so:</p>
2848 <pre class="doc_code">
2849 call void asm sideeffect "eieio", ""()
2852 <p>In some cases inline asms will contain code that will not work unless the
2853 stack is aligned in some way, such as calls or SSE instructions on x86,
2854 yet will not contain code that does that alignment within the asm.
2855 The compiler should make conservative assumptions about what the asm might
2856 contain and should generate its usual stack alignment code in the prologue
2857 if the '<tt>alignstack</tt>' keyword is present:</p>
2859 <pre class="doc_code">
2860 call void asm alignstack "eieio", ""()
2863 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2867 <p>TODO: The format of the asm and constraints string still need to be
2868 documented here. Constraints on what can be done (e.g. duplication, moving,
2869 etc need to be documented). This is probably best done by reference to
2870 another document that covers inline asm from a holistic perspective.</p>
2873 <!-- _______________________________________________________________________ -->
2875 <a name="inlineasm_md">Inline Asm Metadata</a>
2880 <p>The call instructions that wrap inline asm nodes may have a
2881 "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant
2882 integers. If present, the code generator will use the integer as the
2883 location cookie value when report errors through the <tt>LLVMContext</tt>
2884 error reporting mechanisms. This allows a front-end to correlate backend
2885 errors that occur with inline asm back to the source code that produced it.
2888 <pre class="doc_code">
2889 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2891 !42 = !{ i32 1234567 }
2894 <p>It is up to the front-end to make sense of the magic numbers it places in the
2895 IR. If the MDNode contains multiple constants, the code generator will use
2896 the one that corresponds to the line of the asm that the error occurs on.</p>
2902 <!-- ======================================================================= -->
2904 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2909 <p>LLVM IR allows metadata to be attached to instructions in the program that
2910 can convey extra information about the code to the optimizers and code
2911 generator. One example application of metadata is source-level debug
2912 information. There are two metadata primitives: strings and nodes. All
2913 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2914 preceding exclamation point ('<tt>!</tt>').</p>
2916 <p>A metadata string is a string surrounded by double quotes. It can contain
2917 any character by escaping non-printable characters with "<tt>\xx</tt>" where
2918 "<tt>xx</tt>" is the two digit hex code. For example:
2919 "<tt>!"test\00"</tt>".</p>
2921 <p>Metadata nodes are represented with notation similar to structure constants
2922 (a comma separated list of elements, surrounded by braces and preceded by an
2923 exclamation point). Metadata nodes can have any values as their operand. For
2926 <div class="doc_code">
2928 !{ metadata !"test\00", i32 10}
2932 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2933 metadata nodes, which can be looked up in the module symbol table. For
2936 <div class="doc_code">
2938 !foo = metadata !{!4, !3}
2942 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2943 function is using two metadata arguments:</p>
2945 <div class="doc_code">
2947 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2951 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2952 attached to the <tt>add</tt> instruction using the <tt>!dbg</tt>
2955 <div class="doc_code">
2957 %indvar.next = add i64 %indvar, 1, !dbg !21
2961 <p>More information about specific metadata nodes recognized by the optimizers
2962 and code generator is found below.</p>
2964 <!-- _______________________________________________________________________ -->
2966 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
2971 <p>In LLVM IR, memory does not have types, so LLVM's own type system is not
2972 suitable for doing TBAA. Instead, metadata is added to the IR to describe
2973 a type system of a higher level language. This can be used to implement
2974 typical C/C++ TBAA, but it can also be used to implement custom alias
2975 analysis behavior for other languages.</p>
2977 <p>The current metadata format is very simple. TBAA metadata nodes have up to
2978 three fields, e.g.:</p>
2980 <div class="doc_code">
2982 !0 = metadata !{ metadata !"an example type tree" }
2983 !1 = metadata !{ metadata !"int", metadata !0 }
2984 !2 = metadata !{ metadata !"float", metadata !0 }
2985 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2989 <p>The first field is an identity field. It can be any value, usually
2990 a metadata string, which uniquely identifies the type. The most important
2991 name in the tree is the name of the root node. Two trees with
2992 different root node names are entirely disjoint, even if they
2993 have leaves with common names.</p>
2995 <p>The second field identifies the type's parent node in the tree, or
2996 is null or omitted for a root node. A type is considered to alias
2997 all of its descendants and all of its ancestors in the tree. Also,
2998 a type is considered to alias all types in other trees, so that
2999 bitcode produced from multiple front-ends is handled conservatively.</p>
3001 <p>If the third field is present, it's an integer which if equal to 1
3002 indicates that the type is "constant" (meaning
3003 <tt>pointsToConstantMemory</tt> should return true; see
3004 <a href="AliasAnalysis.html#OtherItfs">other useful
3005 <tt>AliasAnalysis</tt> methods</a>).</p>
3009 <!-- _______________________________________________________________________ -->
3011 <a name="fpmath">'<tt>fpmath</tt>' Metadata</a>
3016 <p><tt>fpmath</tt> metadata may be attached to any instruction of floating point
3017 type. It can be used to express the maximum acceptable error in the result of
3018 that instruction, in ULPs, thus potentially allowing the compiler to use a
3019 more efficient but less accurate method of computing it. ULP is defined as
3024 <p>If <tt>x</tt> is a real number that lies between two finite consecutive
3025 floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one
3026 of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the
3027 distance between the two non-equal finite floating-point numbers nearest
3028 <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p>
3032 <p>The metadata node shall consist of a single positive floating point number
3033 representing the maximum relative error, for example:</p>
3035 <div class="doc_code">
3037 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3043 <!-- _______________________________________________________________________ -->
3045 <a name="range">'<tt>range</tt>' Metadata</a>
3049 <p><tt>range</tt> metadata may be attached only to loads of integer types. It
3050 expresses the possible ranges the loaded value is in. The ranges are
3051 represented with a flattened list of integers. The loaded value is known to
3052 be in the union of the ranges defined by each consecutive pair. Each pair
3053 has the following properties:</p>
3055 <li>The type must match the type loaded by the instruction.</li>
3056 <li>The pair <tt>a,b</tt> represents the range <tt>[a,b)</tt>.</li>
3057 <li>Both <tt>a</tt> and <tt>b</tt> are constants.</li>
3058 <li>The range is allowed to wrap.</li>
3059 <li>The range should not represent the full or empty set. That is,
3060 <tt>a!=b</tt>. </li>
3062 <p> In addition, the pairs must be in signed order of the lower bound and
3063 they must be non-contiguous.</p>
3066 <div class="doc_code">
3068 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
3069 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3070 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
3071 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
3073 !0 = metadata !{ i8 0, i8 2 }
3074 !1 = metadata !{ i8 255, i8 2 }
3075 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
3076 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
3084 <!-- *********************************************************************** -->
3086 <a name="module_flags">Module Flags Metadata</a>
3088 <!-- *********************************************************************** -->
3092 <p>Information about the module as a whole is difficult to convey to LLVM's
3093 subsystems. The LLVM IR isn't sufficient to transmit this
3094 information. The <tt>llvm.module.flags</tt> named metadata exists in order to
3095 facilitate this. These flags are in the form of key / value pairs —
3096 much like a dictionary — making it easy for any subsystem who cares
3097 about a flag to look it up.</p>
3099 <p>The <tt>llvm.module.flags</tt> metadata contains a list of metadata
3100 triplets. Each triplet has the following form:</p>
3103 <li>The first element is a <i>behavior</i> flag, which specifies the behavior
3104 when two (or more) modules are merged together, and it encounters two (or
3105 more) metadata with the same ID. The supported behaviors are described
3108 <li>The second element is a metadata string that is a unique ID for the
3109 metadata. How each ID is interpreted is documented below.</li>
3111 <li>The third element is the value of the flag.</li>
3114 <p>When two (or more) modules are merged together, the resulting
3115 <tt>llvm.module.flags</tt> metadata is the union of the
3116 modules' <tt>llvm.module.flags</tt> metadata. The only exception being a flag
3117 with the <i>Override</i> behavior, which may override another flag's value
3120 <p>The following behaviors are supported:</p>
3122 <table border="1" cellspacing="0" cellpadding="4">
3132 <dt><b>Error</b></dt>
3133 <dd>Emits an error if two values disagree. It is an error to have an ID
3134 with both an Error and a Warning behavior.</dd>
3142 <dt><b>Warning</b></dt>
3143 <dd>Emits a warning if two values disagree.</dd>
3151 <dt><b>Require</b></dt>
3152 <dd>Emits an error when the specified value is not present or doesn't
3153 have the specified value. It is an error for two (or more)
3154 <tt>llvm.module.flags</tt> with the same ID to have the Require
3155 behavior but different values. There may be multiple Require flags
3164 <dt><b>Override</b></dt>
3165 <dd>Uses the specified value if the two values disagree. It is an
3166 error for two (or more) <tt>llvm.module.flags</tt> with the same
3167 ID to have the Override behavior but different values.</dd>
3174 <p>An example of module flags:</p>
3176 <pre class="doc_code">
3177 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3178 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3179 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3180 !3 = metadata !{ i32 3, metadata !"qux",
3182 metadata !"foo", i32 1
3185 !llvm.module.flags = !{ !0, !1, !2, !3 }
3189 <li><p>Metadata <tt>!0</tt> has the ID <tt>!"foo"</tt> and the value '1'. The
3190 behavior if two or more <tt>!"foo"</tt> flags are seen is to emit an
3191 error if their values are not equal.</p></li>
3193 <li><p>Metadata <tt>!1</tt> has the ID <tt>!"bar"</tt> and the value '37'. The
3194 behavior if two or more <tt>!"bar"</tt> flags are seen is to use the
3195 value '37' if their values are not equal.</p></li>
3197 <li><p>Metadata <tt>!2</tt> has the ID <tt>!"qux"</tt> and the value '42'. The
3198 behavior if two or more <tt>!"qux"</tt> flags are seen is to emit a
3199 warning if their values are not equal.</p></li>
3201 <li><p>Metadata <tt>!3</tt> has the ID <tt>!"qux"</tt> and the value:</p>
3203 <pre class="doc_code">
3204 metadata !{ metadata !"foo", i32 1 }
3207 <p>The behavior is to emit an error if the <tt>llvm.module.flags</tt> does
3208 not contain a flag with the ID <tt>!"foo"</tt> that has the value
3209 '1'. If two or more <tt>!"qux"</tt> flags exist, then they must have
3210 the same value or an error will be issued.</p></li>
3214 <!-- ======================================================================= -->
3216 <a name="objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a>
3221 <p>On the Mach-O platform, Objective-C stores metadata about garbage collection
3222 in a special section called "image info". The metadata consists of a version
3223 number and a bitmask specifying what types of garbage collection are
3224 supported (if any) by the file. If two or more modules are linked together
3225 their garbage collection metadata needs to be merged rather than appended
3228 <p>The Objective-C garbage collection module flags metadata consists of the
3229 following key-value pairs:</p>
3231 <table border="1" cellspacing="0" cellpadding="4">
3239 <td><tt>Objective-C Version</tt></td>
3240 <td align="left"><b>[Required]</b> — The Objective-C ABI
3241 version. Valid values are 1 and 2.</td>
3244 <td><tt>Objective-C Image Info Version</tt></td>
3245 <td align="left"><b>[Required]</b> — The version of the image info
3246 section. Currently always 0.</td>
3249 <td><tt>Objective-C Image Info Section</tt></td>
3250 <td align="left"><b>[Required]</b> — The section to place the
3251 metadata. Valid values are <tt>"__OBJC, __image_info, regular"</tt> for
3252 Objective-C ABI version 1, and <tt>"__DATA,__objc_imageinfo, regular,
3253 no_dead_strip"</tt> for Objective-C ABI version 2.</td>
3256 <td><tt>Objective-C Garbage Collection</tt></td>
3257 <td align="left"><b>[Required]</b> — Specifies whether garbage
3258 collection is supported or not. Valid values are 0, for no garbage
3259 collection, and 2, for garbage collection supported.</td>
3262 <td><tt>Objective-C GC Only</tt></td>
3263 <td align="left"><b>[Optional]</b> — Specifies that only garbage
3264 collection is supported. If present, its value must be 6. This flag
3265 requires that the <tt>Objective-C Garbage Collection</tt> flag have the
3271 <p>Some important flag interactions:</p>
3274 <li>If a module with <tt>Objective-C Garbage Collection</tt> set to 0 is
3275 merged with a module with <tt>Objective-C Garbage Collection</tt> set to
3276 2, then the resulting module has the <tt>Objective-C Garbage
3277 Collection</tt> flag set to 0.</li>
3279 <li>A module with <tt>Objective-C Garbage Collection</tt> set to 0 cannot be
3280 merged with a module with <tt>Objective-C GC Only</tt> set to 6.</li>
3287 <!-- *********************************************************************** -->
3289 <a name="intrinsic_globals">Intrinsic Global Variables</a>
3291 <!-- *********************************************************************** -->
3293 <p>LLVM has a number of "magic" global variables that contain data that affect
3294 code generation or other IR semantics. These are documented here. All globals
3295 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
3296 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
3299 <!-- ======================================================================= -->
3301 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
3306 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
3307 href="#linkage_appending">appending linkage</a>. This array contains a list of
3308 pointers to global variables and functions which may optionally have a pointer
3309 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
3311 <div class="doc_code">
3316 @llvm.used = appending global [2 x i8*] [
3318 i8* bitcast (i32* @Y to i8*)
3319 ], section "llvm.metadata"
3323 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
3324 compiler, assembler, and linker are required to treat the symbol as if there
3325 is a reference to the global that it cannot see. For example, if a variable
3326 has internal linkage and no references other than that from
3327 the <tt>@llvm.used</tt> list, it cannot be deleted. This is commonly used to
3328 represent references from inline asms and other things the compiler cannot
3329 "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p>
3331 <p>On some targets, the code generator must emit a directive to the assembler or
3332 object file to prevent the assembler and linker from molesting the
3337 <!-- ======================================================================= -->
3339 <a name="intg_compiler_used">
3340 The '<tt>llvm.compiler.used</tt>' Global Variable
3346 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
3347 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
3348 touching the symbol. On targets that support it, this allows an intelligent
3349 linker to optimize references to the symbol without being impeded as it would
3350 be by <tt>@llvm.used</tt>.</p>
3352 <p>This is a rare construct that should only be used in rare circumstances, and
3353 should not be exposed to source languages.</p>
3357 <!-- ======================================================================= -->
3359 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
3364 <div class="doc_code">
3366 %0 = type { i32, void ()* }
3367 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3371 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor
3372 functions and associated priorities. The functions referenced by this array
3373 will be called in ascending order of priority (i.e. lowest first) when the
3374 module is loaded. The order of functions with the same priority is not
3379 <!-- ======================================================================= -->
3381 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3386 <div class="doc_code">
3388 %0 = type { i32, void ()* }
3389 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3393 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions
3394 and associated priorities. The functions referenced by this array will be
3395 called in descending order of priority (i.e. highest first) when the module
3396 is loaded. The order of functions with the same priority is not defined.</p>
3402 <!-- *********************************************************************** -->
3403 <h2><a name="instref">Instruction Reference</a></h2>
3404 <!-- *********************************************************************** -->
3408 <p>The LLVM instruction set consists of several different classifications of
3409 instructions: <a href="#terminators">terminator
3410 instructions</a>, <a href="#binaryops">binary instructions</a>,
3411 <a href="#bitwiseops">bitwise binary instructions</a>,
3412 <a href="#memoryops">memory instructions</a>, and
3413 <a href="#otherops">other instructions</a>.</p>
3415 <!-- ======================================================================= -->
3417 <a name="terminators">Terminator Instructions</a>
3422 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3423 in a program ends with a "Terminator" instruction, which indicates which
3424 block should be executed after the current block is finished. These
3425 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3426 control flow, not values (the one exception being the
3427 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3429 <p>The terminator instructions are:
3430 '<a href="#i_ret"><tt>ret</tt></a>',
3431 '<a href="#i_br"><tt>br</tt></a>',
3432 '<a href="#i_switch"><tt>switch</tt></a>',
3433 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3434 '<a href="#i_invoke"><tt>invoke</tt></a>',
3435 '<a href="#i_resume"><tt>resume</tt></a>', and
3436 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3438 <!-- _______________________________________________________________________ -->
3440 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3447 ret <type> <value> <i>; Return a value from a non-void function</i>
3448 ret void <i>; Return from void function</i>
3452 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3453 a value) from a function back to the caller.</p>
3455 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3456 value and then causes control flow, and one that just causes control flow to
3460 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3461 return value. The type of the return value must be a
3462 '<a href="#t_firstclass">first class</a>' type.</p>
3464 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3465 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3466 value or a return value with a type that does not match its type, or if it
3467 has a void return type and contains a '<tt>ret</tt>' instruction with a
3471 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3472 the calling function's context. If the caller is a
3473 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3474 instruction after the call. If the caller was an
3475 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3476 the beginning of the "normal" destination block. If the instruction returns
3477 a value, that value shall set the call or invoke instruction's return
3482 ret i32 5 <i>; Return an integer value of 5</i>
3483 ret void <i>; Return from a void function</i>
3484 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3488 <!-- _______________________________________________________________________ -->
3490 <a name="i_br">'<tt>br</tt>' Instruction</a>
3497 br i1 <cond>, label <iftrue>, label <iffalse>
3498 br label <dest> <i>; Unconditional branch</i>
3502 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3503 different basic block in the current function. There are two forms of this
3504 instruction, corresponding to a conditional branch and an unconditional
3508 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3509 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3510 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3514 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3515 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3516 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3517 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3522 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3523 br i1 %cond, label %IfEqual, label %IfUnequal
3525 <a href="#i_ret">ret</a> i32 1
3527 <a href="#i_ret">ret</a> i32 0
3532 <!-- _______________________________________________________________________ -->
3534 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3541 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3545 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3546 several different places. It is a generalization of the '<tt>br</tt>'
3547 instruction, allowing a branch to occur to one of many possible
3551 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3552 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3553 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3554 The table is not allowed to contain duplicate constant entries.</p>
3557 <p>The <tt>switch</tt> instruction specifies a table of values and
3558 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3559 is searched for the given value. If the value is found, control flow is
3560 transferred to the corresponding destination; otherwise, control flow is
3561 transferred to the default destination.</p>
3563 <h5>Implementation:</h5>
3564 <p>Depending on properties of the target machine and the particular
3565 <tt>switch</tt> instruction, this instruction may be code generated in
3566 different ways. For example, it could be generated as a series of chained
3567 conditional branches or with a lookup table.</p>
3571 <i>; Emulate a conditional br instruction</i>
3572 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3573 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3575 <i>; Emulate an unconditional br instruction</i>
3576 switch i32 0, label %dest [ ]
3578 <i>; Implement a jump table:</i>
3579 switch i32 %val, label %otherwise [ i32 0, label %onzero
3581 i32 2, label %ontwo ]
3587 <!-- _______________________________________________________________________ -->
3589 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3596 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3601 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3602 within the current function, whose address is specified by
3603 "<tt>address</tt>". Address must be derived from a <a
3604 href="#blockaddress">blockaddress</a> constant.</p>
3608 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3609 rest of the arguments indicate the full set of possible destinations that the
3610 address may point to. Blocks are allowed to occur multiple times in the
3611 destination list, though this isn't particularly useful.</p>
3613 <p>This destination list is required so that dataflow analysis has an accurate
3614 understanding of the CFG.</p>
3618 <p>Control transfers to the block specified in the address argument. All
3619 possible destination blocks must be listed in the label list, otherwise this
3620 instruction has undefined behavior. This implies that jumps to labels
3621 defined in other functions have undefined behavior as well.</p>
3623 <h5>Implementation:</h5>
3625 <p>This is typically implemented with a jump through a register.</p>
3629 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3635 <!-- _______________________________________________________________________ -->
3637 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3644 <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>]
3645 to label <normal label> unwind label <exception label>
3649 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3650 function, with the possibility of control flow transfer to either the
3651 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3652 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3653 control flow will return to the "normal" label. If the callee (or any
3654 indirect callees) returns via the "<a href="#i_resume"><tt>resume</tt></a>"
3655 instruction or other exception handling mechanism, control is interrupted and
3656 continued at the dynamically nearest "exception" label.</p>
3658 <p>The '<tt>exception</tt>' label is a
3659 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3660 exception. As such, '<tt>exception</tt>' label is required to have the
3661 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3662 the information about the behavior of the program after unwinding
3663 happens, as its first non-PHI instruction. The restrictions on the
3664 "<tt>landingpad</tt>" instruction's tightly couples it to the
3665 "<tt>invoke</tt>" instruction, so that the important information contained
3666 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3670 <p>This instruction requires several arguments:</p>
3673 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3674 convention</a> the call should use. If none is specified, the call
3675 defaults to using C calling conventions.</li>
3677 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3678 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3679 '<tt>inreg</tt>' attributes are valid here.</li>
3681 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3682 function value being invoked. In most cases, this is a direct function
3683 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3684 off an arbitrary pointer to function value.</li>
3686 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3687 function to be invoked. </li>
3689 <li>'<tt>function args</tt>': argument list whose types match the function
3690 signature argument types and parameter attributes. All arguments must be
3691 of <a href="#t_firstclass">first class</a> type. If the function
3692 signature indicates the function accepts a variable number of arguments,
3693 the extra arguments can be specified.</li>
3695 <li>'<tt>normal label</tt>': the label reached when the called function
3696 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3698 <li>'<tt>exception label</tt>': the label reached when a callee returns via
3699 the <a href="#i_resume"><tt>resume</tt></a> instruction or other exception
3700 handling mechanism.</li>
3702 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3703 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3704 '<tt>readnone</tt>' attributes are valid here.</li>
3708 <p>This instruction is designed to operate as a standard
3709 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3710 primary difference is that it establishes an association with a label, which
3711 is used by the runtime library to unwind the stack.</p>
3713 <p>This instruction is used in languages with destructors to ensure that proper
3714 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3715 exception. Additionally, this is important for implementation of
3716 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3718 <p>For the purposes of the SSA form, the definition of the value returned by the
3719 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3720 block to the "normal" label. If the callee unwinds then no return value is
3725 %retval = invoke i32 @Test(i32 15) to label %Continue
3726 unwind label %TestCleanup <i>; {i32}:retval set</i>
3727 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3728 unwind label %TestCleanup <i>; {i32}:retval set</i>
3733 <!-- _______________________________________________________________________ -->
3736 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3743 resume <type> <value>
3747 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3751 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3752 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3756 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3757 (in-flight) exception whose unwinding was interrupted with
3758 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3762 resume { i8*, i32 } %exn
3767 <!-- _______________________________________________________________________ -->
3770 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3781 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3782 instruction is used to inform the optimizer that a particular portion of the
3783 code is not reachable. This can be used to indicate that the code after a
3784 no-return function cannot be reached, and other facts.</p>
3787 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3793 <!-- ======================================================================= -->
3795 <a name="binaryops">Binary Operations</a>
3800 <p>Binary operators are used to do most of the computation in a program. They
3801 require two operands of the same type, execute an operation on them, and
3802 produce a single value. The operands might represent multiple data, as is
3803 the case with the <a href="#t_vector">vector</a> data type. The result value
3804 has the same type as its operands.</p>
3806 <p>There are several different binary operators:</p>
3808 <!-- _______________________________________________________________________ -->
3810 <a name="i_add">'<tt>add</tt>' Instruction</a>
3817 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3818 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3819 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3820 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3824 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3827 <p>The two arguments to the '<tt>add</tt>' instruction must
3828 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3829 integer values. Both arguments must have identical types.</p>
3832 <p>The value produced is the integer sum of the two operands.</p>
3834 <p>If the sum has unsigned overflow, the result returned is the mathematical
3835 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3837 <p>Because LLVM integers use a two's complement representation, this instruction
3838 is appropriate for both signed and unsigned integers.</p>
3840 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3841 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3842 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3843 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3844 respectively, occurs.</p>
3848 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3853 <!-- _______________________________________________________________________ -->
3855 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3862 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3866 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3869 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3870 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3871 floating point values. Both arguments must have identical types.</p>
3874 <p>The value produced is the floating point sum of the two operands.</p>
3878 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3883 <!-- _______________________________________________________________________ -->
3885 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3892 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3893 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3894 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3895 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3899 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3902 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3903 '<tt>neg</tt>' instruction present in most other intermediate
3904 representations.</p>
3907 <p>The two arguments to the '<tt>sub</tt>' instruction must
3908 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3909 integer values. Both arguments must have identical types.</p>
3912 <p>The value produced is the integer difference of the two operands.</p>
3914 <p>If the difference has unsigned overflow, the result returned is the
3915 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3918 <p>Because LLVM integers use a two's complement representation, this instruction
3919 is appropriate for both signed and unsigned integers.</p>
3921 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3922 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3923 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3924 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3925 respectively, occurs.</p>
3929 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3930 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3935 <!-- _______________________________________________________________________ -->
3937 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3944 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3948 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3951 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3952 '<tt>fneg</tt>' instruction present in most other intermediate
3953 representations.</p>
3956 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3957 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3958 floating point values. Both arguments must have identical types.</p>
3961 <p>The value produced is the floating point difference of the two operands.</p>
3965 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3966 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3971 <!-- _______________________________________________________________________ -->
3973 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3980 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3981 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3982 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3983 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3987 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3990 <p>The two arguments to the '<tt>mul</tt>' instruction must
3991 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3992 integer values. Both arguments must have identical types.</p>
3995 <p>The value produced is the integer product of the two operands.</p>
3997 <p>If the result of the multiplication has unsigned overflow, the result
3998 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3999 width of the result.</p>
4001 <p>Because LLVM integers use a two's complement representation, and the result
4002 is the same width as the operands, this instruction returns the correct
4003 result for both signed and unsigned integers. If a full product
4004 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
4005 be sign-extended or zero-extended as appropriate to the width of the full
4008 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
4009 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
4010 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
4011 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4012 respectively, occurs.</p>
4016 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
4021 <!-- _______________________________________________________________________ -->
4023 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
4030 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4034 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
4037 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
4038 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4039 floating point values. Both arguments must have identical types.</p>
4042 <p>The value produced is the floating point product of the two operands.</p>
4046 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
4051 <!-- _______________________________________________________________________ -->
4053 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
4060 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4061 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4065 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
4068 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
4069 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4070 values. Both arguments must have identical types.</p>
4073 <p>The value produced is the unsigned integer quotient of the two operands.</p>
4075 <p>Note that unsigned integer division and signed integer division are distinct
4076 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
4078 <p>Division by zero leads to undefined behavior.</p>
4080 <p>If the <tt>exact</tt> keyword is present, the result value of the
4081 <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a
4082 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
4087 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4092 <!-- _______________________________________________________________________ -->
4094 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
4101 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4102 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4106 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
4109 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
4110 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4111 values. Both arguments must have identical types.</p>
4114 <p>The value produced is the signed integer quotient of the two operands rounded
4117 <p>Note that signed integer division and unsigned integer division are distinct
4118 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
4120 <p>Division by zero leads to undefined behavior. Overflow also leads to
4121 undefined behavior; this is a rare case, but can occur, for example, by doing
4122 a 32-bit division of -2147483648 by -1.</p>
4124 <p>If the <tt>exact</tt> keyword is present, the result value of the
4125 <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would
4130 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4135 <!-- _______________________________________________________________________ -->
4137 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
4144 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4148 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
4151 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
4152 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4153 floating point values. Both arguments must have identical types.</p>
4156 <p>The value produced is the floating point quotient of the two operands.</p>
4160 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
4165 <!-- _______________________________________________________________________ -->
4167 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
4174 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4178 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
4179 division of its two arguments.</p>
4182 <p>The two arguments to the '<tt>urem</tt>' instruction must be
4183 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4184 values. Both arguments must have identical types.</p>
4187 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
4188 This instruction always performs an unsigned division to get the
4191 <p>Note that unsigned integer remainder and signed integer remainder are
4192 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
4194 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
4198 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4203 <!-- _______________________________________________________________________ -->
4205 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
4212 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4216 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
4217 division of its two operands. This instruction can also take
4218 <a href="#t_vector">vector</a> versions of the values in which case the
4219 elements must be integers.</p>
4222 <p>The two arguments to the '<tt>srem</tt>' instruction must be
4223 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4224 values. Both arguments must have identical types.</p>
4227 <p>This instruction returns the <i>remainder</i> of a division (where the result
4228 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
4229 <i>modulo</i> operator (where the result is either zero or has the same sign
4230 as the divisor, <tt>op2</tt>) of a value.
4231 For more information about the difference,
4232 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
4233 Math Forum</a>. For a table of how this is implemented in various languages,
4234 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
4235 Wikipedia: modulo operation</a>.</p>
4237 <p>Note that signed integer remainder and unsigned integer remainder are
4238 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
4240 <p>Taking the remainder of a division by zero leads to undefined behavior.
4241 Overflow also leads to undefined behavior; this is a rare case, but can
4242 occur, for example, by taking the remainder of a 32-bit division of
4243 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
4244 lets srem be implemented using instructions that return both the result of
4245 the division and the remainder.)</p>
4249 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4254 <!-- _______________________________________________________________________ -->
4256 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
4263 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4267 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
4268 its two operands.</p>
4271 <p>The two arguments to the '<tt>frem</tt>' instruction must be
4272 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4273 floating point values. Both arguments must have identical types.</p>
4276 <p>This instruction returns the <i>remainder</i> of a division. The remainder
4277 has the same sign as the dividend.</p>
4281 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
4288 <!-- ======================================================================= -->
4290 <a name="bitwiseops">Bitwise Binary Operations</a>
4295 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
4296 program. They are generally very efficient instructions and can commonly be
4297 strength reduced from other instructions. They require two operands of the
4298 same type, execute an operation on them, and produce a single value. The
4299 resulting value is the same type as its operands.</p>
4301 <!-- _______________________________________________________________________ -->
4303 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
4310 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4311 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4312 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4313 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4317 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
4318 a specified number of bits.</p>
4321 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
4322 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4323 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4326 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
4327 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
4328 is (statically or dynamically) negative or equal to or larger than the number
4329 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4330 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4331 shift amount in <tt>op2</tt>.</p>
4333 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
4334 <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits. If
4335 the <tt>nsw</tt> keyword is present, then the shift produces a
4336 <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree
4337 with the resultant sign bit. As such, NUW/NSW have the same semantics as
4338 they would if the shift were expressed as a mul instruction with the same
4339 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
4343 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
4344 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
4345 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
4346 <result> = shl i32 1, 32 <i>; undefined</i>
4347 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4352 <!-- _______________________________________________________________________ -->
4354 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4361 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4362 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4366 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4367 operand shifted to the right a specified number of bits with zero fill.</p>
4370 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4371 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4372 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4375 <p>This instruction always performs a logical shift right operation. The most
4376 significant bits of the result will be filled with zero bits after the shift.
4377 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4378 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4379 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4380 shift amount in <tt>op2</tt>.</p>
4382 <p>If the <tt>exact</tt> keyword is present, the result value of the
4383 <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4384 shifted out are non-zero.</p>
4389 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4390 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4391 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4392 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4393 <result> = lshr i32 1, 32 <i>; undefined</i>
4394 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4399 <!-- _______________________________________________________________________ -->
4401 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4408 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4409 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4413 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4414 operand shifted to the right a specified number of bits with sign
4418 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4419 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4420 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4423 <p>This instruction always performs an arithmetic shift right operation, The
4424 most significant bits of the result will be filled with the sign bit
4425 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4426 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4427 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4428 the corresponding shift amount in <tt>op2</tt>.</p>
4430 <p>If the <tt>exact</tt> keyword is present, the result value of the
4431 <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4432 shifted out are non-zero.</p>
4436 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4437 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4438 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4439 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4440 <result> = ashr i32 1, 32 <i>; undefined</i>
4441 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4446 <!-- _______________________________________________________________________ -->
4448 <a name="i_and">'<tt>and</tt>' Instruction</a>
4455 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4459 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4463 <p>The two arguments to the '<tt>and</tt>' instruction must be
4464 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4465 values. Both arguments must have identical types.</p>
4468 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4470 <table border="1" cellspacing="0" cellpadding="4">
4502 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4503 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4504 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4507 <!-- _______________________________________________________________________ -->
4509 <a name="i_or">'<tt>or</tt>' Instruction</a>
4516 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4520 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4524 <p>The two arguments to the '<tt>or</tt>' instruction must be
4525 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4526 values. Both arguments must have identical types.</p>
4529 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4531 <table border="1" cellspacing="0" cellpadding="4">
4563 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4564 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4565 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4570 <!-- _______________________________________________________________________ -->
4572 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4579 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4583 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4584 its two operands. The <tt>xor</tt> is used to implement the "one's
4585 complement" operation, which is the "~" operator in C.</p>
4588 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4589 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4590 values. Both arguments must have identical types.</p>
4593 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4595 <table border="1" cellspacing="0" cellpadding="4">
4627 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4628 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4629 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4630 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4637 <!-- ======================================================================= -->
4639 <a name="vectorops">Vector Operations</a>
4644 <p>LLVM supports several instructions to represent vector operations in a
4645 target-independent manner. These instructions cover the element-access and
4646 vector-specific operations needed to process vectors effectively. While LLVM
4647 does directly support these vector operations, many sophisticated algorithms
4648 will want to use target-specific intrinsics to take full advantage of a
4649 specific target.</p>
4651 <!-- _______________________________________________________________________ -->
4653 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4660 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4664 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4665 from a vector at a specified index.</p>
4669 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4670 of <a href="#t_vector">vector</a> type. The second operand is an index
4671 indicating the position from which to extract the element. The index may be
4675 <p>The result is a scalar of the same type as the element type of
4676 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4677 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4678 results are undefined.</p>
4682 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4687 <!-- _______________________________________________________________________ -->
4689 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4696 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4700 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4701 vector at a specified index.</p>
4704 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4705 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4706 whose type must equal the element type of the first operand. The third
4707 operand is an index indicating the position at which to insert the value.
4708 The index may be a variable.</p>
4711 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4712 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4713 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4714 results are undefined.</p>
4718 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4723 <!-- _______________________________________________________________________ -->
4725 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4732 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4736 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4737 from two input vectors, returning a vector with the same element type as the
4738 input and length that is the same as the shuffle mask.</p>
4741 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4742 with the same type. The third argument is a shuffle mask whose
4743 element type is always 'i32'. The result of the instruction is a vector
4744 whose length is the same as the shuffle mask and whose element type is the
4745 same as the element type of the first two operands.</p>
4747 <p>The shuffle mask operand is required to be a constant vector with either
4748 constant integer or undef values.</p>
4751 <p>The elements of the two input vectors are numbered from left to right across
4752 both of the vectors. The shuffle mask operand specifies, for each element of
4753 the result vector, which element of the two input vectors the result element
4754 gets. The element selector may be undef (meaning "don't care") and the
4755 second operand may be undef if performing a shuffle from only one vector.</p>
4759 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4760 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4761 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4762 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4763 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4764 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4765 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4766 <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>
4773 <!-- ======================================================================= -->
4775 <a name="aggregateops">Aggregate Operations</a>
4780 <p>LLVM supports several instructions for working with
4781 <a href="#t_aggregate">aggregate</a> values.</p>
4783 <!-- _______________________________________________________________________ -->
4785 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4792 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4796 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4797 from an <a href="#t_aggregate">aggregate</a> value.</p>
4800 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4801 of <a href="#t_struct">struct</a> or
4802 <a href="#t_array">array</a> type. The operands are constant indices to
4803 specify which value to extract in a similar manner as indices in a
4804 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4805 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4807 <li>Since the value being indexed is not a pointer, the first index is
4808 omitted and assumed to be zero.</li>
4809 <li>At least one index must be specified.</li>
4810 <li>Not only struct indices but also array indices must be in
4815 <p>The result is the value at the position in the aggregate specified by the
4820 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4825 <!-- _______________________________________________________________________ -->
4827 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4834 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4838 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4839 in an <a href="#t_aggregate">aggregate</a> value.</p>
4842 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4843 of <a href="#t_struct">struct</a> or
4844 <a href="#t_array">array</a> type. The second operand is a first-class
4845 value to insert. The following operands are constant indices indicating
4846 the position at which to insert the value in a similar manner as indices in a
4847 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4848 value to insert must have the same type as the value identified by the
4852 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4853 that of <tt>val</tt> except that the value at the position specified by the
4854 indices is that of <tt>elt</tt>.</p>
4858 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4859 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4860 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4867 <!-- ======================================================================= -->
4869 <a name="memoryops">Memory Access and Addressing Operations</a>
4874 <p>A key design point of an SSA-based representation is how it represents
4875 memory. In LLVM, no memory locations are in SSA form, which makes things
4876 very simple. This section describes how to read, write, and allocate
4879 <!-- _______________________________________________________________________ -->
4881 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4888 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4892 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4893 currently executing function, to be automatically released when this function
4894 returns to its caller. The object is always allocated in the generic address
4895 space (address space zero).</p>
4898 <p>The '<tt>alloca</tt>' instruction
4899 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4900 runtime stack, returning a pointer of the appropriate type to the program.
4901 If "NumElements" is specified, it is the number of elements allocated,
4902 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4903 specified, the value result of the allocation is guaranteed to be aligned to
4904 at least that boundary. If not specified, or if zero, the target can choose
4905 to align the allocation on any convenient boundary compatible with the
4908 <p>'<tt>type</tt>' may be any sized type.</p>
4911 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4912 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4913 memory is automatically released when the function returns. The
4914 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4915 variables that must have an address available. When the function returns
4916 (either with the <tt><a href="#i_ret">ret</a></tt>
4917 or <tt><a href="#i_resume">resume</a></tt> instructions), the memory is
4918 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4919 The order in which memory is allocated (ie., which way the stack grows) is
4926 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4927 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4928 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4929 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4934 <!-- _______________________________________________________________________ -->
4936 <a name="i_load">'<tt>load</tt>' Instruction</a>
4943 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4944 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4945 !<index> = !{ i32 1 }
4949 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4952 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4953 from which to load. The pointer must point to
4954 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4955 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4956 number or order of execution of this <tt>load</tt> with other <a
4957 href="#volatile">volatile operations</a>.</p>
4959 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
4960 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4961 argument. The <code>release</code> and <code>acq_rel</code> orderings are
4962 not valid on <code>load</code> instructions. Atomic loads produce <a
4963 href="#memorymodel">defined</a> results when they may see multiple atomic
4964 stores. The type of the pointee must be an integer type whose bit width
4965 is a power of two greater than or equal to eight and less than or equal
4966 to a target-specific size limit. <code>align</code> must be explicitly
4967 specified on atomic loads, and the load has undefined behavior if the
4968 alignment is not set to a value which is at least the size in bytes of
4969 the pointee. <code>!nontemporal</code> does not have any defined semantics
4970 for atomic loads.</p>
4972 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4973 operation (that is, the alignment of the memory address). A value of 0 or an
4974 omitted <tt>align</tt> argument means that the operation has the preferential
4975 alignment for the target. It is the responsibility of the code emitter to
4976 ensure that the alignment information is correct. Overestimating the
4977 alignment results in undefined behavior. Underestimating the alignment may
4978 produce less efficient code. An alignment of 1 is always safe.</p>
4980 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4981 metatadata name <index> corresponding to a metadata node with
4982 one <tt>i32</tt> entry of value 1. The existence of
4983 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4984 and code generator that this load is not expected to be reused in the cache.
4985 The code generator may select special instructions to save cache bandwidth,
4986 such as the <tt>MOVNT</tt> instruction on x86.</p>
4988 <p>The optional <tt>!invariant.load</tt> metadata must reference a single
4989 metatadata name <index> corresponding to a metadata node with no
4990 entries. The existence of the <tt>!invariant.load</tt> metatadata on the
4991 instruction tells the optimizer and code generator that this load address
4992 points to memory which does not change value during program execution.
4993 The optimizer may then move this load around, for example, by hoisting it
4994 out of loops using loop invariant code motion.</p>
4997 <p>The location of memory pointed to is loaded. If the value being loaded is of
4998 scalar type then the number of bytes read does not exceed the minimum number
4999 of bytes needed to hold all bits of the type. For example, loading an
5000 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
5001 <tt>i20</tt> with a size that is not an integral number of bytes, the result
5002 is undefined if the value was not originally written using a store of the
5007 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5008 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
5009 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
5014 <!-- _______________________________________________________________________ -->
5016 <a name="i_store">'<tt>store</tt>' Instruction</a>
5023 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
5024 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
5028 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
5031 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
5032 and an address at which to store it. The type of the
5033 '<tt><pointer></tt>' operand must be a pointer to
5034 the <a href="#t_firstclass">first class</a> type of the
5035 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
5036 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
5037 order of execution of this <tt>store</tt> with other <a
5038 href="#volatile">volatile operations</a>.</p>
5040 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
5041 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5042 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
5043 valid on <code>store</code> instructions. Atomic loads produce <a
5044 href="#memorymodel">defined</a> results when they may see multiple atomic
5045 stores. The type of the pointee must be an integer type whose bit width
5046 is a power of two greater than or equal to eight and less than or equal
5047 to a target-specific size limit. <code>align</code> must be explicitly
5048 specified on atomic stores, and the store has undefined behavior if the
5049 alignment is not set to a value which is at least the size in bytes of
5050 the pointee. <code>!nontemporal</code> does not have any defined semantics
5051 for atomic stores.</p>
5053 <p>The optional constant "align" argument specifies the alignment of the
5054 operation (that is, the alignment of the memory address). A value of 0 or an
5055 omitted "align" argument means that the operation has the preferential
5056 alignment for the target. It is the responsibility of the code emitter to
5057 ensure that the alignment information is correct. Overestimating the
5058 alignment results in an undefined behavior. Underestimating the alignment may
5059 produce less efficient code. An alignment of 1 is always safe.</p>
5061 <p>The optional !nontemporal metadata must reference a single metatadata
5062 name <index> corresponding to a metadata node with one i32 entry of
5063 value 1. The existence of the !nontemporal metatadata on the
5064 instruction tells the optimizer and code generator that this load is
5065 not expected to be reused in the cache. The code generator may
5066 select special instructions to save cache bandwidth, such as the
5067 MOVNT instruction on x86.</p>
5071 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
5072 location specified by the '<tt><pointer></tt>' operand. If
5073 '<tt><value></tt>' is of scalar type then the number of bytes written
5074 does not exceed the minimum number of bytes needed to hold all bits of the
5075 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
5076 writing a value of a type like <tt>i20</tt> with a size that is not an
5077 integral number of bytes, it is unspecified what happens to the extra bits
5078 that do not belong to the type, but they will typically be overwritten.</p>
5082 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5083 store i32 3, i32* %ptr <i>; yields {void}</i>
5084 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
5089 <!-- _______________________________________________________________________ -->
5091 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
5098 fence [singlethread] <ordering> <i>; yields {void}</i>
5102 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
5103 between operations.</p>
5105 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
5106 href="#ordering">ordering</a> argument which defines what
5107 <i>synchronizes-with</i> edges they add. They can only be given
5108 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
5109 <code>seq_cst</code> orderings.</p>
5112 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
5113 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
5114 <code>acquire</code> ordering semantics if and only if there exist atomic
5115 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
5116 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
5117 <var>X</var> modifies <var>M</var> (either directly or through some side effect
5118 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
5119 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
5120 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
5121 than an explicit <code>fence</code>, one (but not both) of the atomic operations
5122 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
5123 <code>acquire</code> (resp.) ordering constraint and still
5124 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
5125 <i>happens-before</i> edge.</p>
5127 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
5128 having both <code>acquire</code> and <code>release</code> semantics specified
5129 above, participates in the global program order of other <code>seq_cst</code>
5130 operations and/or fences.</p>
5132 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
5133 specifies that the fence only synchronizes with other fences in the same
5134 thread. (This is useful for interacting with signal handlers.)</p>
5138 fence acquire <i>; yields {void}</i>
5139 fence singlethread seq_cst <i>; yields {void}</i>
5144 <!-- _______________________________________________________________________ -->
5146 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
5153 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
5157 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
5158 It loads a value in memory and compares it to a given value. If they are
5159 equal, it stores a new value into the memory.</p>
5162 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
5163 address to operate on, a value to compare to the value currently be at that
5164 address, and a new value to place at that address if the compared values are
5165 equal. The type of '<var><cmp></var>' must be an integer type whose
5166 bit width is a power of two greater than or equal to eight and less than
5167 or equal to a target-specific size limit. '<var><cmp></var>' and
5168 '<var><new></var>' must have the same type, and the type of
5169 '<var><pointer></var>' must be a pointer to that type. If the
5170 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
5171 optimizer is not allowed to modify the number or order of execution
5172 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
5175 <!-- FIXME: Extend allowed types. -->
5177 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
5178 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
5180 <p>The optional "<code>singlethread</code>" argument declares that the
5181 <code>cmpxchg</code> is only atomic with respect to code (usually signal
5182 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
5183 cmpxchg is atomic with respect to all other code in the system.</p>
5185 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
5186 the size in memory of the operand.
5189 <p>The contents of memory at the location specified by the
5190 '<tt><pointer></tt>' operand is read and compared to
5191 '<tt><cmp></tt>'; if the read value is the equal,
5192 '<tt><new></tt>' is written. The original value at the location
5195 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
5196 purpose of identifying <a href="#release_sequence">release sequences</a>. A
5197 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
5198 parameter determined by dropping any <code>release</code> part of the
5199 <code>cmpxchg</code>'s ordering.</p>
5202 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
5203 optimization work on ARM.)
5205 FIXME: Is a weaker ordering constraint on failure helpful in practice?
5211 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
5212 <a href="#i_br">br</a> label %loop
5215 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
5216 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
5217 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
5218 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
5219 <a href="#i_br">br</a> i1 %success, label %done, label %loop
5227 <!-- _______________________________________________________________________ -->
5229 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
5236 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
5240 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
5243 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
5244 operation to apply, an address whose value to modify, an argument to the
5245 operation. The operation must be one of the following keywords:</p>
5260 <p>The type of '<var><value></var>' must be an integer type whose
5261 bit width is a power of two greater than or equal to eight and less than
5262 or equal to a target-specific size limit. The type of the
5263 '<code><pointer></code>' operand must be a pointer to that type.
5264 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
5265 optimizer is not allowed to modify the number or order of execution of this
5266 <code>atomicrmw</code> with other <a href="#volatile">volatile
5269 <!-- FIXME: Extend allowed types. -->
5272 <p>The contents of memory at the location specified by the
5273 '<tt><pointer></tt>' operand are atomically read, modified, and written
5274 back. The original value at the location is returned. The modification is
5275 specified by the <var>operation</var> argument:</p>
5278 <li>xchg: <code>*ptr = val</code></li>
5279 <li>add: <code>*ptr = *ptr + val</code></li>
5280 <li>sub: <code>*ptr = *ptr - val</code></li>
5281 <li>and: <code>*ptr = *ptr & val</code></li>
5282 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
5283 <li>or: <code>*ptr = *ptr | val</code></li>
5284 <li>xor: <code>*ptr = *ptr ^ val</code></li>
5285 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
5286 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
5287 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
5288 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
5293 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
5298 <!-- _______________________________________________________________________ -->
5300 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
5307 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5308 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5309 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5313 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
5314 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
5315 It performs address calculation only and does not access memory.</p>
5318 <p>The first argument is always a pointer or a vector of pointers,
5319 and forms the basis of the
5320 calculation. The remaining arguments are indices that indicate which of the
5321 elements of the aggregate object are indexed. The interpretation of each
5322 index is dependent on the type being indexed into. The first index always
5323 indexes the pointer value given as the first argument, the second index
5324 indexes a value of the type pointed to (not necessarily the value directly
5325 pointed to, since the first index can be non-zero), etc. The first type
5326 indexed into must be a pointer value, subsequent types can be arrays,
5327 vectors, and structs. Note that subsequent types being indexed into
5328 can never be pointers, since that would require loading the pointer before
5329 continuing calculation.</p>
5331 <p>The type of each index argument depends on the type it is indexing into.
5332 When indexing into a (optionally packed) structure, only <tt>i32</tt>
5333 integer <b>constants</b> are allowed. When indexing into an array, pointer
5334 or vector, integers of any width are allowed, and they are not required to be
5335 constant. These integers are treated as signed values where relevant.</p>
5337 <p>For example, let's consider a C code fragment and how it gets compiled to
5340 <pre class="doc_code">
5352 int *foo(struct ST *s) {
5353 return &s[1].Z.B[5][13];
5357 <p>The LLVM code generated by Clang is:</p>
5359 <pre class="doc_code">
5360 %struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 }
5361 %struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT }
5363 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5365 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5371 <p>In the example above, the first index is indexing into the
5372 '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a
5373 '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a
5374 structure. The second index indexes into the third element of the structure,
5375 yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>'
5376 type, another structure. The third index indexes into the second element of
5377 the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The
5378 two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>'
5379 type. The '<tt>getelementptr</tt>' instruction returns a pointer to this
5380 element, thus computing a value of '<tt>i32*</tt>' type.</p>
5382 <p>Note that it is perfectly legal to index partially through a structure,
5383 returning a pointer to an inner element. Because of this, the LLVM code for
5384 the given testcase is equivalent to:</p>
5386 <pre class="doc_code">
5387 define i32* @foo(%struct.ST* %s) {
5388 %t1 = getelementptr %struct.ST* %s, i32 1 <i>; yields %struct.ST*:%t1</i>
5389 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 <i>; yields %struct.RT*:%t2</i>
5390 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5391 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5392 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5397 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5398 <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the
5399 base pointer is not an <i>in bounds</i> address of an allocated object,
5400 or if any of the addresses that would be formed by successive addition of
5401 the offsets implied by the indices to the base address with infinitely
5402 precise signed arithmetic are not an <i>in bounds</i> address of that
5403 allocated object. The <i>in bounds</i> addresses for an allocated object
5404 are all the addresses that point into the object, plus the address one
5406 In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword
5407 applies to each of the computations element-wise. </p>
5409 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5410 the base address with silently-wrapping two's complement arithmetic. If the
5411 offsets have a different width from the pointer, they are sign-extended or
5412 truncated to the width of the pointer. The result value of the
5413 <tt>getelementptr</tt> may be outside the object pointed to by the base
5414 pointer. The result value may not necessarily be used to access memory
5415 though, even if it happens to point into allocated storage. See the
5416 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5419 <p>The getelementptr instruction is often confusing. For some more insight into
5420 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5424 <i>; yields [12 x i8]*:aptr</i>
5425 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5426 <i>; yields i8*:vptr</i>
5427 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5428 <i>; yields i8*:eptr</i>
5429 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5430 <i>; yields i32*:iptr</i>
5431 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5434 <p>In cases where the pointer argument is a vector of pointers, only a
5435 single index may be used, and the number of vector elements has to be
5436 the same. For example: </p>
5437 <pre class="doc_code">
5438 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5445 <!-- ======================================================================= -->
5447 <a name="convertops">Conversion Operations</a>
5452 <p>The instructions in this category are the conversion instructions (casting)
5453 which all take a single operand and a type. They perform various bit
5454 conversions on the operand.</p>
5456 <!-- _______________________________________________________________________ -->
5458 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5465 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5469 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5470 type <tt>ty2</tt>.</p>
5473 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5474 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5475 of the same number of integers.
5476 The bit size of the <tt>value</tt> must be larger than
5477 the bit size of the destination type, <tt>ty2</tt>.
5478 Equal sized types are not allowed.</p>
5481 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5482 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5483 source size must be larger than the destination size, <tt>trunc</tt> cannot
5484 be a <i>no-op cast</i>. It will always truncate bits.</p>
5488 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5489 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5490 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5491 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5496 <!-- _______________________________________________________________________ -->
5498 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5505 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5509 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5514 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5515 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5516 of the same number of integers.
5517 The bit size of the <tt>value</tt> must be smaller than
5518 the bit size of the destination type,
5522 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5523 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5525 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5529 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5530 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5531 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5536 <!-- _______________________________________________________________________ -->
5538 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5545 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5549 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5552 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5553 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5554 of the same number of integers.
5555 The bit size of the <tt>value</tt> must be smaller than
5556 the bit size of the destination type,
5560 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5561 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5562 of the type <tt>ty2</tt>.</p>
5564 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5568 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5569 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5570 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5575 <!-- _______________________________________________________________________ -->
5577 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5584 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5588 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5592 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5593 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5594 to cast it to. The size of <tt>value</tt> must be larger than the size of
5595 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5596 <i>no-op cast</i>.</p>
5599 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5600 <a href="#t_floating">floating point</a> type to a smaller
5601 <a href="#t_floating">floating point</a> type. If the value cannot fit
5602 within the destination type, <tt>ty2</tt>, then the results are
5607 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5608 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5613 <!-- _______________________________________________________________________ -->
5615 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5622 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5626 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5627 floating point value.</p>
5630 <p>The '<tt>fpext</tt>' instruction takes a
5631 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5632 a <a href="#t_floating">floating point</a> type to cast it to. The source
5633 type must be smaller than the destination type.</p>
5636 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5637 <a href="#t_floating">floating point</a> type to a larger
5638 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5639 used to make a <i>no-op cast</i> because it always changes bits. Use
5640 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5644 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5645 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5650 <!-- _______________________________________________________________________ -->
5652 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5659 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5663 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5664 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5667 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5668 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5669 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5670 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5671 vector integer type with the same number of elements as <tt>ty</tt></p>
5674 <p>The '<tt>fptoui</tt>' instruction converts its
5675 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5676 towards zero) unsigned integer value. If the value cannot fit
5677 in <tt>ty2</tt>, the results are undefined.</p>
5681 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5682 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5683 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5688 <!-- _______________________________________________________________________ -->
5690 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5697 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5701 <p>The '<tt>fptosi</tt>' instruction converts
5702 <a href="#t_floating">floating point</a> <tt>value</tt> to
5703 type <tt>ty2</tt>.</p>
5706 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5707 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5708 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5709 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5710 vector integer type with the same number of elements as <tt>ty</tt></p>
5713 <p>The '<tt>fptosi</tt>' instruction converts its
5714 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5715 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5716 the results are undefined.</p>
5720 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5721 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5722 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5727 <!-- _______________________________________________________________________ -->
5729 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5736 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5740 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5741 integer and converts that value to the <tt>ty2</tt> type.</p>
5744 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5745 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5746 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5747 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5748 floating point type with the same number of elements as <tt>ty</tt></p>
5751 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5752 integer quantity and converts it to the corresponding floating point
5753 value. If the value cannot fit in the floating point value, the results are
5758 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5759 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5764 <!-- _______________________________________________________________________ -->
5766 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5773 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5777 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5778 and converts that value to the <tt>ty2</tt> type.</p>
5781 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5782 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5783 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5784 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5785 floating point type with the same number of elements as <tt>ty</tt></p>
5788 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5789 quantity and converts it to the corresponding floating point value. If the
5790 value cannot fit in the floating point value, the results are undefined.</p>
5794 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5795 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5800 <!-- _______________________________________________________________________ -->
5802 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5809 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5813 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of
5814 pointers <tt>value</tt> to
5815 the integer (or vector of integers) type <tt>ty2</tt>.</p>
5818 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5819 must be a a value of type <a href="#t_pointer">pointer</a> or a vector of
5820 pointers, and a type to cast it to
5821 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector
5822 of integers type.</p>
5825 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5826 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5827 truncating or zero extending that value to the size of the integer type. If
5828 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5829 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5830 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5835 %X = ptrtoint i32* %P to i8 <i>; yields truncation on 32-bit architecture</i>
5836 %Y = ptrtoint i32* %P to i64 <i>; yields zero extension on 32-bit architecture</i>
5837 %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>
5842 <!-- _______________________________________________________________________ -->
5844 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5851 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5855 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5856 pointer type, <tt>ty2</tt>.</p>
5859 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5860 value to cast, and a type to cast it to, which must be a
5861 <a href="#t_pointer">pointer</a> type.</p>
5864 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5865 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5866 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5867 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5868 than the size of a pointer then a zero extension is done. If they are the
5869 same size, nothing is done (<i>no-op cast</i>).</p>
5873 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5874 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5875 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5876 %Z = inttoptr <4 x i32> %G to <4 x i8*><i>; yields truncation of vector G to four pointers</i>
5881 <!-- _______________________________________________________________________ -->
5883 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5890 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5894 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5895 <tt>ty2</tt> without changing any bits.</p>
5898 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5899 non-aggregate first class value, and a type to cast it to, which must also be
5900 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5901 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5902 identical. If the source type is a pointer, the destination type must also be
5903 a pointer. This instruction supports bitwise conversion of vectors to
5904 integers and to vectors of other types (as long as they have the same
5908 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5909 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5910 this conversion. The conversion is done as if the <tt>value</tt> had been
5911 stored to memory and read back as type <tt>ty2</tt>.
5912 Pointer (or vector of pointers) types may only be converted to other pointer
5913 (or vector of pointers) types with this instruction. To convert
5914 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5915 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5919 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
5920 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
5921 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
5922 %Z = bitcast <2 x i32*> %V to <2 x i64*> <i>; yields <2 x i64*></i>
5929 <!-- ======================================================================= -->
5931 <a name="otherops">Other Operations</a>
5936 <p>The instructions in this category are the "miscellaneous" instructions, which
5937 defy better classification.</p>
5939 <!-- _______________________________________________________________________ -->
5941 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5948 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5952 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5953 boolean values based on comparison of its two integer, integer vector,
5954 pointer, or pointer vector operands.</p>
5957 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
5958 the condition code indicating the kind of comparison to perform. It is not a
5959 value, just a keyword. The possible condition code are:</p>
5962 <li><tt>eq</tt>: equal</li>
5963 <li><tt>ne</tt>: not equal </li>
5964 <li><tt>ugt</tt>: unsigned greater than</li>
5965 <li><tt>uge</tt>: unsigned greater or equal</li>
5966 <li><tt>ult</tt>: unsigned less than</li>
5967 <li><tt>ule</tt>: unsigned less or equal</li>
5968 <li><tt>sgt</tt>: signed greater than</li>
5969 <li><tt>sge</tt>: signed greater or equal</li>
5970 <li><tt>slt</tt>: signed less than</li>
5971 <li><tt>sle</tt>: signed less or equal</li>
5974 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
5975 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
5976 typed. They must also be identical types.</p>
5979 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
5980 condition code given as <tt>cond</tt>. The comparison performed always yields
5981 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
5982 result, as follows:</p>
5985 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
5986 <tt>false</tt> otherwise. No sign interpretation is necessary or
5989 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
5990 <tt>false</tt> otherwise. No sign interpretation is necessary or
5993 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
5994 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5996 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
5997 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5998 to <tt>op2</tt>.</li>
6000 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
6001 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6003 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
6004 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6006 <li><tt>sgt</tt>: interprets the operands as signed values and yields
6007 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6009 <li><tt>sge</tt>: interprets the operands as signed values and yields
6010 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6011 to <tt>op2</tt>.</li>
6013 <li><tt>slt</tt>: interprets the operands as signed values and yields
6014 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6016 <li><tt>sle</tt>: interprets the operands as signed values and yields
6017 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6020 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
6021 values are compared as if they were integers.</p>
6023 <p>If the operands are integer vectors, then they are compared element by
6024 element. The result is an <tt>i1</tt> vector with the same number of elements
6025 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
6029 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
6030 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
6031 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
6032 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
6033 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
6034 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
6037 <p>Note that the code generator does not yet support vector types with
6038 the <tt>icmp</tt> instruction.</p>
6042 <!-- _______________________________________________________________________ -->
6044 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
6051 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
6055 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
6056 values based on comparison of its operands.</p>
6058 <p>If the operands are floating point scalars, then the result type is a boolean
6059 (<a href="#t_integer"><tt>i1</tt></a>).</p>
6061 <p>If the operands are floating point vectors, then the result type is a vector
6062 of boolean with the same number of elements as the operands being
6066 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
6067 the condition code indicating the kind of comparison to perform. It is not a
6068 value, just a keyword. The possible condition code are:</p>
6071 <li><tt>false</tt>: no comparison, always returns false</li>
6072 <li><tt>oeq</tt>: ordered and equal</li>
6073 <li><tt>ogt</tt>: ordered and greater than </li>
6074 <li><tt>oge</tt>: ordered and greater than or equal</li>
6075 <li><tt>olt</tt>: ordered and less than </li>
6076 <li><tt>ole</tt>: ordered and less than or equal</li>
6077 <li><tt>one</tt>: ordered and not equal</li>
6078 <li><tt>ord</tt>: ordered (no nans)</li>
6079 <li><tt>ueq</tt>: unordered or equal</li>
6080 <li><tt>ugt</tt>: unordered or greater than </li>
6081 <li><tt>uge</tt>: unordered or greater than or equal</li>
6082 <li><tt>ult</tt>: unordered or less than </li>
6083 <li><tt>ule</tt>: unordered or less than or equal</li>
6084 <li><tt>une</tt>: unordered or not equal</li>
6085 <li><tt>uno</tt>: unordered (either nans)</li>
6086 <li><tt>true</tt>: no comparison, always returns true</li>
6089 <p><i>Ordered</i> means that neither operand is a QNAN while
6090 <i>unordered</i> means that either operand may be a QNAN.</p>
6092 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
6093 a <a href="#t_floating">floating point</a> type or
6094 a <a href="#t_vector">vector</a> of floating point type. They must have
6095 identical types.</p>
6098 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
6099 according to the condition code given as <tt>cond</tt>. If the operands are
6100 vectors, then the vectors are compared element by element. Each comparison
6101 performed always yields an <a href="#t_integer">i1</a> result, as
6105 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
6107 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6108 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6110 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6111 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6113 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6114 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6116 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6117 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6119 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6120 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6122 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6123 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6125 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
6127 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
6128 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6130 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
6131 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6133 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
6134 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6136 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
6137 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6139 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
6140 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6142 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
6143 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6145 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
6147 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
6152 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
6153 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
6154 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
6155 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
6158 <p>Note that the code generator does not yet support vector types with
6159 the <tt>fcmp</tt> instruction.</p>
6163 <!-- _______________________________________________________________________ -->
6165 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
6172 <result> = phi <ty> [ <val0>, <label0>], ...
6176 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
6177 SSA graph representing the function.</p>
6180 <p>The type of the incoming values is specified with the first type field. After
6181 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
6182 one pair for each predecessor basic block of the current block. Only values
6183 of <a href="#t_firstclass">first class</a> type may be used as the value
6184 arguments to the PHI node. Only labels may be used as the label
6187 <p>There must be no non-phi instructions between the start of a basic block and
6188 the PHI instructions: i.e. PHI instructions must be first in a basic
6191 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
6192 occur on the edge from the corresponding predecessor block to the current
6193 block (but after any definition of an '<tt>invoke</tt>' instruction's return
6194 value on the same edge).</p>
6197 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
6198 specified by the pair corresponding to the predecessor basic block that
6199 executed just prior to the current block.</p>
6203 Loop: ; Infinite loop that counts from 0 on up...
6204 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6205 %nextindvar = add i32 %indvar, 1
6211 <!-- _______________________________________________________________________ -->
6213 <a name="i_select">'<tt>select</tt>' Instruction</a>
6220 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
6222 <i>selty</i> is either i1 or {<N x i1>}
6226 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
6227 condition, without branching.</p>
6231 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
6232 values indicating the condition, and two values of the
6233 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
6234 vectors and the condition is a scalar, then entire vectors are selected, not
6235 individual elements.</p>
6238 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
6239 first value argument; otherwise, it returns the second value argument.</p>
6241 <p>If the condition is a vector of i1, then the value arguments must be vectors
6242 of the same size, and the selection is done element by element.</p>
6246 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
6251 <!-- _______________________________________________________________________ -->
6253 <a name="i_call">'<tt>call</tt>' Instruction</a>
6260 <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>]
6264 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
6267 <p>This instruction requires several arguments:</p>
6270 <li>The optional "tail" marker indicates that the callee function does not
6271 access any allocas or varargs in the caller. Note that calls may be
6272 marked "tail" even if they do not occur before
6273 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
6274 present, the function call is eligible for tail call optimization,
6275 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
6276 optimized into a jump</a>. The code generator may optimize calls marked
6277 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
6278 sibling call optimization</a> when the caller and callee have
6279 matching signatures, or 2) forced tail call optimization when the
6280 following extra requirements are met:
6282 <li>Caller and callee both have the calling
6283 convention <tt>fastcc</tt>.</li>
6284 <li>The call is in tail position (ret immediately follows call and ret
6285 uses value of call or is void).</li>
6286 <li>Option <tt>-tailcallopt</tt> is enabled,
6287 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
6288 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
6289 constraints are met.</a></li>
6293 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
6294 convention</a> the call should use. If none is specified, the call
6295 defaults to using C calling conventions. The calling convention of the
6296 call must match the calling convention of the target function, or else the
6297 behavior is undefined.</li>
6299 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
6300 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
6301 '<tt>inreg</tt>' attributes are valid here.</li>
6303 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
6304 type of the return value. Functions that return no value are marked
6305 <tt><a href="#t_void">void</a></tt>.</li>
6307 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
6308 being invoked. The argument types must match the types implied by this
6309 signature. This type can be omitted if the function is not varargs and if
6310 the function type does not return a pointer to a function.</li>
6312 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
6313 be invoked. In most cases, this is a direct function invocation, but
6314 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
6315 to function value.</li>
6317 <li>'<tt>function args</tt>': argument list whose types match the function
6318 signature argument types and parameter attributes. All arguments must be
6319 of <a href="#t_firstclass">first class</a> type. If the function
6320 signature indicates the function accepts a variable number of arguments,
6321 the extra arguments can be specified.</li>
6323 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
6324 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
6325 '<tt>readnone</tt>' attributes are valid here.</li>
6329 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
6330 a specified function, with its incoming arguments bound to the specified
6331 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
6332 function, control flow continues with the instruction after the function
6333 call, and the return value of the function is bound to the result
6338 %retval = call i32 @test(i32 %argc)
6339 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
6340 %X = tail call i32 @foo() <i>; yields i32</i>
6341 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
6342 call void %foo(i8 97 signext)
6344 %struct.A = type { i32, i8 }
6345 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
6346 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
6347 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
6348 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
6349 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
6352 <p>llvm treats calls to some functions with names and arguments that match the
6353 standard C99 library as being the C99 library functions, and may perform
6354 optimizations or generate code for them under that assumption. This is
6355 something we'd like to change in the future to provide better support for
6356 freestanding environments and non-C-based languages.</p>
6360 <!-- _______________________________________________________________________ -->
6362 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
6369 <resultval> = va_arg <va_list*> <arglist>, <argty>
6373 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6374 the "variable argument" area of a function call. It is used to implement the
6375 <tt>va_arg</tt> macro in C.</p>
6378 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6379 argument. It returns a value of the specified argument type and increments
6380 the <tt>va_list</tt> to point to the next argument. The actual type
6381 of <tt>va_list</tt> is target specific.</p>
6384 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6385 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6386 to the next argument. For more information, see the variable argument
6387 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6389 <p>It is legal for this instruction to be called in a function which does not
6390 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6393 <p><tt>va_arg</tt> is an LLVM instruction instead of
6394 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6398 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6400 <p>Note that the code generator does not yet fully support va_arg on many
6401 targets. Also, it does not currently support va_arg with aggregate types on
6406 <!-- _______________________________________________________________________ -->
6408 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6415 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6416 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6418 <clause> := catch <type> <value>
6419 <clause> := filter <array constant type> <array constant>
6423 <p>The '<tt>landingpad</tt>' instruction is used by
6424 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6425 system</a> to specify that a basic block is a landing pad — one where
6426 the exception lands, and corresponds to the code found in the
6427 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6428 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6429 re-entry to the function. The <tt>resultval</tt> has the
6430 type <tt>resultty</tt>.</p>
6433 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6434 function associated with the unwinding mechanism. The optional
6435 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6437 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6438 or <tt>filter</tt> — and contains the global variable representing the
6439 "type" that may be caught or filtered respectively. Unlike the
6440 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6441 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6442 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6443 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6446 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6447 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6448 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6449 calling conventions, how the personality function results are represented in
6450 LLVM IR is target specific.</p>
6452 <p>The clauses are applied in order from top to bottom. If two
6453 <tt>landingpad</tt> instructions are merged together through inlining, the
6454 clauses from the calling function are appended to the list of clauses.
6455 When the call stack is being unwound due to an exception being thrown, the
6456 exception is compared against each <tt>clause</tt> in turn. If it doesn't
6457 match any of the clauses, and the <tt>cleanup</tt> flag is not set, then
6458 unwinding continues further up the call stack.</p>
6460 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6463 <li>A landing pad block is a basic block which is the unwind destination of an
6464 '<tt>invoke</tt>' instruction.</li>
6465 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6466 first non-PHI instruction.</li>
6467 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6469 <li>A basic block that is not a landing pad block may not include a
6470 '<tt>landingpad</tt>' instruction.</li>
6471 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6472 personality function.</li>
6477 ;; A landing pad which can catch an integer.
6478 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6480 ;; A landing pad that is a cleanup.
6481 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6483 ;; A landing pad which can catch an integer and can only throw a double.
6484 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6486 filter [1 x i8**] [@_ZTId]
6495 <!-- *********************************************************************** -->
6496 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6497 <!-- *********************************************************************** -->
6501 <p>LLVM supports the notion of an "intrinsic function". These functions have
6502 well known names and semantics and are required to follow certain
6503 restrictions. Overall, these intrinsics represent an extension mechanism for
6504 the LLVM language that does not require changing all of the transformations
6505 in LLVM when adding to the language (or the bitcode reader/writer, the
6506 parser, etc...).</p>
6508 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6509 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6510 begin with this prefix. Intrinsic functions must always be external
6511 functions: you cannot define the body of intrinsic functions. Intrinsic
6512 functions may only be used in call or invoke instructions: it is illegal to
6513 take the address of an intrinsic function. Additionally, because intrinsic
6514 functions are part of the LLVM language, it is required if any are added that
6515 they be documented here.</p>
6517 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6518 family of functions that perform the same operation but on different data
6519 types. Because LLVM can represent over 8 million different integer types,
6520 overloading is used commonly to allow an intrinsic function to operate on any
6521 integer type. One or more of the argument types or the result type can be
6522 overloaded to accept any integer type. Argument types may also be defined as
6523 exactly matching a previous argument's type or the result type. This allows
6524 an intrinsic function which accepts multiple arguments, but needs all of them
6525 to be of the same type, to only be overloaded with respect to a single
6526 argument or the result.</p>
6528 <p>Overloaded intrinsics will have the names of its overloaded argument types
6529 encoded into its function name, each preceded by a period. Only those types
6530 which are overloaded result in a name suffix. Arguments whose type is matched
6531 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6532 can take an integer of any width and returns an integer of exactly the same
6533 integer width. This leads to a family of functions such as
6534 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6535 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6536 suffix is required. Because the argument's type is matched against the return
6537 type, it does not require its own name suffix.</p>
6539 <p>To learn how to add an intrinsic function, please see the
6540 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6542 <!-- ======================================================================= -->
6544 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6549 <p>Variable argument support is defined in LLVM with
6550 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6551 intrinsic functions. These functions are related to the similarly named
6552 macros defined in the <tt><stdarg.h></tt> header file.</p>
6554 <p>All of these functions operate on arguments that use a target-specific value
6555 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6556 not define what this type is, so all transformations should be prepared to
6557 handle these functions regardless of the type used.</p>
6559 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6560 instruction and the variable argument handling intrinsic functions are
6563 <pre class="doc_code">
6564 define i32 @test(i32 %X, ...) {
6565 ; Initialize variable argument processing
6567 %ap2 = bitcast i8** %ap to i8*
6568 call void @llvm.va_start(i8* %ap2)
6570 ; Read a single integer argument
6571 %tmp = va_arg i8** %ap, i32
6573 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6575 %aq2 = bitcast i8** %aq to i8*
6576 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6577 call void @llvm.va_end(i8* %aq2)
6579 ; Stop processing of arguments.
6580 call void @llvm.va_end(i8* %ap2)
6584 declare void @llvm.va_start(i8*)
6585 declare void @llvm.va_copy(i8*, i8*)
6586 declare void @llvm.va_end(i8*)
6589 <!-- _______________________________________________________________________ -->
6591 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6599 declare void %llvm.va_start(i8* <arglist>)
6603 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6604 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6607 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6610 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6611 macro available in C. In a target-dependent way, it initializes
6612 the <tt>va_list</tt> element to which the argument points, so that the next
6613 call to <tt>va_arg</tt> will produce the first variable argument passed to
6614 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6615 need to know the last argument of the function as the compiler can figure
6620 <!-- _______________________________________________________________________ -->
6622 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6629 declare void @llvm.va_end(i8* <arglist>)
6633 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6634 which has been initialized previously
6635 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6636 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6639 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6642 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6643 macro available in C. In a target-dependent way, it destroys
6644 the <tt>va_list</tt> element to which the argument points. Calls
6645 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6646 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6647 with calls to <tt>llvm.va_end</tt>.</p>
6651 <!-- _______________________________________________________________________ -->
6653 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6660 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6664 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6665 from the source argument list to the destination argument list.</p>
6668 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6669 The second argument is a pointer to a <tt>va_list</tt> element to copy
6673 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6674 macro available in C. In a target-dependent way, it copies the
6675 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6676 element. This intrinsic is necessary because
6677 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6678 arbitrarily complex and require, for example, memory allocation.</p>
6684 <!-- ======================================================================= -->
6686 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6691 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6692 Collection</a> (GC) requires the implementation and generation of these
6693 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6694 roots on the stack</a>, as well as garbage collector implementations that
6695 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6696 barriers. Front-ends for type-safe garbage collected languages should generate
6697 these intrinsics to make use of the LLVM garbage collectors. For more details,
6698 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6701 <p>The garbage collection intrinsics only operate on objects in the generic
6702 address space (address space zero).</p>
6704 <!-- _______________________________________________________________________ -->
6706 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6713 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6717 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6718 the code generator, and allows some metadata to be associated with it.</p>
6721 <p>The first argument specifies the address of a stack object that contains the
6722 root pointer. The second pointer (which must be either a constant or a
6723 global value address) contains the meta-data to be associated with the
6727 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6728 location. At compile-time, the code generator generates information to allow
6729 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6730 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6735 <!-- _______________________________________________________________________ -->
6737 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6744 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6748 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6749 locations, allowing garbage collector implementations that require read
6753 <p>The second argument is the address to read from, which should be an address
6754 allocated from the garbage collector. The first object is a pointer to the
6755 start of the referenced object, if needed by the language runtime (otherwise
6759 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6760 instruction, but may be replaced with substantially more complex code by the
6761 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6762 may only be used in a function which <a href="#gc">specifies a GC
6767 <!-- _______________________________________________________________________ -->
6769 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6776 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6780 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6781 locations, allowing garbage collector implementations that require write
6782 barriers (such as generational or reference counting collectors).</p>
6785 <p>The first argument is the reference to store, the second is the start of the
6786 object to store it to, and the third is the address of the field of Obj to
6787 store to. If the runtime does not require a pointer to the object, Obj may
6791 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6792 instruction, but may be replaced with substantially more complex code by the
6793 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6794 may only be used in a function which <a href="#gc">specifies a GC
6801 <!-- ======================================================================= -->
6803 <a name="int_codegen">Code Generator Intrinsics</a>
6808 <p>These intrinsics are provided by LLVM to expose special features that may
6809 only be implemented with code generator support.</p>
6811 <!-- _______________________________________________________________________ -->
6813 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6820 declare i8 *@llvm.returnaddress(i32 <level>)
6824 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6825 target-specific value indicating the return address of the current function
6826 or one of its callers.</p>
6829 <p>The argument to this intrinsic indicates which function to return the address
6830 for. Zero indicates the calling function, one indicates its caller, etc.
6831 The argument is <b>required</b> to be a constant integer value.</p>
6834 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6835 indicating the return address of the specified call frame, or zero if it
6836 cannot be identified. The value returned by this intrinsic is likely to be
6837 incorrect or 0 for arguments other than zero, so it should only be used for
6838 debugging purposes.</p>
6840 <p>Note that calling this intrinsic does not prevent function inlining or other
6841 aggressive transformations, so the value returned may not be that of the
6842 obvious source-language caller.</p>
6846 <!-- _______________________________________________________________________ -->
6848 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6855 declare i8* @llvm.frameaddress(i32 <level>)
6859 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6860 target-specific frame pointer value for the specified stack frame.</p>
6863 <p>The argument to this intrinsic indicates which function to return the frame
6864 pointer for. Zero indicates the calling function, one indicates its caller,
6865 etc. The argument is <b>required</b> to be a constant integer value.</p>
6868 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6869 indicating the frame address of the specified call frame, or zero if it
6870 cannot be identified. The value returned by this intrinsic is likely to be
6871 incorrect or 0 for arguments other than zero, so it should only be used for
6872 debugging purposes.</p>
6874 <p>Note that calling this intrinsic does not prevent function inlining or other
6875 aggressive transformations, so the value returned may not be that of the
6876 obvious source-language caller.</p>
6880 <!-- _______________________________________________________________________ -->
6882 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6889 declare i8* @llvm.stacksave()
6893 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6894 of the function stack, for use
6895 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6896 useful for implementing language features like scoped automatic variable
6897 sized arrays in C99.</p>
6900 <p>This intrinsic returns a opaque pointer value that can be passed
6901 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6902 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6903 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6904 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6905 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6906 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6910 <!-- _______________________________________________________________________ -->
6912 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6919 declare void @llvm.stackrestore(i8* %ptr)
6923 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6924 the function stack to the state it was in when the
6925 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6926 executed. This is useful for implementing language features like scoped
6927 automatic variable sized arrays in C99.</p>
6930 <p>See the description
6931 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6935 <!-- _______________________________________________________________________ -->
6937 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6944 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6948 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6949 insert a prefetch instruction if supported; otherwise, it is a noop.
6950 Prefetches have no effect on the behavior of the program but can change its
6951 performance characteristics.</p>
6954 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6955 specifier determining if the fetch should be for a read (0) or write (1),
6956 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
6957 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6958 specifies whether the prefetch is performed on the data (1) or instruction (0)
6959 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
6960 must be constant integers.</p>
6963 <p>This intrinsic does not modify the behavior of the program. In particular,
6964 prefetches cannot trap and do not produce a value. On targets that support
6965 this intrinsic, the prefetch can provide hints to the processor cache for
6966 better performance.</p>
6970 <!-- _______________________________________________________________________ -->
6972 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
6979 declare void @llvm.pcmarker(i32 <id>)
6983 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
6984 Counter (PC) in a region of code to simulators and other tools. The method
6985 is target specific, but it is expected that the marker will use exported
6986 symbols to transmit the PC of the marker. The marker makes no guarantees
6987 that it will remain with any specific instruction after optimizations. It is
6988 possible that the presence of a marker will inhibit optimizations. The
6989 intended use is to be inserted after optimizations to allow correlations of
6990 simulation runs.</p>
6993 <p><tt>id</tt> is a numerical id identifying the marker.</p>
6996 <p>This intrinsic does not modify the behavior of the program. Backends that do
6997 not support this intrinsic may ignore it.</p>
7001 <!-- _______________________________________________________________________ -->
7003 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
7010 declare i64 @llvm.readcyclecounter()
7014 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
7015 counter register (or similar low latency, high accuracy clocks) on those
7016 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7017 should map to RPCC. As the backing counters overflow quickly (on the order
7018 of 9 seconds on alpha), this should only be used for small timings.</p>
7021 <p>When directly supported, reading the cycle counter should not modify any
7022 memory. Implementations are allowed to either return a application specific
7023 value or a system wide value. On backends without support, this is lowered
7024 to a constant 0.</p>
7030 <!-- ======================================================================= -->
7032 <a name="int_libc">Standard C Library Intrinsics</a>
7037 <p>LLVM provides intrinsics for a few important standard C library functions.
7038 These intrinsics allow source-language front-ends to pass information about
7039 the alignment of the pointer arguments to the code generator, providing
7040 opportunity for more efficient code generation.</p>
7042 <!-- _______________________________________________________________________ -->
7044 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
7050 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
7051 integer bit width and for different address spaces. Not all targets support
7052 all bit widths however.</p>
7055 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7056 i32 <len>, i32 <align>, i1 <isvolatile>)
7057 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7058 i64 <len>, i32 <align>, i1 <isvolatile>)
7062 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7063 source location to the destination location.</p>
7065 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
7066 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7067 and the pointers can be in specified address spaces.</p>
7071 <p>The first argument is a pointer to the destination, the second is a pointer
7072 to the source. The third argument is an integer argument specifying the
7073 number of bytes to copy, the fourth argument is the alignment of the
7074 source and destination locations, and the fifth is a boolean indicating a
7075 volatile access.</p>
7077 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7078 then the caller guarantees that both the source and destination pointers are
7079 aligned to that boundary.</p>
7081 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7082 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
7083 The detailed access behavior is not very cleanly specified and it is unwise
7084 to depend on it.</p>
7088 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7089 source location to the destination location, which are not allowed to
7090 overlap. It copies "len" bytes of memory over. If the argument is known to
7091 be aligned to some boundary, this can be specified as the fourth argument,
7092 otherwise it should be set to 0 or 1.</p>
7096 <!-- _______________________________________________________________________ -->
7098 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
7104 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
7105 width and for different address space. Not all targets support all bit
7109 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7110 i32 <len>, i32 <align>, i1 <isvolatile>)
7111 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7112 i64 <len>, i32 <align>, i1 <isvolatile>)
7116 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
7117 source location to the destination location. It is similar to the
7118 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
7121 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
7122 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7123 and the pointers can be in specified address spaces.</p>
7127 <p>The first argument is a pointer to the destination, the second is a pointer
7128 to the source. The third argument is an integer argument specifying the
7129 number of bytes to copy, the fourth argument is the alignment of the
7130 source and destination locations, and the fifth is a boolean indicating a
7131 volatile access.</p>
7133 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7134 then the caller guarantees that the source and destination pointers are
7135 aligned to that boundary.</p>
7137 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7138 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
7139 The detailed access behavior is not very cleanly specified and it is unwise
7140 to depend on it.</p>
7144 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
7145 source location to the destination location, which may overlap. It copies
7146 "len" bytes of memory over. If the argument is known to be aligned to some
7147 boundary, this can be specified as the fourth argument, otherwise it should
7148 be set to 0 or 1.</p>
7152 <!-- _______________________________________________________________________ -->
7154 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
7160 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
7161 width and for different address spaces. However, not all targets support all
7165 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7166 i32 <len>, i32 <align>, i1 <isvolatile>)
7167 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7168 i64 <len>, i32 <align>, i1 <isvolatile>)
7172 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
7173 particular byte value.</p>
7175 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
7176 intrinsic does not return a value and takes extra alignment/volatile
7177 arguments. Also, the destination can be in an arbitrary address space.</p>
7180 <p>The first argument is a pointer to the destination to fill, the second is the
7181 byte value with which to fill it, the third argument is an integer argument
7182 specifying the number of bytes to fill, and the fourth argument is the known
7183 alignment of the destination location.</p>
7185 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7186 then the caller guarantees that the destination pointer is aligned to that
7189 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7190 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
7191 The detailed access behavior is not very cleanly specified and it is unwise
7192 to depend on it.</p>
7195 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
7196 at the destination location. If the argument is known to be aligned to some
7197 boundary, this can be specified as the fourth argument, otherwise it should
7198 be set to 0 or 1.</p>
7202 <!-- _______________________________________________________________________ -->
7204 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
7210 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
7211 floating point or vector of floating point type. Not all targets support all
7215 declare float @llvm.sqrt.f32(float %Val)
7216 declare double @llvm.sqrt.f64(double %Val)
7217 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7218 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7219 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7223 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
7224 returning the same value as the libm '<tt>sqrt</tt>' functions would.
7225 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
7226 behavior for negative numbers other than -0.0 (which allows for better
7227 optimization, because there is no need to worry about errno being
7228 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
7231 <p>The argument and return value are floating point numbers of the same
7235 <p>This function returns the sqrt of the specified operand if it is a
7236 nonnegative floating point number.</p>
7240 <!-- _______________________________________________________________________ -->
7242 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
7248 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
7249 floating point or vector of floating point type. Not all targets support all
7253 declare float @llvm.powi.f32(float %Val, i32 %power)
7254 declare double @llvm.powi.f64(double %Val, i32 %power)
7255 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7256 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7257 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7261 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
7262 specified (positive or negative) power. The order of evaluation of
7263 multiplications is not defined. When a vector of floating point type is
7264 used, the second argument remains a scalar integer value.</p>
7267 <p>The second argument is an integer power, and the first is a value to raise to
7271 <p>This function returns the first value raised to the second power with an
7272 unspecified sequence of rounding operations.</p>
7276 <!-- _______________________________________________________________________ -->
7278 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
7284 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
7285 floating point or vector of floating point type. Not all targets support all
7289 declare float @llvm.sin.f32(float %Val)
7290 declare double @llvm.sin.f64(double %Val)
7291 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7292 declare fp128 @llvm.sin.f128(fp128 %Val)
7293 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7297 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
7300 <p>The argument and return value are floating point numbers of the same
7304 <p>This function returns the sine of the specified operand, returning the same
7305 values as the libm <tt>sin</tt> functions would, and handles error conditions
7306 in the same way.</p>
7310 <!-- _______________________________________________________________________ -->
7312 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
7318 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
7319 floating point or vector of floating point type. Not all targets support all
7323 declare float @llvm.cos.f32(float %Val)
7324 declare double @llvm.cos.f64(double %Val)
7325 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7326 declare fp128 @llvm.cos.f128(fp128 %Val)
7327 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7331 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
7334 <p>The argument and return value are floating point numbers of the same
7338 <p>This function returns the cosine of the specified operand, returning the same
7339 values as the libm <tt>cos</tt> functions would, and handles error conditions
7340 in the same way.</p>
7344 <!-- _______________________________________________________________________ -->
7346 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
7352 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
7353 floating point or vector of floating point type. Not all targets support all
7357 declare float @llvm.pow.f32(float %Val, float %Power)
7358 declare double @llvm.pow.f64(double %Val, double %Power)
7359 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7360 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7361 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7365 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
7366 specified (positive or negative) power.</p>
7369 <p>The second argument is a floating point power, and the first is a value to
7370 raise to that power.</p>
7373 <p>This function returns the first value raised to the second power, returning
7374 the same values as the libm <tt>pow</tt> functions would, and handles error
7375 conditions in the same way.</p>
7379 <!-- _______________________________________________________________________ -->
7381 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7387 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7388 floating point or vector of floating point type. Not all targets support all
7392 declare float @llvm.exp.f32(float %Val)
7393 declare double @llvm.exp.f64(double %Val)
7394 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7395 declare fp128 @llvm.exp.f128(fp128 %Val)
7396 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7400 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7403 <p>The argument and return value are floating point numbers of the same
7407 <p>This function returns the same values as the libm <tt>exp</tt> functions
7408 would, and handles error conditions in the same way.</p>
7412 <!-- _______________________________________________________________________ -->
7414 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7420 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7421 floating point or vector of floating point type. Not all targets support all
7425 declare float @llvm.log.f32(float %Val)
7426 declare double @llvm.log.f64(double %Val)
7427 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7428 declare fp128 @llvm.log.f128(fp128 %Val)
7429 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7433 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7436 <p>The argument and return value are floating point numbers of the same
7440 <p>This function returns the same values as the libm <tt>log</tt> functions
7441 would, and handles error conditions in the same way.</p>
7445 <!-- _______________________________________________________________________ -->
7447 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7453 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7454 floating point or vector of floating point type. Not all targets support all
7458 declare float @llvm.fma.f32(float %a, float %b, float %c)
7459 declare double @llvm.fma.f64(double %a, double %b, double %c)
7460 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7461 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7462 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7466 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7470 <p>The argument and return value are floating point numbers of the same
7474 <p>This function returns the same values as the libm <tt>fma</tt> functions
7481 <!-- ======================================================================= -->
7483 <a name="int_manip">Bit Manipulation Intrinsics</a>
7488 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7489 These allow efficient code generation for some algorithms.</p>
7491 <!-- _______________________________________________________________________ -->
7493 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7499 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7500 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7503 declare i16 @llvm.bswap.i16(i16 <id>)
7504 declare i32 @llvm.bswap.i32(i32 <id>)
7505 declare i64 @llvm.bswap.i64(i64 <id>)
7509 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7510 values with an even number of bytes (positive multiple of 16 bits). These
7511 are useful for performing operations on data that is not in the target's
7512 native byte order.</p>
7515 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7516 and low byte of the input i16 swapped. Similarly,
7517 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7518 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7519 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7520 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7521 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7522 more, respectively).</p>
7526 <!-- _______________________________________________________________________ -->
7528 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7534 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7535 width, or on any vector with integer elements. Not all targets support all
7536 bit widths or vector types, however.</p>
7539 declare i8 @llvm.ctpop.i8(i8 <src>)
7540 declare i16 @llvm.ctpop.i16(i16 <src>)
7541 declare i32 @llvm.ctpop.i32(i32 <src>)
7542 declare i64 @llvm.ctpop.i64(i64 <src>)
7543 declare i256 @llvm.ctpop.i256(i256 <src>)
7544 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7548 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7552 <p>The only argument is the value to be counted. The argument may be of any
7553 integer type, or a vector with integer elements.
7554 The return type must match the argument type.</p>
7557 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7558 element of a vector.</p>
7562 <!-- _______________________________________________________________________ -->
7564 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7570 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7571 integer bit width, or any vector whose elements are integers. Not all
7572 targets support all bit widths or vector types, however.</p>
7575 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7576 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7577 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7578 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7579 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7580 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7584 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7585 leading zeros in a variable.</p>
7588 <p>The first argument is the value to be counted. This argument may be of any
7589 integer type, or a vectory with integer element type. The return type
7590 must match the first argument type.</p>
7592 <p>The second argument must be a constant and is a flag to indicate whether the
7593 intrinsic should ensure that a zero as the first argument produces a defined
7594 result. Historically some architectures did not provide a defined result for
7595 zero values as efficiently, and many algorithms are now predicated on
7596 avoiding zero-value inputs.</p>
7599 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7600 zeros in a variable, or within each element of the vector.
7601 If <tt>src == 0</tt> then the result is the size in bits of the type of
7602 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7603 For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7607 <!-- _______________________________________________________________________ -->
7609 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7615 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7616 integer bit width, or any vector of integer elements. Not all targets
7617 support all bit widths or vector types, however.</p>
7620 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7621 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7622 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7623 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7624 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7625 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7629 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7633 <p>The first argument is the value to be counted. This argument may be of any
7634 integer type, or a vectory with integer element type. The return type
7635 must match the first argument type.</p>
7637 <p>The second argument must be a constant and is a flag to indicate whether the
7638 intrinsic should ensure that a zero as the first argument produces a defined
7639 result. Historically some architectures did not provide a defined result for
7640 zero values as efficiently, and many algorithms are now predicated on
7641 avoiding zero-value inputs.</p>
7644 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7645 zeros in a variable, or within each element of a vector.
7646 If <tt>src == 0</tt> then the result is the size in bits of the type of
7647 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7648 For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7654 <!-- ======================================================================= -->
7656 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7661 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7663 <!-- _______________________________________________________________________ -->
7665 <a name="int_sadd_overflow">
7666 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7673 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7674 on any integer bit width.</p>
7677 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7678 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7679 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7683 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7684 a signed addition of the two arguments, and indicate whether an overflow
7685 occurred during the signed summation.</p>
7688 <p>The arguments (%a and %b) and the first element of the result structure may
7689 be of integer types of any bit width, but they must have the same bit
7690 width. The second element of the result structure must be of
7691 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7692 undergo signed addition.</p>
7695 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7696 a signed addition of the two variables. They return a structure — the
7697 first element of which is the signed summation, and the second element of
7698 which is a bit specifying if the signed summation resulted in an
7703 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7704 %sum = extractvalue {i32, i1} %res, 0
7705 %obit = extractvalue {i32, i1} %res, 1
7706 br i1 %obit, label %overflow, label %normal
7711 <!-- _______________________________________________________________________ -->
7713 <a name="int_uadd_overflow">
7714 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7721 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7722 on any integer bit width.</p>
7725 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7726 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7727 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7731 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7732 an unsigned addition of the two arguments, and indicate whether a carry
7733 occurred during the unsigned summation.</p>
7736 <p>The arguments (%a and %b) and the first element of the result structure may
7737 be of integer types of any bit width, but they must have the same bit
7738 width. The second element of the result structure must be of
7739 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7740 undergo unsigned addition.</p>
7743 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7744 an unsigned addition of the two arguments. They return a structure —
7745 the first element of which is the sum, and the second element of which is a
7746 bit specifying if the unsigned summation resulted in a carry.</p>
7750 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7751 %sum = extractvalue {i32, i1} %res, 0
7752 %obit = extractvalue {i32, i1} %res, 1
7753 br i1 %obit, label %carry, label %normal
7758 <!-- _______________________________________________________________________ -->
7760 <a name="int_ssub_overflow">
7761 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7768 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7769 on any integer bit width.</p>
7772 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7773 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7774 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7778 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7779 a signed subtraction of the two arguments, and indicate whether an overflow
7780 occurred during the signed subtraction.</p>
7783 <p>The arguments (%a and %b) and the first element of the result structure may
7784 be of integer types of any bit width, but they must have the same bit
7785 width. The second element of the result structure must be of
7786 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7787 undergo signed subtraction.</p>
7790 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7791 a signed subtraction of the two arguments. They return a structure —
7792 the first element of which is the subtraction, and the second element of
7793 which is a bit specifying if the signed subtraction resulted in an
7798 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7799 %sum = extractvalue {i32, i1} %res, 0
7800 %obit = extractvalue {i32, i1} %res, 1
7801 br i1 %obit, label %overflow, label %normal
7806 <!-- _______________________________________________________________________ -->
7808 <a name="int_usub_overflow">
7809 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7816 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7817 on any integer bit width.</p>
7820 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7821 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7822 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7826 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7827 an unsigned subtraction of the two arguments, and indicate whether an
7828 overflow occurred during the unsigned subtraction.</p>
7831 <p>The arguments (%a and %b) and the first element of the result structure may
7832 be of integer types of any bit width, but they must have the same bit
7833 width. The second element of the result structure must be of
7834 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7835 undergo unsigned subtraction.</p>
7838 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7839 an unsigned subtraction of the two arguments. They return a structure —
7840 the first element of which is the subtraction, and the second element of
7841 which is a bit specifying if the unsigned subtraction resulted in an
7846 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7847 %sum = extractvalue {i32, i1} %res, 0
7848 %obit = extractvalue {i32, i1} %res, 1
7849 br i1 %obit, label %overflow, label %normal
7854 <!-- _______________________________________________________________________ -->
7856 <a name="int_smul_overflow">
7857 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7864 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7865 on any integer bit width.</p>
7868 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7869 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7870 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7875 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7876 a signed multiplication of the two arguments, and indicate whether an
7877 overflow occurred during the signed multiplication.</p>
7880 <p>The arguments (%a and %b) and the first element of the result structure may
7881 be of integer types of any bit width, but they must have the same bit
7882 width. The second element of the result structure must be of
7883 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7884 undergo signed multiplication.</p>
7887 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7888 a signed multiplication of the two arguments. They return a structure —
7889 the first element of which is the multiplication, and the second element of
7890 which is a bit specifying if the signed multiplication resulted in an
7895 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7896 %sum = extractvalue {i32, i1} %res, 0
7897 %obit = extractvalue {i32, i1} %res, 1
7898 br i1 %obit, label %overflow, label %normal
7903 <!-- _______________________________________________________________________ -->
7905 <a name="int_umul_overflow">
7906 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
7913 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
7914 on any integer bit width.</p>
7917 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7918 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7919 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7923 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7924 a unsigned multiplication of the two arguments, and indicate whether an
7925 overflow occurred during the unsigned multiplication.</p>
7928 <p>The arguments (%a and %b) and the first element of the result structure may
7929 be of integer types of any bit width, but they must have the same bit
7930 width. The second element of the result structure must be of
7931 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7932 undergo unsigned multiplication.</p>
7935 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7936 an unsigned multiplication of the two arguments. They return a structure
7937 — the first element of which is the multiplication, and the second
7938 element of which is a bit specifying if the unsigned multiplication resulted
7943 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7944 %sum = extractvalue {i32, i1} %res, 0
7945 %obit = extractvalue {i32, i1} %res, 1
7946 br i1 %obit, label %overflow, label %normal
7953 <!-- ======================================================================= -->
7955 <a name="spec_arithmetic">Specialised Arithmetic Intrinsics</a>
7958 <!-- _______________________________________________________________________ -->
7961 <a name="fmuladd">'<tt>llvm.fmuladd.*</tt>' Intrinsic</a>
7968 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
7969 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
7973 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsic functions represent multiply-add
7974 expressions that can be fused if the code generator determines that the fused
7975 expression would be legal and efficient.</p>
7978 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsics each take three arguments: two
7979 multiplicands, a and b, and an addend c.</p>
7982 <p>The expression:</p>
7984 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
7986 <p>is equivalent to the expression a * b + c, except that rounding will not be
7987 performed between the multiplication and addition steps if the code generator
7988 fuses the operations. Fusion is not guaranteed, even if the target platform
7989 supports it. If a fused multiply-add is required the corresponding llvm.fma.*
7990 intrinsic function should be used instead.</p>
7994 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
7999 <!-- ======================================================================= -->
8001 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
8006 <p>For most target platforms, half precision floating point is a storage-only
8007 format. This means that it is
8008 a dense encoding (in memory) but does not support computation in the
8011 <p>This means that code must first load the half-precision floating point
8012 value as an i16, then convert it to float with <a
8013 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
8014 Computation can then be performed on the float value (including extending to
8015 double etc). To store the value back to memory, it is first converted to
8016 float if needed, then converted to i16 with
8017 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
8018 storing as an i16 value.</p>
8020 <!-- _______________________________________________________________________ -->
8022 <a name="int_convert_to_fp16">
8023 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
8031 declare i16 @llvm.convert.to.fp16(f32 %a)
8035 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8036 a conversion from single precision floating point format to half precision
8037 floating point format.</p>
8040 <p>The intrinsic function contains single argument - the value to be
8044 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8045 a conversion from single precision floating point format to half precision
8046 floating point format. The return value is an <tt>i16</tt> which
8047 contains the converted number.</p>
8051 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8052 store i16 %res, i16* @x, align 2
8057 <!-- _______________________________________________________________________ -->
8059 <a name="int_convert_from_fp16">
8060 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
8068 declare f32 @llvm.convert.from.fp16(i16 %a)
8072 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
8073 a conversion from half precision floating point format to single precision
8074 floating point format.</p>
8077 <p>The intrinsic function contains single argument - the value to be
8081 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
8082 conversion from half single precision floating point format to single
8083 precision floating point format. The input half-float value is represented by
8084 an <tt>i16</tt> value.</p>
8088 %a = load i16* @x, align 2
8089 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8096 <!-- ======================================================================= -->
8098 <a name="int_debugger">Debugger Intrinsics</a>
8103 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
8104 prefix), are described in
8105 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
8106 Level Debugging</a> document.</p>
8110 <!-- ======================================================================= -->
8112 <a name="int_eh">Exception Handling Intrinsics</a>
8117 <p>The LLVM exception handling intrinsics (which all start with
8118 <tt>llvm.eh.</tt> prefix), are described in
8119 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
8120 Handling</a> document.</p>
8124 <!-- ======================================================================= -->
8126 <a name="int_trampoline">Trampoline Intrinsics</a>
8131 <p>These intrinsics make it possible to excise one parameter, marked with
8132 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
8133 The result is a callable
8134 function pointer lacking the nest parameter - the caller does not need to
8135 provide a value for it. Instead, the value to use is stored in advance in a
8136 "trampoline", a block of memory usually allocated on the stack, which also
8137 contains code to splice the nest value into the argument list. This is used
8138 to implement the GCC nested function address extension.</p>
8140 <p>For example, if the function is
8141 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
8142 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
8145 <pre class="doc_code">
8146 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8147 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8148 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8149 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8150 %fp = bitcast i8* %p to i32 (i32, i32)*
8153 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
8154 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
8156 <!-- _______________________________________________________________________ -->
8159 '<tt>llvm.init.trampoline</tt>' Intrinsic
8167 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8171 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
8172 turning it into a trampoline.</p>
8175 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
8176 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
8177 sufficiently aligned block of memory; this memory is written to by the
8178 intrinsic. Note that the size and the alignment are target-specific - LLVM
8179 currently provides no portable way of determining them, so a front-end that
8180 generates this intrinsic needs to have some target-specific knowledge.
8181 The <tt>func</tt> argument must hold a function bitcast to
8182 an <tt>i8*</tt>.</p>
8185 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
8186 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
8187 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
8188 which can be <a href="#int_trampoline">bitcast (to a new function) and
8189 called</a>. The new function's signature is the same as that of
8190 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
8191 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
8192 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
8193 with the same argument list, but with <tt>nval</tt> used for the missing
8194 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
8195 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
8196 to the returned function pointer is undefined.</p>
8199 <!-- _______________________________________________________________________ -->
8202 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
8210 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8214 <p>This performs any required machine-specific adjustment to the address of a
8215 trampoline (passed as <tt>tramp</tt>).</p>
8218 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
8219 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
8223 <p>On some architectures the address of the code to be executed needs to be
8224 different to the address where the trampoline is actually stored. This
8225 intrinsic returns the executable address corresponding to <tt>tramp</tt>
8226 after performing the required machine specific adjustments.
8227 The pointer returned can then be <a href="#int_trampoline"> bitcast and
8235 <!-- ======================================================================= -->
8237 <a name="int_memorymarkers">Memory Use Markers</a>
8242 <p>This class of intrinsics exists to information about the lifetime of memory
8243 objects and ranges where variables are immutable.</p>
8245 <!-- _______________________________________________________________________ -->
8247 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8254 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8258 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8259 object's lifetime.</p>
8262 <p>The first argument is a constant integer representing the size of the
8263 object, or -1 if it is variable sized. The second argument is a pointer to
8267 <p>This intrinsic indicates that before this point in the code, the value of the
8268 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8269 never be used and has an undefined value. A load from the pointer that
8270 precedes this intrinsic can be replaced with
8271 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8275 <!-- _______________________________________________________________________ -->
8277 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8284 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8288 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8289 object's lifetime.</p>
8292 <p>The first argument is a constant integer representing the size of the
8293 object, or -1 if it is variable sized. The second argument is a pointer to
8297 <p>This intrinsic indicates that after this point in the code, the value of the
8298 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8299 never be used and has an undefined value. Any stores into the memory object
8300 following this intrinsic may be removed as dead.
8304 <!-- _______________________________________________________________________ -->
8306 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8313 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8317 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8318 a memory object will not change.</p>
8321 <p>The first argument is a constant integer representing the size of the
8322 object, or -1 if it is variable sized. The second argument is a pointer to
8326 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8327 the return value, the referenced memory location is constant and
8332 <!-- _______________________________________________________________________ -->
8334 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8341 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8345 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8346 a memory object are mutable.</p>
8349 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8350 The second argument is a constant integer representing the size of the
8351 object, or -1 if it is variable sized and the third argument is a pointer
8355 <p>This intrinsic indicates that the memory is mutable again.</p>
8361 <!-- ======================================================================= -->
8363 <a name="int_general">General Intrinsics</a>
8368 <p>This class of intrinsics is designed to be generic and has no specific
8371 <!-- _______________________________________________________________________ -->
8373 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8380 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8384 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8387 <p>The first argument is a pointer to a value, the second is a pointer to a
8388 global string, the third is a pointer to a global string which is the source
8389 file name, and the last argument is the line number.</p>
8392 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8393 This can be useful for special purpose optimizations that want to look for
8394 these annotations. These have no other defined use; they are ignored by code
8395 generation and optimization.</p>
8399 <!-- _______________________________________________________________________ -->
8401 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8407 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8408 any integer bit width.</p>
8411 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8412 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8413 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8414 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8415 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8419 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8422 <p>The first argument is an integer value (result of some expression), the
8423 second is a pointer to a global string, the third is a pointer to a global
8424 string which is the source file name, and the last argument is the line
8425 number. It returns the value of the first argument.</p>
8428 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8429 arbitrary strings. This can be useful for special purpose optimizations that
8430 want to look for these annotations. These have no other defined use; they
8431 are ignored by code generation and optimization.</p>
8435 <!-- _______________________________________________________________________ -->
8437 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8444 declare void @llvm.trap() noreturn nounwind
8448 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8454 <p>This intrinsic is lowered to the target dependent trap instruction. If the
8455 target does not have a trap instruction, this intrinsic will be lowered to
8456 a call of the <tt>abort()</tt> function.</p>
8460 <!-- _______________________________________________________________________ -->
8462 <a name="int_debugtrap">'<tt>llvm.debugtrap</tt>' Intrinsic</a>
8469 declare void @llvm.debugtrap() nounwind
8473 <p>The '<tt>llvm.debugtrap</tt>' intrinsic.</p>
8479 <p>This intrinsic is lowered to code which is intended to cause an execution
8480 trap with the intention of requesting the attention of a debugger.</p>
8484 <!-- _______________________________________________________________________ -->
8486 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8493 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8497 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8498 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8499 ensure that it is placed on the stack before local variables.</p>
8502 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8503 arguments. The first argument is the value loaded from the stack
8504 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8505 that has enough space to hold the value of the guard.</p>
8508 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8509 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8510 stack. This is to ensure that if a local variable on the stack is
8511 overwritten, it will destroy the value of the guard. When the function exits,
8512 the guard on the stack is checked against the original guard. If they are
8513 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8518 <!-- _______________________________________________________________________ -->
8520 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8527 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8528 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8532 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8533 the optimizers to determine at compile time whether a) an operation (like
8534 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8535 runtime check for overflow isn't necessary. An object in this context means
8536 an allocation of a specific class, structure, array, or other object.</p>
8539 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8540 argument is a pointer to or into the <tt>object</tt>. The second argument
8541 is a boolean and determines whether <tt>llvm.objectsize</tt> returns 0 (if
8542 true) or -1 (if false) when the object size is unknown.
8543 The second argument only accepts constants.</p>
8546 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to a constant representing
8547 the size of the object concerned. If the size cannot be determined at compile
8548 time, <tt>llvm.objectsize</tt> returns <tt>i32/i64 -1 or 0</tt>
8549 (depending on the <tt>min</tt> argument).</p>
8552 <!-- _______________________________________________________________________ -->
8554 <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a>
8561 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8562 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8566 <p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the
8567 most probable) value of <tt>val</tt>, which can be used by optimizers.</p>
8570 <p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first
8571 argument is a value. The second argument is an expected value, this needs to
8572 be a constant value, variables are not allowed.</p>
8575 <p>This intrinsic is lowered to the <tt>val</tt>.</p>
8581 <!-- *********************************************************************** -->
8584 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8585 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8586 <a href="http://validator.w3.org/check/referer"><img
8587 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8589 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8590 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8591 Last modified: $Date$