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.</p>
843 <p>A variable may be defined as <tt>thread_local</tt>, which
844 means that it will not be shared by threads (each thread will have a
845 separated copy of the variable). Not all targets support thread-local
846 variables. Optionally, a TLS model may be specified:</p>
849 <dt><b><tt>localdynamic</tt></b>:</dt>
850 <dd>For variables that are only used within the current shared library.</dd>
852 <dt><b><tt>initialexec</tt></b>:</dt>
853 <dd>For variables in modules that will not be loaded dynamically.</dd>
855 <dt><b><tt>localexec</tt></b>:</dt>
856 <dd>For variables defined in the executable and only used within it.</dd>
859 <p>The models correspond to the ELF TLS models; see
860 <a href="http://people.redhat.com/drepper/tls.pdf">ELF
861 Handling For Thread-Local Storage</a> for more information on under which
862 circumstances the different models may be used. The target may choose a
863 different TLS model if the specified model is not supported, or if a better
864 choice of model can be made.</p>
866 <p>A variable may be defined as a global
867 "constant," which indicates that the contents of the variable
868 will <b>never</b> be modified (enabling better optimization, allowing the
869 global data to be placed in the read-only section of an executable, etc).
870 Note that variables that need runtime initialization cannot be marked
871 "constant" as there is a store to the variable.</p>
873 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
874 constant, even if the final definition of the global is not. This capability
875 can be used to enable slightly better optimization of the program, but
876 requires the language definition to guarantee that optimizations based on the
877 'constantness' are valid for the translation units that do not include the
880 <p>As SSA values, global variables define pointer values that are in scope
881 (i.e. they dominate) all basic blocks in the program. Global variables
882 always define a pointer to their "content" type because they describe a
883 region of memory, and all memory objects in LLVM are accessed through
886 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
887 that the address is not significant, only the content. Constants marked
888 like this can be merged with other constants if they have the same
889 initializer. Note that a constant with significant address <em>can</em>
890 be merged with a <tt>unnamed_addr</tt> constant, the result being a
891 constant whose address is significant.</p>
893 <p>A global variable may be declared to reside in a target-specific numbered
894 address space. For targets that support them, address spaces may affect how
895 optimizations are performed and/or what target instructions are used to
896 access the variable. The default address space is zero. The address space
897 qualifier must precede any other attributes.</p>
899 <p>LLVM allows an explicit section to be specified for globals. If the target
900 supports it, it will emit globals to the section specified.</p>
902 <p>An explicit alignment may be specified for a global, which must be a power
903 of 2. If not present, or if the alignment is set to zero, the alignment of
904 the global is set by the target to whatever it feels convenient. If an
905 explicit alignment is specified, the global is forced to have exactly that
906 alignment. Targets and optimizers are not allowed to over-align the global
907 if the global has an assigned section. In this case, the extra alignment
908 could be observable: for example, code could assume that the globals are
909 densely packed in their section and try to iterate over them as an array,
910 alignment padding would break this iteration.</p>
912 <p>For example, the following defines a global in a numbered address space with
913 an initializer, section, and alignment:</p>
915 <pre class="doc_code">
916 @G = addrspace(5) constant float 1.0, section "foo", align 4
919 <p>The following example defines a thread-local global with
920 the <tt>initialexec</tt> TLS model:</p>
922 <pre class="doc_code">
923 @G = thread_local(initialexec) global i32 0, align 4
929 <!-- ======================================================================= -->
931 <a name="functionstructure">Functions</a>
936 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
937 optional <a href="#linkage">linkage type</a>, an optional
938 <a href="#visibility">visibility style</a>, an optional
939 <a href="#callingconv">calling convention</a>,
940 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
941 <a href="#paramattrs">parameter attribute</a> for the return type, a function
942 name, a (possibly empty) argument list (each with optional
943 <a href="#paramattrs">parameter attributes</a>), optional
944 <a href="#fnattrs">function attributes</a>, an optional section, an optional
945 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
946 curly brace, a list of basic blocks, and a closing curly brace.</p>
948 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
949 optional <a href="#linkage">linkage type</a>, an optional
950 <a href="#visibility">visibility style</a>, an optional
951 <a href="#callingconv">calling convention</a>,
952 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
953 <a href="#paramattrs">parameter attribute</a> for the return type, a function
954 name, a possibly empty list of arguments, an optional alignment, and an
955 optional <a href="#gc">garbage collector name</a>.</p>
957 <p>A function definition contains a list of basic blocks, forming the CFG
958 (Control Flow Graph) for the function. Each basic block may optionally start
959 with a label (giving the basic block a symbol table entry), contains a list
960 of instructions, and ends with a <a href="#terminators">terminator</a>
961 instruction (such as a branch or function return).</p>
963 <p>The first basic block in a function is special in two ways: it is immediately
964 executed on entrance to the function, and it is not allowed to have
965 predecessor basic blocks (i.e. there can not be any branches to the entry
966 block of a function). Because the block can have no predecessors, it also
967 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
969 <p>LLVM allows an explicit section to be specified for functions. If the target
970 supports it, it will emit functions to the section specified.</p>
972 <p>An explicit alignment may be specified for a function. If not present, or if
973 the alignment is set to zero, the alignment of the function is set by the
974 target to whatever it feels convenient. If an explicit alignment is
975 specified, the function is forced to have at least that much alignment. All
976 alignments must be a power of 2.</p>
978 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
979 be significant and two identical functions can be merged.</p>
982 <pre class="doc_code">
983 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
984 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
985 <ResultType> @<FunctionName> ([argument list])
986 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
987 [<a href="#gc">gc</a>] { ... }
992 <!-- ======================================================================= -->
994 <a name="aliasstructure">Aliases</a>
999 <p>Aliases act as "second name" for the aliasee value (which can be either
1000 function, global variable, another alias or bitcast of global value). Aliases
1001 may have an optional <a href="#linkage">linkage type</a>, and an
1002 optional <a href="#visibility">visibility style</a>.</p>
1005 <pre class="doc_code">
1006 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
1011 <!-- ======================================================================= -->
1013 <a name="namedmetadatastructure">Named Metadata</a>
1018 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
1019 nodes</a> (but not metadata strings) are the only valid operands for
1020 a named metadata.</p>
1023 <pre class="doc_code">
1024 ; Some unnamed metadata nodes, which are referenced by the named metadata.
1025 !0 = metadata !{metadata !"zero"}
1026 !1 = metadata !{metadata !"one"}
1027 !2 = metadata !{metadata !"two"}
1029 !name = !{!0, !1, !2}
1034 <!-- ======================================================================= -->
1036 <a name="paramattrs">Parameter Attributes</a>
1041 <p>The return type and each parameter of a function type may have a set of
1042 <i>parameter attributes</i> associated with them. Parameter attributes are
1043 used to communicate additional information about the result or parameters of
1044 a function. Parameter attributes are considered to be part of the function,
1045 not of the function type, so functions with different parameter attributes
1046 can have the same function type.</p>
1048 <p>Parameter attributes are simple keywords that follow the type specified. If
1049 multiple parameter attributes are needed, they are space separated. For
1052 <pre class="doc_code">
1053 declare i32 @printf(i8* noalias nocapture, ...)
1054 declare i32 @atoi(i8 zeroext)
1055 declare signext i8 @returns_signed_char()
1058 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1059 <tt>readonly</tt>) come immediately after the argument list.</p>
1061 <p>Currently, only the following parameter attributes are defined:</p>
1064 <dt><tt><b>zeroext</b></tt></dt>
1065 <dd>This indicates to the code generator that the parameter or return value
1066 should be zero-extended to the extent required by the target's ABI (which
1067 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1068 parameter) or the callee (for a return value).</dd>
1070 <dt><tt><b>signext</b></tt></dt>
1071 <dd>This indicates to the code generator that the parameter or return value
1072 should be sign-extended to the extent required by the target's ABI (which
1073 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1076 <dt><tt><b>inreg</b></tt></dt>
1077 <dd>This indicates that this parameter or return value should be treated in a
1078 special target-dependent fashion during while emitting code for a function
1079 call or return (usually, by putting it in a register as opposed to memory,
1080 though some targets use it to distinguish between two different kinds of
1081 registers). Use of this attribute is target-specific.</dd>
1083 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1084 <dd><p>This indicates that the pointer parameter should really be passed by
1085 value to the function. The attribute implies that a hidden copy of the
1087 is made between the caller and the callee, so the callee is unable to
1088 modify the value in the caller. This attribute is only valid on LLVM
1089 pointer arguments. It is generally used to pass structs and arrays by
1090 value, but is also valid on pointers to scalars. The copy is considered
1091 to belong to the caller not the callee (for example,
1092 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1093 <tt>byval</tt> parameters). This is not a valid attribute for return
1096 <p>The byval attribute also supports specifying an alignment with
1097 the align attribute. It indicates the alignment of the stack slot to
1098 form and the known alignment of the pointer specified to the call site. If
1099 the alignment is not specified, then the code generator makes a
1100 target-specific assumption.</p></dd>
1102 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1103 <dd>This indicates that the pointer parameter specifies the address of a
1104 structure that is the return value of the function in the source program.
1105 This pointer must be guaranteed by the caller to be valid: loads and
1106 stores to the structure may be assumed by the callee to not to trap. This
1107 may only be applied to the first parameter. This is not a valid attribute
1108 for return values. </dd>
1110 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1111 <dd>This indicates that pointer values
1112 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1113 value do not alias pointer values which are not <i>based</i> on it,
1114 ignoring certain "irrelevant" dependencies.
1115 For a call to the parent function, dependencies between memory
1116 references from before or after the call and from those during the call
1117 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1118 return value used in that call.
1119 The caller shares the responsibility with the callee for ensuring that
1120 these requirements are met.
1121 For further details, please see the discussion of the NoAlias response in
1122 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1124 Note that this definition of <tt>noalias</tt> is intentionally
1125 similar to the definition of <tt>restrict</tt> in C99 for function
1126 arguments, though it is slightly weaker.
1128 For function return values, C99's <tt>restrict</tt> is not meaningful,
1129 while LLVM's <tt>noalias</tt> is.
1132 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1133 <dd>This indicates that the callee does not make any copies of the pointer
1134 that outlive the callee itself. This is not a valid attribute for return
1137 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1138 <dd>This indicates that the pointer parameter can be excised using the
1139 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1140 attribute for return values.</dd>
1145 <!-- ======================================================================= -->
1147 <a name="gc">Garbage Collector Names</a>
1152 <p>Each function may specify a garbage collector name, which is simply a
1155 <pre class="doc_code">
1156 define void @f() gc "name" { ... }
1159 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1160 collector which will cause the compiler to alter its output in order to
1161 support the named garbage collection algorithm.</p>
1165 <!-- ======================================================================= -->
1167 <a name="fnattrs">Function Attributes</a>
1172 <p>Function attributes are set to communicate additional information about a
1173 function. Function attributes are considered to be part of the function, not
1174 of the function type, so functions with different parameter attributes can
1175 have the same function type.</p>
1177 <p>Function attributes are simple keywords that follow the type specified. If
1178 multiple attributes are needed, they are space separated. For example:</p>
1180 <pre class="doc_code">
1181 define void @f() noinline { ... }
1182 define void @f() alwaysinline { ... }
1183 define void @f() alwaysinline optsize { ... }
1184 define void @f() optsize { ... }
1188 <dt><tt><b>address_safety</b></tt></dt>
1189 <dd>This attribute indicates that the address safety analysis
1190 is enabled for this function. </dd>
1192 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1193 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1194 the backend should forcibly align the stack pointer. Specify the
1195 desired alignment, which must be a power of two, in parentheses.
1197 <dt><tt><b>alwaysinline</b></tt></dt>
1198 <dd>This attribute indicates that the inliner should attempt to inline this
1199 function into callers whenever possible, ignoring any active inlining size
1200 threshold for this caller.</dd>
1202 <dt><tt><b>nonlazybind</b></tt></dt>
1203 <dd>This attribute suppresses lazy symbol binding for the function. This
1204 may make calls to the function faster, at the cost of extra program
1205 startup time if the function is not called during program startup.</dd>
1207 <dt><tt><b>inlinehint</b></tt></dt>
1208 <dd>This attribute indicates that the source code contained a hint that inlining
1209 this function is desirable (such as the "inline" keyword in C/C++). It
1210 is just a hint; it imposes no requirements on the inliner.</dd>
1212 <dt><tt><b>naked</b></tt></dt>
1213 <dd>This attribute disables prologue / epilogue emission for the function.
1214 This can have very system-specific consequences.</dd>
1216 <dt><tt><b>noimplicitfloat</b></tt></dt>
1217 <dd>This attributes disables implicit floating point instructions.</dd>
1219 <dt><tt><b>noinline</b></tt></dt>
1220 <dd>This attribute indicates that the inliner should never inline this
1221 function in any situation. This attribute may not be used together with
1222 the <tt>alwaysinline</tt> attribute.</dd>
1224 <dt><tt><b>noredzone</b></tt></dt>
1225 <dd>This attribute indicates that the code generator should not use a red
1226 zone, even if the target-specific ABI normally permits it.</dd>
1228 <dt><tt><b>noreturn</b></tt></dt>
1229 <dd>This function attribute indicates that the function never returns
1230 normally. This produces undefined behavior at runtime if the function
1231 ever does dynamically return.</dd>
1233 <dt><tt><b>nounwind</b></tt></dt>
1234 <dd>This function attribute indicates that the function never returns with an
1235 unwind or exceptional control flow. If the function does unwind, its
1236 runtime behavior is undefined.</dd>
1238 <dt><tt><b>optsize</b></tt></dt>
1239 <dd>This attribute suggests that optimization passes and code generator passes
1240 make choices that keep the code size of this function low, and otherwise
1241 do optimizations specifically to reduce code size.</dd>
1243 <dt><tt><b>readnone</b></tt></dt>
1244 <dd>This attribute indicates that the function computes its result (or decides
1245 to unwind an exception) based strictly on its arguments, without
1246 dereferencing any pointer arguments or otherwise accessing any mutable
1247 state (e.g. memory, control registers, etc) visible to caller functions.
1248 It does not write through any pointer arguments
1249 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1250 changes any state visible to callers. This means that it cannot unwind
1251 exceptions by calling the <tt>C++</tt> exception throwing methods.</dd>
1253 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1254 <dd>This attribute indicates that the function does not write through any
1255 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1256 arguments) or otherwise modify any state (e.g. memory, control registers,
1257 etc) visible to caller functions. It may dereference pointer arguments
1258 and read state that may be set in the caller. A readonly function always
1259 returns the same value (or unwinds an exception identically) when called
1260 with the same set of arguments and global state. It cannot unwind an
1261 exception by calling the <tt>C++</tt> exception throwing methods.</dd>
1263 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1264 <dd>This attribute indicates that this function can return twice. The
1265 C <code>setjmp</code> is an example of such a function. The compiler
1266 disables some optimizations (like tail calls) in the caller of these
1269 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1270 <dd>This attribute indicates that the function should emit a stack smashing
1271 protector. It is in the form of a "canary"—a random value placed on
1272 the stack before the local variables that's checked upon return from the
1273 function to see if it has been overwritten. A heuristic is used to
1274 determine if a function needs stack protectors or not.<br>
1276 If a function that has an <tt>ssp</tt> attribute is inlined into a
1277 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1278 function will have an <tt>ssp</tt> attribute.</dd>
1280 <dt><tt><b>sspreq</b></tt></dt>
1281 <dd>This attribute indicates that the function should <em>always</em> emit a
1282 stack smashing protector. This overrides
1283 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1285 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1286 function that doesn't have an <tt>sspreq</tt> attribute or which has
1287 an <tt>ssp</tt> attribute, then the resulting function will have
1288 an <tt>sspreq</tt> attribute.</dd>
1290 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1291 <dd>This attribute indicates that the ABI being targeted requires that
1292 an unwind table entry be produce for this function even if we can
1293 show that no exceptions passes by it. This is normally the case for
1294 the ELF x86-64 abi, but it can be disabled for some compilation
1300 <!-- ======================================================================= -->
1302 <a name="moduleasm">Module-Level Inline Assembly</a>
1307 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1308 the GCC "file scope inline asm" blocks. These blocks are internally
1309 concatenated by LLVM and treated as a single unit, but may be separated in
1310 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1312 <pre class="doc_code">
1313 module asm "inline asm code goes here"
1314 module asm "more can go here"
1317 <p>The strings can contain any character by escaping non-printable characters.
1318 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1321 <p>The inline asm code is simply printed to the machine code .s file when
1322 assembly code is generated.</p>
1326 <!-- ======================================================================= -->
1328 <a name="datalayout">Data Layout</a>
1333 <p>A module may specify a target specific data layout string that specifies how
1334 data is to be laid out in memory. The syntax for the data layout is
1337 <pre class="doc_code">
1338 target datalayout = "<i>layout specification</i>"
1341 <p>The <i>layout specification</i> consists of a list of specifications
1342 separated by the minus sign character ('-'). Each specification starts with
1343 a letter and may include other information after the letter to define some
1344 aspect of the data layout. The specifications accepted are as follows:</p>
1348 <dd>Specifies that the target lays out data in big-endian form. That is, the
1349 bits with the most significance have the lowest address location.</dd>
1352 <dd>Specifies that the target lays out data in little-endian form. That is,
1353 the bits with the least significance have the lowest address
1356 <dt><tt>S<i>size</i></tt></dt>
1357 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1358 of stack variables is limited to the natural stack alignment to avoid
1359 dynamic stack realignment. The stack alignment must be a multiple of
1360 8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1361 which does not prevent any alignment promotions.</dd>
1363 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1364 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1365 <i>preferred</i> alignments. All sizes are in bits. Specifying
1366 the <i>pref</i> alignment is optional. If omitted, the
1367 preceding <tt>:</tt> should be omitted too.</dd>
1369 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1370 <dd>This specifies the alignment for an integer type of a given bit
1371 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1373 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1374 <dd>This specifies the alignment for a vector type of a given bit
1377 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1378 <dd>This specifies the alignment for a floating point type of a given bit
1379 <i>size</i>. Only values of <i>size</i> that are supported by the target
1380 will work. 32 (float) and 64 (double) are supported on all targets;
1381 80 or 128 (different flavors of long double) are also supported on some
1384 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1385 <dd>This specifies the alignment for an aggregate type of a given bit
1388 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1389 <dd>This specifies the alignment for a stack object of a given bit
1392 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1393 <dd>This specifies a set of native integer widths for the target CPU
1394 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1395 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1396 this set are considered to support most general arithmetic
1397 operations efficiently.</dd>
1400 <p>When constructing the data layout for a given target, LLVM starts with a
1401 default set of specifications which are then (possibly) overridden by the
1402 specifications in the <tt>datalayout</tt> keyword. The default specifications
1403 are given in this list:</p>
1406 <li><tt>E</tt> - big endian</li>
1407 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1408 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1409 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1410 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1411 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1412 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1413 alignment of 64-bits</li>
1414 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1415 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1416 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1417 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1418 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1419 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1422 <p>When LLVM is determining the alignment for a given type, it uses the
1423 following rules:</p>
1426 <li>If the type sought is an exact match for one of the specifications, that
1427 specification is used.</li>
1429 <li>If no match is found, and the type sought is an integer type, then the
1430 smallest integer type that is larger than the bitwidth of the sought type
1431 is used. If none of the specifications are larger than the bitwidth then
1432 the the largest integer type is used. For example, given the default
1433 specifications above, the i7 type will use the alignment of i8 (next
1434 largest) while both i65 and i256 will use the alignment of i64 (largest
1437 <li>If no match is found, and the type sought is a vector type, then the
1438 largest vector type that is smaller than the sought vector type will be
1439 used as a fall back. This happens because <128 x double> can be
1440 implemented in terms of 64 <2 x double>, for example.</li>
1443 <p>The function of the data layout string may not be what you expect. Notably,
1444 this is not a specification from the frontend of what alignment the code
1445 generator should use.</p>
1447 <p>Instead, if specified, the target data layout is required to match what the
1448 ultimate <em>code generator</em> expects. This string is used by the
1449 mid-level optimizers to
1450 improve code, and this only works if it matches what the ultimate code
1451 generator uses. If you would like to generate IR that does not embed this
1452 target-specific detail into the IR, then you don't have to specify the
1453 string. This will disable some optimizations that require precise layout
1454 information, but this also prevents those optimizations from introducing
1455 target specificity into the IR.</p>
1461 <!-- ======================================================================= -->
1463 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1468 <p>Any memory access must be done through a pointer value associated
1469 with an address range of the memory access, otherwise the behavior
1470 is undefined. Pointer values are associated with address ranges
1471 according to the following rules:</p>
1474 <li>A pointer value is associated with the addresses associated with
1475 any value it is <i>based</i> on.
1476 <li>An address of a global variable is associated with the address
1477 range of the variable's storage.</li>
1478 <li>The result value of an allocation instruction is associated with
1479 the address range of the allocated storage.</li>
1480 <li>A null pointer in the default address-space is associated with
1482 <li>An integer constant other than zero or a pointer value returned
1483 from a function not defined within LLVM may be associated with address
1484 ranges allocated through mechanisms other than those provided by
1485 LLVM. Such ranges shall not overlap with any ranges of addresses
1486 allocated by mechanisms provided by LLVM.</li>
1489 <p>A pointer value is <i>based</i> on another pointer value according
1490 to the following rules:</p>
1493 <li>A pointer value formed from a
1494 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1495 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1496 <li>The result value of a
1497 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1498 of the <tt>bitcast</tt>.</li>
1499 <li>A pointer value formed by an
1500 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1501 pointer values that contribute (directly or indirectly) to the
1502 computation of the pointer's value.</li>
1503 <li>The "<i>based</i> on" relationship is transitive.</li>
1506 <p>Note that this definition of <i>"based"</i> is intentionally
1507 similar to the definition of <i>"based"</i> in C99, though it is
1508 slightly weaker.</p>
1510 <p>LLVM IR does not associate types with memory. The result type of a
1511 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1512 alignment of the memory from which to load, as well as the
1513 interpretation of the value. The first operand type of a
1514 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1515 and alignment of the store.</p>
1517 <p>Consequently, type-based alias analysis, aka TBAA, aka
1518 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1519 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1520 additional information which specialized optimization passes may use
1521 to implement type-based alias analysis.</p>
1525 <!-- ======================================================================= -->
1527 <a name="volatile">Volatile Memory Accesses</a>
1532 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1533 href="#i_store"><tt>store</tt></a>s, and <a
1534 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1535 The optimizers must not change the number of volatile operations or change their
1536 order of execution relative to other volatile operations. The optimizers
1537 <i>may</i> change the order of volatile operations relative to non-volatile
1538 operations. This is not Java's "volatile" and has no cross-thread
1539 synchronization behavior.</p>
1543 <!-- ======================================================================= -->
1545 <a name="memmodel">Memory Model for Concurrent Operations</a>
1550 <p>The LLVM IR does not define any way to start parallel threads of execution
1551 or to register signal handlers. Nonetheless, there are platform-specific
1552 ways to create them, and we define LLVM IR's behavior in their presence. This
1553 model is inspired by the C++0x memory model.</p>
1555 <p>For a more informal introduction to this model, see the
1556 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1558 <p>We define a <i>happens-before</i> partial order as the least partial order
1561 <li>Is a superset of single-thread program order, and</li>
1562 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1563 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1564 by platform-specific techniques, like pthread locks, thread
1565 creation, thread joining, etc., and by atomic instructions.
1566 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1570 <p>Note that program order does not introduce <i>happens-before</i> edges
1571 between a thread and signals executing inside that thread.</p>
1573 <p>Every (defined) read operation (load instructions, memcpy, atomic
1574 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1575 (defined) write operations (store instructions, atomic
1576 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1577 initialized globals are considered to have a write of the initializer which is
1578 atomic and happens before any other read or write of the memory in question.
1579 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1580 any write to the same byte, except:</p>
1583 <li>If <var>write<sub>1</sub></var> happens before
1584 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1585 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1586 does not see <var>write<sub>1</sub></var>.
1587 <li>If <var>R<sub>byte</sub></var> happens before
1588 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1589 see <var>write<sub>3</sub></var>.
1592 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1594 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1595 is supposed to give guarantees which can support
1596 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1597 addresses which do not behave like normal memory. It does not generally
1598 provide cross-thread synchronization.)
1599 <li>Otherwise, if there is no write to the same byte that happens before
1600 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1601 <tt>undef</tt> for that byte.
1602 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1603 <var>R<sub>byte</sub></var> returns the value written by that
1605 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1606 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1607 values written. See the <a href="#ordering">Atomic Memory Ordering
1608 Constraints</a> section for additional constraints on how the choice
1610 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1613 <p><var>R</var> returns the value composed of the series of bytes it read.
1614 This implies that some bytes within the value may be <tt>undef</tt>
1615 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1616 defines the semantics of the operation; it doesn't mean that targets will
1617 emit more than one instruction to read the series of bytes.</p>
1619 <p>Note that in cases where none of the atomic intrinsics are used, this model
1620 places only one restriction on IR transformations on top of what is required
1621 for single-threaded execution: introducing a store to a byte which might not
1622 otherwise be stored is not allowed in general. (Specifically, in the case
1623 where another thread might write to and read from an address, introducing a
1624 store can change a load that may see exactly one write into a load that may
1625 see multiple writes.)</p>
1627 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1628 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1629 none of the backends currently in the tree fall into this category; however,
1630 there might be targets which care. If there are, we want a paragraph
1633 Targets may specify that stores narrower than a certain width are not
1634 available; on such a target, for the purposes of this model, treat any
1635 non-atomic write with an alignment or width less than the minimum width
1636 as if it writes to the relevant surrounding bytes.
1641 <!-- ======================================================================= -->
1643 <a name="ordering">Atomic Memory Ordering Constraints</a>
1648 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1649 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1650 <a href="#i_fence"><code>fence</code></a>,
1651 <a href="#i_load"><code>atomic load</code></a>, and
1652 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1653 that determines which other atomic instructions on the same address they
1654 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1655 but are somewhat more colloquial. If these descriptions aren't precise enough,
1656 check those specs (see spec references in the
1657 <a href="Atomics.html#introduction">atomics guide</a>).
1658 <a href="#i_fence"><code>fence</code></a> instructions
1659 treat these orderings somewhat differently since they don't take an address.
1660 See that instruction's documentation for details.</p>
1662 <p>For a simpler introduction to the ordering constraints, see the
1663 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1666 <dt><code>unordered</code></dt>
1667 <dd>The set of values that can be read is governed by the happens-before
1668 partial order. A value cannot be read unless some operation wrote it.
1669 This is intended to provide a guarantee strong enough to model Java's
1670 non-volatile shared variables. This ordering cannot be specified for
1671 read-modify-write operations; it is not strong enough to make them atomic
1672 in any interesting way.</dd>
1673 <dt><code>monotonic</code></dt>
1674 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1675 total order for modifications by <code>monotonic</code> operations on each
1676 address. All modification orders must be compatible with the happens-before
1677 order. There is no guarantee that the modification orders can be combined to
1678 a global total order for the whole program (and this often will not be
1679 possible). The read in an atomic read-modify-write operation
1680 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1681 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1682 reads the value in the modification order immediately before the value it
1683 writes. If one atomic read happens before another atomic read of the same
1684 address, the later read must see the same value or a later value in the
1685 address's modification order. This disallows reordering of
1686 <code>monotonic</code> (or stronger) operations on the same address. If an
1687 address is written <code>monotonic</code>ally by one thread, and other threads
1688 <code>monotonic</code>ally read that address repeatedly, the other threads must
1689 eventually see the write. This corresponds to the C++0x/C1x
1690 <code>memory_order_relaxed</code>.</dd>
1691 <dt><code>acquire</code></dt>
1692 <dd>In addition to the guarantees of <code>monotonic</code>,
1693 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1694 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1695 <dt><code>release</code></dt>
1696 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1697 writes a value which is subsequently read by an <code>acquire</code> operation,
1698 it <i>synchronizes-with</i> that operation. (This isn't a complete
1699 description; see the C++0x definition of a release sequence.) This corresponds
1700 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1701 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1702 <code>acquire</code> and <code>release</code> operation on its address.
1703 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1704 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1705 <dd>In addition to the guarantees of <code>acq_rel</code>
1706 (<code>acquire</code> for an operation which only reads, <code>release</code>
1707 for an operation which only writes), there is a global total order on all
1708 sequentially-consistent operations on all addresses, which is consistent with
1709 the <i>happens-before</i> partial order and with the modification orders of
1710 all the affected addresses. Each sequentially-consistent read sees the last
1711 preceding write to the same address in this global order. This corresponds
1712 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1715 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1716 it only <i>synchronizes with</i> or participates in modification and seq_cst
1717 total orderings with other operations running in the same thread (for example,
1718 in signal handlers).</p>
1724 <!-- *********************************************************************** -->
1725 <h2><a name="typesystem">Type System</a></h2>
1726 <!-- *********************************************************************** -->
1730 <p>The LLVM type system is one of the most important features of the
1731 intermediate representation. Being typed enables a number of optimizations
1732 to be performed on the intermediate representation directly, without having
1733 to do extra analyses on the side before the transformation. A strong type
1734 system makes it easier to read the generated code and enables novel analyses
1735 and transformations that are not feasible to perform on normal three address
1736 code representations.</p>
1738 <!-- ======================================================================= -->
1740 <a name="t_classifications">Type Classifications</a>
1745 <p>The types fall into a few useful classifications:</p>
1747 <table border="1" cellspacing="0" cellpadding="4">
1749 <tr><th>Classification</th><th>Types</th></tr>
1751 <td><a href="#t_integer">integer</a></td>
1752 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1755 <td><a href="#t_floating">floating point</a></td>
1756 <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1759 <td><a name="t_firstclass">first class</a></td>
1760 <td><a href="#t_integer">integer</a>,
1761 <a href="#t_floating">floating point</a>,
1762 <a href="#t_pointer">pointer</a>,
1763 <a href="#t_vector">vector</a>,
1764 <a href="#t_struct">structure</a>,
1765 <a href="#t_array">array</a>,
1766 <a href="#t_label">label</a>,
1767 <a href="#t_metadata">metadata</a>.
1771 <td><a href="#t_primitive">primitive</a></td>
1772 <td><a href="#t_label">label</a>,
1773 <a href="#t_void">void</a>,
1774 <a href="#t_integer">integer</a>,
1775 <a href="#t_floating">floating point</a>,
1776 <a href="#t_x86mmx">x86mmx</a>,
1777 <a href="#t_metadata">metadata</a>.</td>
1780 <td><a href="#t_derived">derived</a></td>
1781 <td><a href="#t_array">array</a>,
1782 <a href="#t_function">function</a>,
1783 <a href="#t_pointer">pointer</a>,
1784 <a href="#t_struct">structure</a>,
1785 <a href="#t_vector">vector</a>,
1786 <a href="#t_opaque">opaque</a>.
1792 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1793 important. Values of these types are the only ones which can be produced by
1798 <!-- ======================================================================= -->
1800 <a name="t_primitive">Primitive Types</a>
1805 <p>The primitive types are the fundamental building blocks of the LLVM
1808 <!-- _______________________________________________________________________ -->
1810 <a name="t_integer">Integer Type</a>
1816 <p>The integer type is a very simple type that simply specifies an arbitrary
1817 bit width for the integer type desired. Any bit width from 1 bit to
1818 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1825 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1829 <table class="layout">
1831 <td class="left"><tt>i1</tt></td>
1832 <td class="left">a single-bit integer.</td>
1835 <td class="left"><tt>i32</tt></td>
1836 <td class="left">a 32-bit integer.</td>
1839 <td class="left"><tt>i1942652</tt></td>
1840 <td class="left">a really big integer of over 1 million bits.</td>
1846 <!-- _______________________________________________________________________ -->
1848 <a name="t_floating">Floating Point Types</a>
1855 <tr><th>Type</th><th>Description</th></tr>
1856 <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr>
1857 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1858 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1859 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1860 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1861 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1867 <!-- _______________________________________________________________________ -->
1869 <a name="t_x86mmx">X86mmx Type</a>
1875 <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>
1884 <!-- _______________________________________________________________________ -->
1886 <a name="t_void">Void Type</a>
1892 <p>The void type does not represent any value and has no size.</p>
1901 <!-- _______________________________________________________________________ -->
1903 <a name="t_label">Label Type</a>
1909 <p>The label type represents code labels.</p>
1918 <!-- _______________________________________________________________________ -->
1920 <a name="t_metadata">Metadata Type</a>
1926 <p>The metadata type represents embedded metadata. No derived types may be
1927 created from metadata except for <a href="#t_function">function</a>
1939 <!-- ======================================================================= -->
1941 <a name="t_derived">Derived Types</a>
1946 <p>The real power in LLVM comes from the derived types in the system. This is
1947 what allows a programmer to represent arrays, functions, pointers, and other
1948 useful types. Each of these types contain one or more element types which
1949 may be a primitive type, or another derived type. For example, it is
1950 possible to have a two dimensional array, using an array as the element type
1951 of another array.</p>
1953 <!-- _______________________________________________________________________ -->
1955 <a name="t_aggregate">Aggregate Types</a>
1960 <p>Aggregate Types are a subset of derived types that can contain multiple
1961 member types. <a href="#t_array">Arrays</a> and
1962 <a href="#t_struct">structs</a> are aggregate types.
1963 <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p>
1967 <!-- _______________________________________________________________________ -->
1969 <a name="t_array">Array Type</a>
1975 <p>The array type is a very simple derived type that arranges elements
1976 sequentially in memory. The array type requires a size (number of elements)
1977 and an underlying data type.</p>
1981 [<# elements> x <elementtype>]
1984 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1985 be any type with a size.</p>
1988 <table class="layout">
1990 <td class="left"><tt>[40 x i32]</tt></td>
1991 <td class="left">Array of 40 32-bit integer values.</td>
1994 <td class="left"><tt>[41 x i32]</tt></td>
1995 <td class="left">Array of 41 32-bit integer values.</td>
1998 <td class="left"><tt>[4 x i8]</tt></td>
1999 <td class="left">Array of 4 8-bit integer values.</td>
2002 <p>Here are some examples of multidimensional arrays:</p>
2003 <table class="layout">
2005 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
2006 <td class="left">3x4 array of 32-bit integer values.</td>
2009 <td class="left"><tt>[12 x [10 x float]]</tt></td>
2010 <td class="left">12x10 array of single precision floating point values.</td>
2013 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
2014 <td class="left">2x3x4 array of 16-bit integer values.</td>
2018 <p>There is no restriction on indexing beyond the end of the array implied by
2019 a static type (though there are restrictions on indexing beyond the bounds
2020 of an allocated object in some cases). This means that single-dimension
2021 'variable sized array' addressing can be implemented in LLVM with a zero
2022 length array type. An implementation of 'pascal style arrays' in LLVM could
2023 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
2027 <!-- _______________________________________________________________________ -->
2029 <a name="t_function">Function Type</a>
2035 <p>The function type can be thought of as a function signature. It consists of
2036 a return type and a list of formal parameter types. The return type of a
2037 function type is a first class type or a void type.</p>
2041 <returntype> (<parameter list>)
2044 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
2045 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
2046 which indicates that the function takes a variable number of arguments.
2047 Variable argument functions can access their arguments with
2048 the <a href="#int_varargs">variable argument handling intrinsic</a>
2049 functions. '<tt><returntype></tt>' is any type except
2050 <a href="#t_label">label</a>.</p>
2053 <table class="layout">
2055 <td class="left"><tt>i32 (i32)</tt></td>
2056 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2058 </tr><tr class="layout">
2059 <td class="left"><tt>float (i16, i32 *) *
2061 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2062 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2063 returning <tt>float</tt>.
2065 </tr><tr class="layout">
2066 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2067 <td class="left">A vararg function that takes at least one
2068 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2069 which returns an integer. This is the signature for <tt>printf</tt> in
2072 </tr><tr class="layout">
2073 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2074 <td class="left">A function taking an <tt>i32</tt>, returning a
2075 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2082 <!-- _______________________________________________________________________ -->
2084 <a name="t_struct">Structure Type</a>
2090 <p>The structure type is used to represent a collection of data members together
2091 in memory. The elements of a structure may be any type that has a size.</p>
2093 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2094 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2095 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2096 Structures in registers are accessed using the
2097 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2098 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2100 <p>Structures may optionally be "packed" structures, which indicate that the
2101 alignment of the struct is one byte, and that there is no padding between
2102 the elements. In non-packed structs, padding between field types is inserted
2103 as defined by the TargetData string in the module, which is required to match
2104 what the underlying code generator expects.</p>
2106 <p>Structures can either be "literal" or "identified". A literal structure is
2107 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2108 types are always defined at the top level with a name. Literal types are
2109 uniqued by their contents and can never be recursive or opaque since there is
2110 no way to write one. Identified types can be recursive, can be opaqued, and are
2116 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2117 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2121 <table class="layout">
2123 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2124 <td class="left">A triple of three <tt>i32</tt> values</td>
2127 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2128 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2129 second element is a <a href="#t_pointer">pointer</a> to a
2130 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2131 an <tt>i32</tt>.</td>
2134 <td class="left"><tt><{ i8, i32 }></tt></td>
2135 <td class="left">A packed struct known to be 5 bytes in size.</td>
2141 <!-- _______________________________________________________________________ -->
2143 <a name="t_opaque">Opaque Structure Types</a>
2149 <p>Opaque structure types are used to represent named structure types that do
2150 not have a body specified. This corresponds (for example) to the C notion of
2151 a forward declared structure.</p>
2160 <table class="layout">
2162 <td class="left"><tt>opaque</tt></td>
2163 <td class="left">An opaque type.</td>
2171 <!-- _______________________________________________________________________ -->
2173 <a name="t_pointer">Pointer Type</a>
2179 <p>The pointer type is used to specify memory locations.
2180 Pointers are commonly used to reference objects in memory.</p>
2182 <p>Pointer types may have an optional address space attribute defining the
2183 numbered address space where the pointed-to object resides. The default
2184 address space is number zero. The semantics of non-zero address
2185 spaces are target-specific.</p>
2187 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2188 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2196 <table class="layout">
2198 <td class="left"><tt>[4 x i32]*</tt></td>
2199 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2200 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2203 <td class="left"><tt>i32 (i32*) *</tt></td>
2204 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2205 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2209 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2210 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2211 that resides in address space #5.</td>
2217 <!-- _______________________________________________________________________ -->
2219 <a name="t_vector">Vector Type</a>
2225 <p>A vector type is a simple derived type that represents a vector of elements.
2226 Vector types are used when multiple primitive data are operated in parallel
2227 using a single instruction (SIMD). A vector type requires a size (number of
2228 elements) and an underlying primitive data type. Vector types are considered
2229 <a href="#t_firstclass">first class</a>.</p>
2233 < <# elements> x <elementtype> >
2236 <p>The number of elements is a constant integer value larger than 0; elementtype
2237 may be any integer or floating point type, or a pointer to these types.
2238 Vectors of size zero are not allowed. </p>
2241 <table class="layout">
2243 <td class="left"><tt><4 x i32></tt></td>
2244 <td class="left">Vector of 4 32-bit integer values.</td>
2247 <td class="left"><tt><8 x float></tt></td>
2248 <td class="left">Vector of 8 32-bit floating-point values.</td>
2251 <td class="left"><tt><2 x i64></tt></td>
2252 <td class="left">Vector of 2 64-bit integer values.</td>
2255 <td class="left"><tt><4 x i64*></tt></td>
2256 <td class="left">Vector of 4 pointers to 64-bit integer values.</td>
2266 <!-- *********************************************************************** -->
2267 <h2><a name="constants">Constants</a></h2>
2268 <!-- *********************************************************************** -->
2272 <p>LLVM has several different basic types of constants. This section describes
2273 them all and their syntax.</p>
2275 <!-- ======================================================================= -->
2277 <a name="simpleconstants">Simple Constants</a>
2283 <dt><b>Boolean constants</b></dt>
2284 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2285 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2287 <dt><b>Integer constants</b></dt>
2288 <dd>Standard integers (such as '4') are constants of
2289 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2290 with integer types.</dd>
2292 <dt><b>Floating point constants</b></dt>
2293 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2294 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2295 notation (see below). The assembler requires the exact decimal value of a
2296 floating-point constant. For example, the assembler accepts 1.25 but
2297 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2298 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2300 <dt><b>Null pointer constants</b></dt>
2301 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2302 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2305 <p>The one non-intuitive notation for constants is the hexadecimal form of
2306 floating point constants. For example, the form '<tt>double
2307 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2308 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2309 constants are required (and the only time that they are generated by the
2310 disassembler) is when a floating point constant must be emitted but it cannot
2311 be represented as a decimal floating point number in a reasonable number of
2312 digits. For example, NaN's, infinities, and other special values are
2313 represented in their IEEE hexadecimal format so that assembly and disassembly
2314 do not cause any bits to change in the constants.</p>
2316 <p>When using the hexadecimal form, constants of types half, float, and double are
2317 represented using the 16-digit form shown above (which matches the IEEE754
2318 representation for double); half and float values must, however, be exactly
2319 representable as IEE754 half and single precision, respectively.
2320 Hexadecimal format is always used
2321 for long double, and there are three forms of long double. The 80-bit format
2322 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2323 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2324 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2325 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2326 currently supported target uses this format. Long doubles will only work if
2327 they match the long double format on your target. The IEEE 16-bit format
2328 (half precision) is represented by <tt>0xH</tt> followed by 4 hexadecimal
2329 digits. All hexadecimal formats are big-endian (sign bit at the left).</p>
2331 <p>There are no constants of type x86mmx.</p>
2334 <!-- ======================================================================= -->
2336 <a name="aggregateconstants"></a> <!-- old anchor -->
2337 <a name="complexconstants">Complex Constants</a>
2342 <p>Complex constants are a (potentially recursive) combination of simple
2343 constants and smaller complex constants.</p>
2346 <dt><b>Structure constants</b></dt>
2347 <dd>Structure constants are represented with notation similar to structure
2348 type definitions (a comma separated list of elements, surrounded by braces
2349 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2350 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2351 Structure constants must have <a href="#t_struct">structure type</a>, and
2352 the number and types of elements must match those specified by the
2355 <dt><b>Array constants</b></dt>
2356 <dd>Array constants are represented with notation similar to array type
2357 definitions (a comma separated list of elements, surrounded by square
2358 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2359 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2360 the number and types of elements must match those specified by the
2363 <dt><b>Vector constants</b></dt>
2364 <dd>Vector constants are represented with notation similar to vector type
2365 definitions (a comma separated list of elements, surrounded by
2366 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2367 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2368 have <a href="#t_vector">vector type</a>, and the number and types of
2369 elements must match those specified by the type.</dd>
2371 <dt><b>Zero initialization</b></dt>
2372 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2373 value to zero of <em>any</em> type, including scalar and
2374 <a href="#t_aggregate">aggregate</a> types.
2375 This is often used to avoid having to print large zero initializers
2376 (e.g. for large arrays) and is always exactly equivalent to using explicit
2377 zero initializers.</dd>
2379 <dt><b>Metadata node</b></dt>
2380 <dd>A metadata node is a structure-like constant with
2381 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2382 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2383 be interpreted as part of the instruction stream, metadata is a place to
2384 attach additional information such as debug info.</dd>
2389 <!-- ======================================================================= -->
2391 <a name="globalconstants">Global Variable and Function Addresses</a>
2396 <p>The addresses of <a href="#globalvars">global variables</a>
2397 and <a href="#functionstructure">functions</a> are always implicitly valid
2398 (link-time) constants. These constants are explicitly referenced when
2399 the <a href="#identifiers">identifier for the global</a> is used and always
2400 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2401 legal LLVM file:</p>
2403 <pre class="doc_code">
2406 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2411 <!-- ======================================================================= -->
2413 <a name="undefvalues">Undefined Values</a>
2418 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2419 indicates that the user of the value may receive an unspecified bit-pattern.
2420 Undefined values may be of any type (other than '<tt>label</tt>'
2421 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2423 <p>Undefined values are useful because they indicate to the compiler that the
2424 program is well defined no matter what value is used. This gives the
2425 compiler more freedom to optimize. Here are some examples of (potentially
2426 surprising) transformations that are valid (in pseudo IR):</p>
2429 <pre class="doc_code">
2439 <p>This is safe because all of the output bits are affected by the undef bits.
2440 Any output bit can have a zero or one depending on the input bits.</p>
2442 <pre class="doc_code">
2453 <p>These logical operations have bits that are not always affected by the input.
2454 For example, if <tt>%X</tt> has a zero bit, then the output of the
2455 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2456 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2457 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2458 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2459 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2460 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2461 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2463 <pre class="doc_code">
2464 %A = select undef, %X, %Y
2465 %B = select undef, 42, %Y
2466 %C = select %X, %Y, undef
2477 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2478 branch) conditions can go <em>either way</em>, but they have to come from one
2479 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2480 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2481 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2482 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2483 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2486 <pre class="doc_code">
2487 %A = xor undef, undef
2505 <p>This example points out that two '<tt>undef</tt>' operands are not
2506 necessarily the same. This can be surprising to people (and also matches C
2507 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2508 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2509 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2510 its value over its "live range". This is true because the variable doesn't
2511 actually <em>have a live range</em>. Instead, the value is logically read
2512 from arbitrary registers that happen to be around when needed, so the value
2513 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2514 need to have the same semantics or the core LLVM "replace all uses with"
2515 concept would not hold.</p>
2517 <pre class="doc_code">
2525 <p>These examples show the crucial difference between an <em>undefined
2526 value</em> and <em>undefined behavior</em>. An undefined value (like
2527 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2528 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2529 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2530 defined on SNaN's. However, in the second example, we can make a more
2531 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2532 arbitrary value, we are allowed to assume that it could be zero. Since a
2533 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2534 the operation does not execute at all. This allows us to delete the divide and
2535 all code after it. Because the undefined operation "can't happen", the
2536 optimizer can assume that it occurs in dead code.</p>
2538 <pre class="doc_code">
2539 a: store undef -> %X
2540 b: store %X -> undef
2546 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2547 undefined value can be assumed to not have any effect; we can assume that the
2548 value is overwritten with bits that happen to match what was already there.
2549 However, a store <em>to</em> an undefined location could clobber arbitrary
2550 memory, therefore, it has undefined behavior.</p>
2554 <!-- ======================================================================= -->
2556 <a name="poisonvalues">Poison Values</a>
2561 <p>Poison values are similar to <a href="#undefvalues">undef values</a>, however
2562 they also represent the fact that an instruction or constant expression which
2563 cannot evoke side effects has nevertheless detected a condition which results
2564 in undefined behavior.</p>
2566 <p>There is currently no way of representing a poison value in the IR; they
2567 only exist when produced by operations such as
2568 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2570 <p>Poison value behavior is defined in terms of value <i>dependence</i>:</p>
2573 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2574 their operands.</li>
2576 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2577 to their dynamic predecessor basic block.</li>
2579 <li>Function arguments depend on the corresponding actual argument values in
2580 the dynamic callers of their functions.</li>
2582 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2583 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2584 control back to them.</li>
2586 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2587 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>,
2588 or exception-throwing call instructions that dynamically transfer control
2591 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2592 referenced memory addresses, following the order in the IR
2593 (including loads and stores implied by intrinsics such as
2594 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2596 <!-- TODO: In the case of multiple threads, this only applies if the store
2597 "happens-before" the load or store. -->
2599 <!-- TODO: floating-point exception state -->
2601 <li>An instruction with externally visible side effects depends on the most
2602 recent preceding instruction with externally visible side effects, following
2603 the order in the IR. (This includes
2604 <a href="#volatile">volatile operations</a>.)</li>
2606 <li>An instruction <i>control-depends</i> on a
2607 <a href="#terminators">terminator instruction</a>
2608 if the terminator instruction has multiple successors and the instruction
2609 is always executed when control transfers to one of the successors, and
2610 may not be executed when control is transferred to another.</li>
2612 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2613 instruction if the set of instructions it otherwise depends on would be
2614 different if the terminator had transferred control to a different
2617 <li>Dependence is transitive.</li>
2621 <p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>,
2622 with the additional affect that any instruction which has a <i>dependence</i>
2623 on a poison value has undefined behavior.</p>
2625 <p>Here are some examples:</p>
2627 <pre class="doc_code">
2629 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2630 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2631 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2632 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2634 store i32 %poison, i32* @g ; Poison value stored to memory.
2635 %poison2 = load i32* @g ; Poison value loaded back from memory.
2637 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2639 %narrowaddr = bitcast i32* @g to i16*
2640 %wideaddr = bitcast i32* @g to i64*
2641 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2642 %poison4 = load i64* %wideaddr ; Returns a poison value.
2644 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2645 br i1 %cmp, label %true, label %end ; Branch to either destination.
2648 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2649 ; it has undefined behavior.
2653 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2654 ; Both edges into this PHI are
2655 ; control-dependent on %cmp, so this
2656 ; always results in a poison value.
2658 store volatile i32 0, i32* @g ; This would depend on the store in %true
2659 ; if %cmp is true, or the store in %entry
2660 ; otherwise, so this is undefined behavior.
2662 br i1 %cmp, label %second_true, label %second_end
2663 ; The same branch again, but this time the
2664 ; true block doesn't have side effects.
2671 store volatile i32 0, i32* @g ; This time, the instruction always depends
2672 ; on the store in %end. Also, it is
2673 ; control-equivalent to %end, so this is
2674 ; well-defined (ignoring earlier undefined
2675 ; behavior in this example).
2680 <!-- ======================================================================= -->
2682 <a name="blockaddress">Addresses of Basic Blocks</a>
2687 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2689 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2690 basic block in the specified function, and always has an i8* type. Taking
2691 the address of the entry block is illegal.</p>
2693 <p>This value only has defined behavior when used as an operand to the
2694 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2695 comparisons against null. Pointer equality tests between labels addresses
2696 results in undefined behavior — though, again, comparison against null
2697 is ok, and no label is equal to the null pointer. This may be passed around
2698 as an opaque pointer sized value as long as the bits are not inspected. This
2699 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2700 long as the original value is reconstituted before the <tt>indirectbr</tt>
2703 <p>Finally, some targets may provide defined semantics when using the value as
2704 the operand to an inline assembly, but that is target specific.</p>
2709 <!-- ======================================================================= -->
2711 <a name="constantexprs">Constant Expressions</a>
2716 <p>Constant expressions are used to allow expressions involving other constants
2717 to be used as constants. Constant expressions may be of
2718 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2719 operation that does not have side effects (e.g. load and call are not
2720 supported). The following is the syntax for constant expressions:</p>
2723 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2724 <dd>Truncate a constant to another type. The bit size of CST must be larger
2725 than the bit size of TYPE. Both types must be integers.</dd>
2727 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2728 <dd>Zero extend a constant to another type. The bit size of CST must be
2729 smaller than the bit size of TYPE. Both types must be integers.</dd>
2731 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2732 <dd>Sign extend a constant to another type. The bit size of CST must be
2733 smaller than the bit size of TYPE. Both types must be integers.</dd>
2735 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2736 <dd>Truncate a floating point constant to another floating point type. The
2737 size of CST must be larger than the size of TYPE. Both types must be
2738 floating point.</dd>
2740 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2741 <dd>Floating point extend a constant to another type. The size of CST must be
2742 smaller or equal to the size of TYPE. Both types must be floating
2745 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2746 <dd>Convert a floating point constant to the corresponding unsigned integer
2747 constant. TYPE must be a scalar or vector integer type. CST must be of
2748 scalar or vector floating point type. Both CST and TYPE must be scalars,
2749 or vectors of the same number of elements. If the value won't fit in the
2750 integer type, the results are undefined.</dd>
2752 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2753 <dd>Convert a floating point constant to the corresponding signed integer
2754 constant. TYPE must be a scalar or vector integer type. CST must be of
2755 scalar or vector floating point type. Both CST and TYPE must be scalars,
2756 or vectors of the same number of elements. If the value won't fit in the
2757 integer type, the results are undefined.</dd>
2759 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2760 <dd>Convert an unsigned integer constant to the corresponding floating point
2761 constant. TYPE must be a scalar or vector floating point type. CST must be
2762 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2763 vectors of the same number of elements. If the value won't fit in the
2764 floating point type, the results are undefined.</dd>
2766 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2767 <dd>Convert a signed integer constant to the corresponding floating point
2768 constant. TYPE must be a scalar or vector floating point type. CST must be
2769 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2770 vectors of the same number of elements. If the value won't fit in the
2771 floating point type, the results are undefined.</dd>
2773 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2774 <dd>Convert a pointer typed constant to the corresponding integer constant
2775 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2776 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2777 make it fit in <tt>TYPE</tt>.</dd>
2779 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2780 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2781 type. CST must be of integer type. The CST value is zero extended,
2782 truncated, or unchanged to make it fit in a pointer size. This one is
2783 <i>really</i> dangerous!</dd>
2785 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2786 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2787 are the same as those for the <a href="#i_bitcast">bitcast
2788 instruction</a>.</dd>
2790 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2791 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2792 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2793 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2794 instruction, the index list may have zero or more indexes, which are
2795 required to make sense for the type of "CSTPTR".</dd>
2797 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2798 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2800 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2801 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2803 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2804 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2806 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2807 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2810 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2811 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2814 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2815 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2818 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2819 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2820 constants. The index list is interpreted in a similar manner as indices in
2821 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2822 index value must be specified.</dd>
2824 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2825 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2826 constants. The index list is interpreted in a similar manner as indices in
2827 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2828 index value must be specified.</dd>
2830 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2831 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2832 be any of the <a href="#binaryops">binary</a>
2833 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2834 on operands are the same as those for the corresponding instruction
2835 (e.g. no bitwise operations on floating point values are allowed).</dd>
2842 <!-- *********************************************************************** -->
2843 <h2><a name="othervalues">Other Values</a></h2>
2844 <!-- *********************************************************************** -->
2846 <!-- ======================================================================= -->
2848 <a name="inlineasm">Inline Assembler Expressions</a>
2853 <p>LLVM supports inline assembler expressions (as opposed
2854 to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of
2855 a special value. This value represents the inline assembler as a string
2856 (containing the instructions to emit), a list of operand constraints (stored
2857 as a string), a flag that indicates whether or not the inline asm
2858 expression has side effects, and a flag indicating whether the function
2859 containing the asm needs to align its stack conservatively. An example
2860 inline assembler expression is:</p>
2862 <pre class="doc_code">
2863 i32 (i32) asm "bswap $0", "=r,r"
2866 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2867 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2870 <pre class="doc_code">
2871 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2874 <p>Inline asms with side effects not visible in the constraint list must be
2875 marked as having side effects. This is done through the use of the
2876 '<tt>sideeffect</tt>' keyword, like so:</p>
2878 <pre class="doc_code">
2879 call void asm sideeffect "eieio", ""()
2882 <p>In some cases inline asms will contain code that will not work unless the
2883 stack is aligned in some way, such as calls or SSE instructions on x86,
2884 yet will not contain code that does that alignment within the asm.
2885 The compiler should make conservative assumptions about what the asm might
2886 contain and should generate its usual stack alignment code in the prologue
2887 if the '<tt>alignstack</tt>' keyword is present:</p>
2889 <pre class="doc_code">
2890 call void asm alignstack "eieio", ""()
2893 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2897 <p>TODO: The format of the asm and constraints string still need to be
2898 documented here. Constraints on what can be done (e.g. duplication, moving,
2899 etc need to be documented). This is probably best done by reference to
2900 another document that covers inline asm from a holistic perspective.</p>
2903 <!-- _______________________________________________________________________ -->
2905 <a name="inlineasm_md">Inline Asm Metadata</a>
2910 <p>The call instructions that wrap inline asm nodes may have a
2911 "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant
2912 integers. If present, the code generator will use the integer as the
2913 location cookie value when report errors through the <tt>LLVMContext</tt>
2914 error reporting mechanisms. This allows a front-end to correlate backend
2915 errors that occur with inline asm back to the source code that produced it.
2918 <pre class="doc_code">
2919 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2921 !42 = !{ i32 1234567 }
2924 <p>It is up to the front-end to make sense of the magic numbers it places in the
2925 IR. If the MDNode contains multiple constants, the code generator will use
2926 the one that corresponds to the line of the asm that the error occurs on.</p>
2932 <!-- ======================================================================= -->
2934 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2939 <p>LLVM IR allows metadata to be attached to instructions in the program that
2940 can convey extra information about the code to the optimizers and code
2941 generator. One example application of metadata is source-level debug
2942 information. There are two metadata primitives: strings and nodes. All
2943 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2944 preceding exclamation point ('<tt>!</tt>').</p>
2946 <p>A metadata string is a string surrounded by double quotes. It can contain
2947 any character by escaping non-printable characters with "<tt>\xx</tt>" where
2948 "<tt>xx</tt>" is the two digit hex code. For example:
2949 "<tt>!"test\00"</tt>".</p>
2951 <p>Metadata nodes are represented with notation similar to structure constants
2952 (a comma separated list of elements, surrounded by braces and preceded by an
2953 exclamation point). Metadata nodes can have any values as their operand. For
2956 <div class="doc_code">
2958 !{ metadata !"test\00", i32 10}
2962 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2963 metadata nodes, which can be looked up in the module symbol table. For
2966 <div class="doc_code">
2968 !foo = metadata !{!4, !3}
2972 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2973 function is using two metadata arguments:</p>
2975 <div class="doc_code">
2977 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2981 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2982 attached to the <tt>add</tt> instruction using the <tt>!dbg</tt>
2985 <div class="doc_code">
2987 %indvar.next = add i64 %indvar, 1, !dbg !21
2991 <p>More information about specific metadata nodes recognized by the optimizers
2992 and code generator is found below.</p>
2994 <!-- _______________________________________________________________________ -->
2996 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
3001 <p>In LLVM IR, memory does not have types, so LLVM's own type system is not
3002 suitable for doing TBAA. Instead, metadata is added to the IR to describe
3003 a type system of a higher level language. This can be used to implement
3004 typical C/C++ TBAA, but it can also be used to implement custom alias
3005 analysis behavior for other languages.</p>
3007 <p>The current metadata format is very simple. TBAA metadata nodes have up to
3008 three fields, e.g.:</p>
3010 <div class="doc_code">
3012 !0 = metadata !{ metadata !"an example type tree" }
3013 !1 = metadata !{ metadata !"int", metadata !0 }
3014 !2 = metadata !{ metadata !"float", metadata !0 }
3015 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
3019 <p>The first field is an identity field. It can be any value, usually
3020 a metadata string, which uniquely identifies the type. The most important
3021 name in the tree is the name of the root node. Two trees with
3022 different root node names are entirely disjoint, even if they
3023 have leaves with common names.</p>
3025 <p>The second field identifies the type's parent node in the tree, or
3026 is null or omitted for a root node. A type is considered to alias
3027 all of its descendants and all of its ancestors in the tree. Also,
3028 a type is considered to alias all types in other trees, so that
3029 bitcode produced from multiple front-ends is handled conservatively.</p>
3031 <p>If the third field is present, it's an integer which if equal to 1
3032 indicates that the type is "constant" (meaning
3033 <tt>pointsToConstantMemory</tt> should return true; see
3034 <a href="AliasAnalysis.html#OtherItfs">other useful
3035 <tt>AliasAnalysis</tt> methods</a>).</p>
3039 <!-- _______________________________________________________________________ -->
3041 <a name="fpmath">'<tt>fpmath</tt>' Metadata</a>
3046 <p><tt>fpmath</tt> metadata may be attached to any instruction of floating point
3047 type. It can be used to express the maximum acceptable error in the result of
3048 that instruction, in ULPs, thus potentially allowing the compiler to use a
3049 more efficient but less accurate method of computing it. ULP is defined as
3054 <p>If <tt>x</tt> is a real number that lies between two finite consecutive
3055 floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one
3056 of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the
3057 distance between the two non-equal finite floating-point numbers nearest
3058 <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p>
3062 <p>The metadata node shall consist of a single positive floating point number
3063 representing the maximum relative error, for example:</p>
3065 <div class="doc_code">
3067 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3073 <!-- _______________________________________________________________________ -->
3075 <a name="range">'<tt>range</tt>' Metadata</a>
3079 <p><tt>range</tt> metadata may be attached only to loads of integer types. It
3080 expresses the possible ranges the loaded value is in. The ranges are
3081 represented with a flattened list of integers. The loaded value is known to
3082 be in the union of the ranges defined by each consecutive pair. Each pair
3083 has the following properties:</p>
3085 <li>The type must match the type loaded by the instruction.</li>
3086 <li>The pair <tt>a,b</tt> represents the range <tt>[a,b)</tt>.</li>
3087 <li>Both <tt>a</tt> and <tt>b</tt> are constants.</li>
3088 <li>The range is allowed to wrap.</li>
3089 <li>The range should not represent the full or empty set. That is,
3090 <tt>a!=b</tt>. </li>
3092 <p> In addition, the pairs must be in signed order of the lower bound and
3093 they must be non-contiguous.</p>
3096 <div class="doc_code">
3098 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
3099 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3100 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
3101 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
3103 !0 = metadata !{ i8 0, i8 2 }
3104 !1 = metadata !{ i8 255, i8 2 }
3105 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
3106 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
3114 <!-- *********************************************************************** -->
3116 <a name="module_flags">Module Flags Metadata</a>
3118 <!-- *********************************************************************** -->
3122 <p>Information about the module as a whole is difficult to convey to LLVM's
3123 subsystems. The LLVM IR isn't sufficient to transmit this
3124 information. The <tt>llvm.module.flags</tt> named metadata exists in order to
3125 facilitate this. These flags are in the form of key / value pairs —
3126 much like a dictionary — making it easy for any subsystem who cares
3127 about a flag to look it up.</p>
3129 <p>The <tt>llvm.module.flags</tt> metadata contains a list of metadata
3130 triplets. Each triplet has the following form:</p>
3133 <li>The first element is a <i>behavior</i> flag, which specifies the behavior
3134 when two (or more) modules are merged together, and it encounters two (or
3135 more) metadata with the same ID. The supported behaviors are described
3138 <li>The second element is a metadata string that is a unique ID for the
3139 metadata. How each ID is interpreted is documented below.</li>
3141 <li>The third element is the value of the flag.</li>
3144 <p>When two (or more) modules are merged together, the resulting
3145 <tt>llvm.module.flags</tt> metadata is the union of the
3146 modules' <tt>llvm.module.flags</tt> metadata. The only exception being a flag
3147 with the <i>Override</i> behavior, which may override another flag's value
3150 <p>The following behaviors are supported:</p>
3152 <table border="1" cellspacing="0" cellpadding="4">
3162 <dt><b>Error</b></dt>
3163 <dd>Emits an error if two values disagree. It is an error to have an ID
3164 with both an Error and a Warning behavior.</dd>
3172 <dt><b>Warning</b></dt>
3173 <dd>Emits a warning if two values disagree.</dd>
3181 <dt><b>Require</b></dt>
3182 <dd>Emits an error when the specified value is not present or doesn't
3183 have the specified value. It is an error for two (or more)
3184 <tt>llvm.module.flags</tt> with the same ID to have the Require
3185 behavior but different values. There may be multiple Require flags
3194 <dt><b>Override</b></dt>
3195 <dd>Uses the specified value if the two values disagree. It is an
3196 error for two (or more) <tt>llvm.module.flags</tt> with the same
3197 ID to have the Override behavior but different values.</dd>
3204 <p>An example of module flags:</p>
3206 <pre class="doc_code">
3207 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3208 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3209 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3210 !3 = metadata !{ i32 3, metadata !"qux",
3212 metadata !"foo", i32 1
3215 !llvm.module.flags = !{ !0, !1, !2, !3 }
3219 <li><p>Metadata <tt>!0</tt> has the ID <tt>!"foo"</tt> and the value '1'. The
3220 behavior if two or more <tt>!"foo"</tt> flags are seen is to emit an
3221 error if their values are not equal.</p></li>
3223 <li><p>Metadata <tt>!1</tt> has the ID <tt>!"bar"</tt> and the value '37'. The
3224 behavior if two or more <tt>!"bar"</tt> flags are seen is to use the
3225 value '37' if their values are not equal.</p></li>
3227 <li><p>Metadata <tt>!2</tt> has the ID <tt>!"qux"</tt> and the value '42'. The
3228 behavior if two or more <tt>!"qux"</tt> flags are seen is to emit a
3229 warning if their values are not equal.</p></li>
3231 <li><p>Metadata <tt>!3</tt> has the ID <tt>!"qux"</tt> and the value:</p>
3233 <pre class="doc_code">
3234 metadata !{ metadata !"foo", i32 1 }
3237 <p>The behavior is to emit an error if the <tt>llvm.module.flags</tt> does
3238 not contain a flag with the ID <tt>!"foo"</tt> that has the value
3239 '1'. If two or more <tt>!"qux"</tt> flags exist, then they must have
3240 the same value or an error will be issued.</p></li>
3244 <!-- ======================================================================= -->
3246 <a name="objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a>
3251 <p>On the Mach-O platform, Objective-C stores metadata about garbage collection
3252 in a special section called "image info". The metadata consists of a version
3253 number and a bitmask specifying what types of garbage collection are
3254 supported (if any) by the file. If two or more modules are linked together
3255 their garbage collection metadata needs to be merged rather than appended
3258 <p>The Objective-C garbage collection module flags metadata consists of the
3259 following key-value pairs:</p>
3261 <table border="1" cellspacing="0" cellpadding="4">
3269 <td><tt>Objective-C Version</tt></td>
3270 <td align="left"><b>[Required]</b> — The Objective-C ABI
3271 version. Valid values are 1 and 2.</td>
3274 <td><tt>Objective-C Image Info Version</tt></td>
3275 <td align="left"><b>[Required]</b> — The version of the image info
3276 section. Currently always 0.</td>
3279 <td><tt>Objective-C Image Info Section</tt></td>
3280 <td align="left"><b>[Required]</b> — The section to place the
3281 metadata. Valid values are <tt>"__OBJC, __image_info, regular"</tt> for
3282 Objective-C ABI version 1, and <tt>"__DATA,__objc_imageinfo, regular,
3283 no_dead_strip"</tt> for Objective-C ABI version 2.</td>
3286 <td><tt>Objective-C Garbage Collection</tt></td>
3287 <td align="left"><b>[Required]</b> — Specifies whether garbage
3288 collection is supported or not. Valid values are 0, for no garbage
3289 collection, and 2, for garbage collection supported.</td>
3292 <td><tt>Objective-C GC Only</tt></td>
3293 <td align="left"><b>[Optional]</b> — Specifies that only garbage
3294 collection is supported. If present, its value must be 6. This flag
3295 requires that the <tt>Objective-C Garbage Collection</tt> flag have the
3301 <p>Some important flag interactions:</p>
3304 <li>If a module with <tt>Objective-C Garbage Collection</tt> set to 0 is
3305 merged with a module with <tt>Objective-C Garbage Collection</tt> set to
3306 2, then the resulting module has the <tt>Objective-C Garbage
3307 Collection</tt> flag set to 0.</li>
3309 <li>A module with <tt>Objective-C Garbage Collection</tt> set to 0 cannot be
3310 merged with a module with <tt>Objective-C GC Only</tt> set to 6.</li>
3317 <!-- *********************************************************************** -->
3319 <a name="intrinsic_globals">Intrinsic Global Variables</a>
3321 <!-- *********************************************************************** -->
3323 <p>LLVM has a number of "magic" global variables that contain data that affect
3324 code generation or other IR semantics. These are documented here. All globals
3325 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
3326 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
3329 <!-- ======================================================================= -->
3331 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
3336 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
3337 href="#linkage_appending">appending linkage</a>. This array contains a list of
3338 pointers to global variables and functions which may optionally have a pointer
3339 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
3341 <div class="doc_code">
3346 @llvm.used = appending global [2 x i8*] [
3348 i8* bitcast (i32* @Y to i8*)
3349 ], section "llvm.metadata"
3353 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
3354 compiler, assembler, and linker are required to treat the symbol as if there
3355 is a reference to the global that it cannot see. For example, if a variable
3356 has internal linkage and no references other than that from
3357 the <tt>@llvm.used</tt> list, it cannot be deleted. This is commonly used to
3358 represent references from inline asms and other things the compiler cannot
3359 "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p>
3361 <p>On some targets, the code generator must emit a directive to the assembler or
3362 object file to prevent the assembler and linker from molesting the
3367 <!-- ======================================================================= -->
3369 <a name="intg_compiler_used">
3370 The '<tt>llvm.compiler.used</tt>' Global Variable
3376 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
3377 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
3378 touching the symbol. On targets that support it, this allows an intelligent
3379 linker to optimize references to the symbol without being impeded as it would
3380 be by <tt>@llvm.used</tt>.</p>
3382 <p>This is a rare construct that should only be used in rare circumstances, and
3383 should not be exposed to source languages.</p>
3387 <!-- ======================================================================= -->
3389 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
3394 <div class="doc_code">
3396 %0 = type { i32, void ()* }
3397 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3401 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor
3402 functions and associated priorities. The functions referenced by this array
3403 will be called in ascending order of priority (i.e. lowest first) when the
3404 module is loaded. The order of functions with the same priority is not
3409 <!-- ======================================================================= -->
3411 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3416 <div class="doc_code">
3418 %0 = type { i32, void ()* }
3419 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3423 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions
3424 and associated priorities. The functions referenced by this array will be
3425 called in descending order of priority (i.e. highest first) when the module
3426 is loaded. The order of functions with the same priority is not defined.</p>
3432 <!-- *********************************************************************** -->
3433 <h2><a name="instref">Instruction Reference</a></h2>
3434 <!-- *********************************************************************** -->
3438 <p>The LLVM instruction set consists of several different classifications of
3439 instructions: <a href="#terminators">terminator
3440 instructions</a>, <a href="#binaryops">binary instructions</a>,
3441 <a href="#bitwiseops">bitwise binary instructions</a>,
3442 <a href="#memoryops">memory instructions</a>, and
3443 <a href="#otherops">other instructions</a>.</p>
3445 <!-- ======================================================================= -->
3447 <a name="terminators">Terminator Instructions</a>
3452 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3453 in a program ends with a "Terminator" instruction, which indicates which
3454 block should be executed after the current block is finished. These
3455 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3456 control flow, not values (the one exception being the
3457 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3459 <p>The terminator instructions are:
3460 '<a href="#i_ret"><tt>ret</tt></a>',
3461 '<a href="#i_br"><tt>br</tt></a>',
3462 '<a href="#i_switch"><tt>switch</tt></a>',
3463 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3464 '<a href="#i_invoke"><tt>invoke</tt></a>',
3465 '<a href="#i_resume"><tt>resume</tt></a>', and
3466 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3468 <!-- _______________________________________________________________________ -->
3470 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3477 ret <type> <value> <i>; Return a value from a non-void function</i>
3478 ret void <i>; Return from void function</i>
3482 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3483 a value) from a function back to the caller.</p>
3485 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3486 value and then causes control flow, and one that just causes control flow to
3490 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3491 return value. The type of the return value must be a
3492 '<a href="#t_firstclass">first class</a>' type.</p>
3494 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3495 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3496 value or a return value with a type that does not match its type, or if it
3497 has a void return type and contains a '<tt>ret</tt>' instruction with a
3501 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3502 the calling function's context. If the caller is a
3503 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3504 instruction after the call. If the caller was an
3505 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3506 the beginning of the "normal" destination block. If the instruction returns
3507 a value, that value shall set the call or invoke instruction's return
3512 ret i32 5 <i>; Return an integer value of 5</i>
3513 ret void <i>; Return from a void function</i>
3514 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3518 <!-- _______________________________________________________________________ -->
3520 <a name="i_br">'<tt>br</tt>' Instruction</a>
3527 br i1 <cond>, label <iftrue>, label <iffalse>
3528 br label <dest> <i>; Unconditional branch</i>
3532 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3533 different basic block in the current function. There are two forms of this
3534 instruction, corresponding to a conditional branch and an unconditional
3538 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3539 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3540 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3544 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3545 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3546 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3547 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3552 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3553 br i1 %cond, label %IfEqual, label %IfUnequal
3555 <a href="#i_ret">ret</a> i32 1
3557 <a href="#i_ret">ret</a> i32 0
3562 <!-- _______________________________________________________________________ -->
3564 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3571 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3575 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3576 several different places. It is a generalization of the '<tt>br</tt>'
3577 instruction, allowing a branch to occur to one of many possible
3581 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3582 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3583 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3584 The table is not allowed to contain duplicate constant entries.</p>
3587 <p>The <tt>switch</tt> instruction specifies a table of values and
3588 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3589 is searched for the given value. If the value is found, control flow is
3590 transferred to the corresponding destination; otherwise, control flow is
3591 transferred to the default destination.</p>
3593 <h5>Implementation:</h5>
3594 <p>Depending on properties of the target machine and the particular
3595 <tt>switch</tt> instruction, this instruction may be code generated in
3596 different ways. For example, it could be generated as a series of chained
3597 conditional branches or with a lookup table.</p>
3601 <i>; Emulate a conditional br instruction</i>
3602 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3603 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3605 <i>; Emulate an unconditional br instruction</i>
3606 switch i32 0, label %dest [ ]
3608 <i>; Implement a jump table:</i>
3609 switch i32 %val, label %otherwise [ i32 0, label %onzero
3611 i32 2, label %ontwo ]
3617 <!-- _______________________________________________________________________ -->
3619 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3626 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3631 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3632 within the current function, whose address is specified by
3633 "<tt>address</tt>". Address must be derived from a <a
3634 href="#blockaddress">blockaddress</a> constant.</p>
3638 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3639 rest of the arguments indicate the full set of possible destinations that the
3640 address may point to. Blocks are allowed to occur multiple times in the
3641 destination list, though this isn't particularly useful.</p>
3643 <p>This destination list is required so that dataflow analysis has an accurate
3644 understanding of the CFG.</p>
3648 <p>Control transfers to the block specified in the address argument. All
3649 possible destination blocks must be listed in the label list, otherwise this
3650 instruction has undefined behavior. This implies that jumps to labels
3651 defined in other functions have undefined behavior as well.</p>
3653 <h5>Implementation:</h5>
3655 <p>This is typically implemented with a jump through a register.</p>
3659 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3665 <!-- _______________________________________________________________________ -->
3667 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3674 <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>]
3675 to label <normal label> unwind label <exception label>
3679 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3680 function, with the possibility of control flow transfer to either the
3681 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3682 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3683 control flow will return to the "normal" label. If the callee (or any
3684 indirect callees) returns via the "<a href="#i_resume"><tt>resume</tt></a>"
3685 instruction or other exception handling mechanism, control is interrupted and
3686 continued at the dynamically nearest "exception" label.</p>
3688 <p>The '<tt>exception</tt>' label is a
3689 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3690 exception. As such, '<tt>exception</tt>' label is required to have the
3691 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3692 the information about the behavior of the program after unwinding
3693 happens, as its first non-PHI instruction. The restrictions on the
3694 "<tt>landingpad</tt>" instruction's tightly couples it to the
3695 "<tt>invoke</tt>" instruction, so that the important information contained
3696 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3700 <p>This instruction requires several arguments:</p>
3703 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3704 convention</a> the call should use. If none is specified, the call
3705 defaults to using C calling conventions.</li>
3707 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3708 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3709 '<tt>inreg</tt>' attributes are valid here.</li>
3711 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3712 function value being invoked. In most cases, this is a direct function
3713 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3714 off an arbitrary pointer to function value.</li>
3716 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3717 function to be invoked. </li>
3719 <li>'<tt>function args</tt>': argument list whose types match the function
3720 signature argument types and parameter attributes. All arguments must be
3721 of <a href="#t_firstclass">first class</a> type. If the function
3722 signature indicates the function accepts a variable number of arguments,
3723 the extra arguments can be specified.</li>
3725 <li>'<tt>normal label</tt>': the label reached when the called function
3726 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3728 <li>'<tt>exception label</tt>': the label reached when a callee returns via
3729 the <a href="#i_resume"><tt>resume</tt></a> instruction or other exception
3730 handling mechanism.</li>
3732 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3733 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3734 '<tt>readnone</tt>' attributes are valid here.</li>
3738 <p>This instruction is designed to operate as a standard
3739 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3740 primary difference is that it establishes an association with a label, which
3741 is used by the runtime library to unwind the stack.</p>
3743 <p>This instruction is used in languages with destructors to ensure that proper
3744 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3745 exception. Additionally, this is important for implementation of
3746 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3748 <p>For the purposes of the SSA form, the definition of the value returned by the
3749 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3750 block to the "normal" label. If the callee unwinds then no return value is
3755 %retval = invoke i32 @Test(i32 15) to label %Continue
3756 unwind label %TestCleanup <i>; {i32}:retval set</i>
3757 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3758 unwind label %TestCleanup <i>; {i32}:retval set</i>
3763 <!-- _______________________________________________________________________ -->
3766 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3773 resume <type> <value>
3777 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3781 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3782 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3786 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3787 (in-flight) exception whose unwinding was interrupted with
3788 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3792 resume { i8*, i32 } %exn
3797 <!-- _______________________________________________________________________ -->
3800 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3811 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3812 instruction is used to inform the optimizer that a particular portion of the
3813 code is not reachable. This can be used to indicate that the code after a
3814 no-return function cannot be reached, and other facts.</p>
3817 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3823 <!-- ======================================================================= -->
3825 <a name="binaryops">Binary Operations</a>
3830 <p>Binary operators are used to do most of the computation in a program. They
3831 require two operands of the same type, execute an operation on them, and
3832 produce a single value. The operands might represent multiple data, as is
3833 the case with the <a href="#t_vector">vector</a> data type. The result value
3834 has the same type as its operands.</p>
3836 <p>There are several different binary operators:</p>
3838 <!-- _______________________________________________________________________ -->
3840 <a name="i_add">'<tt>add</tt>' Instruction</a>
3847 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3848 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3849 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3850 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3854 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3857 <p>The two arguments to the '<tt>add</tt>' instruction must
3858 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3859 integer values. Both arguments must have identical types.</p>
3862 <p>The value produced is the integer sum of the two operands.</p>
3864 <p>If the sum has unsigned overflow, the result returned is the mathematical
3865 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3867 <p>Because LLVM integers use a two's complement representation, this instruction
3868 is appropriate for both signed and unsigned integers.</p>
3870 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3871 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3872 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3873 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3874 respectively, occurs.</p>
3878 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3883 <!-- _______________________________________________________________________ -->
3885 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3892 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3896 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3899 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3900 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3901 floating point values. Both arguments must have identical types.</p>
3904 <p>The value produced is the floating point sum of the two operands.</p>
3908 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3913 <!-- _______________________________________________________________________ -->
3915 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3922 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3923 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3924 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3925 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3929 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3932 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3933 '<tt>neg</tt>' instruction present in most other intermediate
3934 representations.</p>
3937 <p>The two arguments to the '<tt>sub</tt>' instruction must
3938 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3939 integer values. Both arguments must have identical types.</p>
3942 <p>The value produced is the integer difference of the two operands.</p>
3944 <p>If the difference has unsigned overflow, the result returned is the
3945 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3948 <p>Because LLVM integers use a two's complement representation, this instruction
3949 is appropriate for both signed and unsigned integers.</p>
3951 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3952 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3953 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3954 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3955 respectively, occurs.</p>
3959 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3960 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3965 <!-- _______________________________________________________________________ -->
3967 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3974 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3978 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3981 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3982 '<tt>fneg</tt>' instruction present in most other intermediate
3983 representations.</p>
3986 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3987 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3988 floating point values. Both arguments must have identical types.</p>
3991 <p>The value produced is the floating point difference of the two operands.</p>
3995 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3996 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
4001 <!-- _______________________________________________________________________ -->
4003 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
4010 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4011 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4012 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4013 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4017 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
4020 <p>The two arguments to the '<tt>mul</tt>' instruction must
4021 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4022 integer values. Both arguments must have identical types.</p>
4025 <p>The value produced is the integer product of the two operands.</p>
4027 <p>If the result of the multiplication has unsigned overflow, the result
4028 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
4029 width of the result.</p>
4031 <p>Because LLVM integers use a two's complement representation, and the result
4032 is the same width as the operands, this instruction returns the correct
4033 result for both signed and unsigned integers. If a full product
4034 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
4035 be sign-extended or zero-extended as appropriate to the width of the full
4038 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
4039 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
4040 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
4041 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4042 respectively, occurs.</p>
4046 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
4051 <!-- _______________________________________________________________________ -->
4053 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
4060 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4064 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
4067 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
4068 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4069 floating point values. Both arguments must have identical types.</p>
4072 <p>The value produced is the floating point product of the two operands.</p>
4076 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
4081 <!-- _______________________________________________________________________ -->
4083 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
4090 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4091 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4095 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
4098 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
4099 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4100 values. Both arguments must have identical types.</p>
4103 <p>The value produced is the unsigned integer quotient of the two operands.</p>
4105 <p>Note that unsigned integer division and signed integer division are distinct
4106 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
4108 <p>Division by zero leads to undefined behavior.</p>
4110 <p>If the <tt>exact</tt> keyword is present, the result value of the
4111 <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a
4112 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
4117 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4122 <!-- _______________________________________________________________________ -->
4124 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
4131 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4132 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4136 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
4139 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
4140 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4141 values. Both arguments must have identical types.</p>
4144 <p>The value produced is the signed integer quotient of the two operands rounded
4147 <p>Note that signed integer division and unsigned integer division are distinct
4148 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
4150 <p>Division by zero leads to undefined behavior. Overflow also leads to
4151 undefined behavior; this is a rare case, but can occur, for example, by doing
4152 a 32-bit division of -2147483648 by -1.</p>
4154 <p>If the <tt>exact</tt> keyword is present, the result value of the
4155 <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would
4160 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4165 <!-- _______________________________________________________________________ -->
4167 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
4174 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4178 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
4181 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
4182 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4183 floating point values. Both arguments must have identical types.</p>
4186 <p>The value produced is the floating point quotient of the two operands.</p>
4190 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
4195 <!-- _______________________________________________________________________ -->
4197 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
4204 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4208 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
4209 division of its two arguments.</p>
4212 <p>The two arguments to the '<tt>urem</tt>' instruction must be
4213 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4214 values. Both arguments must have identical types.</p>
4217 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
4218 This instruction always performs an unsigned division to get the
4221 <p>Note that unsigned integer remainder and signed integer remainder are
4222 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
4224 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
4228 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4233 <!-- _______________________________________________________________________ -->
4235 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
4242 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4246 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
4247 division of its two operands. This instruction can also take
4248 <a href="#t_vector">vector</a> versions of the values in which case the
4249 elements must be integers.</p>
4252 <p>The two arguments to the '<tt>srem</tt>' instruction must be
4253 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4254 values. Both arguments must have identical types.</p>
4257 <p>This instruction returns the <i>remainder</i> of a division (where the result
4258 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
4259 <i>modulo</i> operator (where the result is either zero or has the same sign
4260 as the divisor, <tt>op2</tt>) of a value.
4261 For more information about the difference,
4262 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
4263 Math Forum</a>. For a table of how this is implemented in various languages,
4264 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
4265 Wikipedia: modulo operation</a>.</p>
4267 <p>Note that signed integer remainder and unsigned integer remainder are
4268 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
4270 <p>Taking the remainder of a division by zero leads to undefined behavior.
4271 Overflow also leads to undefined behavior; this is a rare case, but can
4272 occur, for example, by taking the remainder of a 32-bit division of
4273 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
4274 lets srem be implemented using instructions that return both the result of
4275 the division and the remainder.)</p>
4279 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4284 <!-- _______________________________________________________________________ -->
4286 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
4293 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4297 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
4298 its two operands.</p>
4301 <p>The two arguments to the '<tt>frem</tt>' instruction must be
4302 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4303 floating point values. Both arguments must have identical types.</p>
4306 <p>This instruction returns the <i>remainder</i> of a division. The remainder
4307 has the same sign as the dividend.</p>
4311 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
4318 <!-- ======================================================================= -->
4320 <a name="bitwiseops">Bitwise Binary Operations</a>
4325 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
4326 program. They are generally very efficient instructions and can commonly be
4327 strength reduced from other instructions. They require two operands of the
4328 same type, execute an operation on them, and produce a single value. The
4329 resulting value is the same type as its operands.</p>
4331 <!-- _______________________________________________________________________ -->
4333 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
4340 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4341 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4342 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4343 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4347 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
4348 a specified number of bits.</p>
4351 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
4352 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4353 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4356 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
4357 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
4358 is (statically or dynamically) negative or equal to or larger than the number
4359 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4360 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4361 shift amount in <tt>op2</tt>.</p>
4363 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
4364 <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits. If
4365 the <tt>nsw</tt> keyword is present, then the shift produces a
4366 <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree
4367 with the resultant sign bit. As such, NUW/NSW have the same semantics as
4368 they would if the shift were expressed as a mul instruction with the same
4369 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
4373 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
4374 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
4375 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
4376 <result> = shl i32 1, 32 <i>; undefined</i>
4377 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4382 <!-- _______________________________________________________________________ -->
4384 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4391 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4392 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4396 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4397 operand shifted to the right a specified number of bits with zero fill.</p>
4400 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4401 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4402 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4405 <p>This instruction always performs a logical shift right operation. The most
4406 significant bits of the result will be filled with zero bits after the shift.
4407 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4408 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4409 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4410 shift amount in <tt>op2</tt>.</p>
4412 <p>If the <tt>exact</tt> keyword is present, the result value of the
4413 <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4414 shifted out are non-zero.</p>
4419 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4420 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4421 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4422 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4423 <result> = lshr i32 1, 32 <i>; undefined</i>
4424 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4429 <!-- _______________________________________________________________________ -->
4431 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4438 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4439 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4443 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4444 operand shifted to the right a specified number of bits with sign
4448 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4449 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4450 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4453 <p>This instruction always performs an arithmetic shift right operation, The
4454 most significant bits of the result will be filled with the sign bit
4455 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4456 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4457 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4458 the corresponding shift amount in <tt>op2</tt>.</p>
4460 <p>If the <tt>exact</tt> keyword is present, the result value of the
4461 <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4462 shifted out are non-zero.</p>
4466 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4467 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4468 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4469 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4470 <result> = ashr i32 1, 32 <i>; undefined</i>
4471 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4476 <!-- _______________________________________________________________________ -->
4478 <a name="i_and">'<tt>and</tt>' Instruction</a>
4485 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4489 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4493 <p>The two arguments to the '<tt>and</tt>' instruction must be
4494 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4495 values. Both arguments must have identical types.</p>
4498 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4500 <table border="1" cellspacing="0" cellpadding="4">
4532 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4533 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4534 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4537 <!-- _______________________________________________________________________ -->
4539 <a name="i_or">'<tt>or</tt>' Instruction</a>
4546 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4550 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4554 <p>The two arguments to the '<tt>or</tt>' instruction must be
4555 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4556 values. Both arguments must have identical types.</p>
4559 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4561 <table border="1" cellspacing="0" cellpadding="4">
4593 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4594 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4595 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4600 <!-- _______________________________________________________________________ -->
4602 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4609 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4613 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4614 its two operands. The <tt>xor</tt> is used to implement the "one's
4615 complement" operation, which is the "~" operator in C.</p>
4618 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4619 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4620 values. Both arguments must have identical types.</p>
4623 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4625 <table border="1" cellspacing="0" cellpadding="4">
4657 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4658 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4659 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4660 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4667 <!-- ======================================================================= -->
4669 <a name="vectorops">Vector Operations</a>
4674 <p>LLVM supports several instructions to represent vector operations in a
4675 target-independent manner. These instructions cover the element-access and
4676 vector-specific operations needed to process vectors effectively. While LLVM
4677 does directly support these vector operations, many sophisticated algorithms
4678 will want to use target-specific intrinsics to take full advantage of a
4679 specific target.</p>
4681 <!-- _______________________________________________________________________ -->
4683 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4690 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4694 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4695 from a vector at a specified index.</p>
4699 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4700 of <a href="#t_vector">vector</a> type. The second operand is an index
4701 indicating the position from which to extract the element. The index may be
4705 <p>The result is a scalar of the same type as the element type of
4706 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4707 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4708 results are undefined.</p>
4712 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4717 <!-- _______________________________________________________________________ -->
4719 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4726 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4730 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4731 vector at a specified index.</p>
4734 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4735 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4736 whose type must equal the element type of the first operand. The third
4737 operand is an index indicating the position at which to insert the value.
4738 The index may be a variable.</p>
4741 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4742 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4743 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4744 results are undefined.</p>
4748 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4753 <!-- _______________________________________________________________________ -->
4755 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4762 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4766 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4767 from two input vectors, returning a vector with the same element type as the
4768 input and length that is the same as the shuffle mask.</p>
4771 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4772 with the same type. The third argument is a shuffle mask whose
4773 element type is always 'i32'. The result of the instruction is a vector
4774 whose length is the same as the shuffle mask and whose element type is the
4775 same as the element type of the first two operands.</p>
4777 <p>The shuffle mask operand is required to be a constant vector with either
4778 constant integer or undef values.</p>
4781 <p>The elements of the two input vectors are numbered from left to right across
4782 both of the vectors. The shuffle mask operand specifies, for each element of
4783 the result vector, which element of the two input vectors the result element
4784 gets. The element selector may be undef (meaning "don't care") and the
4785 second operand may be undef if performing a shuffle from only one vector.</p>
4789 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4790 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4791 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4792 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4793 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4794 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4795 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4796 <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>
4803 <!-- ======================================================================= -->
4805 <a name="aggregateops">Aggregate Operations</a>
4810 <p>LLVM supports several instructions for working with
4811 <a href="#t_aggregate">aggregate</a> values.</p>
4813 <!-- _______________________________________________________________________ -->
4815 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4822 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4826 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4827 from an <a href="#t_aggregate">aggregate</a> value.</p>
4830 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4831 of <a href="#t_struct">struct</a> or
4832 <a href="#t_array">array</a> type. The operands are constant indices to
4833 specify which value to extract in a similar manner as indices in a
4834 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4835 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4837 <li>Since the value being indexed is not a pointer, the first index is
4838 omitted and assumed to be zero.</li>
4839 <li>At least one index must be specified.</li>
4840 <li>Not only struct indices but also array indices must be in
4845 <p>The result is the value at the position in the aggregate specified by the
4850 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4855 <!-- _______________________________________________________________________ -->
4857 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4864 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4868 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4869 in an <a href="#t_aggregate">aggregate</a> value.</p>
4872 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4873 of <a href="#t_struct">struct</a> or
4874 <a href="#t_array">array</a> type. The second operand is a first-class
4875 value to insert. The following operands are constant indices indicating
4876 the position at which to insert the value in a similar manner as indices in a
4877 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4878 value to insert must have the same type as the value identified by the
4882 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4883 that of <tt>val</tt> except that the value at the position specified by the
4884 indices is that of <tt>elt</tt>.</p>
4888 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4889 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4890 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4897 <!-- ======================================================================= -->
4899 <a name="memoryops">Memory Access and Addressing Operations</a>
4904 <p>A key design point of an SSA-based representation is how it represents
4905 memory. In LLVM, no memory locations are in SSA form, which makes things
4906 very simple. This section describes how to read, write, and allocate
4909 <!-- _______________________________________________________________________ -->
4911 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4918 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4922 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4923 currently executing function, to be automatically released when this function
4924 returns to its caller. The object is always allocated in the generic address
4925 space (address space zero).</p>
4928 <p>The '<tt>alloca</tt>' instruction
4929 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4930 runtime stack, returning a pointer of the appropriate type to the program.
4931 If "NumElements" is specified, it is the number of elements allocated,
4932 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4933 specified, the value result of the allocation is guaranteed to be aligned to
4934 at least that boundary. If not specified, or if zero, the target can choose
4935 to align the allocation on any convenient boundary compatible with the
4938 <p>'<tt>type</tt>' may be any sized type.</p>
4941 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4942 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4943 memory is automatically released when the function returns. The
4944 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4945 variables that must have an address available. When the function returns
4946 (either with the <tt><a href="#i_ret">ret</a></tt>
4947 or <tt><a href="#i_resume">resume</a></tt> instructions), the memory is
4948 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4949 The order in which memory is allocated (ie., which way the stack grows) is
4956 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4957 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4958 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4959 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4964 <!-- _______________________________________________________________________ -->
4966 <a name="i_load">'<tt>load</tt>' Instruction</a>
4973 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4974 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4975 !<index> = !{ i32 1 }
4979 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4982 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4983 from which to load. The pointer must point to
4984 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4985 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4986 number or order of execution of this <tt>load</tt> with other <a
4987 href="#volatile">volatile operations</a>.</p>
4989 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
4990 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4991 argument. The <code>release</code> and <code>acq_rel</code> orderings are
4992 not valid on <code>load</code> instructions. Atomic loads produce <a
4993 href="#memorymodel">defined</a> results when they may see multiple atomic
4994 stores. The type of the pointee must be an integer type whose bit width
4995 is a power of two greater than or equal to eight and less than or equal
4996 to a target-specific size limit. <code>align</code> must be explicitly
4997 specified on atomic loads, and the load has undefined behavior if the
4998 alignment is not set to a value which is at least the size in bytes of
4999 the pointee. <code>!nontemporal</code> does not have any defined semantics
5000 for atomic loads.</p>
5002 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
5003 operation (that is, the alignment of the memory address). A value of 0 or an
5004 omitted <tt>align</tt> argument means that the operation has the preferential
5005 alignment for the target. It is the responsibility of the code emitter to
5006 ensure that the alignment information is correct. Overestimating the
5007 alignment results in undefined behavior. Underestimating the alignment may
5008 produce less efficient code. An alignment of 1 is always safe.</p>
5010 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
5011 metatadata name <index> corresponding to a metadata node with
5012 one <tt>i32</tt> entry of value 1. The existence of
5013 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
5014 and code generator that this load is not expected to be reused in the cache.
5015 The code generator may select special instructions to save cache bandwidth,
5016 such as the <tt>MOVNT</tt> instruction on x86.</p>
5018 <p>The optional <tt>!invariant.load</tt> metadata must reference a single
5019 metatadata name <index> corresponding to a metadata node with no
5020 entries. The existence of the <tt>!invariant.load</tt> metatadata on the
5021 instruction tells the optimizer and code generator that this load address
5022 points to memory which does not change value during program execution.
5023 The optimizer may then move this load around, for example, by hoisting it
5024 out of loops using loop invariant code motion.</p>
5027 <p>The location of memory pointed to is loaded. If the value being loaded is of
5028 scalar type then the number of bytes read does not exceed the minimum number
5029 of bytes needed to hold all bits of the type. For example, loading an
5030 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
5031 <tt>i20</tt> with a size that is not an integral number of bytes, the result
5032 is undefined if the value was not originally written using a store of the
5037 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5038 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
5039 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
5044 <!-- _______________________________________________________________________ -->
5046 <a name="i_store">'<tt>store</tt>' Instruction</a>
5053 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
5054 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
5058 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
5061 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
5062 and an address at which to store it. The type of the
5063 '<tt><pointer></tt>' operand must be a pointer to
5064 the <a href="#t_firstclass">first class</a> type of the
5065 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
5066 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
5067 order of execution of this <tt>store</tt> with other <a
5068 href="#volatile">volatile operations</a>.</p>
5070 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
5071 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5072 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
5073 valid on <code>store</code> instructions. Atomic loads produce <a
5074 href="#memorymodel">defined</a> results when they may see multiple atomic
5075 stores. The type of the pointee must be an integer type whose bit width
5076 is a power of two greater than or equal to eight and less than or equal
5077 to a target-specific size limit. <code>align</code> must be explicitly
5078 specified on atomic stores, and the store has undefined behavior if the
5079 alignment is not set to a value which is at least the size in bytes of
5080 the pointee. <code>!nontemporal</code> does not have any defined semantics
5081 for atomic stores.</p>
5083 <p>The optional constant "align" argument specifies the alignment of the
5084 operation (that is, the alignment of the memory address). A value of 0 or an
5085 omitted "align" argument means that the operation has the preferential
5086 alignment for the target. It is the responsibility of the code emitter to
5087 ensure that the alignment information is correct. Overestimating the
5088 alignment results in an undefined behavior. Underestimating the alignment may
5089 produce less efficient code. An alignment of 1 is always safe.</p>
5091 <p>The optional !nontemporal metadata must reference a single metatadata
5092 name <index> corresponding to a metadata node with one i32 entry of
5093 value 1. The existence of the !nontemporal metatadata on the
5094 instruction tells the optimizer and code generator that this load is
5095 not expected to be reused in the cache. The code generator may
5096 select special instructions to save cache bandwidth, such as the
5097 MOVNT instruction on x86.</p>
5101 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
5102 location specified by the '<tt><pointer></tt>' operand. If
5103 '<tt><value></tt>' is of scalar type then the number of bytes written
5104 does not exceed the minimum number of bytes needed to hold all bits of the
5105 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
5106 writing a value of a type like <tt>i20</tt> with a size that is not an
5107 integral number of bytes, it is unspecified what happens to the extra bits
5108 that do not belong to the type, but they will typically be overwritten.</p>
5112 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5113 store i32 3, i32* %ptr <i>; yields {void}</i>
5114 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
5119 <!-- _______________________________________________________________________ -->
5121 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
5128 fence [singlethread] <ordering> <i>; yields {void}</i>
5132 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
5133 between operations.</p>
5135 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
5136 href="#ordering">ordering</a> argument which defines what
5137 <i>synchronizes-with</i> edges they add. They can only be given
5138 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
5139 <code>seq_cst</code> orderings.</p>
5142 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
5143 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
5144 <code>acquire</code> ordering semantics if and only if there exist atomic
5145 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
5146 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
5147 <var>X</var> modifies <var>M</var> (either directly or through some side effect
5148 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
5149 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
5150 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
5151 than an explicit <code>fence</code>, one (but not both) of the atomic operations
5152 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
5153 <code>acquire</code> (resp.) ordering constraint and still
5154 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
5155 <i>happens-before</i> edge.</p>
5157 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
5158 having both <code>acquire</code> and <code>release</code> semantics specified
5159 above, participates in the global program order of other <code>seq_cst</code>
5160 operations and/or fences.</p>
5162 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
5163 specifies that the fence only synchronizes with other fences in the same
5164 thread. (This is useful for interacting with signal handlers.)</p>
5168 fence acquire <i>; yields {void}</i>
5169 fence singlethread seq_cst <i>; yields {void}</i>
5174 <!-- _______________________________________________________________________ -->
5176 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
5183 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
5187 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
5188 It loads a value in memory and compares it to a given value. If they are
5189 equal, it stores a new value into the memory.</p>
5192 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
5193 address to operate on, a value to compare to the value currently be at that
5194 address, and a new value to place at that address if the compared values are
5195 equal. The type of '<var><cmp></var>' must be an integer type whose
5196 bit width is a power of two greater than or equal to eight and less than
5197 or equal to a target-specific size limit. '<var><cmp></var>' and
5198 '<var><new></var>' must have the same type, and the type of
5199 '<var><pointer></var>' must be a pointer to that type. If the
5200 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
5201 optimizer is not allowed to modify the number or order of execution
5202 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
5205 <!-- FIXME: Extend allowed types. -->
5207 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
5208 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
5210 <p>The optional "<code>singlethread</code>" argument declares that the
5211 <code>cmpxchg</code> is only atomic with respect to code (usually signal
5212 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
5213 cmpxchg is atomic with respect to all other code in the system.</p>
5215 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
5216 the size in memory of the operand.
5219 <p>The contents of memory at the location specified by the
5220 '<tt><pointer></tt>' operand is read and compared to
5221 '<tt><cmp></tt>'; if the read value is the equal,
5222 '<tt><new></tt>' is written. The original value at the location
5225 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
5226 purpose of identifying <a href="#release_sequence">release sequences</a>. A
5227 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
5228 parameter determined by dropping any <code>release</code> part of the
5229 <code>cmpxchg</code>'s ordering.</p>
5232 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
5233 optimization work on ARM.)
5235 FIXME: Is a weaker ordering constraint on failure helpful in practice?
5241 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
5242 <a href="#i_br">br</a> label %loop
5245 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
5246 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
5247 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
5248 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
5249 <a href="#i_br">br</a> i1 %success, label %done, label %loop
5257 <!-- _______________________________________________________________________ -->
5259 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
5266 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
5270 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
5273 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
5274 operation to apply, an address whose value to modify, an argument to the
5275 operation. The operation must be one of the following keywords:</p>
5290 <p>The type of '<var><value></var>' must be an integer type whose
5291 bit width is a power of two greater than or equal to eight and less than
5292 or equal to a target-specific size limit. The type of the
5293 '<code><pointer></code>' operand must be a pointer to that type.
5294 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
5295 optimizer is not allowed to modify the number or order of execution of this
5296 <code>atomicrmw</code> with other <a href="#volatile">volatile
5299 <!-- FIXME: Extend allowed types. -->
5302 <p>The contents of memory at the location specified by the
5303 '<tt><pointer></tt>' operand are atomically read, modified, and written
5304 back. The original value at the location is returned. The modification is
5305 specified by the <var>operation</var> argument:</p>
5308 <li>xchg: <code>*ptr = val</code></li>
5309 <li>add: <code>*ptr = *ptr + val</code></li>
5310 <li>sub: <code>*ptr = *ptr - val</code></li>
5311 <li>and: <code>*ptr = *ptr & val</code></li>
5312 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
5313 <li>or: <code>*ptr = *ptr | val</code></li>
5314 <li>xor: <code>*ptr = *ptr ^ val</code></li>
5315 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
5316 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
5317 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
5318 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
5323 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
5328 <!-- _______________________________________________________________________ -->
5330 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
5337 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5338 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5339 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5343 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
5344 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
5345 It performs address calculation only and does not access memory.</p>
5348 <p>The first argument is always a pointer or a vector of pointers,
5349 and forms the basis of the
5350 calculation. The remaining arguments are indices that indicate which of the
5351 elements of the aggregate object are indexed. The interpretation of each
5352 index is dependent on the type being indexed into. The first index always
5353 indexes the pointer value given as the first argument, the second index
5354 indexes a value of the type pointed to (not necessarily the value directly
5355 pointed to, since the first index can be non-zero), etc. The first type
5356 indexed into must be a pointer value, subsequent types can be arrays,
5357 vectors, and structs. Note that subsequent types being indexed into
5358 can never be pointers, since that would require loading the pointer before
5359 continuing calculation.</p>
5361 <p>The type of each index argument depends on the type it is indexing into.
5362 When indexing into a (optionally packed) structure, only <tt>i32</tt>
5363 integer <b>constants</b> are allowed. When indexing into an array, pointer
5364 or vector, integers of any width are allowed, and they are not required to be
5365 constant. These integers are treated as signed values where relevant.</p>
5367 <p>For example, let's consider a C code fragment and how it gets compiled to
5370 <pre class="doc_code">
5382 int *foo(struct ST *s) {
5383 return &s[1].Z.B[5][13];
5387 <p>The LLVM code generated by Clang is:</p>
5389 <pre class="doc_code">
5390 %struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 }
5391 %struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT }
5393 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5395 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5401 <p>In the example above, the first index is indexing into the
5402 '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a
5403 '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a
5404 structure. The second index indexes into the third element of the structure,
5405 yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>'
5406 type, another structure. The third index indexes into the second element of
5407 the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The
5408 two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>'
5409 type. The '<tt>getelementptr</tt>' instruction returns a pointer to this
5410 element, thus computing a value of '<tt>i32*</tt>' type.</p>
5412 <p>Note that it is perfectly legal to index partially through a structure,
5413 returning a pointer to an inner element. Because of this, the LLVM code for
5414 the given testcase is equivalent to:</p>
5416 <pre class="doc_code">
5417 define i32* @foo(%struct.ST* %s) {
5418 %t1 = getelementptr %struct.ST* %s, i32 1 <i>; yields %struct.ST*:%t1</i>
5419 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 <i>; yields %struct.RT*:%t2</i>
5420 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5421 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5422 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5427 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5428 <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the
5429 base pointer is not an <i>in bounds</i> address of an allocated object,
5430 or if any of the addresses that would be formed by successive addition of
5431 the offsets implied by the indices to the base address with infinitely
5432 precise signed arithmetic are not an <i>in bounds</i> address of that
5433 allocated object. The <i>in bounds</i> addresses for an allocated object
5434 are all the addresses that point into the object, plus the address one
5436 In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword
5437 applies to each of the computations element-wise. </p>
5439 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5440 the base address with silently-wrapping two's complement arithmetic. If the
5441 offsets have a different width from the pointer, they are sign-extended or
5442 truncated to the width of the pointer. The result value of the
5443 <tt>getelementptr</tt> may be outside the object pointed to by the base
5444 pointer. The result value may not necessarily be used to access memory
5445 though, even if it happens to point into allocated storage. See the
5446 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5449 <p>The getelementptr instruction is often confusing. For some more insight into
5450 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5454 <i>; yields [12 x i8]*:aptr</i>
5455 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5456 <i>; yields i8*:vptr</i>
5457 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5458 <i>; yields i8*:eptr</i>
5459 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5460 <i>; yields i32*:iptr</i>
5461 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5464 <p>In cases where the pointer argument is a vector of pointers, only a
5465 single index may be used, and the number of vector elements has to be
5466 the same. For example: </p>
5467 <pre class="doc_code">
5468 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5475 <!-- ======================================================================= -->
5477 <a name="convertops">Conversion Operations</a>
5482 <p>The instructions in this category are the conversion instructions (casting)
5483 which all take a single operand and a type. They perform various bit
5484 conversions on the operand.</p>
5486 <!-- _______________________________________________________________________ -->
5488 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5495 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5499 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5500 type <tt>ty2</tt>.</p>
5503 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5504 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5505 of the same number of integers.
5506 The bit size of the <tt>value</tt> must be larger than
5507 the bit size of the destination type, <tt>ty2</tt>.
5508 Equal sized types are not allowed.</p>
5511 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5512 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5513 source size must be larger than the destination size, <tt>trunc</tt> cannot
5514 be a <i>no-op cast</i>. It will always truncate bits.</p>
5518 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5519 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5520 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5521 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5526 <!-- _______________________________________________________________________ -->
5528 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5535 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5539 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5544 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5545 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5546 of the same number of integers.
5547 The bit size of the <tt>value</tt> must be smaller than
5548 the bit size of the destination type,
5552 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5553 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5555 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5559 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5560 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5561 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5566 <!-- _______________________________________________________________________ -->
5568 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5575 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5579 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5582 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5583 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5584 of the same number of integers.
5585 The bit size of the <tt>value</tt> must be smaller than
5586 the bit size of the destination type,
5590 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5591 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5592 of the type <tt>ty2</tt>.</p>
5594 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5598 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5599 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5600 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5605 <!-- _______________________________________________________________________ -->
5607 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5614 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5618 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5622 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5623 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5624 to cast it to. The size of <tt>value</tt> must be larger than the size of
5625 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5626 <i>no-op cast</i>.</p>
5629 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5630 <a href="#t_floating">floating point</a> type to a smaller
5631 <a href="#t_floating">floating point</a> type. If the value cannot fit
5632 within the destination type, <tt>ty2</tt>, then the results are
5637 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5638 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5643 <!-- _______________________________________________________________________ -->
5645 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5652 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5656 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5657 floating point value.</p>
5660 <p>The '<tt>fpext</tt>' instruction takes a
5661 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5662 a <a href="#t_floating">floating point</a> type to cast it to. The source
5663 type must be smaller than the destination type.</p>
5666 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5667 <a href="#t_floating">floating point</a> type to a larger
5668 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5669 used to make a <i>no-op cast</i> because it always changes bits. Use
5670 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5674 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5675 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5680 <!-- _______________________________________________________________________ -->
5682 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5689 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5693 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5694 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5697 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5698 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5699 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5700 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5701 vector integer type with the same number of elements as <tt>ty</tt></p>
5704 <p>The '<tt>fptoui</tt>' instruction converts its
5705 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5706 towards zero) unsigned integer value. If the value cannot fit
5707 in <tt>ty2</tt>, the results are undefined.</p>
5711 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5712 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5713 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5718 <!-- _______________________________________________________________________ -->
5720 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5727 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5731 <p>The '<tt>fptosi</tt>' instruction converts
5732 <a href="#t_floating">floating point</a> <tt>value</tt> to
5733 type <tt>ty2</tt>.</p>
5736 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5737 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5738 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5739 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5740 vector integer type with the same number of elements as <tt>ty</tt></p>
5743 <p>The '<tt>fptosi</tt>' instruction converts its
5744 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5745 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5746 the results are undefined.</p>
5750 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5751 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5752 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5757 <!-- _______________________________________________________________________ -->
5759 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5766 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5770 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5771 integer and converts that value to the <tt>ty2</tt> type.</p>
5774 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5775 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5776 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5777 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5778 floating point type with the same number of elements as <tt>ty</tt></p>
5781 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5782 integer quantity and converts it to the corresponding floating point
5783 value. If the value cannot fit in the floating point value, the results are
5788 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5789 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5794 <!-- _______________________________________________________________________ -->
5796 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5803 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5807 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5808 and converts that value to the <tt>ty2</tt> type.</p>
5811 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5812 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5813 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5814 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5815 floating point type with the same number of elements as <tt>ty</tt></p>
5818 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5819 quantity and converts it to the corresponding floating point value. If the
5820 value cannot fit in the floating point value, the results are undefined.</p>
5824 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5825 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5830 <!-- _______________________________________________________________________ -->
5832 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5839 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5843 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of
5844 pointers <tt>value</tt> to
5845 the integer (or vector of integers) type <tt>ty2</tt>.</p>
5848 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5849 must be a a value of type <a href="#t_pointer">pointer</a> or a vector of
5850 pointers, and a type to cast it to
5851 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector
5852 of integers type.</p>
5855 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5856 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5857 truncating or zero extending that value to the size of the integer type. If
5858 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5859 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5860 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5865 %X = ptrtoint i32* %P to i8 <i>; yields truncation on 32-bit architecture</i>
5866 %Y = ptrtoint i32* %P to i64 <i>; yields zero extension on 32-bit architecture</i>
5867 %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>
5872 <!-- _______________________________________________________________________ -->
5874 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5881 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5885 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5886 pointer type, <tt>ty2</tt>.</p>
5889 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5890 value to cast, and a type to cast it to, which must be a
5891 <a href="#t_pointer">pointer</a> type.</p>
5894 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5895 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5896 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5897 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5898 than the size of a pointer then a zero extension is done. If they are the
5899 same size, nothing is done (<i>no-op cast</i>).</p>
5903 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5904 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5905 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5906 %Z = inttoptr <4 x i32> %G to <4 x i8*><i>; yields truncation of vector G to four pointers</i>
5911 <!-- _______________________________________________________________________ -->
5913 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5920 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5924 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5925 <tt>ty2</tt> without changing any bits.</p>
5928 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5929 non-aggregate first class value, and a type to cast it to, which must also be
5930 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5931 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5932 identical. If the source type is a pointer, the destination type must also be
5933 a pointer. This instruction supports bitwise conversion of vectors to
5934 integers and to vectors of other types (as long as they have the same
5938 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5939 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5940 this conversion. The conversion is done as if the <tt>value</tt> had been
5941 stored to memory and read back as type <tt>ty2</tt>.
5942 Pointer (or vector of pointers) types may only be converted to other pointer
5943 (or vector of pointers) types with this instruction. To convert
5944 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5945 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5949 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
5950 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
5951 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
5952 %Z = bitcast <2 x i32*> %V to <2 x i64*> <i>; yields <2 x i64*></i>
5959 <!-- ======================================================================= -->
5961 <a name="otherops">Other Operations</a>
5966 <p>The instructions in this category are the "miscellaneous" instructions, which
5967 defy better classification.</p>
5969 <!-- _______________________________________________________________________ -->
5971 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5978 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5982 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5983 boolean values based on comparison of its two integer, integer vector,
5984 pointer, or pointer vector operands.</p>
5987 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
5988 the condition code indicating the kind of comparison to perform. It is not a
5989 value, just a keyword. The possible condition code are:</p>
5992 <li><tt>eq</tt>: equal</li>
5993 <li><tt>ne</tt>: not equal </li>
5994 <li><tt>ugt</tt>: unsigned greater than</li>
5995 <li><tt>uge</tt>: unsigned greater or equal</li>
5996 <li><tt>ult</tt>: unsigned less than</li>
5997 <li><tt>ule</tt>: unsigned less or equal</li>
5998 <li><tt>sgt</tt>: signed greater than</li>
5999 <li><tt>sge</tt>: signed greater or equal</li>
6000 <li><tt>slt</tt>: signed less than</li>
6001 <li><tt>sle</tt>: signed less or equal</li>
6004 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
6005 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
6006 typed. They must also be identical types.</p>
6009 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
6010 condition code given as <tt>cond</tt>. The comparison performed always yields
6011 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
6012 result, as follows:</p>
6015 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
6016 <tt>false</tt> otherwise. No sign interpretation is necessary or
6019 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
6020 <tt>false</tt> otherwise. No sign interpretation is necessary or
6023 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
6024 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6026 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
6027 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6028 to <tt>op2</tt>.</li>
6030 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
6031 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6033 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
6034 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6036 <li><tt>sgt</tt>: interprets the operands as signed values and yields
6037 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6039 <li><tt>sge</tt>: interprets the operands as signed values and yields
6040 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6041 to <tt>op2</tt>.</li>
6043 <li><tt>slt</tt>: interprets the operands as signed values and yields
6044 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6046 <li><tt>sle</tt>: interprets the operands as signed values and yields
6047 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6050 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
6051 values are compared as if they were integers.</p>
6053 <p>If the operands are integer vectors, then they are compared element by
6054 element. The result is an <tt>i1</tt> vector with the same number of elements
6055 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
6059 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
6060 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
6061 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
6062 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
6063 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
6064 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
6067 <p>Note that the code generator does not yet support vector types with
6068 the <tt>icmp</tt> instruction.</p>
6072 <!-- _______________________________________________________________________ -->
6074 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
6081 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
6085 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
6086 values based on comparison of its operands.</p>
6088 <p>If the operands are floating point scalars, then the result type is a boolean
6089 (<a href="#t_integer"><tt>i1</tt></a>).</p>
6091 <p>If the operands are floating point vectors, then the result type is a vector
6092 of boolean with the same number of elements as the operands being
6096 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
6097 the condition code indicating the kind of comparison to perform. It is not a
6098 value, just a keyword. The possible condition code are:</p>
6101 <li><tt>false</tt>: no comparison, always returns false</li>
6102 <li><tt>oeq</tt>: ordered and equal</li>
6103 <li><tt>ogt</tt>: ordered and greater than </li>
6104 <li><tt>oge</tt>: ordered and greater than or equal</li>
6105 <li><tt>olt</tt>: ordered and less than </li>
6106 <li><tt>ole</tt>: ordered and less than or equal</li>
6107 <li><tt>one</tt>: ordered and not equal</li>
6108 <li><tt>ord</tt>: ordered (no nans)</li>
6109 <li><tt>ueq</tt>: unordered or equal</li>
6110 <li><tt>ugt</tt>: unordered or greater than </li>
6111 <li><tt>uge</tt>: unordered or greater than or equal</li>
6112 <li><tt>ult</tt>: unordered or less than </li>
6113 <li><tt>ule</tt>: unordered or less than or equal</li>
6114 <li><tt>une</tt>: unordered or not equal</li>
6115 <li><tt>uno</tt>: unordered (either nans)</li>
6116 <li><tt>true</tt>: no comparison, always returns true</li>
6119 <p><i>Ordered</i> means that neither operand is a QNAN while
6120 <i>unordered</i> means that either operand may be a QNAN.</p>
6122 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
6123 a <a href="#t_floating">floating point</a> type or
6124 a <a href="#t_vector">vector</a> of floating point type. They must have
6125 identical types.</p>
6128 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
6129 according to the condition code given as <tt>cond</tt>. If the operands are
6130 vectors, then the vectors are compared element by element. Each comparison
6131 performed always yields an <a href="#t_integer">i1</a> result, as
6135 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
6137 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6138 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6140 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6141 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6143 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6144 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6146 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6147 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6149 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6150 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6152 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6153 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6155 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
6157 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
6158 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6160 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
6161 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6163 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
6164 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6166 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
6167 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6169 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
6170 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6172 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
6173 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6175 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
6177 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
6182 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
6183 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
6184 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
6185 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
6188 <p>Note that the code generator does not yet support vector types with
6189 the <tt>fcmp</tt> instruction.</p>
6193 <!-- _______________________________________________________________________ -->
6195 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
6202 <result> = phi <ty> [ <val0>, <label0>], ...
6206 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
6207 SSA graph representing the function.</p>
6210 <p>The type of the incoming values is specified with the first type field. After
6211 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
6212 one pair for each predecessor basic block of the current block. Only values
6213 of <a href="#t_firstclass">first class</a> type may be used as the value
6214 arguments to the PHI node. Only labels may be used as the label
6217 <p>There must be no non-phi instructions between the start of a basic block and
6218 the PHI instructions: i.e. PHI instructions must be first in a basic
6221 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
6222 occur on the edge from the corresponding predecessor block to the current
6223 block (but after any definition of an '<tt>invoke</tt>' instruction's return
6224 value on the same edge).</p>
6227 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
6228 specified by the pair corresponding to the predecessor basic block that
6229 executed just prior to the current block.</p>
6233 Loop: ; Infinite loop that counts from 0 on up...
6234 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6235 %nextindvar = add i32 %indvar, 1
6241 <!-- _______________________________________________________________________ -->
6243 <a name="i_select">'<tt>select</tt>' Instruction</a>
6250 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
6252 <i>selty</i> is either i1 or {<N x i1>}
6256 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
6257 condition, without branching.</p>
6261 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
6262 values indicating the condition, and two values of the
6263 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
6264 vectors and the condition is a scalar, then entire vectors are selected, not
6265 individual elements.</p>
6268 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
6269 first value argument; otherwise, it returns the second value argument.</p>
6271 <p>If the condition is a vector of i1, then the value arguments must be vectors
6272 of the same size, and the selection is done element by element.</p>
6276 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
6281 <!-- _______________________________________________________________________ -->
6283 <a name="i_call">'<tt>call</tt>' Instruction</a>
6290 <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>]
6294 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
6297 <p>This instruction requires several arguments:</p>
6300 <li>The optional "tail" marker indicates that the callee function does not
6301 access any allocas or varargs in the caller. Note that calls may be
6302 marked "tail" even if they do not occur before
6303 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
6304 present, the function call is eligible for tail call optimization,
6305 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
6306 optimized into a jump</a>. The code generator may optimize calls marked
6307 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
6308 sibling call optimization</a> when the caller and callee have
6309 matching signatures, or 2) forced tail call optimization when the
6310 following extra requirements are met:
6312 <li>Caller and callee both have the calling
6313 convention <tt>fastcc</tt>.</li>
6314 <li>The call is in tail position (ret immediately follows call and ret
6315 uses value of call or is void).</li>
6316 <li>Option <tt>-tailcallopt</tt> is enabled,
6317 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
6318 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
6319 constraints are met.</a></li>
6323 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
6324 convention</a> the call should use. If none is specified, the call
6325 defaults to using C calling conventions. The calling convention of the
6326 call must match the calling convention of the target function, or else the
6327 behavior is undefined.</li>
6329 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
6330 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
6331 '<tt>inreg</tt>' attributes are valid here.</li>
6333 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
6334 type of the return value. Functions that return no value are marked
6335 <tt><a href="#t_void">void</a></tt>.</li>
6337 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
6338 being invoked. The argument types must match the types implied by this
6339 signature. This type can be omitted if the function is not varargs and if
6340 the function type does not return a pointer to a function.</li>
6342 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
6343 be invoked. In most cases, this is a direct function invocation, but
6344 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
6345 to function value.</li>
6347 <li>'<tt>function args</tt>': argument list whose types match the function
6348 signature argument types and parameter attributes. All arguments must be
6349 of <a href="#t_firstclass">first class</a> type. If the function
6350 signature indicates the function accepts a variable number of arguments,
6351 the extra arguments can be specified.</li>
6353 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
6354 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
6355 '<tt>readnone</tt>' attributes are valid here.</li>
6359 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
6360 a specified function, with its incoming arguments bound to the specified
6361 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
6362 function, control flow continues with the instruction after the function
6363 call, and the return value of the function is bound to the result
6368 %retval = call i32 @test(i32 %argc)
6369 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
6370 %X = tail call i32 @foo() <i>; yields i32</i>
6371 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
6372 call void %foo(i8 97 signext)
6374 %struct.A = type { i32, i8 }
6375 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
6376 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
6377 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
6378 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
6379 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
6382 <p>llvm treats calls to some functions with names and arguments that match the
6383 standard C99 library as being the C99 library functions, and may perform
6384 optimizations or generate code for them under that assumption. This is
6385 something we'd like to change in the future to provide better support for
6386 freestanding environments and non-C-based languages.</p>
6390 <!-- _______________________________________________________________________ -->
6392 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
6399 <resultval> = va_arg <va_list*> <arglist>, <argty>
6403 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6404 the "variable argument" area of a function call. It is used to implement the
6405 <tt>va_arg</tt> macro in C.</p>
6408 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6409 argument. It returns a value of the specified argument type and increments
6410 the <tt>va_list</tt> to point to the next argument. The actual type
6411 of <tt>va_list</tt> is target specific.</p>
6414 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6415 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6416 to the next argument. For more information, see the variable argument
6417 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6419 <p>It is legal for this instruction to be called in a function which does not
6420 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6423 <p><tt>va_arg</tt> is an LLVM instruction instead of
6424 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6428 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6430 <p>Note that the code generator does not yet fully support va_arg on many
6431 targets. Also, it does not currently support va_arg with aggregate types on
6436 <!-- _______________________________________________________________________ -->
6438 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6445 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6446 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6448 <clause> := catch <type> <value>
6449 <clause> := filter <array constant type> <array constant>
6453 <p>The '<tt>landingpad</tt>' instruction is used by
6454 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6455 system</a> to specify that a basic block is a landing pad — one where
6456 the exception lands, and corresponds to the code found in the
6457 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6458 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6459 re-entry to the function. The <tt>resultval</tt> has the
6460 type <tt>resultty</tt>.</p>
6463 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6464 function associated with the unwinding mechanism. The optional
6465 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6467 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6468 or <tt>filter</tt> — and contains the global variable representing the
6469 "type" that may be caught or filtered respectively. Unlike the
6470 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6471 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6472 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6473 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6476 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6477 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6478 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6479 calling conventions, how the personality function results are represented in
6480 LLVM IR is target specific.</p>
6482 <p>The clauses are applied in order from top to bottom. If two
6483 <tt>landingpad</tt> instructions are merged together through inlining, the
6484 clauses from the calling function are appended to the list of clauses.
6485 When the call stack is being unwound due to an exception being thrown, the
6486 exception is compared against each <tt>clause</tt> in turn. If it doesn't
6487 match any of the clauses, and the <tt>cleanup</tt> flag is not set, then
6488 unwinding continues further up the call stack.</p>
6490 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6493 <li>A landing pad block is a basic block which is the unwind destination of an
6494 '<tt>invoke</tt>' instruction.</li>
6495 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6496 first non-PHI instruction.</li>
6497 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6499 <li>A basic block that is not a landing pad block may not include a
6500 '<tt>landingpad</tt>' instruction.</li>
6501 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6502 personality function.</li>
6507 ;; A landing pad which can catch an integer.
6508 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6510 ;; A landing pad that is a cleanup.
6511 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6513 ;; A landing pad which can catch an integer and can only throw a double.
6514 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6516 filter [1 x i8**] [@_ZTId]
6525 <!-- *********************************************************************** -->
6526 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6527 <!-- *********************************************************************** -->
6531 <p>LLVM supports the notion of an "intrinsic function". These functions have
6532 well known names and semantics and are required to follow certain
6533 restrictions. Overall, these intrinsics represent an extension mechanism for
6534 the LLVM language that does not require changing all of the transformations
6535 in LLVM when adding to the language (or the bitcode reader/writer, the
6536 parser, etc...).</p>
6538 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6539 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6540 begin with this prefix. Intrinsic functions must always be external
6541 functions: you cannot define the body of intrinsic functions. Intrinsic
6542 functions may only be used in call or invoke instructions: it is illegal to
6543 take the address of an intrinsic function. Additionally, because intrinsic
6544 functions are part of the LLVM language, it is required if any are added that
6545 they be documented here.</p>
6547 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6548 family of functions that perform the same operation but on different data
6549 types. Because LLVM can represent over 8 million different integer types,
6550 overloading is used commonly to allow an intrinsic function to operate on any
6551 integer type. One or more of the argument types or the result type can be
6552 overloaded to accept any integer type. Argument types may also be defined as
6553 exactly matching a previous argument's type or the result type. This allows
6554 an intrinsic function which accepts multiple arguments, but needs all of them
6555 to be of the same type, to only be overloaded with respect to a single
6556 argument or the result.</p>
6558 <p>Overloaded intrinsics will have the names of its overloaded argument types
6559 encoded into its function name, each preceded by a period. Only those types
6560 which are overloaded result in a name suffix. Arguments whose type is matched
6561 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6562 can take an integer of any width and returns an integer of exactly the same
6563 integer width. This leads to a family of functions such as
6564 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6565 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6566 suffix is required. Because the argument's type is matched against the return
6567 type, it does not require its own name suffix.</p>
6569 <p>To learn how to add an intrinsic function, please see the
6570 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6572 <!-- ======================================================================= -->
6574 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6579 <p>Variable argument support is defined in LLVM with
6580 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6581 intrinsic functions. These functions are related to the similarly named
6582 macros defined in the <tt><stdarg.h></tt> header file.</p>
6584 <p>All of these functions operate on arguments that use a target-specific value
6585 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6586 not define what this type is, so all transformations should be prepared to
6587 handle these functions regardless of the type used.</p>
6589 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6590 instruction and the variable argument handling intrinsic functions are
6593 <pre class="doc_code">
6594 define i32 @test(i32 %X, ...) {
6595 ; Initialize variable argument processing
6597 %ap2 = bitcast i8** %ap to i8*
6598 call void @llvm.va_start(i8* %ap2)
6600 ; Read a single integer argument
6601 %tmp = va_arg i8** %ap, i32
6603 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6605 %aq2 = bitcast i8** %aq to i8*
6606 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6607 call void @llvm.va_end(i8* %aq2)
6609 ; Stop processing of arguments.
6610 call void @llvm.va_end(i8* %ap2)
6614 declare void @llvm.va_start(i8*)
6615 declare void @llvm.va_copy(i8*, i8*)
6616 declare void @llvm.va_end(i8*)
6619 <!-- _______________________________________________________________________ -->
6621 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6629 declare void %llvm.va_start(i8* <arglist>)
6633 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6634 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6637 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6640 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6641 macro available in C. In a target-dependent way, it initializes
6642 the <tt>va_list</tt> element to which the argument points, so that the next
6643 call to <tt>va_arg</tt> will produce the first variable argument passed to
6644 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6645 need to know the last argument of the function as the compiler can figure
6650 <!-- _______________________________________________________________________ -->
6652 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6659 declare void @llvm.va_end(i8* <arglist>)
6663 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6664 which has been initialized previously
6665 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6666 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6669 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6672 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6673 macro available in C. In a target-dependent way, it destroys
6674 the <tt>va_list</tt> element to which the argument points. Calls
6675 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6676 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6677 with calls to <tt>llvm.va_end</tt>.</p>
6681 <!-- _______________________________________________________________________ -->
6683 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6690 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6694 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6695 from the source argument list to the destination argument list.</p>
6698 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6699 The second argument is a pointer to a <tt>va_list</tt> element to copy
6703 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6704 macro available in C. In a target-dependent way, it copies the
6705 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6706 element. This intrinsic is necessary because
6707 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6708 arbitrarily complex and require, for example, memory allocation.</p>
6714 <!-- ======================================================================= -->
6716 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6721 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6722 Collection</a> (GC) requires the implementation and generation of these
6723 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6724 roots on the stack</a>, as well as garbage collector implementations that
6725 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6726 barriers. Front-ends for type-safe garbage collected languages should generate
6727 these intrinsics to make use of the LLVM garbage collectors. For more details,
6728 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6731 <p>The garbage collection intrinsics only operate on objects in the generic
6732 address space (address space zero).</p>
6734 <!-- _______________________________________________________________________ -->
6736 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6743 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6747 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6748 the code generator, and allows some metadata to be associated with it.</p>
6751 <p>The first argument specifies the address of a stack object that contains the
6752 root pointer. The second pointer (which must be either a constant or a
6753 global value address) contains the meta-data to be associated with the
6757 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6758 location. At compile-time, the code generator generates information to allow
6759 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6760 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6765 <!-- _______________________________________________________________________ -->
6767 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6774 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6778 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6779 locations, allowing garbage collector implementations that require read
6783 <p>The second argument is the address to read from, which should be an address
6784 allocated from the garbage collector. The first object is a pointer to the
6785 start of the referenced object, if needed by the language runtime (otherwise
6789 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6790 instruction, but may be replaced with substantially more complex code by the
6791 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6792 may only be used in a function which <a href="#gc">specifies a GC
6797 <!-- _______________________________________________________________________ -->
6799 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6806 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6810 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6811 locations, allowing garbage collector implementations that require write
6812 barriers (such as generational or reference counting collectors).</p>
6815 <p>The first argument is the reference to store, the second is the start of the
6816 object to store it to, and the third is the address of the field of Obj to
6817 store to. If the runtime does not require a pointer to the object, Obj may
6821 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6822 instruction, but may be replaced with substantially more complex code by the
6823 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6824 may only be used in a function which <a href="#gc">specifies a GC
6831 <!-- ======================================================================= -->
6833 <a name="int_codegen">Code Generator Intrinsics</a>
6838 <p>These intrinsics are provided by LLVM to expose special features that may
6839 only be implemented with code generator support.</p>
6841 <!-- _______________________________________________________________________ -->
6843 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6850 declare i8 *@llvm.returnaddress(i32 <level>)
6854 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6855 target-specific value indicating the return address of the current function
6856 or one of its callers.</p>
6859 <p>The argument to this intrinsic indicates which function to return the address
6860 for. Zero indicates the calling function, one indicates its caller, etc.
6861 The argument is <b>required</b> to be a constant integer value.</p>
6864 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6865 indicating the return address of the specified call frame, or zero if it
6866 cannot be identified. The value returned by this intrinsic is likely to be
6867 incorrect or 0 for arguments other than zero, so it should only be used for
6868 debugging purposes.</p>
6870 <p>Note that calling this intrinsic does not prevent function inlining or other
6871 aggressive transformations, so the value returned may not be that of the
6872 obvious source-language caller.</p>
6876 <!-- _______________________________________________________________________ -->
6878 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6885 declare i8* @llvm.frameaddress(i32 <level>)
6889 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6890 target-specific frame pointer value for the specified stack frame.</p>
6893 <p>The argument to this intrinsic indicates which function to return the frame
6894 pointer for. Zero indicates the calling function, one indicates its caller,
6895 etc. The argument is <b>required</b> to be a constant integer value.</p>
6898 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6899 indicating the frame address of the specified call frame, or zero if it
6900 cannot be identified. The value returned by this intrinsic is likely to be
6901 incorrect or 0 for arguments other than zero, so it should only be used for
6902 debugging purposes.</p>
6904 <p>Note that calling this intrinsic does not prevent function inlining or other
6905 aggressive transformations, so the value returned may not be that of the
6906 obvious source-language caller.</p>
6910 <!-- _______________________________________________________________________ -->
6912 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6919 declare i8* @llvm.stacksave()
6923 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6924 of the function stack, for use
6925 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6926 useful for implementing language features like scoped automatic variable
6927 sized arrays in C99.</p>
6930 <p>This intrinsic returns a opaque pointer value that can be passed
6931 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6932 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6933 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6934 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6935 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6936 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6940 <!-- _______________________________________________________________________ -->
6942 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6949 declare void @llvm.stackrestore(i8* %ptr)
6953 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6954 the function stack to the state it was in when the
6955 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6956 executed. This is useful for implementing language features like scoped
6957 automatic variable sized arrays in C99.</p>
6960 <p>See the description
6961 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6965 <!-- _______________________________________________________________________ -->
6967 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6974 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6978 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6979 insert a prefetch instruction if supported; otherwise, it is a noop.
6980 Prefetches have no effect on the behavior of the program but can change its
6981 performance characteristics.</p>
6984 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6985 specifier determining if the fetch should be for a read (0) or write (1),
6986 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
6987 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6988 specifies whether the prefetch is performed on the data (1) or instruction (0)
6989 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
6990 must be constant integers.</p>
6993 <p>This intrinsic does not modify the behavior of the program. In particular,
6994 prefetches cannot trap and do not produce a value. On targets that support
6995 this intrinsic, the prefetch can provide hints to the processor cache for
6996 better performance.</p>
7000 <!-- _______________________________________________________________________ -->
7002 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
7009 declare void @llvm.pcmarker(i32 <id>)
7013 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
7014 Counter (PC) in a region of code to simulators and other tools. The method
7015 is target specific, but it is expected that the marker will use exported
7016 symbols to transmit the PC of the marker. The marker makes no guarantees
7017 that it will remain with any specific instruction after optimizations. It is
7018 possible that the presence of a marker will inhibit optimizations. The
7019 intended use is to be inserted after optimizations to allow correlations of
7020 simulation runs.</p>
7023 <p><tt>id</tt> is a numerical id identifying the marker.</p>
7026 <p>This intrinsic does not modify the behavior of the program. Backends that do
7027 not support this intrinsic may ignore it.</p>
7031 <!-- _______________________________________________________________________ -->
7033 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
7040 declare i64 @llvm.readcyclecounter()
7044 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
7045 counter register (or similar low latency, high accuracy clocks) on those
7046 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7047 should map to RPCC. As the backing counters overflow quickly (on the order
7048 of 9 seconds on alpha), this should only be used for small timings.</p>
7051 <p>When directly supported, reading the cycle counter should not modify any
7052 memory. Implementations are allowed to either return a application specific
7053 value or a system wide value. On backends without support, this is lowered
7054 to a constant 0.</p>
7060 <!-- ======================================================================= -->
7062 <a name="int_libc">Standard C Library Intrinsics</a>
7067 <p>LLVM provides intrinsics for a few important standard C library functions.
7068 These intrinsics allow source-language front-ends to pass information about
7069 the alignment of the pointer arguments to the code generator, providing
7070 opportunity for more efficient code generation.</p>
7072 <!-- _______________________________________________________________________ -->
7074 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
7080 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
7081 integer bit width and for different address spaces. Not all targets support
7082 all bit widths however.</p>
7085 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7086 i32 <len>, i32 <align>, i1 <isvolatile>)
7087 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7088 i64 <len>, i32 <align>, i1 <isvolatile>)
7092 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7093 source location to the destination location.</p>
7095 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
7096 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7097 and the pointers can be in specified address spaces.</p>
7101 <p>The first argument is a pointer to the destination, the second is a pointer
7102 to the source. The third argument is an integer argument specifying the
7103 number of bytes to copy, the fourth argument is the alignment of the
7104 source and destination locations, and the fifth is a boolean indicating a
7105 volatile access.</p>
7107 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7108 then the caller guarantees that both the source and destination pointers are
7109 aligned to that boundary.</p>
7111 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7112 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
7113 The detailed access behavior is not very cleanly specified and it is unwise
7114 to depend on it.</p>
7118 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7119 source location to the destination location, which are not allowed to
7120 overlap. It copies "len" bytes of memory over. If the argument is known to
7121 be aligned to some boundary, this can be specified as the fourth argument,
7122 otherwise it should be set to 0 or 1.</p>
7126 <!-- _______________________________________________________________________ -->
7128 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
7134 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
7135 width and for different address space. Not all targets support all bit
7139 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7140 i32 <len>, i32 <align>, i1 <isvolatile>)
7141 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7142 i64 <len>, i32 <align>, i1 <isvolatile>)
7146 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
7147 source location to the destination location. It is similar to the
7148 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
7151 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
7152 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7153 and the pointers can be in specified address spaces.</p>
7157 <p>The first argument is a pointer to the destination, the second is a pointer
7158 to the source. The third argument is an integer argument specifying the
7159 number of bytes to copy, the fourth argument is the alignment of the
7160 source and destination locations, and the fifth is a boolean indicating a
7161 volatile access.</p>
7163 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7164 then the caller guarantees that the source and destination pointers are
7165 aligned to that boundary.</p>
7167 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7168 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
7169 The detailed access behavior is not very cleanly specified and it is unwise
7170 to depend on it.</p>
7174 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
7175 source location to the destination location, which may overlap. It copies
7176 "len" bytes of memory over. If the argument is known to be aligned to some
7177 boundary, this can be specified as the fourth argument, otherwise it should
7178 be set to 0 or 1.</p>
7182 <!-- _______________________________________________________________________ -->
7184 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
7190 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
7191 width and for different address spaces. However, not all targets support all
7195 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7196 i32 <len>, i32 <align>, i1 <isvolatile>)
7197 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7198 i64 <len>, i32 <align>, i1 <isvolatile>)
7202 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
7203 particular byte value.</p>
7205 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
7206 intrinsic does not return a value and takes extra alignment/volatile
7207 arguments. Also, the destination can be in an arbitrary address space.</p>
7210 <p>The first argument is a pointer to the destination to fill, the second is the
7211 byte value with which to fill it, the third argument is an integer argument
7212 specifying the number of bytes to fill, and the fourth argument is the known
7213 alignment of the destination location.</p>
7215 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7216 then the caller guarantees that the destination pointer is aligned to that
7219 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7220 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
7221 The detailed access behavior is not very cleanly specified and it is unwise
7222 to depend on it.</p>
7225 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
7226 at the destination location. If the argument is known to be aligned to some
7227 boundary, this can be specified as the fourth argument, otherwise it should
7228 be set to 0 or 1.</p>
7232 <!-- _______________________________________________________________________ -->
7234 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
7240 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
7241 floating point or vector of floating point type. Not all targets support all
7245 declare float @llvm.sqrt.f32(float %Val)
7246 declare double @llvm.sqrt.f64(double %Val)
7247 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7248 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7249 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7253 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
7254 returning the same value as the libm '<tt>sqrt</tt>' functions would.
7255 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
7256 behavior for negative numbers other than -0.0 (which allows for better
7257 optimization, because there is no need to worry about errno being
7258 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
7261 <p>The argument and return value are floating point numbers of the same
7265 <p>This function returns the sqrt of the specified operand if it is a
7266 nonnegative floating point number.</p>
7270 <!-- _______________________________________________________________________ -->
7272 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
7278 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
7279 floating point or vector of floating point type. Not all targets support all
7283 declare float @llvm.powi.f32(float %Val, i32 %power)
7284 declare double @llvm.powi.f64(double %Val, i32 %power)
7285 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7286 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7287 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7291 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
7292 specified (positive or negative) power. The order of evaluation of
7293 multiplications is not defined. When a vector of floating point type is
7294 used, the second argument remains a scalar integer value.</p>
7297 <p>The second argument is an integer power, and the first is a value to raise to
7301 <p>This function returns the first value raised to the second power with an
7302 unspecified sequence of rounding operations.</p>
7306 <!-- _______________________________________________________________________ -->
7308 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
7314 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
7315 floating point or vector of floating point type. Not all targets support all
7319 declare float @llvm.sin.f32(float %Val)
7320 declare double @llvm.sin.f64(double %Val)
7321 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7322 declare fp128 @llvm.sin.f128(fp128 %Val)
7323 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7327 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
7330 <p>The argument and return value are floating point numbers of the same
7334 <p>This function returns the sine of the specified operand, returning the same
7335 values as the libm <tt>sin</tt> functions would, and handles error conditions
7336 in the same way.</p>
7340 <!-- _______________________________________________________________________ -->
7342 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
7348 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
7349 floating point or vector of floating point type. Not all targets support all
7353 declare float @llvm.cos.f32(float %Val)
7354 declare double @llvm.cos.f64(double %Val)
7355 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7356 declare fp128 @llvm.cos.f128(fp128 %Val)
7357 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7361 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
7364 <p>The argument and return value are floating point numbers of the same
7368 <p>This function returns the cosine of the specified operand, returning the same
7369 values as the libm <tt>cos</tt> functions would, and handles error conditions
7370 in the same way.</p>
7374 <!-- _______________________________________________________________________ -->
7376 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
7382 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
7383 floating point or vector of floating point type. Not all targets support all
7387 declare float @llvm.pow.f32(float %Val, float %Power)
7388 declare double @llvm.pow.f64(double %Val, double %Power)
7389 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7390 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7391 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7395 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
7396 specified (positive or negative) power.</p>
7399 <p>The second argument is a floating point power, and the first is a value to
7400 raise to that power.</p>
7403 <p>This function returns the first value raised to the second power, returning
7404 the same values as the libm <tt>pow</tt> functions would, and handles error
7405 conditions in the same way.</p>
7409 <!-- _______________________________________________________________________ -->
7411 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7417 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7418 floating point or vector of floating point type. Not all targets support all
7422 declare float @llvm.exp.f32(float %Val)
7423 declare double @llvm.exp.f64(double %Val)
7424 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7425 declare fp128 @llvm.exp.f128(fp128 %Val)
7426 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7430 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7433 <p>The argument and return value are floating point numbers of the same
7437 <p>This function returns the same values as the libm <tt>exp</tt> functions
7438 would, and handles error conditions in the same way.</p>
7442 <!-- _______________________________________________________________________ -->
7444 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7450 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7451 floating point or vector of floating point type. Not all targets support all
7455 declare float @llvm.log.f32(float %Val)
7456 declare double @llvm.log.f64(double %Val)
7457 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7458 declare fp128 @llvm.log.f128(fp128 %Val)
7459 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7463 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7466 <p>The argument and return value are floating point numbers of the same
7470 <p>This function returns the same values as the libm <tt>log</tt> functions
7471 would, and handles error conditions in the same way.</p>
7475 <!-- _______________________________________________________________________ -->
7477 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7483 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7484 floating point or vector of floating point type. Not all targets support all
7488 declare float @llvm.fma.f32(float %a, float %b, float %c)
7489 declare double @llvm.fma.f64(double %a, double %b, double %c)
7490 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7491 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7492 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7496 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7500 <p>The argument and return value are floating point numbers of the same
7504 <p>This function returns the same values as the libm <tt>fma</tt> functions
7511 <!-- ======================================================================= -->
7513 <a name="int_manip">Bit Manipulation Intrinsics</a>
7518 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7519 These allow efficient code generation for some algorithms.</p>
7521 <!-- _______________________________________________________________________ -->
7523 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7529 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7530 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7533 declare i16 @llvm.bswap.i16(i16 <id>)
7534 declare i32 @llvm.bswap.i32(i32 <id>)
7535 declare i64 @llvm.bswap.i64(i64 <id>)
7539 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7540 values with an even number of bytes (positive multiple of 16 bits). These
7541 are useful for performing operations on data that is not in the target's
7542 native byte order.</p>
7545 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7546 and low byte of the input i16 swapped. Similarly,
7547 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7548 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7549 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7550 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7551 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7552 more, respectively).</p>
7556 <!-- _______________________________________________________________________ -->
7558 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7564 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7565 width, or on any vector with integer elements. Not all targets support all
7566 bit widths or vector types, however.</p>
7569 declare i8 @llvm.ctpop.i8(i8 <src>)
7570 declare i16 @llvm.ctpop.i16(i16 <src>)
7571 declare i32 @llvm.ctpop.i32(i32 <src>)
7572 declare i64 @llvm.ctpop.i64(i64 <src>)
7573 declare i256 @llvm.ctpop.i256(i256 <src>)
7574 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7578 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7582 <p>The only argument is the value to be counted. The argument may be of any
7583 integer type, or a vector with integer elements.
7584 The return type must match the argument type.</p>
7587 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7588 element of a vector.</p>
7592 <!-- _______________________________________________________________________ -->
7594 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7600 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7601 integer bit width, or any vector whose elements are integers. Not all
7602 targets support all bit widths or vector types, however.</p>
7605 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7606 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7607 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7608 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7609 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7610 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7614 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7615 leading zeros in a variable.</p>
7618 <p>The first argument is the value to be counted. This argument may be of any
7619 integer type, or a vectory with integer element type. The return type
7620 must match the first argument type.</p>
7622 <p>The second argument must be a constant and is a flag to indicate whether the
7623 intrinsic should ensure that a zero as the first argument produces a defined
7624 result. Historically some architectures did not provide a defined result for
7625 zero values as efficiently, and many algorithms are now predicated on
7626 avoiding zero-value inputs.</p>
7629 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7630 zeros in a variable, or within each element of the vector.
7631 If <tt>src == 0</tt> then the result is the size in bits of the type of
7632 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7633 For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7637 <!-- _______________________________________________________________________ -->
7639 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7645 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7646 integer bit width, or any vector of integer elements. Not all targets
7647 support all bit widths or vector types, however.</p>
7650 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7651 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7652 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7653 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7654 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7655 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7659 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7663 <p>The first argument is the value to be counted. This argument may be of any
7664 integer type, or a vectory with integer element type. The return type
7665 must match the first argument type.</p>
7667 <p>The second argument must be a constant and is a flag to indicate whether the
7668 intrinsic should ensure that a zero as the first argument produces a defined
7669 result. Historically some architectures did not provide a defined result for
7670 zero values as efficiently, and many algorithms are now predicated on
7671 avoiding zero-value inputs.</p>
7674 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7675 zeros in a variable, or within each element of a vector.
7676 If <tt>src == 0</tt> then the result is the size in bits of the type of
7677 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7678 For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7684 <!-- ======================================================================= -->
7686 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7691 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7693 <!-- _______________________________________________________________________ -->
7695 <a name="int_sadd_overflow">
7696 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7703 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7704 on any integer bit width.</p>
7707 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7708 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7709 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7713 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7714 a signed addition of the two arguments, and indicate whether an overflow
7715 occurred during the signed summation.</p>
7718 <p>The arguments (%a and %b) and the first element of the result structure may
7719 be of integer types of any bit width, but they must have the same bit
7720 width. The second element of the result structure must be of
7721 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7722 undergo signed addition.</p>
7725 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7726 a signed addition of the two variables. They return a structure — the
7727 first element of which is the signed summation, and the second element of
7728 which is a bit specifying if the signed summation resulted in an
7733 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7734 %sum = extractvalue {i32, i1} %res, 0
7735 %obit = extractvalue {i32, i1} %res, 1
7736 br i1 %obit, label %overflow, label %normal
7741 <!-- _______________________________________________________________________ -->
7743 <a name="int_uadd_overflow">
7744 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7751 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7752 on any integer bit width.</p>
7755 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7756 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7757 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7761 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7762 an unsigned addition of the two arguments, and indicate whether a carry
7763 occurred during the unsigned summation.</p>
7766 <p>The arguments (%a and %b) and the first element of the result structure may
7767 be of integer types of any bit width, but they must have the same bit
7768 width. The second element of the result structure must be of
7769 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7770 undergo unsigned addition.</p>
7773 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7774 an unsigned addition of the two arguments. They return a structure —
7775 the first element of which is the sum, and the second element of which is a
7776 bit specifying if the unsigned summation resulted in a carry.</p>
7780 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7781 %sum = extractvalue {i32, i1} %res, 0
7782 %obit = extractvalue {i32, i1} %res, 1
7783 br i1 %obit, label %carry, label %normal
7788 <!-- _______________________________________________________________________ -->
7790 <a name="int_ssub_overflow">
7791 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7798 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7799 on any integer bit width.</p>
7802 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7803 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7804 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7808 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7809 a signed subtraction of the two arguments, and indicate whether an overflow
7810 occurred during the signed subtraction.</p>
7813 <p>The arguments (%a and %b) and the first element of the result structure may
7814 be of integer types of any bit width, but they must have the same bit
7815 width. The second element of the result structure must be of
7816 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7817 undergo signed subtraction.</p>
7820 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7821 a signed subtraction of the two arguments. They return a structure —
7822 the first element of which is the subtraction, and the second element of
7823 which is a bit specifying if the signed subtraction resulted in an
7828 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7829 %sum = extractvalue {i32, i1} %res, 0
7830 %obit = extractvalue {i32, i1} %res, 1
7831 br i1 %obit, label %overflow, label %normal
7836 <!-- _______________________________________________________________________ -->
7838 <a name="int_usub_overflow">
7839 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7846 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7847 on any integer bit width.</p>
7850 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7851 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7852 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7856 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7857 an unsigned subtraction of the two arguments, and indicate whether an
7858 overflow occurred during the unsigned subtraction.</p>
7861 <p>The arguments (%a and %b) and the first element of the result structure may
7862 be of integer types of any bit width, but they must have the same bit
7863 width. The second element of the result structure must be of
7864 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7865 undergo unsigned subtraction.</p>
7868 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7869 an unsigned subtraction of the two arguments. They return a structure —
7870 the first element of which is the subtraction, and the second element of
7871 which is a bit specifying if the unsigned subtraction resulted in an
7876 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7877 %sum = extractvalue {i32, i1} %res, 0
7878 %obit = extractvalue {i32, i1} %res, 1
7879 br i1 %obit, label %overflow, label %normal
7884 <!-- _______________________________________________________________________ -->
7886 <a name="int_smul_overflow">
7887 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7894 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7895 on any integer bit width.</p>
7898 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7899 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7900 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7905 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7906 a signed multiplication of the two arguments, and indicate whether an
7907 overflow occurred during the signed multiplication.</p>
7910 <p>The arguments (%a and %b) and the first element of the result structure may
7911 be of integer types of any bit width, but they must have the same bit
7912 width. The second element of the result structure must be of
7913 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7914 undergo signed multiplication.</p>
7917 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7918 a signed multiplication of the two arguments. They return a structure —
7919 the first element of which is the multiplication, and the second element of
7920 which is a bit specifying if the signed multiplication resulted in an
7925 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7926 %sum = extractvalue {i32, i1} %res, 0
7927 %obit = extractvalue {i32, i1} %res, 1
7928 br i1 %obit, label %overflow, label %normal
7933 <!-- _______________________________________________________________________ -->
7935 <a name="int_umul_overflow">
7936 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
7943 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
7944 on any integer bit width.</p>
7947 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7948 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7949 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7953 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7954 a unsigned multiplication of the two arguments, and indicate whether an
7955 overflow occurred during the unsigned multiplication.</p>
7958 <p>The arguments (%a and %b) and the first element of the result structure may
7959 be of integer types of any bit width, but they must have the same bit
7960 width. The second element of the result structure must be of
7961 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7962 undergo unsigned multiplication.</p>
7965 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7966 an unsigned multiplication of the two arguments. They return a structure
7967 — the first element of which is the multiplication, and the second
7968 element of which is a bit specifying if the unsigned multiplication resulted
7973 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7974 %sum = extractvalue {i32, i1} %res, 0
7975 %obit = extractvalue {i32, i1} %res, 1
7976 br i1 %obit, label %overflow, label %normal
7983 <!-- ======================================================================= -->
7985 <a name="spec_arithmetic">Specialised Arithmetic Intrinsics</a>
7988 <!-- _______________________________________________________________________ -->
7991 <a name="fmuladd">'<tt>llvm.fmuladd.*</tt>' Intrinsic</a>
7998 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
7999 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8003 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsic functions represent multiply-add
8004 expressions that can be fused if the code generator determines that the fused
8005 expression would be legal and efficient.</p>
8008 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsics each take three arguments: two
8009 multiplicands, a and b, and an addend c.</p>
8012 <p>The expression:</p>
8014 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8016 <p>is equivalent to the expression a * b + c, except that rounding will not be
8017 performed between the multiplication and addition steps if the code generator
8018 fuses the operations. Fusion is not guaranteed, even if the target platform
8019 supports it. If a fused multiply-add is required the corresponding llvm.fma.*
8020 intrinsic function should be used instead.</p>
8024 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8029 <!-- ======================================================================= -->
8031 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
8036 <p>For most target platforms, half precision floating point is a storage-only
8037 format. This means that it is
8038 a dense encoding (in memory) but does not support computation in the
8041 <p>This means that code must first load the half-precision floating point
8042 value as an i16, then convert it to float with <a
8043 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
8044 Computation can then be performed on the float value (including extending to
8045 double etc). To store the value back to memory, it is first converted to
8046 float if needed, then converted to i16 with
8047 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
8048 storing as an i16 value.</p>
8050 <!-- _______________________________________________________________________ -->
8052 <a name="int_convert_to_fp16">
8053 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
8061 declare i16 @llvm.convert.to.fp16(f32 %a)
8065 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8066 a conversion from single precision floating point format to half precision
8067 floating point format.</p>
8070 <p>The intrinsic function contains single argument - the value to be
8074 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8075 a conversion from single precision floating point format to half precision
8076 floating point format. The return value is an <tt>i16</tt> which
8077 contains the converted number.</p>
8081 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8082 store i16 %res, i16* @x, align 2
8087 <!-- _______________________________________________________________________ -->
8089 <a name="int_convert_from_fp16">
8090 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
8098 declare f32 @llvm.convert.from.fp16(i16 %a)
8102 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
8103 a conversion from half precision floating point format to single precision
8104 floating point format.</p>
8107 <p>The intrinsic function contains single argument - the value to be
8111 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
8112 conversion from half single precision floating point format to single
8113 precision floating point format. The input half-float value is represented by
8114 an <tt>i16</tt> value.</p>
8118 %a = load i16* @x, align 2
8119 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8126 <!-- ======================================================================= -->
8128 <a name="int_debugger">Debugger Intrinsics</a>
8133 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
8134 prefix), are described in
8135 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
8136 Level Debugging</a> document.</p>
8140 <!-- ======================================================================= -->
8142 <a name="int_eh">Exception Handling Intrinsics</a>
8147 <p>The LLVM exception handling intrinsics (which all start with
8148 <tt>llvm.eh.</tt> prefix), are described in
8149 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
8150 Handling</a> document.</p>
8154 <!-- ======================================================================= -->
8156 <a name="int_trampoline">Trampoline Intrinsics</a>
8161 <p>These intrinsics make it possible to excise one parameter, marked with
8162 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
8163 The result is a callable
8164 function pointer lacking the nest parameter - the caller does not need to
8165 provide a value for it. Instead, the value to use is stored in advance in a
8166 "trampoline", a block of memory usually allocated on the stack, which also
8167 contains code to splice the nest value into the argument list. This is used
8168 to implement the GCC nested function address extension.</p>
8170 <p>For example, if the function is
8171 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
8172 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
8175 <pre class="doc_code">
8176 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8177 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8178 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8179 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8180 %fp = bitcast i8* %p to i32 (i32, i32)*
8183 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
8184 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
8186 <!-- _______________________________________________________________________ -->
8189 '<tt>llvm.init.trampoline</tt>' Intrinsic
8197 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8201 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
8202 turning it into a trampoline.</p>
8205 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
8206 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
8207 sufficiently aligned block of memory; this memory is written to by the
8208 intrinsic. Note that the size and the alignment are target-specific - LLVM
8209 currently provides no portable way of determining them, so a front-end that
8210 generates this intrinsic needs to have some target-specific knowledge.
8211 The <tt>func</tt> argument must hold a function bitcast to
8212 an <tt>i8*</tt>.</p>
8215 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
8216 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
8217 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
8218 which can be <a href="#int_trampoline">bitcast (to a new function) and
8219 called</a>. The new function's signature is the same as that of
8220 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
8221 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
8222 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
8223 with the same argument list, but with <tt>nval</tt> used for the missing
8224 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
8225 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
8226 to the returned function pointer is undefined.</p>
8229 <!-- _______________________________________________________________________ -->
8232 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
8240 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8244 <p>This performs any required machine-specific adjustment to the address of a
8245 trampoline (passed as <tt>tramp</tt>).</p>
8248 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
8249 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
8253 <p>On some architectures the address of the code to be executed needs to be
8254 different to the address where the trampoline is actually stored. This
8255 intrinsic returns the executable address corresponding to <tt>tramp</tt>
8256 after performing the required machine specific adjustments.
8257 The pointer returned can then be <a href="#int_trampoline"> bitcast and
8265 <!-- ======================================================================= -->
8267 <a name="int_memorymarkers">Memory Use Markers</a>
8272 <p>This class of intrinsics exists to information about the lifetime of memory
8273 objects and ranges where variables are immutable.</p>
8275 <!-- _______________________________________________________________________ -->
8277 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8284 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8288 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start 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 before 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. A load from the pointer that
8300 precedes this intrinsic can be replaced with
8301 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8305 <!-- _______________________________________________________________________ -->
8307 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8314 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8318 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8319 object's lifetime.</p>
8322 <p>The first argument is a constant integer representing the size of the
8323 object, or -1 if it is variable sized. The second argument is a pointer to
8327 <p>This intrinsic indicates that after this point in the code, the value of the
8328 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8329 never be used and has an undefined value. Any stores into the memory object
8330 following this intrinsic may be removed as dead.
8334 <!-- _______________________________________________________________________ -->
8336 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8343 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8347 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8348 a memory object will not change.</p>
8351 <p>The first argument is a constant integer representing the size of the
8352 object, or -1 if it is variable sized. The second argument is a pointer to
8356 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8357 the return value, the referenced memory location is constant and
8362 <!-- _______________________________________________________________________ -->
8364 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8371 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8375 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8376 a memory object are mutable.</p>
8379 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8380 The second argument is a constant integer representing the size of the
8381 object, or -1 if it is variable sized and the third argument is a pointer
8385 <p>This intrinsic indicates that the memory is mutable again.</p>
8391 <!-- ======================================================================= -->
8393 <a name="int_general">General Intrinsics</a>
8398 <p>This class of intrinsics is designed to be generic and has no specific
8401 <!-- _______________________________________________________________________ -->
8403 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8410 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8414 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8417 <p>The first argument is a pointer to a value, the second is a pointer to a
8418 global string, the third is a pointer to a global string which is the source
8419 file name, and the last argument is the line number.</p>
8422 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8423 This can be useful for special purpose optimizations that want to look for
8424 these annotations. These have no other defined use; they are ignored by code
8425 generation and optimization.</p>
8429 <!-- _______________________________________________________________________ -->
8431 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8437 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8438 any integer bit width.</p>
8441 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8442 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8443 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8444 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8445 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8449 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8452 <p>The first argument is an integer value (result of some expression), the
8453 second is a pointer to a global string, the third is a pointer to a global
8454 string which is the source file name, and the last argument is the line
8455 number. It returns the value of the first argument.</p>
8458 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8459 arbitrary strings. This can be useful for special purpose optimizations that
8460 want to look for these annotations. These have no other defined use; they
8461 are ignored by code generation and optimization.</p>
8465 <!-- _______________________________________________________________________ -->
8467 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8474 declare void @llvm.trap() noreturn nounwind
8478 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8484 <p>This intrinsic is lowered to the target dependent trap instruction. If the
8485 target does not have a trap instruction, this intrinsic will be lowered to
8486 a call of the <tt>abort()</tt> function.</p>
8490 <!-- _______________________________________________________________________ -->
8492 <a name="int_debugtrap">'<tt>llvm.debugtrap</tt>' Intrinsic</a>
8499 declare void @llvm.debugtrap() nounwind
8503 <p>The '<tt>llvm.debugtrap</tt>' intrinsic.</p>
8509 <p>This intrinsic is lowered to code which is intended to cause an execution
8510 trap with the intention of requesting the attention of a debugger.</p>
8514 <!-- _______________________________________________________________________ -->
8516 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8523 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8527 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8528 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8529 ensure that it is placed on the stack before local variables.</p>
8532 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8533 arguments. The first argument is the value loaded from the stack
8534 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8535 that has enough space to hold the value of the guard.</p>
8538 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8539 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8540 stack. This is to ensure that if a local variable on the stack is
8541 overwritten, it will destroy the value of the guard. When the function exits,
8542 the guard on the stack is checked against the original guard. If they are
8543 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8548 <!-- _______________________________________________________________________ -->
8550 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8557 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8558 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8562 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8563 the optimizers to determine at compile time whether a) an operation (like
8564 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8565 runtime check for overflow isn't necessary. An object in this context means
8566 an allocation of a specific class, structure, array, or other object.</p>
8569 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8570 argument is a pointer to or into the <tt>object</tt>. The second argument
8571 is a boolean and determines whether <tt>llvm.objectsize</tt> returns 0 (if
8572 true) or -1 (if false) when the object size is unknown.
8573 The second argument only accepts constants.</p>
8576 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to a constant representing
8577 the size of the object concerned. If the size cannot be determined at compile
8578 time, <tt>llvm.objectsize</tt> returns <tt>i32/i64 -1 or 0</tt>
8579 (depending on the <tt>min</tt> argument).</p>
8582 <!-- _______________________________________________________________________ -->
8584 <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a>
8591 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8592 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8596 <p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the
8597 most probable) value of <tt>val</tt>, which can be used by optimizers.</p>
8600 <p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first
8601 argument is a value. The second argument is an expected value, this needs to
8602 be a constant value, variables are not allowed.</p>
8605 <p>This intrinsic is lowered to the <tt>val</tt>.</p>
8611 <!-- *********************************************************************** -->
8614 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8615 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8616 <a href="http://validator.w3.org/check/referer"><img
8617 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8619 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8620 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8621 Last modified: $Date$