1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <h1>LLVM Language Reference Manual</h1>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a>
25 <li><a href="#linkage_private">'<tt>private</tt>' Linkage</a></li>
26 <li><a href="#linkage_linker_private">'<tt>linker_private</tt>' Linkage</a></li>
27 <li><a href="#linkage_linker_private_weak">'<tt>linker_private_weak</tt>' Linkage</a></li>
28 <li><a href="#linkage_linker_private_weak_def_auto">'<tt>linker_private_weak_def_auto</tt>' Linkage</a></li>
29 <li><a href="#linkage_internal">'<tt>internal</tt>' Linkage</a></li>
30 <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
31 <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
32 <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
33 <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
34 <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
35 <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
36 <li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
37 <li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
38 <li><a href="#linkage_external">'<tt>external</tt>' Linkage</a></li>
39 <li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
40 <li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
43 <li><a href="#callingconv">Calling Conventions</a></li>
44 <li><a href="#namedtypes">Named Types</a></li>
45 <li><a href="#globalvars">Global Variables</a></li>
46 <li><a href="#functionstructure">Functions</a></li>
47 <li><a href="#aliasstructure">Aliases</a></li>
48 <li><a href="#namedmetadatastructure">Named Metadata</a></li>
49 <li><a href="#paramattrs">Parameter Attributes</a></li>
50 <li><a href="#fnattrs">Function Attributes</a></li>
51 <li><a href="#gc">Garbage Collector Names</a></li>
52 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
53 <li><a href="#datalayout">Data Layout</a></li>
54 <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
55 <li><a href="#volatile">Volatile Memory Accesses</a></li>
56 <li><a href="#memmodel">Memory Model for Concurrent Operations</a></li>
57 <li><a href="#ordering">Atomic Memory Ordering Constraints</a></li>
60 <li><a href="#typesystem">Type System</a>
62 <li><a href="#t_classifications">Type Classifications</a></li>
63 <li><a href="#t_primitive">Primitive Types</a>
65 <li><a href="#t_integer">Integer Type</a></li>
66 <li><a href="#t_floating">Floating Point Types</a></li>
67 <li><a href="#t_x86mmx">X86mmx Type</a></li>
68 <li><a href="#t_void">Void Type</a></li>
69 <li><a href="#t_label">Label Type</a></li>
70 <li><a href="#t_metadata">Metadata Type</a></li>
73 <li><a href="#t_derived">Derived Types</a>
75 <li><a href="#t_aggregate">Aggregate Types</a>
77 <li><a href="#t_array">Array Type</a></li>
78 <li><a href="#t_struct">Structure Type</a></li>
79 <li><a href="#t_opaque">Opaque Structure Types</a></li>
80 <li><a href="#t_vector">Vector Type</a></li>
83 <li><a href="#t_function">Function Type</a></li>
84 <li><a href="#t_pointer">Pointer Type</a></li>
89 <li><a href="#constants">Constants</a>
91 <li><a href="#simpleconstants">Simple Constants</a></li>
92 <li><a href="#complexconstants">Complex Constants</a></li>
93 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
94 <li><a href="#undefvalues">Undefined Values</a></li>
95 <li><a href="#poisonvalues">Poison Values</a></li>
96 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
97 <li><a href="#constantexprs">Constant Expressions</a></li>
100 <li><a href="#othervalues">Other Values</a>
102 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
103 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a>
105 <li><a href="#tbaa">'<tt>tbaa</tt>' Metadata</a></li>
106 <li><a href="#fpaccuracy">'<tt>fpaccuracy</tt>' Metadata</a></li>
111 <li><a href="#module_flags">Module Flags Metadata</a>
113 <li><a href="#objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a></li>
116 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
118 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
119 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
120 Global Variable</a></li>
121 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
122 Global Variable</a></li>
123 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
124 Global Variable</a></li>
127 <li><a href="#instref">Instruction Reference</a>
129 <li><a href="#terminators">Terminator Instructions</a>
131 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
132 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
133 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
134 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
135 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
136 <li><a href="#i_resume">'<tt>resume</tt>' Instruction</a></li>
137 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
140 <li><a href="#binaryops">Binary Operations</a>
142 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
143 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
144 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
145 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
146 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
147 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
148 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
149 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
150 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
151 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
152 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
153 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
156 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
158 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
159 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
160 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
161 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
162 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
163 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
166 <li><a href="#vectorops">Vector Operations</a>
168 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
169 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
170 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
173 <li><a href="#aggregateops">Aggregate Operations</a>
175 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
176 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
179 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
181 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
182 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
183 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
184 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
185 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
186 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
187 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
190 <li><a href="#convertops">Conversion Operations</a>
192 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
193 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
194 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
195 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
196 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
197 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
198 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
199 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
200 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
201 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
202 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
203 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
206 <li><a href="#otherops">Other Operations</a>
208 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
209 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
210 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
211 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
212 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
213 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
214 <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
219 <li><a href="#intrinsics">Intrinsic Functions</a>
221 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
223 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
224 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
225 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
228 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
230 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
231 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
232 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
235 <li><a href="#int_codegen">Code Generator Intrinsics</a>
237 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
238 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
239 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
240 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
241 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
242 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
243 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
246 <li><a href="#int_libc">Standard C Library Intrinsics</a>
248 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
249 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
250 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
251 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
252 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
253 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
254 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
255 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
256 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
257 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
258 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
261 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
263 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
264 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
265 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
266 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
269 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
271 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
272 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
273 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
274 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
275 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
276 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
279 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
281 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
282 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
285 <li><a href="#int_debugger">Debugger intrinsics</a></li>
286 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
287 <li><a href="#int_trampoline">Trampoline Intrinsics</a>
289 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
290 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
293 <li><a href="#int_memorymarkers">Memory Use Markers</a>
295 <li><a href="#int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a></li>
296 <li><a href="#int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a></li>
297 <li><a href="#int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a></li>
298 <li><a href="#int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a></li>
301 <li><a href="#int_general">General intrinsics</a>
303 <li><a href="#int_var_annotation">
304 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
305 <li><a href="#int_annotation">
306 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
307 <li><a href="#int_trap">
308 '<tt>llvm.trap</tt>' Intrinsic</a></li>
309 <li><a href="#int_stackprotector">
310 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
311 <li><a href="#int_objectsize">
312 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
313 <li><a href="#int_expect">
314 '<tt>llvm.expect</tt>' Intrinsic</a></li>
321 <div class="doc_author">
322 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
323 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
326 <!-- *********************************************************************** -->
327 <h2><a name="abstract">Abstract</a></h2>
328 <!-- *********************************************************************** -->
332 <p>This document is a reference manual for the LLVM assembly language. LLVM is
333 a Static Single Assignment (SSA) based representation that provides type
334 safety, low-level operations, flexibility, and the capability of representing
335 'all' high-level languages cleanly. It is the common code representation
336 used throughout all phases of the LLVM compilation strategy.</p>
340 <!-- *********************************************************************** -->
341 <h2><a name="introduction">Introduction</a></h2>
342 <!-- *********************************************************************** -->
346 <p>The LLVM code representation is designed to be used in three different forms:
347 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
348 for fast loading by a Just-In-Time compiler), and as a human readable
349 assembly language representation. This allows LLVM to provide a powerful
350 intermediate representation for efficient compiler transformations and
351 analysis, while providing a natural means to debug and visualize the
352 transformations. The three different forms of LLVM are all equivalent. This
353 document describes the human readable representation and notation.</p>
355 <p>The LLVM representation aims to be light-weight and low-level while being
356 expressive, typed, and extensible at the same time. It aims to be a
357 "universal IR" of sorts, by being at a low enough level that high-level ideas
358 may be cleanly mapped to it (similar to how microprocessors are "universal
359 IR's", allowing many source languages to be mapped to them). By providing
360 type information, LLVM can be used as the target of optimizations: for
361 example, through pointer analysis, it can be proven that a C automatic
362 variable is never accessed outside of the current function, allowing it to
363 be promoted to a simple SSA value instead of a memory location.</p>
365 <!-- _______________________________________________________________________ -->
367 <a name="wellformed">Well-Formedness</a>
372 <p>It is important to note that this document describes 'well formed' LLVM
373 assembly language. There is a difference between what the parser accepts and
374 what is considered 'well formed'. For example, the following instruction is
375 syntactically okay, but not well formed:</p>
377 <pre class="doc_code">
378 %x = <a href="#i_add">add</a> i32 1, %x
381 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
382 LLVM infrastructure provides a verification pass that may be used to verify
383 that an LLVM module is well formed. This pass is automatically run by the
384 parser after parsing input assembly and by the optimizer before it outputs
385 bitcode. The violations pointed out by the verifier pass indicate bugs in
386 transformation passes or input to the parser.</p>
392 <!-- Describe the typesetting conventions here. -->
394 <!-- *********************************************************************** -->
395 <h2><a name="identifiers">Identifiers</a></h2>
396 <!-- *********************************************************************** -->
400 <p>LLVM identifiers come in two basic types: global and local. Global
401 identifiers (functions, global variables) begin with the <tt>'@'</tt>
402 character. Local identifiers (register names, types) begin with
403 the <tt>'%'</tt> character. Additionally, there are three different formats
404 for identifiers, for different purposes:</p>
407 <li>Named values are represented as a string of characters with their prefix.
408 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
409 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
410 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
411 other characters in their names can be surrounded with quotes. Special
412 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
413 ASCII code for the character in hexadecimal. In this way, any character
414 can be used in a name value, even quotes themselves.</li>
416 <li>Unnamed values are represented as an unsigned numeric value with their
417 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
419 <li>Constants, which are described in a <a href="#constants">section about
420 constants</a>, below.</li>
423 <p>LLVM requires that values start with a prefix for two reasons: Compilers
424 don't need to worry about name clashes with reserved words, and the set of
425 reserved words may be expanded in the future without penalty. Additionally,
426 unnamed identifiers allow a compiler to quickly come up with a temporary
427 variable without having to avoid symbol table conflicts.</p>
429 <p>Reserved words in LLVM are very similar to reserved words in other
430 languages. There are keywords for different opcodes
431 ('<tt><a href="#i_add">add</a></tt>',
432 '<tt><a href="#i_bitcast">bitcast</a></tt>',
433 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
434 ('<tt><a href="#t_void">void</a></tt>',
435 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
436 reserved words cannot conflict with variable names, because none of them
437 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
439 <p>Here is an example of LLVM code to multiply the integer variable
440 '<tt>%X</tt>' by 8:</p>
444 <pre class="doc_code">
445 %result = <a href="#i_mul">mul</a> i32 %X, 8
448 <p>After strength reduction:</p>
450 <pre class="doc_code">
451 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
454 <p>And the hard way:</p>
456 <pre class="doc_code">
457 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
458 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
459 %result = <a href="#i_add">add</a> i32 %1, %1
462 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
463 lexical features of LLVM:</p>
466 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
469 <li>Unnamed temporaries are created when the result of a computation is not
470 assigned to a named value.</li>
472 <li>Unnamed temporaries are numbered sequentially</li>
475 <p>It also shows a convention that we follow in this document. When
476 demonstrating instructions, we will follow an instruction with a comment that
477 defines the type and name of value produced. Comments are shown in italic
482 <!-- *********************************************************************** -->
483 <h2><a name="highlevel">High Level Structure</a></h2>
484 <!-- *********************************************************************** -->
486 <!-- ======================================================================= -->
488 <a name="modulestructure">Module Structure</a>
493 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
494 of the input programs. Each module consists of functions, global variables,
495 and symbol table entries. Modules may be combined together with the LLVM
496 linker, which merges function (and global variable) definitions, resolves
497 forward declarations, and merges symbol table entries. Here is an example of
498 the "hello world" module:</p>
500 <pre class="doc_code">
501 <i>; Declare the string constant as a global constant.</i>
502 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
504 <i>; External declaration of the puts function</i>
505 <a href="#functionstructure">declare</a> i32 @puts(i8*) <i>; i32 (i8*)* </i>
507 <i>; Definition of main function</i>
508 define i32 @main() { <i>; i32()* </i>
509 <i>; Convert [13 x i8]* to i8 *...</i>
510 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8*</i>
512 <i>; Call puts function to write out the string to stdout.</i>
513 <a href="#i_call">call</a> i32 @puts(i8* %cast210) <i>; i32</i>
514 <a href="#i_ret">ret</a> i32 0
517 <i>; Named metadata</i>
518 !1 = metadata !{i32 41}
522 <p>This example is made up of a <a href="#globalvars">global variable</a> named
523 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
524 a <a href="#functionstructure">function definition</a> for
525 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
528 <p>In general, a module is made up of a list of global values, where both
529 functions and global variables are global values. Global values are
530 represented by a pointer to a memory location (in this case, a pointer to an
531 array of char, and a pointer to a function), and have one of the
532 following <a href="#linkage">linkage types</a>.</p>
536 <!-- ======================================================================= -->
538 <a name="linkage">Linkage Types</a>
543 <p>All Global Variables and Functions have one of the following types of
547 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
548 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
549 by objects in the current module. In particular, linking code into a
550 module with an private global value may cause the private to be renamed as
551 necessary to avoid collisions. Because the symbol is private to the
552 module, all references can be updated. This doesn't show up in any symbol
553 table in the object file.</dd>
555 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
556 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
557 assembler and evaluated by the linker. Unlike normal strong symbols, they
558 are removed by the linker from the final linked image (executable or
559 dynamic library).</dd>
561 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
562 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
563 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
564 linker. The symbols are removed by the linker from the final linked image
565 (executable or dynamic library).</dd>
567 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
568 <dd>Similar to "<tt>linker_private_weak</tt>", but it's known that the address
569 of the object is not taken. For instance, functions that had an inline
570 definition, but the compiler decided not to inline it. Note,
571 unlike <tt>linker_private</tt> and <tt>linker_private_weak</tt>,
572 <tt>linker_private_weak_def_auto</tt> may have only <tt>default</tt>
573 visibility. The symbols are removed by the linker from the final linked
574 image (executable or dynamic library).</dd>
576 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
577 <dd>Similar to private, but the value shows as a local symbol
578 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
579 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
581 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
582 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
583 into the object file corresponding to the LLVM module. They exist to
584 allow inlining and other optimizations to take place given knowledge of
585 the definition of the global, which is known to be somewhere outside the
586 module. Globals with <tt>available_externally</tt> linkage are allowed to
587 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
588 This linkage type is only allowed on definitions, not declarations.</dd>
590 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
591 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
592 the same name when linkage occurs. This can be used to implement
593 some forms of inline functions, templates, or other code which must be
594 generated in each translation unit that uses it, but where the body may
595 be overridden with a more definitive definition later. Unreferenced
596 <tt>linkonce</tt> globals are allowed to be discarded. Note that
597 <tt>linkonce</tt> linkage does not actually allow the optimizer to
598 inline the body of this function into callers because it doesn't know if
599 this definition of the function is the definitive definition within the
600 program or whether it will be overridden by a stronger definition.
601 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
604 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
605 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
606 <tt>linkonce</tt> linkage, except that unreferenced globals with
607 <tt>weak</tt> linkage may not be discarded. This is used for globals that
608 are declared "weak" in C source code.</dd>
610 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
611 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
612 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
614 Symbols with "<tt>common</tt>" linkage are merged in the same way as
615 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
616 <tt>common</tt> symbols may not have an explicit section,
617 must have a zero initializer, and may not be marked '<a
618 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
619 have common linkage.</dd>
622 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
623 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
624 pointer to array type. When two global variables with appending linkage
625 are linked together, the two global arrays are appended together. This is
626 the LLVM, typesafe, equivalent of having the system linker append together
627 "sections" with identical names when .o files are linked.</dd>
629 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
630 <dd>The semantics of this linkage follow the ELF object file model: the symbol
631 is weak until linked, if not linked, the symbol becomes null instead of
632 being an undefined reference.</dd>
634 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
635 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
636 <dd>Some languages allow differing globals to be merged, such as two functions
637 with different semantics. Other languages, such as <tt>C++</tt>, ensure
638 that only equivalent globals are ever merged (the "one definition rule"
639 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
640 and <tt>weak_odr</tt> linkage types to indicate that the global will only
641 be merged with equivalent globals. These linkage types are otherwise the
642 same as their non-<tt>odr</tt> versions.</dd>
644 <dt><tt><b><a name="linkage_external">external</a></b></tt></dt>
645 <dd>If none of the above identifiers are used, the global is externally
646 visible, meaning that it participates in linkage and can be used to
647 resolve external symbol references.</dd>
650 <p>The next two types of linkage are targeted for Microsoft Windows platform
651 only. They are designed to support importing (exporting) symbols from (to)
652 DLLs (Dynamic Link Libraries).</p>
655 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
656 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
657 or variable via a global pointer to a pointer that is set up by the DLL
658 exporting the symbol. On Microsoft Windows targets, the pointer name is
659 formed by combining <code>__imp_</code> and the function or variable
662 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
663 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
664 pointer to a pointer in a DLL, so that it can be referenced with the
665 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
666 name is formed by combining <code>__imp_</code> and the function or
670 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
671 another module defined a "<tt>.LC0</tt>" variable and was linked with this
672 one, one of the two would be renamed, preventing a collision. Since
673 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
674 declarations), they are accessible outside of the current module.</p>
676 <p>It is illegal for a function <i>declaration</i> to have any linkage type
677 other than <tt>external</tt>, <tt>dllimport</tt>
678 or <tt>extern_weak</tt>.</p>
680 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
681 or <tt>weak_odr</tt> linkages.</p>
685 <!-- ======================================================================= -->
687 <a name="callingconv">Calling Conventions</a>
692 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
693 and <a href="#i_invoke">invokes</a> can all have an optional calling
694 convention specified for the call. The calling convention of any pair of
695 dynamic caller/callee must match, or the behavior of the program is
696 undefined. The following calling conventions are supported by LLVM, and more
697 may be added in the future:</p>
700 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
701 <dd>This calling convention (the default if no other calling convention is
702 specified) matches the target C calling conventions. This calling
703 convention supports varargs function calls and tolerates some mismatch in
704 the declared prototype and implemented declaration of the function (as
707 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
708 <dd>This calling convention attempts to make calls as fast as possible
709 (e.g. by passing things in registers). This calling convention allows the
710 target to use whatever tricks it wants to produce fast code for the
711 target, without having to conform to an externally specified ABI
712 (Application Binary Interface).
713 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
714 when this or the GHC convention is used.</a> This calling convention
715 does not support varargs and requires the prototype of all callees to
716 exactly match the prototype of the function definition.</dd>
718 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
719 <dd>This calling convention attempts to make code in the caller as efficient
720 as possible under the assumption that the call is not commonly executed.
721 As such, these calls often preserve all registers so that the call does
722 not break any live ranges in the caller side. 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>cc <em>10</em></tt>" - GHC convention</b>:</dt>
727 <dd>This calling convention has been implemented specifically for use by the
728 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
729 It passes everything in registers, going to extremes to achieve this by
730 disabling callee save registers. This calling convention should not be
731 used lightly but only for specific situations such as an alternative to
732 the <em>register pinning</em> performance technique often used when
733 implementing functional programming languages.At the moment only X86
734 supports this convention and it has the following limitations:
736 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
737 floating point types are supported.</li>
738 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
739 6 floating point parameters.</li>
741 This calling convention supports
742 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
743 requires both the caller and callee are using it.
746 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
747 <dd>Any calling convention may be specified by number, allowing
748 target-specific calling conventions to be used. Target specific calling
749 conventions start at 64.</dd>
752 <p>More calling conventions can be added/defined on an as-needed basis, to
753 support Pascal conventions or any other well-known target-independent
758 <!-- ======================================================================= -->
760 <a name="visibility">Visibility Styles</a>
765 <p>All Global Variables and Functions have one of the following visibility
769 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
770 <dd>On targets that use the ELF object file format, default visibility means
771 that the declaration is visible to other modules and, in shared libraries,
772 means that the declared entity may be overridden. On Darwin, default
773 visibility means that the declaration is visible to other modules. Default
774 visibility corresponds to "external linkage" in the language.</dd>
776 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
777 <dd>Two declarations of an object with hidden visibility refer to the same
778 object if they are in the same shared object. Usually, hidden visibility
779 indicates that the symbol will not be placed into the dynamic symbol
780 table, so no other module (executable or shared library) can reference it
783 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
784 <dd>On ELF, protected visibility indicates that the symbol will be placed in
785 the dynamic symbol table, but that references within the defining module
786 will bind to the local symbol. That is, the symbol cannot be overridden by
792 <!-- ======================================================================= -->
794 <a name="namedtypes">Named Types</a>
799 <p>LLVM IR allows you to specify name aliases for certain types. This can make
800 it easier to read the IR and make the IR more condensed (particularly when
801 recursive types are involved). An example of a name specification is:</p>
803 <pre class="doc_code">
804 %mytype = type { %mytype*, i32 }
807 <p>You may give a name to any <a href="#typesystem">type</a> except
808 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
809 is expected with the syntax "%mytype".</p>
811 <p>Note that type names are aliases for the structural type that they indicate,
812 and that you can therefore specify multiple names for the same type. This
813 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
814 uses structural typing, the name is not part of the type. When printing out
815 LLVM IR, the printer will pick <em>one name</em> to render all types of a
816 particular shape. This means that if you have code where two different
817 source types end up having the same LLVM type, that the dumper will sometimes
818 print the "wrong" or unexpected type. This is an important design point and
819 isn't going to change.</p>
823 <!-- ======================================================================= -->
825 <a name="globalvars">Global Variables</a>
830 <p>Global variables define regions of memory allocated at compilation time
831 instead of run-time. Global variables may optionally be initialized, may
832 have an explicit section to be placed in, and may have an optional explicit
833 alignment specified. A variable may be defined as "thread_local", which
834 means that it will not be shared by threads (each thread will have a
835 separated copy of the variable). A variable may be defined as a global
836 "constant," which indicates that the contents of the variable
837 will <b>never</b> be modified (enabling better optimization, allowing the
838 global data to be placed in the read-only section of an executable, etc).
839 Note that variables that need runtime initialization cannot be marked
840 "constant" as there is a store to the variable.</p>
842 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
843 constant, even if the final definition of the global is not. This capability
844 can be used to enable slightly better optimization of the program, but
845 requires the language definition to guarantee that optimizations based on the
846 'constantness' are valid for the translation units that do not include the
849 <p>As SSA values, global variables define pointer values that are in scope
850 (i.e. they dominate) all basic blocks in the program. Global variables
851 always define a pointer to their "content" type because they describe a
852 region of memory, and all memory objects in LLVM are accessed through
855 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
856 that the address is not significant, only the content. Constants marked
857 like this can be merged with other constants if they have the same
858 initializer. Note that a constant with significant address <em>can</em>
859 be merged with a <tt>unnamed_addr</tt> constant, the result being a
860 constant whose address is significant.</p>
862 <p>A global variable may be declared to reside in a target-specific numbered
863 address space. For targets that support them, address spaces may affect how
864 optimizations are performed and/or what target instructions are used to
865 access the variable. The default address space is zero. The address space
866 qualifier must precede any other attributes.</p>
868 <p>LLVM allows an explicit section to be specified for globals. If the target
869 supports it, it will emit globals to the section specified.</p>
871 <p>An explicit alignment may be specified for a global, which must be a power
872 of 2. If not present, or if the alignment is set to zero, the alignment of
873 the global is set by the target to whatever it feels convenient. If an
874 explicit alignment is specified, the global is forced to have exactly that
875 alignment. Targets and optimizers are not allowed to over-align the global
876 if the global has an assigned section. In this case, the extra alignment
877 could be observable: for example, code could assume that the globals are
878 densely packed in their section and try to iterate over them as an array,
879 alignment padding would break this iteration.</p>
881 <p>For example, the following defines a global in a numbered address space with
882 an initializer, section, and alignment:</p>
884 <pre class="doc_code">
885 @G = addrspace(5) constant float 1.0, section "foo", align 4
891 <!-- ======================================================================= -->
893 <a name="functionstructure">Functions</a>
898 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
899 optional <a href="#linkage">linkage type</a>, an optional
900 <a href="#visibility">visibility style</a>, an optional
901 <a href="#callingconv">calling convention</a>,
902 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
903 <a href="#paramattrs">parameter attribute</a> for the return type, a function
904 name, a (possibly empty) argument list (each with optional
905 <a href="#paramattrs">parameter attributes</a>), optional
906 <a href="#fnattrs">function attributes</a>, an optional section, an optional
907 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
908 curly brace, a list of basic blocks, and a closing curly brace.</p>
910 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
911 optional <a href="#linkage">linkage type</a>, an optional
912 <a href="#visibility">visibility style</a>, an optional
913 <a href="#callingconv">calling convention</a>,
914 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
915 <a href="#paramattrs">parameter attribute</a> for the return type, a function
916 name, a possibly empty list of arguments, an optional alignment, and an
917 optional <a href="#gc">garbage collector name</a>.</p>
919 <p>A function definition contains a list of basic blocks, forming the CFG
920 (Control Flow Graph) for the function. Each basic block may optionally start
921 with a label (giving the basic block a symbol table entry), contains a list
922 of instructions, and ends with a <a href="#terminators">terminator</a>
923 instruction (such as a branch or function return).</p>
925 <p>The first basic block in a function is special in two ways: it is immediately
926 executed on entrance to the function, and it is not allowed to have
927 predecessor basic blocks (i.e. there can not be any branches to the entry
928 block of a function). Because the block can have no predecessors, it also
929 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
931 <p>LLVM allows an explicit section to be specified for functions. If the target
932 supports it, it will emit functions to the section specified.</p>
934 <p>An explicit alignment may be specified for a function. If not present, or if
935 the alignment is set to zero, the alignment of the function is set by the
936 target to whatever it feels convenient. If an explicit alignment is
937 specified, the function is forced to have at least that much alignment. All
938 alignments must be a power of 2.</p>
940 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
941 be significant and two identical functions can be merged.</p>
944 <pre class="doc_code">
945 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
946 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
947 <ResultType> @<FunctionName> ([argument list])
948 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
949 [<a href="#gc">gc</a>] { ... }
954 <!-- ======================================================================= -->
956 <a name="aliasstructure">Aliases</a>
961 <p>Aliases act as "second name" for the aliasee value (which can be either
962 function, global variable, another alias or bitcast of global value). Aliases
963 may have an optional <a href="#linkage">linkage type</a>, and an
964 optional <a href="#visibility">visibility style</a>.</p>
967 <pre class="doc_code">
968 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
973 <!-- ======================================================================= -->
975 <a name="namedmetadatastructure">Named Metadata</a>
980 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
981 nodes</a> (but not metadata strings) are the only valid operands for
982 a named metadata.</p>
985 <pre class="doc_code">
986 ; Some unnamed metadata nodes, which are referenced by the named metadata.
987 !0 = metadata !{metadata !"zero"}
988 !1 = metadata !{metadata !"one"}
989 !2 = metadata !{metadata !"two"}
991 !name = !{!0, !1, !2}
996 <!-- ======================================================================= -->
998 <a name="paramattrs">Parameter Attributes</a>
1003 <p>The return type and each parameter of a function type may have a set of
1004 <i>parameter attributes</i> associated with them. Parameter attributes are
1005 used to communicate additional information about the result or parameters of
1006 a function. Parameter attributes are considered to be part of the function,
1007 not of the function type, so functions with different parameter attributes
1008 can have the same function type.</p>
1010 <p>Parameter attributes are simple keywords that follow the type specified. If
1011 multiple parameter attributes are needed, they are space separated. For
1014 <pre class="doc_code">
1015 declare i32 @printf(i8* noalias nocapture, ...)
1016 declare i32 @atoi(i8 zeroext)
1017 declare signext i8 @returns_signed_char()
1020 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1021 <tt>readonly</tt>) come immediately after the argument list.</p>
1023 <p>Currently, only the following parameter attributes are defined:</p>
1026 <dt><tt><b>zeroext</b></tt></dt>
1027 <dd>This indicates to the code generator that the parameter or return value
1028 should be zero-extended to the extent required by the target's ABI (which
1029 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1030 parameter) or the callee (for a return value).</dd>
1032 <dt><tt><b>signext</b></tt></dt>
1033 <dd>This indicates to the code generator that the parameter or return value
1034 should be sign-extended to the extent required by the target's ABI (which
1035 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1038 <dt><tt><b>inreg</b></tt></dt>
1039 <dd>This indicates that this parameter or return value should be treated in a
1040 special target-dependent fashion during while emitting code for a function
1041 call or return (usually, by putting it in a register as opposed to memory,
1042 though some targets use it to distinguish between two different kinds of
1043 registers). Use of this attribute is target-specific.</dd>
1045 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1046 <dd><p>This indicates that the pointer parameter should really be passed by
1047 value to the function. The attribute implies that a hidden copy of the
1049 is made between the caller and the callee, so the callee is unable to
1050 modify the value in the callee. This attribute is only valid on LLVM
1051 pointer arguments. It is generally used to pass structs and arrays by
1052 value, but is also valid on pointers to scalars. The copy is considered
1053 to belong to the caller not the callee (for example,
1054 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1055 <tt>byval</tt> parameters). This is not a valid attribute for return
1058 <p>The byval attribute also supports specifying an alignment with
1059 the align attribute. It indicates the alignment of the stack slot to
1060 form and the known alignment of the pointer specified to the call site. If
1061 the alignment is not specified, then the code generator makes a
1062 target-specific assumption.</p></dd>
1064 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1065 <dd>This indicates that the pointer parameter specifies the address of a
1066 structure that is the return value of the function in the source program.
1067 This pointer must be guaranteed by the caller to be valid: loads and
1068 stores to the structure may be assumed by the callee to not to trap. This
1069 may only be applied to the first parameter. This is not a valid attribute
1070 for return values. </dd>
1072 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1073 <dd>This indicates that pointer values
1074 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1075 value do not alias pointer values which are not <i>based</i> on it,
1076 ignoring certain "irrelevant" dependencies.
1077 For a call to the parent function, dependencies between memory
1078 references from before or after the call and from those during the call
1079 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1080 return value used in that call.
1081 The caller shares the responsibility with the callee for ensuring that
1082 these requirements are met.
1083 For further details, please see the discussion of the NoAlias response in
1084 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1086 Note that this definition of <tt>noalias</tt> is intentionally
1087 similar to the definition of <tt>restrict</tt> in C99 for function
1088 arguments, though it is slightly weaker.
1090 For function return values, C99's <tt>restrict</tt> is not meaningful,
1091 while LLVM's <tt>noalias</tt> is.
1094 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1095 <dd>This indicates that the callee does not make any copies of the pointer
1096 that outlive the callee itself. This is not a valid attribute for return
1099 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1100 <dd>This indicates that the pointer parameter can be excised using the
1101 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1102 attribute for return values.</dd>
1107 <!-- ======================================================================= -->
1109 <a name="gc">Garbage Collector Names</a>
1114 <p>Each function may specify a garbage collector name, which is simply a
1117 <pre class="doc_code">
1118 define void @f() gc "name" { ... }
1121 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1122 collector which will cause the compiler to alter its output in order to
1123 support the named garbage collection algorithm.</p>
1127 <!-- ======================================================================= -->
1129 <a name="fnattrs">Function Attributes</a>
1134 <p>Function attributes are set to communicate additional information about a
1135 function. Function attributes are considered to be part of the function, not
1136 of the function type, so functions with different parameter attributes can
1137 have the same function type.</p>
1139 <p>Function attributes are simple keywords that follow the type specified. If
1140 multiple attributes are needed, they are space separated. For example:</p>
1142 <pre class="doc_code">
1143 define void @f() noinline { ... }
1144 define void @f() alwaysinline { ... }
1145 define void @f() alwaysinline optsize { ... }
1146 define void @f() optsize { ... }
1150 <dt><tt><b>address_safety</b></tt></dt>
1151 <dd>This attribute indicates that the address safety analysis
1152 is enabled for this function. </dd>
1154 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1155 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1156 the backend should forcibly align the stack pointer. Specify the
1157 desired alignment, which must be a power of two, in parentheses.
1159 <dt><tt><b>alwaysinline</b></tt></dt>
1160 <dd>This attribute indicates that the inliner should attempt to inline this
1161 function into callers whenever possible, ignoring any active inlining size
1162 threshold for this caller.</dd>
1164 <dt><tt><b>nonlazybind</b></tt></dt>
1165 <dd>This attribute suppresses lazy symbol binding for the function. This
1166 may make calls to the function faster, at the cost of extra program
1167 startup time if the function is not called during program startup.</dd>
1169 <dt><tt><b>inlinehint</b></tt></dt>
1170 <dd>This attribute indicates that the source code contained a hint that inlining
1171 this function is desirable (such as the "inline" keyword in C/C++). It
1172 is just a hint; it imposes no requirements on the inliner.</dd>
1174 <dt><tt><b>naked</b></tt></dt>
1175 <dd>This attribute disables prologue / epilogue emission for the function.
1176 This can have very system-specific consequences.</dd>
1178 <dt><tt><b>noimplicitfloat</b></tt></dt>
1179 <dd>This attributes disables implicit floating point instructions.</dd>
1181 <dt><tt><b>noinline</b></tt></dt>
1182 <dd>This attribute indicates that the inliner should never inline this
1183 function in any situation. This attribute may not be used together with
1184 the <tt>alwaysinline</tt> attribute.</dd>
1186 <dt><tt><b>noredzone</b></tt></dt>
1187 <dd>This attribute indicates that the code generator should not use a red
1188 zone, even if the target-specific ABI normally permits it.</dd>
1190 <dt><tt><b>noreturn</b></tt></dt>
1191 <dd>This function attribute indicates that the function never returns
1192 normally. This produces undefined behavior at runtime if the function
1193 ever does dynamically return.</dd>
1195 <dt><tt><b>nounwind</b></tt></dt>
1196 <dd>This function attribute indicates that the function never returns with an
1197 unwind or exceptional control flow. If the function does unwind, its
1198 runtime behavior is undefined.</dd>
1200 <dt><tt><b>optsize</b></tt></dt>
1201 <dd>This attribute suggests that optimization passes and code generator passes
1202 make choices that keep the code size of this function low, and otherwise
1203 do optimizations specifically to reduce code size.</dd>
1205 <dt><tt><b>readnone</b></tt></dt>
1206 <dd>This attribute indicates that the function computes its result (or decides
1207 to unwind an exception) based strictly on its arguments, without
1208 dereferencing any pointer arguments or otherwise accessing any mutable
1209 state (e.g. memory, control registers, etc) visible to caller functions.
1210 It does not write through any pointer arguments
1211 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1212 changes any state visible to callers. This means that it cannot unwind
1213 exceptions by calling the <tt>C++</tt> exception throwing methods.</dd>
1215 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1216 <dd>This attribute indicates that the function does not write through any
1217 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1218 arguments) or otherwise modify any state (e.g. memory, control registers,
1219 etc) visible to caller functions. It may dereference pointer arguments
1220 and read state that may be set in the caller. A readonly function always
1221 returns the same value (or unwinds an exception identically) when called
1222 with the same set of arguments and global state. It cannot unwind an
1223 exception by calling the <tt>C++</tt> exception throwing methods.</dd>
1225 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1226 <dd>This attribute indicates that this function can return twice. The
1227 C <code>setjmp</code> is an example of such a function. The compiler
1228 disables some optimizations (like tail calls) in the caller of these
1231 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1232 <dd>This attribute indicates that the function should emit a stack smashing
1233 protector. It is in the form of a "canary"—a random value placed on
1234 the stack before the local variables that's checked upon return from the
1235 function to see if it has been overwritten. A heuristic is used to
1236 determine if a function needs stack protectors or not.<br>
1238 If a function that has an <tt>ssp</tt> attribute is inlined into a
1239 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1240 function will have an <tt>ssp</tt> attribute.</dd>
1242 <dt><tt><b>sspreq</b></tt></dt>
1243 <dd>This attribute indicates that the function should <em>always</em> emit a
1244 stack smashing protector. This overrides
1245 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1247 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1248 function that doesn't have an <tt>sspreq</tt> attribute or which has
1249 an <tt>ssp</tt> attribute, then the resulting function will have
1250 an <tt>sspreq</tt> attribute.</dd>
1252 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1253 <dd>This attribute indicates that the ABI being targeted requires that
1254 an unwind table entry be produce for this function even if we can
1255 show that no exceptions passes by it. This is normally the case for
1256 the ELF x86-64 abi, but it can be disabled for some compilation
1262 <!-- ======================================================================= -->
1264 <a name="moduleasm">Module-Level Inline Assembly</a>
1269 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1270 the GCC "file scope inline asm" blocks. These blocks are internally
1271 concatenated by LLVM and treated as a single unit, but may be separated in
1272 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1274 <pre class="doc_code">
1275 module asm "inline asm code goes here"
1276 module asm "more can go here"
1279 <p>The strings can contain any character by escaping non-printable characters.
1280 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1283 <p>The inline asm code is simply printed to the machine code .s file when
1284 assembly code is generated.</p>
1288 <!-- ======================================================================= -->
1290 <a name="datalayout">Data Layout</a>
1295 <p>A module may specify a target specific data layout string that specifies how
1296 data is to be laid out in memory. The syntax for the data layout is
1299 <pre class="doc_code">
1300 target datalayout = "<i>layout specification</i>"
1303 <p>The <i>layout specification</i> consists of a list of specifications
1304 separated by the minus sign character ('-'). Each specification starts with
1305 a letter and may include other information after the letter to define some
1306 aspect of the data layout. The specifications accepted are as follows:</p>
1310 <dd>Specifies that the target lays out data in big-endian form. That is, the
1311 bits with the most significance have the lowest address location.</dd>
1314 <dd>Specifies that the target lays out data in little-endian form. That is,
1315 the bits with the least significance have the lowest address
1318 <dt><tt>S<i>size</i></tt></dt>
1319 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1320 of stack variables is limited to the natural stack alignment to avoid
1321 dynamic stack realignment. The stack alignment must be a multiple of
1322 8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1323 which does not prevent any alignment promotions.</dd>
1325 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1326 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1327 <i>preferred</i> alignments. All sizes are in bits. Specifying
1328 the <i>pref</i> alignment is optional. If omitted, the
1329 preceding <tt>:</tt> should be omitted too.</dd>
1331 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1332 <dd>This specifies the alignment for an integer type of a given bit
1333 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1335 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1336 <dd>This specifies the alignment for a vector type of a given bit
1339 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1340 <dd>This specifies the alignment for a floating point type of a given bit
1341 <i>size</i>. Only values of <i>size</i> that are supported by the target
1342 will work. 32 (float) and 64 (double) are supported on all targets;
1343 80 or 128 (different flavors of long double) are also supported on some
1346 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1347 <dd>This specifies the alignment for an aggregate type of a given bit
1350 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1351 <dd>This specifies the alignment for a stack object of a given bit
1354 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1355 <dd>This specifies a set of native integer widths for the target CPU
1356 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1357 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1358 this set are considered to support most general arithmetic
1359 operations efficiently.</dd>
1362 <p>When constructing the data layout for a given target, LLVM starts with a
1363 default set of specifications which are then (possibly) overridden by the
1364 specifications in the <tt>datalayout</tt> keyword. The default specifications
1365 are given in this list:</p>
1368 <li><tt>E</tt> - big endian</li>
1369 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1370 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1371 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1372 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1373 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1374 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1375 alignment of 64-bits</li>
1376 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1377 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1378 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1379 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1380 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1381 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1384 <p>When LLVM is determining the alignment for a given type, it uses the
1385 following rules:</p>
1388 <li>If the type sought is an exact match for one of the specifications, that
1389 specification is used.</li>
1391 <li>If no match is found, and the type sought is an integer type, then the
1392 smallest integer type that is larger than the bitwidth of the sought type
1393 is used. If none of the specifications are larger than the bitwidth then
1394 the the largest integer type is used. For example, given the default
1395 specifications above, the i7 type will use the alignment of i8 (next
1396 largest) while both i65 and i256 will use the alignment of i64 (largest
1399 <li>If no match is found, and the type sought is a vector type, then the
1400 largest vector type that is smaller than the sought vector type will be
1401 used as a fall back. This happens because <128 x double> can be
1402 implemented in terms of 64 <2 x double>, for example.</li>
1405 <p>The function of the data layout string may not be what you expect. Notably,
1406 this is not a specification from the frontend of what alignment the code
1407 generator should use.</p>
1409 <p>Instead, if specified, the target data layout is required to match what the
1410 ultimate <em>code generator</em> expects. This string is used by the
1411 mid-level optimizers to
1412 improve code, and this only works if it matches what the ultimate code
1413 generator uses. If you would like to generate IR that does not embed this
1414 target-specific detail into the IR, then you don't have to specify the
1415 string. This will disable some optimizations that require precise layout
1416 information, but this also prevents those optimizations from introducing
1417 target specificity into the IR.</p>
1423 <!-- ======================================================================= -->
1425 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1430 <p>Any memory access must be done through a pointer value associated
1431 with an address range of the memory access, otherwise the behavior
1432 is undefined. Pointer values are associated with address ranges
1433 according to the following rules:</p>
1436 <li>A pointer value is associated with the addresses associated with
1437 any value it is <i>based</i> on.
1438 <li>An address of a global variable is associated with the address
1439 range of the variable's storage.</li>
1440 <li>The result value of an allocation instruction is associated with
1441 the address range of the allocated storage.</li>
1442 <li>A null pointer in the default address-space is associated with
1444 <li>An integer constant other than zero or a pointer value returned
1445 from a function not defined within LLVM may be associated with address
1446 ranges allocated through mechanisms other than those provided by
1447 LLVM. Such ranges shall not overlap with any ranges of addresses
1448 allocated by mechanisms provided by LLVM.</li>
1451 <p>A pointer value is <i>based</i> on another pointer value according
1452 to the following rules:</p>
1455 <li>A pointer value formed from a
1456 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1457 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1458 <li>The result value of a
1459 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1460 of the <tt>bitcast</tt>.</li>
1461 <li>A pointer value formed by an
1462 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1463 pointer values that contribute (directly or indirectly) to the
1464 computation of the pointer's value.</li>
1465 <li>The "<i>based</i> on" relationship is transitive.</li>
1468 <p>Note that this definition of <i>"based"</i> is intentionally
1469 similar to the definition of <i>"based"</i> in C99, though it is
1470 slightly weaker.</p>
1472 <p>LLVM IR does not associate types with memory. The result type of a
1473 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1474 alignment of the memory from which to load, as well as the
1475 interpretation of the value. The first operand type of a
1476 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1477 and alignment of the store.</p>
1479 <p>Consequently, type-based alias analysis, aka TBAA, aka
1480 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1481 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1482 additional information which specialized optimization passes may use
1483 to implement type-based alias analysis.</p>
1487 <!-- ======================================================================= -->
1489 <a name="volatile">Volatile Memory Accesses</a>
1494 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1495 href="#i_store"><tt>store</tt></a>s, and <a
1496 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1497 The optimizers must not change the number of volatile operations or change their
1498 order of execution relative to other volatile operations. The optimizers
1499 <i>may</i> change the order of volatile operations relative to non-volatile
1500 operations. This is not Java's "volatile" and has no cross-thread
1501 synchronization behavior.</p>
1505 <!-- ======================================================================= -->
1507 <a name="memmodel">Memory Model for Concurrent Operations</a>
1512 <p>The LLVM IR does not define any way to start parallel threads of execution
1513 or to register signal handlers. Nonetheless, there are platform-specific
1514 ways to create them, and we define LLVM IR's behavior in their presence. This
1515 model is inspired by the C++0x memory model.</p>
1517 <p>For a more informal introduction to this model, see the
1518 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1520 <p>We define a <i>happens-before</i> partial order as the least partial order
1523 <li>Is a superset of single-thread program order, and</li>
1524 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1525 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1526 by platform-specific techniques, like pthread locks, thread
1527 creation, thread joining, etc., and by atomic instructions.
1528 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1532 <p>Note that program order does not introduce <i>happens-before</i> edges
1533 between a thread and signals executing inside that thread.</p>
1535 <p>Every (defined) read operation (load instructions, memcpy, atomic
1536 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1537 (defined) write operations (store instructions, atomic
1538 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1539 initialized globals are considered to have a write of the initializer which is
1540 atomic and happens before any other read or write of the memory in question.
1541 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1542 any write to the same byte, except:</p>
1545 <li>If <var>write<sub>1</sub></var> happens before
1546 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1547 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1548 does not see <var>write<sub>1</sub></var>.
1549 <li>If <var>R<sub>byte</sub></var> happens before
1550 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1551 see <var>write<sub>3</sub></var>.
1554 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1556 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1557 is supposed to give guarantees which can support
1558 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1559 addresses which do not behave like normal memory. It does not generally
1560 provide cross-thread synchronization.)
1561 <li>Otherwise, if there is no write to the same byte that happens before
1562 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1563 <tt>undef</tt> for that byte.
1564 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1565 <var>R<sub>byte</sub></var> returns the value written by that
1567 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1568 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1569 values written. See the <a href="#ordering">Atomic Memory Ordering
1570 Constraints</a> section for additional constraints on how the choice
1572 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1575 <p><var>R</var> returns the value composed of the series of bytes it read.
1576 This implies that some bytes within the value may be <tt>undef</tt>
1577 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1578 defines the semantics of the operation; it doesn't mean that targets will
1579 emit more than one instruction to read the series of bytes.</p>
1581 <p>Note that in cases where none of the atomic intrinsics are used, this model
1582 places only one restriction on IR transformations on top of what is required
1583 for single-threaded execution: introducing a store to a byte which might not
1584 otherwise be stored is not allowed in general. (Specifically, in the case
1585 where another thread might write to and read from an address, introducing a
1586 store can change a load that may see exactly one write into a load that may
1587 see multiple writes.)</p>
1589 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1590 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1591 none of the backends currently in the tree fall into this category; however,
1592 there might be targets which care. If there are, we want a paragraph
1595 Targets may specify that stores narrower than a certain width are not
1596 available; on such a target, for the purposes of this model, treat any
1597 non-atomic write with an alignment or width less than the minimum width
1598 as if it writes to the relevant surrounding bytes.
1603 <!-- ======================================================================= -->
1605 <a name="ordering">Atomic Memory Ordering Constraints</a>
1610 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1611 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1612 <a href="#i_fence"><code>fence</code></a>,
1613 <a href="#i_load"><code>atomic load</code></a>, and
1614 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1615 that determines which other atomic instructions on the same address they
1616 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1617 but are somewhat more colloquial. If these descriptions aren't precise enough,
1618 check those specs (see spec references in the
1619 <a href="Atomics.html#introduction">atomics guide</a>).
1620 <a href="#i_fence"><code>fence</code></a> instructions
1621 treat these orderings somewhat differently since they don't take an address.
1622 See that instruction's documentation for details.</p>
1624 <p>For a simpler introduction to the ordering constraints, see the
1625 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1628 <dt><code>unordered</code></dt>
1629 <dd>The set of values that can be read is governed by the happens-before
1630 partial order. A value cannot be read unless some operation wrote it.
1631 This is intended to provide a guarantee strong enough to model Java's
1632 non-volatile shared variables. This ordering cannot be specified for
1633 read-modify-write operations; it is not strong enough to make them atomic
1634 in any interesting way.</dd>
1635 <dt><code>monotonic</code></dt>
1636 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1637 total order for modifications by <code>monotonic</code> operations on each
1638 address. All modification orders must be compatible with the happens-before
1639 order. There is no guarantee that the modification orders can be combined to
1640 a global total order for the whole program (and this often will not be
1641 possible). The read in an atomic read-modify-write operation
1642 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1643 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1644 reads the value in the modification order immediately before the value it
1645 writes. If one atomic read happens before another atomic read of the same
1646 address, the later read must see the same value or a later value in the
1647 address's modification order. This disallows reordering of
1648 <code>monotonic</code> (or stronger) operations on the same address. If an
1649 address is written <code>monotonic</code>ally by one thread, and other threads
1650 <code>monotonic</code>ally read that address repeatedly, the other threads must
1651 eventually see the write. This corresponds to the C++0x/C1x
1652 <code>memory_order_relaxed</code>.</dd>
1653 <dt><code>acquire</code></dt>
1654 <dd>In addition to the guarantees of <code>monotonic</code>,
1655 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1656 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1657 <dt><code>release</code></dt>
1658 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1659 writes a value which is subsequently read by an <code>acquire</code> operation,
1660 it <i>synchronizes-with</i> that operation. (This isn't a complete
1661 description; see the C++0x definition of a release sequence.) This corresponds
1662 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1663 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1664 <code>acquire</code> and <code>release</code> operation on its address.
1665 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1666 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1667 <dd>In addition to the guarantees of <code>acq_rel</code>
1668 (<code>acquire</code> for an operation which only reads, <code>release</code>
1669 for an operation which only writes), there is a global total order on all
1670 sequentially-consistent operations on all addresses, which is consistent with
1671 the <i>happens-before</i> partial order and with the modification orders of
1672 all the affected addresses. Each sequentially-consistent read sees the last
1673 preceding write to the same address in this global order. This corresponds
1674 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1677 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1678 it only <i>synchronizes with</i> or participates in modification and seq_cst
1679 total orderings with other operations running in the same thread (for example,
1680 in signal handlers).</p>
1686 <!-- *********************************************************************** -->
1687 <h2><a name="typesystem">Type System</a></h2>
1688 <!-- *********************************************************************** -->
1692 <p>The LLVM type system is one of the most important features of the
1693 intermediate representation. Being typed enables a number of optimizations
1694 to be performed on the intermediate representation directly, without having
1695 to do extra analyses on the side before the transformation. A strong type
1696 system makes it easier to read the generated code and enables novel analyses
1697 and transformations that are not feasible to perform on normal three address
1698 code representations.</p>
1700 <!-- ======================================================================= -->
1702 <a name="t_classifications">Type Classifications</a>
1707 <p>The types fall into a few useful classifications:</p>
1709 <table border="1" cellspacing="0" cellpadding="4">
1711 <tr><th>Classification</th><th>Types</th></tr>
1713 <td><a href="#t_integer">integer</a></td>
1714 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1717 <td><a href="#t_floating">floating point</a></td>
1718 <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1721 <td><a name="t_firstclass">first class</a></td>
1722 <td><a href="#t_integer">integer</a>,
1723 <a href="#t_floating">floating point</a>,
1724 <a href="#t_pointer">pointer</a>,
1725 <a href="#t_vector">vector</a>,
1726 <a href="#t_struct">structure</a>,
1727 <a href="#t_array">array</a>,
1728 <a href="#t_label">label</a>,
1729 <a href="#t_metadata">metadata</a>.
1733 <td><a href="#t_primitive">primitive</a></td>
1734 <td><a href="#t_label">label</a>,
1735 <a href="#t_void">void</a>,
1736 <a href="#t_integer">integer</a>,
1737 <a href="#t_floating">floating point</a>,
1738 <a href="#t_x86mmx">x86mmx</a>,
1739 <a href="#t_metadata">metadata</a>.</td>
1742 <td><a href="#t_derived">derived</a></td>
1743 <td><a href="#t_array">array</a>,
1744 <a href="#t_function">function</a>,
1745 <a href="#t_pointer">pointer</a>,
1746 <a href="#t_struct">structure</a>,
1747 <a href="#t_vector">vector</a>,
1748 <a href="#t_opaque">opaque</a>.
1754 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1755 important. Values of these types are the only ones which can be produced by
1760 <!-- ======================================================================= -->
1762 <a name="t_primitive">Primitive Types</a>
1767 <p>The primitive types are the fundamental building blocks of the LLVM
1770 <!-- _______________________________________________________________________ -->
1772 <a name="t_integer">Integer Type</a>
1778 <p>The integer type is a very simple type that simply specifies an arbitrary
1779 bit width for the integer type desired. Any bit width from 1 bit to
1780 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1787 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1791 <table class="layout">
1793 <td class="left"><tt>i1</tt></td>
1794 <td class="left">a single-bit integer.</td>
1797 <td class="left"><tt>i32</tt></td>
1798 <td class="left">a 32-bit integer.</td>
1801 <td class="left"><tt>i1942652</tt></td>
1802 <td class="left">a really big integer of over 1 million bits.</td>
1808 <!-- _______________________________________________________________________ -->
1810 <a name="t_floating">Floating Point Types</a>
1817 <tr><th>Type</th><th>Description</th></tr>
1818 <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr>
1819 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1820 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1821 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1822 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1823 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1829 <!-- _______________________________________________________________________ -->
1831 <a name="t_x86mmx">X86mmx Type</a>
1837 <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>
1846 <!-- _______________________________________________________________________ -->
1848 <a name="t_void">Void Type</a>
1854 <p>The void type does not represent any value and has no size.</p>
1863 <!-- _______________________________________________________________________ -->
1865 <a name="t_label">Label Type</a>
1871 <p>The label type represents code labels.</p>
1880 <!-- _______________________________________________________________________ -->
1882 <a name="t_metadata">Metadata Type</a>
1888 <p>The metadata type represents embedded metadata. No derived types may be
1889 created from metadata except for <a href="#t_function">function</a>
1901 <!-- ======================================================================= -->
1903 <a name="t_derived">Derived Types</a>
1908 <p>The real power in LLVM comes from the derived types in the system. This is
1909 what allows a programmer to represent arrays, functions, pointers, and other
1910 useful types. Each of these types contain one or more element types which
1911 may be a primitive type, or another derived type. For example, it is
1912 possible to have a two dimensional array, using an array as the element type
1913 of another array.</p>
1915 <!-- _______________________________________________________________________ -->
1917 <a name="t_aggregate">Aggregate Types</a>
1922 <p>Aggregate Types are a subset of derived types that can contain multiple
1923 member types. <a href="#t_array">Arrays</a> and
1924 <a href="#t_struct">structs</a> are aggregate types.
1925 <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p>
1929 <!-- _______________________________________________________________________ -->
1931 <a name="t_array">Array Type</a>
1937 <p>The array type is a very simple derived type that arranges elements
1938 sequentially in memory. The array type requires a size (number of elements)
1939 and an underlying data type.</p>
1943 [<# elements> x <elementtype>]
1946 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1947 be any type with a size.</p>
1950 <table class="layout">
1952 <td class="left"><tt>[40 x i32]</tt></td>
1953 <td class="left">Array of 40 32-bit integer values.</td>
1956 <td class="left"><tt>[41 x i32]</tt></td>
1957 <td class="left">Array of 41 32-bit integer values.</td>
1960 <td class="left"><tt>[4 x i8]</tt></td>
1961 <td class="left">Array of 4 8-bit integer values.</td>
1964 <p>Here are some examples of multidimensional arrays:</p>
1965 <table class="layout">
1967 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1968 <td class="left">3x4 array of 32-bit integer values.</td>
1971 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1972 <td class="left">12x10 array of single precision floating point values.</td>
1975 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1976 <td class="left">2x3x4 array of 16-bit integer values.</td>
1980 <p>There is no restriction on indexing beyond the end of the array implied by
1981 a static type (though there are restrictions on indexing beyond the bounds
1982 of an allocated object in some cases). This means that single-dimension
1983 'variable sized array' addressing can be implemented in LLVM with a zero
1984 length array type. An implementation of 'pascal style arrays' in LLVM could
1985 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1989 <!-- _______________________________________________________________________ -->
1991 <a name="t_function">Function Type</a>
1997 <p>The function type can be thought of as a function signature. It consists of
1998 a return type and a list of formal parameter types. The return type of a
1999 function type is a first class type or a void type.</p>
2003 <returntype> (<parameter list>)
2006 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
2007 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
2008 which indicates that the function takes a variable number of arguments.
2009 Variable argument functions can access their arguments with
2010 the <a href="#int_varargs">variable argument handling intrinsic</a>
2011 functions. '<tt><returntype></tt>' is any type except
2012 <a href="#t_label">label</a>.</p>
2015 <table class="layout">
2017 <td class="left"><tt>i32 (i32)</tt></td>
2018 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2020 </tr><tr class="layout">
2021 <td class="left"><tt>float (i16, i32 *) *
2023 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2024 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2025 returning <tt>float</tt>.
2027 </tr><tr class="layout">
2028 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2029 <td class="left">A vararg function that takes at least one
2030 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2031 which returns an integer. This is the signature for <tt>printf</tt> in
2034 </tr><tr class="layout">
2035 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2036 <td class="left">A function taking an <tt>i32</tt>, returning a
2037 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2044 <!-- _______________________________________________________________________ -->
2046 <a name="t_struct">Structure Type</a>
2052 <p>The structure type is used to represent a collection of data members together
2053 in memory. The elements of a structure may be any type that has a size.</p>
2055 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2056 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2057 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2058 Structures in registers are accessed using the
2059 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2060 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2062 <p>Structures may optionally be "packed" structures, which indicate that the
2063 alignment of the struct is one byte, and that there is no padding between
2064 the elements. In non-packed structs, padding between field types is inserted
2065 as defined by the TargetData string in the module, which is required to match
2066 what the underlying code generator expects.</p>
2068 <p>Structures can either be "literal" or "identified". A literal structure is
2069 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2070 types are always defined at the top level with a name. Literal types are
2071 uniqued by their contents and can never be recursive or opaque since there is
2072 no way to write one. Identified types can be recursive, can be opaqued, and are
2078 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2079 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2083 <table class="layout">
2085 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2086 <td class="left">A triple of three <tt>i32</tt> values</td>
2089 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2090 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2091 second element is a <a href="#t_pointer">pointer</a> to a
2092 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2093 an <tt>i32</tt>.</td>
2096 <td class="left"><tt><{ i8, i32 }></tt></td>
2097 <td class="left">A packed struct known to be 5 bytes in size.</td>
2103 <!-- _______________________________________________________________________ -->
2105 <a name="t_opaque">Opaque Structure Types</a>
2111 <p>Opaque structure types are used to represent named structure types that do
2112 not have a body specified. This corresponds (for example) to the C notion of
2113 a forward declared structure.</p>
2122 <table class="layout">
2124 <td class="left"><tt>opaque</tt></td>
2125 <td class="left">An opaque type.</td>
2133 <!-- _______________________________________________________________________ -->
2135 <a name="t_pointer">Pointer Type</a>
2141 <p>The pointer type is used to specify memory locations.
2142 Pointers are commonly used to reference objects in memory.</p>
2144 <p>Pointer types may have an optional address space attribute defining the
2145 numbered address space where the pointed-to object resides. The default
2146 address space is number zero. The semantics of non-zero address
2147 spaces are target-specific.</p>
2149 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2150 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2158 <table class="layout">
2160 <td class="left"><tt>[4 x i32]*</tt></td>
2161 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2162 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2165 <td class="left"><tt>i32 (i32*) *</tt></td>
2166 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2167 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2171 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2172 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2173 that resides in address space #5.</td>
2179 <!-- _______________________________________________________________________ -->
2181 <a name="t_vector">Vector Type</a>
2187 <p>A vector type is a simple derived type that represents a vector of elements.
2188 Vector types are used when multiple primitive data are operated in parallel
2189 using a single instruction (SIMD). A vector type requires a size (number of
2190 elements) and an underlying primitive data type. Vector types are considered
2191 <a href="#t_firstclass">first class</a>.</p>
2195 < <# elements> x <elementtype> >
2198 <p>The number of elements is a constant integer value larger than 0; elementtype
2199 may be any integer or floating point type, or a pointer to these types.
2200 Vectors of size zero are not allowed. </p>
2203 <table class="layout">
2205 <td class="left"><tt><4 x i32></tt></td>
2206 <td class="left">Vector of 4 32-bit integer values.</td>
2209 <td class="left"><tt><8 x float></tt></td>
2210 <td class="left">Vector of 8 32-bit floating-point values.</td>
2213 <td class="left"><tt><2 x i64></tt></td>
2214 <td class="left">Vector of 2 64-bit integer values.</td>
2217 <td class="left"><tt><4 x i64*></tt></td>
2218 <td class="left">Vector of 4 pointers to 64-bit integer values.</td>
2228 <!-- *********************************************************************** -->
2229 <h2><a name="constants">Constants</a></h2>
2230 <!-- *********************************************************************** -->
2234 <p>LLVM has several different basic types of constants. This section describes
2235 them all and their syntax.</p>
2237 <!-- ======================================================================= -->
2239 <a name="simpleconstants">Simple Constants</a>
2245 <dt><b>Boolean constants</b></dt>
2246 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2247 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2249 <dt><b>Integer constants</b></dt>
2250 <dd>Standard integers (such as '4') are constants of
2251 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2252 with integer types.</dd>
2254 <dt><b>Floating point constants</b></dt>
2255 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2256 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2257 notation (see below). The assembler requires the exact decimal value of a
2258 floating-point constant. For example, the assembler accepts 1.25 but
2259 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2260 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2262 <dt><b>Null pointer constants</b></dt>
2263 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2264 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2267 <p>The one non-intuitive notation for constants is the hexadecimal form of
2268 floating point constants. For example, the form '<tt>double
2269 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2270 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2271 constants are required (and the only time that they are generated by the
2272 disassembler) is when a floating point constant must be emitted but it cannot
2273 be represented as a decimal floating point number in a reasonable number of
2274 digits. For example, NaN's, infinities, and other special values are
2275 represented in their IEEE hexadecimal format so that assembly and disassembly
2276 do not cause any bits to change in the constants.</p>
2278 <p>When using the hexadecimal form, constants of types half, float, and double are
2279 represented using the 16-digit form shown above (which matches the IEEE754
2280 representation for double); half and float values must, however, be exactly
2281 representable as IEE754 half and single precision, respectively.
2282 Hexadecimal format is always used
2283 for long double, and there are three forms of long double. The 80-bit format
2284 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2285 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2286 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2287 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2288 currently supported target uses this format. Long doubles will only work if
2289 they match the long double format on your target. All hexadecimal formats
2290 are big-endian (sign bit at the left).</p>
2292 <p>There are no constants of type x86mmx.</p>
2295 <!-- ======================================================================= -->
2297 <a name="aggregateconstants"></a> <!-- old anchor -->
2298 <a name="complexconstants">Complex Constants</a>
2303 <p>Complex constants are a (potentially recursive) combination of simple
2304 constants and smaller complex constants.</p>
2307 <dt><b>Structure constants</b></dt>
2308 <dd>Structure constants are represented with notation similar to structure
2309 type definitions (a comma separated list of elements, surrounded by braces
2310 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2311 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2312 Structure constants must have <a href="#t_struct">structure type</a>, and
2313 the number and types of elements must match those specified by the
2316 <dt><b>Array constants</b></dt>
2317 <dd>Array constants are represented with notation similar to array type
2318 definitions (a comma separated list of elements, surrounded by square
2319 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2320 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2321 the number and types of elements must match those specified by the
2324 <dt><b>Vector constants</b></dt>
2325 <dd>Vector constants are represented with notation similar to vector type
2326 definitions (a comma separated list of elements, surrounded by
2327 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2328 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2329 have <a href="#t_vector">vector type</a>, and the number and types of
2330 elements must match those specified by the type.</dd>
2332 <dt><b>Zero initialization</b></dt>
2333 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2334 value to zero of <em>any</em> type, including scalar and
2335 <a href="#t_aggregate">aggregate</a> types.
2336 This is often used to avoid having to print large zero initializers
2337 (e.g. for large arrays) and is always exactly equivalent to using explicit
2338 zero initializers.</dd>
2340 <dt><b>Metadata node</b></dt>
2341 <dd>A metadata node is a structure-like constant with
2342 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2343 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2344 be interpreted as part of the instruction stream, metadata is a place to
2345 attach additional information such as debug info.</dd>
2350 <!-- ======================================================================= -->
2352 <a name="globalconstants">Global Variable and Function Addresses</a>
2357 <p>The addresses of <a href="#globalvars">global variables</a>
2358 and <a href="#functionstructure">functions</a> are always implicitly valid
2359 (link-time) constants. These constants are explicitly referenced when
2360 the <a href="#identifiers">identifier for the global</a> is used and always
2361 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2362 legal LLVM file:</p>
2364 <pre class="doc_code">
2367 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2372 <!-- ======================================================================= -->
2374 <a name="undefvalues">Undefined Values</a>
2379 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2380 indicates that the user of the value may receive an unspecified bit-pattern.
2381 Undefined values may be of any type (other than '<tt>label</tt>'
2382 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2384 <p>Undefined values are useful because they indicate to the compiler that the
2385 program is well defined no matter what value is used. This gives the
2386 compiler more freedom to optimize. Here are some examples of (potentially
2387 surprising) transformations that are valid (in pseudo IR):</p>
2390 <pre class="doc_code">
2400 <p>This is safe because all of the output bits are affected by the undef bits.
2401 Any output bit can have a zero or one depending on the input bits.</p>
2403 <pre class="doc_code">
2414 <p>These logical operations have bits that are not always affected by the input.
2415 For example, if <tt>%X</tt> has a zero bit, then the output of the
2416 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2417 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2418 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2419 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2420 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2421 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2422 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2424 <pre class="doc_code">
2425 %A = select undef, %X, %Y
2426 %B = select undef, 42, %Y
2427 %C = select %X, %Y, undef
2438 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2439 branch) conditions can go <em>either way</em>, but they have to come from one
2440 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2441 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2442 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2443 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2444 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2447 <pre class="doc_code">
2448 %A = xor undef, undef
2466 <p>This example points out that two '<tt>undef</tt>' operands are not
2467 necessarily the same. This can be surprising to people (and also matches C
2468 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2469 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2470 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2471 its value over its "live range". This is true because the variable doesn't
2472 actually <em>have a live range</em>. Instead, the value is logically read
2473 from arbitrary registers that happen to be around when needed, so the value
2474 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2475 need to have the same semantics or the core LLVM "replace all uses with"
2476 concept would not hold.</p>
2478 <pre class="doc_code">
2486 <p>These examples show the crucial difference between an <em>undefined
2487 value</em> and <em>undefined behavior</em>. An undefined value (like
2488 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2489 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2490 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2491 defined on SNaN's. However, in the second example, we can make a more
2492 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2493 arbitrary value, we are allowed to assume that it could be zero. Since a
2494 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2495 the operation does not execute at all. This allows us to delete the divide and
2496 all code after it. Because the undefined operation "can't happen", the
2497 optimizer can assume that it occurs in dead code.</p>
2499 <pre class="doc_code">
2500 a: store undef -> %X
2501 b: store %X -> undef
2507 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2508 undefined value can be assumed to not have any effect; we can assume that the
2509 value is overwritten with bits that happen to match what was already there.
2510 However, a store <em>to</em> an undefined location could clobber arbitrary
2511 memory, therefore, it has undefined behavior.</p>
2515 <!-- ======================================================================= -->
2517 <a name="poisonvalues">Poison Values</a>
2522 <p>Poison values are similar to <a href="#undefvalues">undef values</a>, however
2523 they also represent the fact that an instruction or constant expression which
2524 cannot evoke side effects has nevertheless detected a condition which results
2525 in undefined behavior.</p>
2527 <p>There is currently no way of representing a poison value in the IR; they
2528 only exist when produced by operations such as
2529 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2531 <p>Poison value behavior is defined in terms of value <i>dependence</i>:</p>
2534 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2535 their operands.</li>
2537 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2538 to their dynamic predecessor basic block.</li>
2540 <li>Function arguments depend on the corresponding actual argument values in
2541 the dynamic callers of their functions.</li>
2543 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2544 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2545 control back to them.</li>
2547 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2548 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>,
2549 or exception-throwing call instructions that dynamically transfer control
2552 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2553 referenced memory addresses, following the order in the IR
2554 (including loads and stores implied by intrinsics such as
2555 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2557 <!-- TODO: In the case of multiple threads, this only applies if the store
2558 "happens-before" the load or store. -->
2560 <!-- TODO: floating-point exception state -->
2562 <li>An instruction with externally visible side effects depends on the most
2563 recent preceding instruction with externally visible side effects, following
2564 the order in the IR. (This includes
2565 <a href="#volatile">volatile operations</a>.)</li>
2567 <li>An instruction <i>control-depends</i> on a
2568 <a href="#terminators">terminator instruction</a>
2569 if the terminator instruction has multiple successors and the instruction
2570 is always executed when control transfers to one of the successors, and
2571 may not be executed when control is transferred to another.</li>
2573 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2574 instruction if the set of instructions it otherwise depends on would be
2575 different if the terminator had transferred control to a different
2578 <li>Dependence is transitive.</li>
2582 <p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>,
2583 with the additional affect that any instruction which has a <i>dependence</i>
2584 on a poison value has undefined behavior.</p>
2586 <p>Here are some examples:</p>
2588 <pre class="doc_code">
2590 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2591 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2592 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2593 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2595 store i32 %poison, i32* @g ; Poison value stored to memory.
2596 %poison2 = load i32* @g ; Poison value loaded back from memory.
2598 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2600 %narrowaddr = bitcast i32* @g to i16*
2601 %wideaddr = bitcast i32* @g to i64*
2602 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2603 %poison4 = load i64* %wideaddr ; Returns a poison value.
2605 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2606 br i1 %cmp, label %true, label %end ; Branch to either destination.
2609 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2610 ; it has undefined behavior.
2614 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2615 ; Both edges into this PHI are
2616 ; control-dependent on %cmp, so this
2617 ; always results in a poison value.
2619 store volatile i32 0, i32* @g ; This would depend on the store in %true
2620 ; if %cmp is true, or the store in %entry
2621 ; otherwise, so this is undefined behavior.
2623 br i1 %cmp, label %second_true, label %second_end
2624 ; The same branch again, but this time the
2625 ; true block doesn't have side effects.
2632 store volatile i32 0, i32* @g ; This time, the instruction always depends
2633 ; on the store in %end. Also, it is
2634 ; control-equivalent to %end, so this is
2635 ; well-defined (ignoring earlier undefined
2636 ; behavior in this example).
2641 <!-- ======================================================================= -->
2643 <a name="blockaddress">Addresses of Basic Blocks</a>
2648 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2650 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2651 basic block in the specified function, and always has an i8* type. Taking
2652 the address of the entry block is illegal.</p>
2654 <p>This value only has defined behavior when used as an operand to the
2655 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2656 comparisons against null. Pointer equality tests between labels addresses
2657 results in undefined behavior — though, again, comparison against null
2658 is ok, and no label is equal to the null pointer. This may be passed around
2659 as an opaque pointer sized value as long as the bits are not inspected. This
2660 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2661 long as the original value is reconstituted before the <tt>indirectbr</tt>
2664 <p>Finally, some targets may provide defined semantics when using the value as
2665 the operand to an inline assembly, but that is target specific.</p>
2670 <!-- ======================================================================= -->
2672 <a name="constantexprs">Constant Expressions</a>
2677 <p>Constant expressions are used to allow expressions involving other constants
2678 to be used as constants. Constant expressions may be of
2679 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2680 operation that does not have side effects (e.g. load and call are not
2681 supported). The following is the syntax for constant expressions:</p>
2684 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2685 <dd>Truncate a constant to another type. The bit size of CST must be larger
2686 than the bit size of TYPE. Both types must be integers.</dd>
2688 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2689 <dd>Zero extend a constant to another type. The bit size of CST must be
2690 smaller than the bit size of TYPE. Both types must be integers.</dd>
2692 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2693 <dd>Sign extend a constant to another type. The bit size of CST must be
2694 smaller than the bit size of TYPE. Both types must be integers.</dd>
2696 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2697 <dd>Truncate a floating point constant to another floating point type. The
2698 size of CST must be larger than the size of TYPE. Both types must be
2699 floating point.</dd>
2701 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2702 <dd>Floating point extend a constant to another type. The size of CST must be
2703 smaller or equal to the size of TYPE. Both types must be floating
2706 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2707 <dd>Convert a floating point constant to the corresponding unsigned integer
2708 constant. TYPE must be a scalar or vector integer type. CST must be of
2709 scalar or vector floating point type. Both CST and TYPE must be scalars,
2710 or vectors of the same number of elements. If the value won't fit in the
2711 integer type, the results are undefined.</dd>
2713 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2714 <dd>Convert a floating point constant to the corresponding signed integer
2715 constant. TYPE must be a scalar or vector integer type. CST must be of
2716 scalar or vector floating point type. Both CST and TYPE must be scalars,
2717 or vectors of the same number of elements. If the value won't fit in the
2718 integer type, the results are undefined.</dd>
2720 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2721 <dd>Convert an unsigned integer constant to the corresponding floating point
2722 constant. TYPE must be a scalar or vector floating point type. CST must be
2723 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2724 vectors of the same number of elements. If the value won't fit in the
2725 floating point type, the results are undefined.</dd>
2727 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2728 <dd>Convert a signed integer constant to the corresponding floating point
2729 constant. TYPE must be a scalar or vector floating point type. CST must be
2730 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2731 vectors of the same number of elements. If the value won't fit in the
2732 floating point type, the results are undefined.</dd>
2734 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2735 <dd>Convert a pointer typed constant to the corresponding integer constant
2736 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2737 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2738 make it fit in <tt>TYPE</tt>.</dd>
2740 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2741 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2742 type. CST must be of integer type. The CST value is zero extended,
2743 truncated, or unchanged to make it fit in a pointer size. This one is
2744 <i>really</i> dangerous!</dd>
2746 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2747 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2748 are the same as those for the <a href="#i_bitcast">bitcast
2749 instruction</a>.</dd>
2751 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2752 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2753 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2754 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2755 instruction, the index list may have zero or more indexes, which are
2756 required to make sense for the type of "CSTPTR".</dd>
2758 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2759 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2761 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2762 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2764 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2765 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2767 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2768 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2771 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2772 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2775 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2776 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2779 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2780 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2781 constants. The index list is interpreted in a similar manner as indices in
2782 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2783 index value must be specified.</dd>
2785 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2786 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2787 constants. The index list is interpreted in a similar manner as indices in
2788 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2789 index value must be specified.</dd>
2791 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2792 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2793 be any of the <a href="#binaryops">binary</a>
2794 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2795 on operands are the same as those for the corresponding instruction
2796 (e.g. no bitwise operations on floating point values are allowed).</dd>
2803 <!-- *********************************************************************** -->
2804 <h2><a name="othervalues">Other Values</a></h2>
2805 <!-- *********************************************************************** -->
2807 <!-- ======================================================================= -->
2809 <a name="inlineasm">Inline Assembler Expressions</a>
2814 <p>LLVM supports inline assembler expressions (as opposed
2815 to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of
2816 a special value. This value represents the inline assembler as a string
2817 (containing the instructions to emit), a list of operand constraints (stored
2818 as a string), a flag that indicates whether or not the inline asm
2819 expression has side effects, and a flag indicating whether the function
2820 containing the asm needs to align its stack conservatively. An example
2821 inline assembler expression is:</p>
2823 <pre class="doc_code">
2824 i32 (i32) asm "bswap $0", "=r,r"
2827 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2828 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2831 <pre class="doc_code">
2832 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2835 <p>Inline asms with side effects not visible in the constraint list must be
2836 marked as having side effects. This is done through the use of the
2837 '<tt>sideeffect</tt>' keyword, like so:</p>
2839 <pre class="doc_code">
2840 call void asm sideeffect "eieio", ""()
2843 <p>In some cases inline asms will contain code that will not work unless the
2844 stack is aligned in some way, such as calls or SSE instructions on x86,
2845 yet will not contain code that does that alignment within the asm.
2846 The compiler should make conservative assumptions about what the asm might
2847 contain and should generate its usual stack alignment code in the prologue
2848 if the '<tt>alignstack</tt>' keyword is present:</p>
2850 <pre class="doc_code">
2851 call void asm alignstack "eieio", ""()
2854 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2858 <p>TODO: The format of the asm and constraints string still need to be
2859 documented here. Constraints on what can be done (e.g. duplication, moving,
2860 etc need to be documented). This is probably best done by reference to
2861 another document that covers inline asm from a holistic perspective.</p>
2864 <!-- _______________________________________________________________________ -->
2866 <a name="inlineasm_md">Inline Asm Metadata</a>
2871 <p>The call instructions that wrap inline asm nodes may have a
2872 "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant
2873 integers. If present, the code generator will use the integer as the
2874 location cookie value when report errors through the <tt>LLVMContext</tt>
2875 error reporting mechanisms. This allows a front-end to correlate backend
2876 errors that occur with inline asm back to the source code that produced it.
2879 <pre class="doc_code">
2880 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2882 !42 = !{ i32 1234567 }
2885 <p>It is up to the front-end to make sense of the magic numbers it places in the
2886 IR. If the MDNode contains multiple constants, the code generator will use
2887 the one that corresponds to the line of the asm that the error occurs on.</p>
2893 <!-- ======================================================================= -->
2895 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2900 <p>LLVM IR allows metadata to be attached to instructions in the program that
2901 can convey extra information about the code to the optimizers and code
2902 generator. One example application of metadata is source-level debug
2903 information. There are two metadata primitives: strings and nodes. All
2904 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2905 preceding exclamation point ('<tt>!</tt>').</p>
2907 <p>A metadata string is a string surrounded by double quotes. It can contain
2908 any character by escaping non-printable characters with "<tt>\xx</tt>" where
2909 "<tt>xx</tt>" is the two digit hex code. For example:
2910 "<tt>!"test\00"</tt>".</p>
2912 <p>Metadata nodes are represented with notation similar to structure constants
2913 (a comma separated list of elements, surrounded by braces and preceded by an
2914 exclamation point). Metadata nodes can have any values as their operand. For
2917 <div class="doc_code">
2919 !{ metadata !"test\00", i32 10}
2923 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2924 metadata nodes, which can be looked up in the module symbol table. For
2927 <div class="doc_code">
2929 !foo = metadata !{!4, !3}
2933 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2934 function is using two metadata arguments:</p>
2936 <div class="doc_code">
2938 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2942 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2943 attached to the <tt>add</tt> instruction using the <tt>!dbg</tt>
2946 <div class="doc_code">
2948 %indvar.next = add i64 %indvar, 1, !dbg !21
2952 <p>More information about specific metadata nodes recognized by the optimizers
2953 and code generator is found below.</p>
2955 <!-- _______________________________________________________________________ -->
2957 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
2962 <p>In LLVM IR, memory does not have types, so LLVM's own type system is not
2963 suitable for doing TBAA. Instead, metadata is added to the IR to describe
2964 a type system of a higher level language. This can be used to implement
2965 typical C/C++ TBAA, but it can also be used to implement custom alias
2966 analysis behavior for other languages.</p>
2968 <p>The current metadata format is very simple. TBAA metadata nodes have up to
2969 three fields, e.g.:</p>
2971 <div class="doc_code">
2973 !0 = metadata !{ metadata !"an example type tree" }
2974 !1 = metadata !{ metadata !"int", metadata !0 }
2975 !2 = metadata !{ metadata !"float", metadata !0 }
2976 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2980 <p>The first field is an identity field. It can be any value, usually
2981 a metadata string, which uniquely identifies the type. The most important
2982 name in the tree is the name of the root node. Two trees with
2983 different root node names are entirely disjoint, even if they
2984 have leaves with common names.</p>
2986 <p>The second field identifies the type's parent node in the tree, or
2987 is null or omitted for a root node. A type is considered to alias
2988 all of its descendants and all of its ancestors in the tree. Also,
2989 a type is considered to alias all types in other trees, so that
2990 bitcode produced from multiple front-ends is handled conservatively.</p>
2992 <p>If the third field is present, it's an integer which if equal to 1
2993 indicates that the type is "constant" (meaning
2994 <tt>pointsToConstantMemory</tt> should return true; see
2995 <a href="AliasAnalysis.html#OtherItfs">other useful
2996 <tt>AliasAnalysis</tt> methods</a>).</p>
3000 <!-- _______________________________________________________________________ -->
3002 <a name="fpaccuracy">'<tt>fpaccuracy</tt>' Metadata</a>
3007 <p><tt>fpaccuracy</tt> metadata may be attached to any instruction of floating
3008 point type. It expresses the maximum relative error of the result of
3009 that instruction, in ULPs. ULP is defined as follows:</p>
3013 <p>If <tt>x</tt> is a real number that lies between two finite consecutive
3014 floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one
3015 of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the
3016 distance between the two non-equal finite floating-point numbers nearest
3017 <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p>
3021 <p>The maximum relative error may be any rational number. The metadata node
3022 shall consist of a pair of unsigned integers respectively representing
3023 the numerator and denominator. For example, 2.5 ULP:</p>
3025 <div class="doc_code">
3027 !0 = metadata !{ i32 5, i32 2 }
3037 <!-- *********************************************************************** -->
3039 <a name="module_flags">Module Flags Metadata</a>
3041 <!-- *********************************************************************** -->
3045 <p>Information about the module as a whole is difficult to convey to LLVM's
3046 subsystems. The LLVM IR isn't sufficient to transmit this
3047 information. The <tt>llvm.module.flags</tt> named metadata exists in order to
3048 facilitate this. These flags are in the form of key / value pairs —
3049 much like a dictionary — making it easy for any subsystem who cares
3050 about a flag to look it up.</p>
3052 <p>The <tt>llvm.module.flags</tt> metadata contains a list of metadata
3053 triplets. Each triplet has the following form:</p>
3056 <li>The first element is a <i>behavior</i> flag, which specifies the behavior
3057 when two (or more) modules are merged together, and it encounters two (or
3058 more) metadata with the same ID. The supported behaviors are described
3061 <li>The second element is a metadata string that is a unique ID for the
3062 metadata. How each ID is interpreted is documented below.</li>
3064 <li>The third element is the value of the flag.</li>
3067 <p>When two (or more) modules are merged together, the resulting
3068 <tt>llvm.module.flags</tt> metadata is the union of the
3069 modules' <tt>llvm.module.flags</tt> metadata. The only exception being a flag
3070 with the <i>Override</i> behavior, which may override another flag's value
3073 <p>The following behaviors are supported:</p>
3075 <table border="1" cellspacing="0" cellpadding="4">
3084 <dt><b>Error</b></dt>
3085 <dd>Emits an error if two values disagree. It is an error to have an ID
3086 with both an Error and a Warning behavior.</dd>
3092 <dt><b>Warning</b></dt>
3093 <dd>Emits a warning if two values disagree.</dd>
3099 <dt><b>Require</b></dt>
3100 <dd>Emits an error when the specified value is not present or doesn't
3101 have the specified value. It is an error for two (or more)
3102 <tt>llvm.module.flags</tt> with the same ID to have the Require
3103 behavior but different values. There may be multiple Require flags
3110 <dt><b>Override</b></dt>
3111 <dd>Uses the specified value if the two values disagree. It is an error
3112 for two (or more) <tt>llvm.module.flags</tt> with the same ID to
3113 have the Override behavior but different values.</dd>
3119 <p>An example of module flags:</p>
3121 <pre class="doc_code">
3122 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3123 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3124 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3125 !3 = metadata !{ i32 3, metadata !"qux",
3127 metadata !"foo", i32 1
3130 !llvm.module.flags = !{ !0, !1, !2, !3 }
3134 <li><p>Metadata <tt>!0</tt> has the ID <tt>!"foo"</tt> and the value '1'. The
3135 behavior if two or more <tt>!"foo"</tt> flags are seen is to emit an
3136 error if their values are not equal.</p></li>
3138 <li><p>Metadata <tt>!1</tt> has the ID <tt>!"bar"</tt> and the value '37'. The
3139 behavior if two or more <tt>!"bar"</tt> flags are seen is to use the
3140 value '37' if their values are not equal.</p></li>
3142 <li><p>Metadata <tt>!2</tt> has the ID <tt>!"qux"</tt> and the value '42'. The
3143 behavior if two or more <tt>!"qux"</tt> flags are seen is to emit a
3144 warning if their values are not equal.</p></li>
3146 <li><p>Metadata <tt>!3</tt> has the ID <tt>!"qux"</tt> and the value:</p>
3148 <pre class="doc_code">
3149 metadata !{ metadata !"foo", i32 1 }
3152 <p>The behavior is to emit an error if the <tt>llvm.module.flags</tt> does
3153 not contain a flag with the ID <tt>!"foo"</tt> that has the value
3154 '1'. If two or more <tt>!"qux"</tt> flags exist, then they must have
3155 the same value or an error will be issued.</p></li>
3159 <!-- ======================================================================= -->
3161 <a name="objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a>
3166 <p>On the Mach-O platform, Objective-C stores metadata about garbage collection
3167 in a special section called "image info". The metadata consists of a version
3168 number and a bitmask specifying what types of garbage collection are
3169 supported (if any) by the file. If two or more modules are linked together
3170 their garbage collection metadata needs to be merged rather than appended
3173 <p>The Objective-C garbage collection module flags metadata consists of the
3174 following key-value pairs:</p>
3176 <table border="1" cellspacing="0" cellpadding="4">
3179 <th width="30%">Key</th>
3183 <td><tt>Objective-C Version</tt></td>
3184 <td align="left"><b>[Required]</b> — The Objective-C ABI
3185 version. Valid values are 1 and 2.</td>
3188 <td><tt>Objective-C Image Info Version</tt></td>
3189 <td align="left"><b>[Required]</b> — The version of the image info
3190 section. Currently always 0.</td>
3193 <td><tt>Objective-C Image Info Section</tt></td>
3194 <td align="left"><b>[Required]</b> — The section to place the
3195 metadata. Valid values are <tt>"__OBJC, __image_info, regular"</tt> for
3196 Objective-C ABI version 1, and <tt>"__DATA,__objc_imageinfo, regular,
3197 no_dead_strip"</tt> for Objective-C ABI version 2.</td>
3200 <td><tt>Objective-C Garbage Collection</tt></td>
3201 <td align="left"><b>[Required]</b> — Specifies whether garbage
3202 collection is supported or not. Valid values are 0, for no garbage
3203 collection, and 2, for garbage collection supported.</td>
3206 <td><tt>Objective-C GC Only</tt></td>
3207 <td align="left"><b>[Optional]</b> — Specifies that only garbage
3208 collection is supported. If present, its value must be 6. This flag
3209 requires that the <tt>Objective-C Garbage Collection</tt> flag have the
3215 <p>Some important flag interactions:</p>
3218 <li>If a module with <tt>Objective-C Garbage Collection</tt> set to 0 is
3219 merged with a module with <tt>Objective-C Garbage Collection</tt> set to
3220 2, then the resulting module has the <tt>Objective-C Garbage
3221 Collection</tt> flag set to 0.</li>
3223 <li>A module with <tt>Objective-C Garbage Collection</tt> set to 0 cannot be
3224 merged with a module with <tt>Objective-C GC Only</tt> set to 6.</li>
3231 <!-- *********************************************************************** -->
3233 <a name="intrinsic_globals">Intrinsic Global Variables</a>
3235 <!-- *********************************************************************** -->
3237 <p>LLVM has a number of "magic" global variables that contain data that affect
3238 code generation or other IR semantics. These are documented here. All globals
3239 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
3240 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
3243 <!-- ======================================================================= -->
3245 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
3250 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
3251 href="#linkage_appending">appending linkage</a>. This array contains a list of
3252 pointers to global variables and functions which may optionally have a pointer
3253 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
3255 <div class="doc_code">
3260 @llvm.used = appending global [2 x i8*] [
3262 i8* bitcast (i32* @Y to i8*)
3263 ], section "llvm.metadata"
3267 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
3268 compiler, assembler, and linker are required to treat the symbol as if there
3269 is a reference to the global that it cannot see. For example, if a variable
3270 has internal linkage and no references other than that from
3271 the <tt>@llvm.used</tt> list, it cannot be deleted. This is commonly used to
3272 represent references from inline asms and other things the compiler cannot
3273 "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p>
3275 <p>On some targets, the code generator must emit a directive to the assembler or
3276 object file to prevent the assembler and linker from molesting the
3281 <!-- ======================================================================= -->
3283 <a name="intg_compiler_used">
3284 The '<tt>llvm.compiler.used</tt>' Global Variable
3290 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
3291 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
3292 touching the symbol. On targets that support it, this allows an intelligent
3293 linker to optimize references to the symbol without being impeded as it would
3294 be by <tt>@llvm.used</tt>.</p>
3296 <p>This is a rare construct that should only be used in rare circumstances, and
3297 should not be exposed to source languages.</p>
3301 <!-- ======================================================================= -->
3303 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
3308 <div class="doc_code">
3310 %0 = type { i32, void ()* }
3311 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3315 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor
3316 functions and associated priorities. The functions referenced by this array
3317 will be called in ascending order of priority (i.e. lowest first) when the
3318 module is loaded. The order of functions with the same priority is not
3323 <!-- ======================================================================= -->
3325 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3330 <div class="doc_code">
3332 %0 = type { i32, void ()* }
3333 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3337 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions
3338 and associated priorities. The functions referenced by this array will be
3339 called in descending order of priority (i.e. highest first) when the module
3340 is loaded. The order of functions with the same priority is not defined.</p>
3346 <!-- *********************************************************************** -->
3347 <h2><a name="instref">Instruction Reference</a></h2>
3348 <!-- *********************************************************************** -->
3352 <p>The LLVM instruction set consists of several different classifications of
3353 instructions: <a href="#terminators">terminator
3354 instructions</a>, <a href="#binaryops">binary instructions</a>,
3355 <a href="#bitwiseops">bitwise binary instructions</a>,
3356 <a href="#memoryops">memory instructions</a>, and
3357 <a href="#otherops">other instructions</a>.</p>
3359 <!-- ======================================================================= -->
3361 <a name="terminators">Terminator Instructions</a>
3366 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3367 in a program ends with a "Terminator" instruction, which indicates which
3368 block should be executed after the current block is finished. These
3369 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3370 control flow, not values (the one exception being the
3371 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3373 <p>The terminator instructions are:
3374 '<a href="#i_ret"><tt>ret</tt></a>',
3375 '<a href="#i_br"><tt>br</tt></a>',
3376 '<a href="#i_switch"><tt>switch</tt></a>',
3377 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3378 '<a href="#i_invoke"><tt>invoke</tt></a>',
3379 '<a href="#i_resume"><tt>resume</tt></a>', and
3380 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3382 <!-- _______________________________________________________________________ -->
3384 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3391 ret <type> <value> <i>; Return a value from a non-void function</i>
3392 ret void <i>; Return from void function</i>
3396 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3397 a value) from a function back to the caller.</p>
3399 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3400 value and then causes control flow, and one that just causes control flow to
3404 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3405 return value. The type of the return value must be a
3406 '<a href="#t_firstclass">first class</a>' type.</p>
3408 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3409 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3410 value or a return value with a type that does not match its type, or if it
3411 has a void return type and contains a '<tt>ret</tt>' instruction with a
3415 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3416 the calling function's context. If the caller is a
3417 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3418 instruction after the call. If the caller was an
3419 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3420 the beginning of the "normal" destination block. If the instruction returns
3421 a value, that value shall set the call or invoke instruction's return
3426 ret i32 5 <i>; Return an integer value of 5</i>
3427 ret void <i>; Return from a void function</i>
3428 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3432 <!-- _______________________________________________________________________ -->
3434 <a name="i_br">'<tt>br</tt>' Instruction</a>
3441 br i1 <cond>, label <iftrue>, label <iffalse>
3442 br label <dest> <i>; Unconditional branch</i>
3446 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3447 different basic block in the current function. There are two forms of this
3448 instruction, corresponding to a conditional branch and an unconditional
3452 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3453 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3454 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3458 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3459 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3460 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3461 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3466 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3467 br i1 %cond, label %IfEqual, label %IfUnequal
3469 <a href="#i_ret">ret</a> i32 1
3471 <a href="#i_ret">ret</a> i32 0
3476 <!-- _______________________________________________________________________ -->
3478 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3485 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3489 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3490 several different places. It is a generalization of the '<tt>br</tt>'
3491 instruction, allowing a branch to occur to one of many possible
3495 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3496 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3497 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3498 The table is not allowed to contain duplicate constant entries.</p>
3501 <p>The <tt>switch</tt> instruction specifies a table of values and
3502 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3503 is searched for the given value. If the value is found, control flow is
3504 transferred to the corresponding destination; otherwise, control flow is
3505 transferred to the default destination.</p>
3507 <h5>Implementation:</h5>
3508 <p>Depending on properties of the target machine and the particular
3509 <tt>switch</tt> instruction, this instruction may be code generated in
3510 different ways. For example, it could be generated as a series of chained
3511 conditional branches or with a lookup table.</p>
3515 <i>; Emulate a conditional br instruction</i>
3516 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3517 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3519 <i>; Emulate an unconditional br instruction</i>
3520 switch i32 0, label %dest [ ]
3522 <i>; Implement a jump table:</i>
3523 switch i32 %val, label %otherwise [ i32 0, label %onzero
3525 i32 2, label %ontwo ]
3531 <!-- _______________________________________________________________________ -->
3533 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3540 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3545 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3546 within the current function, whose address is specified by
3547 "<tt>address</tt>". Address must be derived from a <a
3548 href="#blockaddress">blockaddress</a> constant.</p>
3552 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3553 rest of the arguments indicate the full set of possible destinations that the
3554 address may point to. Blocks are allowed to occur multiple times in the
3555 destination list, though this isn't particularly useful.</p>
3557 <p>This destination list is required so that dataflow analysis has an accurate
3558 understanding of the CFG.</p>
3562 <p>Control transfers to the block specified in the address argument. All
3563 possible destination blocks must be listed in the label list, otherwise this
3564 instruction has undefined behavior. This implies that jumps to labels
3565 defined in other functions have undefined behavior as well.</p>
3567 <h5>Implementation:</h5>
3569 <p>This is typically implemented with a jump through a register.</p>
3573 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3579 <!-- _______________________________________________________________________ -->
3581 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3588 <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>]
3589 to label <normal label> unwind label <exception label>
3593 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3594 function, with the possibility of control flow transfer to either the
3595 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3596 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3597 control flow will return to the "normal" label. If the callee (or any
3598 indirect callees) returns via the "<a href="#i_resume"><tt>resume</tt></a>"
3599 instruction or other exception handling mechanism, control is interrupted and
3600 continued at the dynamically nearest "exception" label.</p>
3602 <p>The '<tt>exception</tt>' label is a
3603 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3604 exception. As such, '<tt>exception</tt>' label is required to have the
3605 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3606 the information about the behavior of the program after unwinding
3607 happens, as its first non-PHI instruction. The restrictions on the
3608 "<tt>landingpad</tt>" instruction's tightly couples it to the
3609 "<tt>invoke</tt>" instruction, so that the important information contained
3610 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3614 <p>This instruction requires several arguments:</p>
3617 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3618 convention</a> the call should use. If none is specified, the call
3619 defaults to using C calling conventions.</li>
3621 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3622 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3623 '<tt>inreg</tt>' attributes are valid here.</li>
3625 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3626 function value being invoked. In most cases, this is a direct function
3627 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3628 off an arbitrary pointer to function value.</li>
3630 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3631 function to be invoked. </li>
3633 <li>'<tt>function args</tt>': argument list whose types match the function
3634 signature argument types and parameter attributes. All arguments must be
3635 of <a href="#t_firstclass">first class</a> type. If the function
3636 signature indicates the function accepts a variable number of arguments,
3637 the extra arguments can be specified.</li>
3639 <li>'<tt>normal label</tt>': the label reached when the called function
3640 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3642 <li>'<tt>exception label</tt>': the label reached when a callee returns via
3643 the <a href="#i_resume"><tt>resume</tt></a> instruction or other exception
3644 handling mechanism.</li>
3646 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3647 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3648 '<tt>readnone</tt>' attributes are valid here.</li>
3652 <p>This instruction is designed to operate as a standard
3653 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3654 primary difference is that it establishes an association with a label, which
3655 is used by the runtime library to unwind the stack.</p>
3657 <p>This instruction is used in languages with destructors to ensure that proper
3658 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3659 exception. Additionally, this is important for implementation of
3660 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3662 <p>For the purposes of the SSA form, the definition of the value returned by the
3663 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3664 block to the "normal" label. If the callee unwinds then no return value is
3669 %retval = invoke i32 @Test(i32 15) to label %Continue
3670 unwind label %TestCleanup <i>; {i32}:retval set</i>
3671 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3672 unwind label %TestCleanup <i>; {i32}:retval set</i>
3677 <!-- _______________________________________________________________________ -->
3680 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3687 resume <type> <value>
3691 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3695 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3696 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3700 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3701 (in-flight) exception whose unwinding was interrupted with
3702 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3706 resume { i8*, i32 } %exn
3711 <!-- _______________________________________________________________________ -->
3714 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3725 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3726 instruction is used to inform the optimizer that a particular portion of the
3727 code is not reachable. This can be used to indicate that the code after a
3728 no-return function cannot be reached, and other facts.</p>
3731 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3737 <!-- ======================================================================= -->
3739 <a name="binaryops">Binary Operations</a>
3744 <p>Binary operators are used to do most of the computation in a program. They
3745 require two operands of the same type, execute an operation on them, and
3746 produce a single value. The operands might represent multiple data, as is
3747 the case with the <a href="#t_vector">vector</a> data type. The result value
3748 has the same type as its operands.</p>
3750 <p>There are several different binary operators:</p>
3752 <!-- _______________________________________________________________________ -->
3754 <a name="i_add">'<tt>add</tt>' Instruction</a>
3761 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3762 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3763 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3764 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3768 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3771 <p>The two arguments to the '<tt>add</tt>' instruction must
3772 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3773 integer values. Both arguments must have identical types.</p>
3776 <p>The value produced is the integer sum of the two operands.</p>
3778 <p>If the sum has unsigned overflow, the result returned is the mathematical
3779 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3781 <p>Because LLVM integers use a two's complement representation, this instruction
3782 is appropriate for both signed and unsigned integers.</p>
3784 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3785 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3786 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3787 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3788 respectively, occurs.</p>
3792 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3797 <!-- _______________________________________________________________________ -->
3799 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3806 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3810 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3813 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3814 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3815 floating point values. Both arguments must have identical types.</p>
3818 <p>The value produced is the floating point sum of the two operands.</p>
3822 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3827 <!-- _______________________________________________________________________ -->
3829 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3836 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3837 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3838 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3839 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3843 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3846 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3847 '<tt>neg</tt>' instruction present in most other intermediate
3848 representations.</p>
3851 <p>The two arguments to the '<tt>sub</tt>' instruction must
3852 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3853 integer values. Both arguments must have identical types.</p>
3856 <p>The value produced is the integer difference of the two operands.</p>
3858 <p>If the difference has unsigned overflow, the result returned is the
3859 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3862 <p>Because LLVM integers use a two's complement representation, this instruction
3863 is appropriate for both signed and unsigned integers.</p>
3865 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3866 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3867 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3868 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3869 respectively, occurs.</p>
3873 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3874 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3879 <!-- _______________________________________________________________________ -->
3881 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3888 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3892 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3895 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3896 '<tt>fneg</tt>' instruction present in most other intermediate
3897 representations.</p>
3900 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3901 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3902 floating point values. Both arguments must have identical types.</p>
3905 <p>The value produced is the floating point difference of the two operands.</p>
3909 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3910 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3915 <!-- _______________________________________________________________________ -->
3917 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3924 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3925 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3926 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3927 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3931 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3934 <p>The two arguments to the '<tt>mul</tt>' instruction must
3935 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3936 integer values. Both arguments must have identical types.</p>
3939 <p>The value produced is the integer product of the two operands.</p>
3941 <p>If the result of the multiplication has unsigned overflow, the result
3942 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3943 width of the result.</p>
3945 <p>Because LLVM integers use a two's complement representation, and the result
3946 is the same width as the operands, this instruction returns the correct
3947 result for both signed and unsigned integers. If a full product
3948 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3949 be sign-extended or zero-extended as appropriate to the width of the full
3952 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3953 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3954 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3955 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3956 respectively, occurs.</p>
3960 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3965 <!-- _______________________________________________________________________ -->
3967 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3974 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3978 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3981 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3982 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3983 floating point values. Both arguments must have identical types.</p>
3986 <p>The value produced is the floating point product of the two operands.</p>
3990 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3995 <!-- _______________________________________________________________________ -->
3997 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
4004 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4005 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4009 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
4012 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
4013 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4014 values. Both arguments must have identical types.</p>
4017 <p>The value produced is the unsigned integer quotient of the two operands.</p>
4019 <p>Note that unsigned integer division and signed integer division are distinct
4020 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
4022 <p>Division by zero leads to undefined behavior.</p>
4024 <p>If the <tt>exact</tt> keyword is present, the result value of the
4025 <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a
4026 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
4031 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4036 <!-- _______________________________________________________________________ -->
4038 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
4045 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4046 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4050 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
4053 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
4054 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4055 values. Both arguments must have identical types.</p>
4058 <p>The value produced is the signed integer quotient of the two operands rounded
4061 <p>Note that signed integer division and unsigned integer division are distinct
4062 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
4064 <p>Division by zero leads to undefined behavior. Overflow also leads to
4065 undefined behavior; this is a rare case, but can occur, for example, by doing
4066 a 32-bit division of -2147483648 by -1.</p>
4068 <p>If the <tt>exact</tt> keyword is present, the result value of the
4069 <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would
4074 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4079 <!-- _______________________________________________________________________ -->
4081 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
4088 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4092 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
4095 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
4096 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4097 floating point values. Both arguments must have identical types.</p>
4100 <p>The value produced is the floating point quotient of the two operands.</p>
4104 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
4109 <!-- _______________________________________________________________________ -->
4111 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
4118 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4122 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
4123 division of its two arguments.</p>
4126 <p>The two arguments to the '<tt>urem</tt>' instruction must be
4127 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4128 values. Both arguments must have identical types.</p>
4131 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
4132 This instruction always performs an unsigned division to get the
4135 <p>Note that unsigned integer remainder and signed integer remainder are
4136 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
4138 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
4142 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4147 <!-- _______________________________________________________________________ -->
4149 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
4156 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4160 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
4161 division of its two operands. This instruction can also take
4162 <a href="#t_vector">vector</a> versions of the values in which case the
4163 elements must be integers.</p>
4166 <p>The two arguments to the '<tt>srem</tt>' instruction must be
4167 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4168 values. Both arguments must have identical types.</p>
4171 <p>This instruction returns the <i>remainder</i> of a division (where the result
4172 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
4173 <i>modulo</i> operator (where the result is either zero or has the same sign
4174 as the divisor, <tt>op2</tt>) of a value.
4175 For more information about the difference,
4176 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
4177 Math Forum</a>. For a table of how this is implemented in various languages,
4178 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
4179 Wikipedia: modulo operation</a>.</p>
4181 <p>Note that signed integer remainder and unsigned integer remainder are
4182 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
4184 <p>Taking the remainder of a division by zero leads to undefined behavior.
4185 Overflow also leads to undefined behavior; this is a rare case, but can
4186 occur, for example, by taking the remainder of a 32-bit division of
4187 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
4188 lets srem be implemented using instructions that return both the result of
4189 the division and the remainder.)</p>
4193 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4198 <!-- _______________________________________________________________________ -->
4200 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
4207 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4211 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
4212 its two operands.</p>
4215 <p>The two arguments to the '<tt>frem</tt>' instruction must be
4216 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4217 floating point values. Both arguments must have identical types.</p>
4220 <p>This instruction returns the <i>remainder</i> of a division. The remainder
4221 has the same sign as the dividend.</p>
4225 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
4232 <!-- ======================================================================= -->
4234 <a name="bitwiseops">Bitwise Binary Operations</a>
4239 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
4240 program. They are generally very efficient instructions and can commonly be
4241 strength reduced from other instructions. They require two operands of the
4242 same type, execute an operation on them, and produce a single value. The
4243 resulting value is the same type as its operands.</p>
4245 <!-- _______________________________________________________________________ -->
4247 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
4254 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4255 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4256 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4257 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4261 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
4262 a specified number of bits.</p>
4265 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
4266 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4267 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4270 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
4271 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
4272 is (statically or dynamically) negative or equal to or larger than the number
4273 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4274 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4275 shift amount in <tt>op2</tt>.</p>
4277 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
4278 <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits. If
4279 the <tt>nsw</tt> keyword is present, then the shift produces a
4280 <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree
4281 with the resultant sign bit. As such, NUW/NSW have the same semantics as
4282 they would if the shift were expressed as a mul instruction with the same
4283 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
4287 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
4288 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
4289 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
4290 <result> = shl i32 1, 32 <i>; undefined</i>
4291 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4296 <!-- _______________________________________________________________________ -->
4298 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4305 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4306 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4310 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4311 operand shifted to the right a specified number of bits with zero fill.</p>
4314 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4315 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4316 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4319 <p>This instruction always performs a logical shift right operation. The most
4320 significant bits of the result will be filled with zero bits after the shift.
4321 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4322 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4323 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4324 shift amount in <tt>op2</tt>.</p>
4326 <p>If the <tt>exact</tt> keyword is present, the result value of the
4327 <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4328 shifted out are non-zero.</p>
4333 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4334 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4335 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4336 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4337 <result> = lshr i32 1, 32 <i>; undefined</i>
4338 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4343 <!-- _______________________________________________________________________ -->
4345 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4352 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4353 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4357 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4358 operand shifted to the right a specified number of bits with sign
4362 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4363 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4364 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4367 <p>This instruction always performs an arithmetic shift right operation, The
4368 most significant bits of the result will be filled with the sign bit
4369 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4370 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4371 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4372 the corresponding shift amount in <tt>op2</tt>.</p>
4374 <p>If the <tt>exact</tt> keyword is present, the result value of the
4375 <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4376 shifted out are non-zero.</p>
4380 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4381 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4382 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4383 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4384 <result> = ashr i32 1, 32 <i>; undefined</i>
4385 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4390 <!-- _______________________________________________________________________ -->
4392 <a name="i_and">'<tt>and</tt>' Instruction</a>
4399 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4403 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4407 <p>The two arguments to the '<tt>and</tt>' instruction must be
4408 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4409 values. Both arguments must have identical types.</p>
4412 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4414 <table border="1" cellspacing="0" cellpadding="4">
4446 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4447 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4448 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4451 <!-- _______________________________________________________________________ -->
4453 <a name="i_or">'<tt>or</tt>' Instruction</a>
4460 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4464 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4468 <p>The two arguments to the '<tt>or</tt>' instruction must be
4469 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4470 values. Both arguments must have identical types.</p>
4473 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4475 <table border="1" cellspacing="0" cellpadding="4">
4507 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4508 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4509 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4514 <!-- _______________________________________________________________________ -->
4516 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4523 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4527 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4528 its two operands. The <tt>xor</tt> is used to implement the "one's
4529 complement" operation, which is the "~" operator in C.</p>
4532 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4533 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4534 values. Both arguments must have identical types.</p>
4537 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4539 <table border="1" cellspacing="0" cellpadding="4">
4571 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4572 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4573 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4574 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4581 <!-- ======================================================================= -->
4583 <a name="vectorops">Vector Operations</a>
4588 <p>LLVM supports several instructions to represent vector operations in a
4589 target-independent manner. These instructions cover the element-access and
4590 vector-specific operations needed to process vectors effectively. While LLVM
4591 does directly support these vector operations, many sophisticated algorithms
4592 will want to use target-specific intrinsics to take full advantage of a
4593 specific target.</p>
4595 <!-- _______________________________________________________________________ -->
4597 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4604 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4608 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4609 from a vector at a specified index.</p>
4613 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4614 of <a href="#t_vector">vector</a> type. The second operand is an index
4615 indicating the position from which to extract the element. The index may be
4619 <p>The result is a scalar of the same type as the element type of
4620 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4621 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4622 results are undefined.</p>
4626 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4631 <!-- _______________________________________________________________________ -->
4633 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4640 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4644 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4645 vector at a specified index.</p>
4648 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4649 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4650 whose type must equal the element type of the first operand. The third
4651 operand is an index indicating the position at which to insert the value.
4652 The index may be a variable.</p>
4655 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4656 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4657 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4658 results are undefined.</p>
4662 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4667 <!-- _______________________________________________________________________ -->
4669 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4676 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4680 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4681 from two input vectors, returning a vector with the same element type as the
4682 input and length that is the same as the shuffle mask.</p>
4685 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4686 with types that match each other. The third argument is a shuffle mask whose
4687 element type is always 'i32'. The result of the instruction is a vector
4688 whose length is the same as the shuffle mask and whose element type is the
4689 same as the element type of the first two operands.</p>
4691 <p>The shuffle mask operand is required to be a constant vector with either
4692 constant integer or undef values.</p>
4695 <p>The elements of the two input vectors are numbered from left to right across
4696 both of the vectors. The shuffle mask operand specifies, for each element of
4697 the result vector, which element of the two input vectors the result element
4698 gets. The element selector may be undef (meaning "don't care") and the
4699 second operand may be undef if performing a shuffle from only one vector.</p>
4703 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4704 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4705 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4706 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4707 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4708 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4709 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4710 <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>
4717 <!-- ======================================================================= -->
4719 <a name="aggregateops">Aggregate Operations</a>
4724 <p>LLVM supports several instructions for working with
4725 <a href="#t_aggregate">aggregate</a> values.</p>
4727 <!-- _______________________________________________________________________ -->
4729 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4736 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4740 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4741 from an <a href="#t_aggregate">aggregate</a> value.</p>
4744 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4745 of <a href="#t_struct">struct</a> or
4746 <a href="#t_array">array</a> type. The operands are constant indices to
4747 specify which value to extract in a similar manner as indices in a
4748 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4749 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4751 <li>Since the value being indexed is not a pointer, the first index is
4752 omitted and assumed to be zero.</li>
4753 <li>At least one index must be specified.</li>
4754 <li>Not only struct indices but also array indices must be in
4759 <p>The result is the value at the position in the aggregate specified by the
4764 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4769 <!-- _______________________________________________________________________ -->
4771 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4778 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4782 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4783 in an <a href="#t_aggregate">aggregate</a> value.</p>
4786 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4787 of <a href="#t_struct">struct</a> or
4788 <a href="#t_array">array</a> type. The second operand is a first-class
4789 value to insert. The following operands are constant indices indicating
4790 the position at which to insert the value in a similar manner as indices in a
4791 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4792 value to insert must have the same type as the value identified by the
4796 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4797 that of <tt>val</tt> except that the value at the position specified by the
4798 indices is that of <tt>elt</tt>.</p>
4802 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4803 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4804 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4811 <!-- ======================================================================= -->
4813 <a name="memoryops">Memory Access and Addressing Operations</a>
4818 <p>A key design point of an SSA-based representation is how it represents
4819 memory. In LLVM, no memory locations are in SSA form, which makes things
4820 very simple. This section describes how to read, write, and allocate
4823 <!-- _______________________________________________________________________ -->
4825 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4832 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4836 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4837 currently executing function, to be automatically released when this function
4838 returns to its caller. The object is always allocated in the generic address
4839 space (address space zero).</p>
4842 <p>The '<tt>alloca</tt>' instruction
4843 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4844 runtime stack, returning a pointer of the appropriate type to the program.
4845 If "NumElements" is specified, it is the number of elements allocated,
4846 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4847 specified, the value result of the allocation is guaranteed to be aligned to
4848 at least that boundary. If not specified, or if zero, the target can choose
4849 to align the allocation on any convenient boundary compatible with the
4852 <p>'<tt>type</tt>' may be any sized type.</p>
4855 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4856 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4857 memory is automatically released when the function returns. The
4858 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4859 variables that must have an address available. When the function returns
4860 (either with the <tt><a href="#i_ret">ret</a></tt>
4861 or <tt><a href="#i_resume">resume</a></tt> instructions), the memory is
4862 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4866 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4867 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4868 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4869 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4874 <!-- _______________________________________________________________________ -->
4876 <a name="i_load">'<tt>load</tt>' Instruction</a>
4883 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4884 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4885 !<index> = !{ i32 1 }
4889 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4892 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4893 from which to load. The pointer must point to
4894 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4895 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4896 number or order of execution of this <tt>load</tt> with other <a
4897 href="#volatile">volatile operations</a>.</p>
4899 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
4900 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4901 argument. The <code>release</code> and <code>acq_rel</code> orderings are
4902 not valid on <code>load</code> instructions. Atomic loads produce <a
4903 href="#memorymodel">defined</a> results when they may see multiple atomic
4904 stores. The type of the pointee must be an integer type whose bit width
4905 is a power of two greater than or equal to eight and less than or equal
4906 to a target-specific size limit. <code>align</code> must be explicitly
4907 specified on atomic loads, and the load has undefined behavior if the
4908 alignment is not set to a value which is at least the size in bytes of
4909 the pointee. <code>!nontemporal</code> does not have any defined semantics
4910 for atomic loads.</p>
4912 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4913 operation (that is, the alignment of the memory address). A value of 0 or an
4914 omitted <tt>align</tt> argument means that the operation has the preferential
4915 alignment for the target. It is the responsibility of the code emitter to
4916 ensure that the alignment information is correct. Overestimating the
4917 alignment results in undefined behavior. Underestimating the alignment may
4918 produce less efficient code. An alignment of 1 is always safe.</p>
4920 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4921 metatadata name <index> corresponding to a metadata node with
4922 one <tt>i32</tt> entry of value 1. The existence of
4923 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4924 and code generator that this load is not expected to be reused in the cache.
4925 The code generator may select special instructions to save cache bandwidth,
4926 such as the <tt>MOVNT</tt> instruction on x86.</p>
4928 <p>The optional <tt>!invariant.load</tt> metadata must reference a single
4929 metatadata name <index> corresponding to a metadata node with no
4930 entries. The existence of the <tt>!invariant.load</tt> metatadata on the
4931 instruction tells the optimizer and code generator that this load address
4932 points to memory which does not change value during program execution.
4933 The optimizer may then move this load around, for example, by hoisting it
4934 out of loops using loop invariant code motion.</p>
4937 <p>The location of memory pointed to is loaded. If the value being loaded is of
4938 scalar type then the number of bytes read does not exceed the minimum number
4939 of bytes needed to hold all bits of the type. For example, loading an
4940 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4941 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4942 is undefined if the value was not originally written using a store of the
4947 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4948 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4949 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4954 <!-- _______________________________________________________________________ -->
4956 <a name="i_store">'<tt>store</tt>' Instruction</a>
4963 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4964 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
4968 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4971 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4972 and an address at which to store it. The type of the
4973 '<tt><pointer></tt>' operand must be a pointer to
4974 the <a href="#t_firstclass">first class</a> type of the
4975 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
4976 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
4977 order of execution of this <tt>store</tt> with other <a
4978 href="#volatile">volatile operations</a>.</p>
4980 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
4981 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4982 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
4983 valid on <code>store</code> instructions. Atomic loads produce <a
4984 href="#memorymodel">defined</a> results when they may see multiple atomic
4985 stores. The type of the pointee must be an integer type whose bit width
4986 is a power of two greater than or equal to eight and less than or equal
4987 to a target-specific size limit. <code>align</code> must be explicitly
4988 specified on atomic stores, and the store has undefined behavior if the
4989 alignment is not set to a value which is at least the size in bytes of
4990 the pointee. <code>!nontemporal</code> does not have any defined semantics
4991 for atomic stores.</p>
4993 <p>The optional constant "align" argument specifies the alignment of the
4994 operation (that is, the alignment of the memory address). A value of 0 or an
4995 omitted "align" argument means that the operation has the preferential
4996 alignment for the target. It is the responsibility of the code emitter to
4997 ensure that the alignment information is correct. Overestimating the
4998 alignment results in an undefined behavior. Underestimating the alignment may
4999 produce less efficient code. An alignment of 1 is always safe.</p>
5001 <p>The optional !nontemporal metadata must reference a single metatadata
5002 name <index> corresponding to a metadata node with one i32 entry of
5003 value 1. The existence of the !nontemporal metatadata on the
5004 instruction tells the optimizer and code generator that this load is
5005 not expected to be reused in the cache. The code generator may
5006 select special instructions to save cache bandwidth, such as the
5007 MOVNT instruction on x86.</p>
5011 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
5012 location specified by the '<tt><pointer></tt>' operand. If
5013 '<tt><value></tt>' is of scalar type then the number of bytes written
5014 does not exceed the minimum number of bytes needed to hold all bits of the
5015 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
5016 writing a value of a type like <tt>i20</tt> with a size that is not an
5017 integral number of bytes, it is unspecified what happens to the extra bits
5018 that do not belong to the type, but they will typically be overwritten.</p>
5022 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5023 store i32 3, i32* %ptr <i>; yields {void}</i>
5024 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
5029 <!-- _______________________________________________________________________ -->
5031 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
5038 fence [singlethread] <ordering> <i>; yields {void}</i>
5042 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
5043 between operations.</p>
5045 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
5046 href="#ordering">ordering</a> argument which defines what
5047 <i>synchronizes-with</i> edges they add. They can only be given
5048 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
5049 <code>seq_cst</code> orderings.</p>
5052 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
5053 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
5054 <code>acquire</code> ordering semantics if and only if there exist atomic
5055 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
5056 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
5057 <var>X</var> modifies <var>M</var> (either directly or through some side effect
5058 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
5059 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
5060 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
5061 than an explicit <code>fence</code>, one (but not both) of the atomic operations
5062 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
5063 <code>acquire</code> (resp.) ordering constraint and still
5064 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
5065 <i>happens-before</i> edge.</p>
5067 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
5068 having both <code>acquire</code> and <code>release</code> semantics specified
5069 above, participates in the global program order of other <code>seq_cst</code>
5070 operations and/or fences.</p>
5072 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
5073 specifies that the fence only synchronizes with other fences in the same
5074 thread. (This is useful for interacting with signal handlers.)</p>
5078 fence acquire <i>; yields {void}</i>
5079 fence singlethread seq_cst <i>; yields {void}</i>
5084 <!-- _______________________________________________________________________ -->
5086 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
5093 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
5097 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
5098 It loads a value in memory and compares it to a given value. If they are
5099 equal, it stores a new value into the memory.</p>
5102 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
5103 address to operate on, a value to compare to the value currently be at that
5104 address, and a new value to place at that address if the compared values are
5105 equal. The type of '<var><cmp></var>' must be an integer type whose
5106 bit width is a power of two greater than or equal to eight and less than
5107 or equal to a target-specific size limit. '<var><cmp></var>' and
5108 '<var><new></var>' must have the same type, and the type of
5109 '<var><pointer></var>' must be a pointer to that type. If the
5110 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
5111 optimizer is not allowed to modify the number or order of execution
5112 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
5115 <!-- FIXME: Extend allowed types. -->
5117 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
5118 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
5120 <p>The optional "<code>singlethread</code>" argument declares that the
5121 <code>cmpxchg</code> is only atomic with respect to code (usually signal
5122 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
5123 cmpxchg is atomic with respect to all other code in the system.</p>
5125 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
5126 the size in memory of the operand.
5129 <p>The contents of memory at the location specified by the
5130 '<tt><pointer></tt>' operand is read and compared to
5131 '<tt><cmp></tt>'; if the read value is the equal,
5132 '<tt><new></tt>' is written. The original value at the location
5135 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
5136 purpose of identifying <a href="#release_sequence">release sequences</a>. A
5137 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
5138 parameter determined by dropping any <code>release</code> part of the
5139 <code>cmpxchg</code>'s ordering.</p>
5142 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
5143 optimization work on ARM.)
5145 FIXME: Is a weaker ordering constraint on failure helpful in practice?
5151 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
5152 <a href="#i_br">br</a> label %loop
5155 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
5156 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
5157 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
5158 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
5159 <a href="#i_br">br</a> i1 %success, label %done, label %loop
5167 <!-- _______________________________________________________________________ -->
5169 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
5176 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
5180 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
5183 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
5184 operation to apply, an address whose value to modify, an argument to the
5185 operation. The operation must be one of the following keywords:</p>
5200 <p>The type of '<var><value></var>' must be an integer type whose
5201 bit width is a power of two greater than or equal to eight and less than
5202 or equal to a target-specific size limit. The type of the
5203 '<code><pointer></code>' operand must be a pointer to that type.
5204 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
5205 optimizer is not allowed to modify the number or order of execution of this
5206 <code>atomicrmw</code> with other <a href="#volatile">volatile
5209 <!-- FIXME: Extend allowed types. -->
5212 <p>The contents of memory at the location specified by the
5213 '<tt><pointer></tt>' operand are atomically read, modified, and written
5214 back. The original value at the location is returned. The modification is
5215 specified by the <var>operation</var> argument:</p>
5218 <li>xchg: <code>*ptr = val</code></li>
5219 <li>add: <code>*ptr = *ptr + val</code></li>
5220 <li>sub: <code>*ptr = *ptr - val</code></li>
5221 <li>and: <code>*ptr = *ptr & val</code></li>
5222 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
5223 <li>or: <code>*ptr = *ptr | val</code></li>
5224 <li>xor: <code>*ptr = *ptr ^ val</code></li>
5225 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
5226 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
5227 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
5228 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
5233 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
5238 <!-- _______________________________________________________________________ -->
5240 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
5247 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5248 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5249 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5253 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
5254 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
5255 It performs address calculation only and does not access memory.</p>
5258 <p>The first argument is always a pointer or a vector of pointers,
5259 and forms the basis of the
5260 calculation. The remaining arguments are indices that indicate which of the
5261 elements of the aggregate object are indexed. The interpretation of each
5262 index is dependent on the type being indexed into. The first index always
5263 indexes the pointer value given as the first argument, the second index
5264 indexes a value of the type pointed to (not necessarily the value directly
5265 pointed to, since the first index can be non-zero), etc. The first type
5266 indexed into must be a pointer value, subsequent types can be arrays,
5267 vectors, and structs. Note that subsequent types being indexed into
5268 can never be pointers, since that would require loading the pointer before
5269 continuing calculation.</p>
5271 <p>The type of each index argument depends on the type it is indexing into.
5272 When indexing into a (optionally packed) structure, only <tt>i32</tt>
5273 integer <b>constants</b> are allowed. When indexing into an array, pointer
5274 or vector, integers of any width are allowed, and they are not required to be
5275 constant. These integers are treated as signed values where relevant.</p>
5277 <p>For example, let's consider a C code fragment and how it gets compiled to
5280 <pre class="doc_code">
5292 int *foo(struct ST *s) {
5293 return &s[1].Z.B[5][13];
5297 <p>The LLVM code generated by Clang is:</p>
5299 <pre class="doc_code">
5300 %struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 }
5301 %struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT }
5303 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5305 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5311 <p>In the example above, the first index is indexing into the
5312 '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a
5313 '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a
5314 structure. The second index indexes into the third element of the structure,
5315 yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>'
5316 type, another structure. The third index indexes into the second element of
5317 the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The
5318 two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>'
5319 type. The '<tt>getelementptr</tt>' instruction returns a pointer to this
5320 element, thus computing a value of '<tt>i32*</tt>' type.</p>
5322 <p>Note that it is perfectly legal to index partially through a structure,
5323 returning a pointer to an inner element. Because of this, the LLVM code for
5324 the given testcase is equivalent to:</p>
5326 <pre class="doc_code">
5327 define i32* @foo(%struct.ST* %s) {
5328 %t1 = getelementptr %struct.ST* %s, i32 1 <i>; yields %struct.ST*:%t1</i>
5329 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 <i>; yields %struct.RT*:%t2</i>
5330 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5331 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5332 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5337 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5338 <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the
5339 base pointer is not an <i>in bounds</i> address of an allocated object,
5340 or if any of the addresses that would be formed by successive addition of
5341 the offsets implied by the indices to the base address with infinitely
5342 precise signed arithmetic are not an <i>in bounds</i> address of that
5343 allocated object. The <i>in bounds</i> addresses for an allocated object
5344 are all the addresses that point into the object, plus the address one
5346 In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword
5347 applies to each of the computations element-wise. </p>
5349 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5350 the base address with silently-wrapping two's complement arithmetic. If the
5351 offsets have a different width from the pointer, they are sign-extended or
5352 truncated to the width of the pointer. The result value of the
5353 <tt>getelementptr</tt> may be outside the object pointed to by the base
5354 pointer. The result value may not necessarily be used to access memory
5355 though, even if it happens to point into allocated storage. See the
5356 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5359 <p>The getelementptr instruction is often confusing. For some more insight into
5360 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5364 <i>; yields [12 x i8]*:aptr</i>
5365 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5366 <i>; yields i8*:vptr</i>
5367 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5368 <i>; yields i8*:eptr</i>
5369 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5370 <i>; yields i32*:iptr</i>
5371 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5374 <p>In cases where the pointer argument is a vector of pointers, only a
5375 single index may be used, and the number of vector elements has to be
5376 the same. For example: </p>
5377 <pre class="doc_code">
5378 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5385 <!-- ======================================================================= -->
5387 <a name="convertops">Conversion Operations</a>
5392 <p>The instructions in this category are the conversion instructions (casting)
5393 which all take a single operand and a type. They perform various bit
5394 conversions on the operand.</p>
5396 <!-- _______________________________________________________________________ -->
5398 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5405 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5409 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5410 type <tt>ty2</tt>.</p>
5413 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5414 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5415 of the same number of integers.
5416 The bit size of the <tt>value</tt> must be larger than
5417 the bit size of the destination type, <tt>ty2</tt>.
5418 Equal sized types are not allowed.</p>
5421 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5422 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5423 source size must be larger than the destination size, <tt>trunc</tt> cannot
5424 be a <i>no-op cast</i>. It will always truncate bits.</p>
5428 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5429 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5430 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5431 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5436 <!-- _______________________________________________________________________ -->
5438 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5445 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5449 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5454 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5455 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5456 of the same number of integers.
5457 The bit size of the <tt>value</tt> must be smaller than
5458 the bit size of the destination type,
5462 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5463 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5465 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5469 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5470 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5471 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5476 <!-- _______________________________________________________________________ -->
5478 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5485 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5489 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5492 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5493 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5494 of the same number of integers.
5495 The bit size of the <tt>value</tt> must be smaller than
5496 the bit size of the destination type,
5500 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5501 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5502 of the type <tt>ty2</tt>.</p>
5504 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5508 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5509 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5510 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5515 <!-- _______________________________________________________________________ -->
5517 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5524 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5528 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5532 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5533 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5534 to cast it to. The size of <tt>value</tt> must be larger than the size of
5535 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5536 <i>no-op cast</i>.</p>
5539 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5540 <a href="#t_floating">floating point</a> type to a smaller
5541 <a href="#t_floating">floating point</a> type. If the value cannot fit
5542 within the destination type, <tt>ty2</tt>, then the results are
5547 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5548 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5553 <!-- _______________________________________________________________________ -->
5555 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5562 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5566 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5567 floating point value.</p>
5570 <p>The '<tt>fpext</tt>' instruction takes a
5571 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5572 a <a href="#t_floating">floating point</a> type to cast it to. The source
5573 type must be smaller than the destination type.</p>
5576 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5577 <a href="#t_floating">floating point</a> type to a larger
5578 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5579 used to make a <i>no-op cast</i> because it always changes bits. Use
5580 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5584 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5585 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5590 <!-- _______________________________________________________________________ -->
5592 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5599 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5603 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5604 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5607 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5608 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5609 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5610 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5611 vector integer type with the same number of elements as <tt>ty</tt></p>
5614 <p>The '<tt>fptoui</tt>' instruction converts its
5615 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5616 towards zero) unsigned integer value. If the value cannot fit
5617 in <tt>ty2</tt>, the results are undefined.</p>
5621 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5622 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5623 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5628 <!-- _______________________________________________________________________ -->
5630 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5637 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5641 <p>The '<tt>fptosi</tt>' instruction converts
5642 <a href="#t_floating">floating point</a> <tt>value</tt> to
5643 type <tt>ty2</tt>.</p>
5646 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5647 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5648 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5649 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5650 vector integer type with the same number of elements as <tt>ty</tt></p>
5653 <p>The '<tt>fptosi</tt>' instruction converts its
5654 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5655 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5656 the results are undefined.</p>
5660 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5661 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5662 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5667 <!-- _______________________________________________________________________ -->
5669 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5676 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5680 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5681 integer and converts that value to the <tt>ty2</tt> type.</p>
5684 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5685 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5686 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5687 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5688 floating point type with the same number of elements as <tt>ty</tt></p>
5691 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5692 integer quantity and converts it to the corresponding floating point
5693 value. If the value cannot fit in the floating point value, the results are
5698 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5699 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5704 <!-- _______________________________________________________________________ -->
5706 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5713 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5717 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5718 and converts that value to the <tt>ty2</tt> type.</p>
5721 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5722 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5723 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5724 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5725 floating point type with the same number of elements as <tt>ty</tt></p>
5728 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5729 quantity and converts it to the corresponding floating point value. If the
5730 value cannot fit in the floating point value, the results are undefined.</p>
5734 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5735 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5740 <!-- _______________________________________________________________________ -->
5742 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5749 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5753 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of
5754 pointers <tt>value</tt> to
5755 the integer (or vector of integers) type <tt>ty2</tt>.</p>
5758 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5759 must be a a value of type <a href="#t_pointer">pointer</a> or a vector of
5760 pointers, and a type to cast it to
5761 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector
5762 of integers type.</p>
5765 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5766 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5767 truncating or zero extending that value to the size of the integer type. If
5768 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5769 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5770 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5775 %X = ptrtoint i32* %P to i8 <i>; yields truncation on 32-bit architecture</i>
5776 %Y = ptrtoint i32* %P to i64 <i>; yields zero extension on 32-bit architecture</i>
5777 %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>
5782 <!-- _______________________________________________________________________ -->
5784 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5791 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5795 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5796 pointer type, <tt>ty2</tt>.</p>
5799 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5800 value to cast, and a type to cast it to, which must be a
5801 <a href="#t_pointer">pointer</a> type.</p>
5804 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5805 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5806 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5807 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5808 than the size of a pointer then a zero extension is done. If they are the
5809 same size, nothing is done (<i>no-op cast</i>).</p>
5813 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5814 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5815 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5816 %Z = inttoptr <4 x i32> %G to <4 x i8*><i>; yields truncation of vector G to four pointers</i>
5821 <!-- _______________________________________________________________________ -->
5823 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5830 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5834 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5835 <tt>ty2</tt> without changing any bits.</p>
5838 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5839 non-aggregate first class value, and a type to cast it to, which must also be
5840 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5841 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5842 identical. If the source type is a pointer, the destination type must also be
5843 a pointer. This instruction supports bitwise conversion of vectors to
5844 integers and to vectors of other types (as long as they have the same
5848 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5849 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5850 this conversion. The conversion is done as if the <tt>value</tt> had been
5851 stored to memory and read back as type <tt>ty2</tt>.
5852 Pointer (or vector of pointers) types may only be converted to other pointer
5853 (or vector of pointers) types with this instruction. To convert
5854 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5855 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5859 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
5860 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
5861 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
5862 %Z = bitcast <2 x i32*> %V to <2 x i64*> <i>; yields <2 x i64*></i>
5869 <!-- ======================================================================= -->
5871 <a name="otherops">Other Operations</a>
5876 <p>The instructions in this category are the "miscellaneous" instructions, which
5877 defy better classification.</p>
5879 <!-- _______________________________________________________________________ -->
5881 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5888 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5892 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5893 boolean values based on comparison of its two integer, integer vector,
5894 pointer, or pointer vector operands.</p>
5897 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
5898 the condition code indicating the kind of comparison to perform. It is not a
5899 value, just a keyword. The possible condition code are:</p>
5902 <li><tt>eq</tt>: equal</li>
5903 <li><tt>ne</tt>: not equal </li>
5904 <li><tt>ugt</tt>: unsigned greater than</li>
5905 <li><tt>uge</tt>: unsigned greater or equal</li>
5906 <li><tt>ult</tt>: unsigned less than</li>
5907 <li><tt>ule</tt>: unsigned less or equal</li>
5908 <li><tt>sgt</tt>: signed greater than</li>
5909 <li><tt>sge</tt>: signed greater or equal</li>
5910 <li><tt>slt</tt>: signed less than</li>
5911 <li><tt>sle</tt>: signed less or equal</li>
5914 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
5915 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
5916 typed. They must also be identical types.</p>
5919 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
5920 condition code given as <tt>cond</tt>. The comparison performed always yields
5921 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
5922 result, as follows:</p>
5925 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
5926 <tt>false</tt> otherwise. No sign interpretation is necessary or
5929 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
5930 <tt>false</tt> otherwise. No sign interpretation is necessary or
5933 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
5934 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5936 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
5937 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5938 to <tt>op2</tt>.</li>
5940 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
5941 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5943 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
5944 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5946 <li><tt>sgt</tt>: interprets the operands as signed values and yields
5947 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5949 <li><tt>sge</tt>: interprets the operands as signed values and yields
5950 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5951 to <tt>op2</tt>.</li>
5953 <li><tt>slt</tt>: interprets the operands as signed values and yields
5954 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5956 <li><tt>sle</tt>: interprets the operands as signed values and yields
5957 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5960 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
5961 values are compared as if they were integers.</p>
5963 <p>If the operands are integer vectors, then they are compared element by
5964 element. The result is an <tt>i1</tt> vector with the same number of elements
5965 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
5969 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
5970 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
5971 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
5972 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
5973 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
5974 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
5977 <p>Note that the code generator does not yet support vector types with
5978 the <tt>icmp</tt> instruction.</p>
5982 <!-- _______________________________________________________________________ -->
5984 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
5991 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5995 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
5996 values based on comparison of its operands.</p>
5998 <p>If the operands are floating point scalars, then the result type is a boolean
5999 (<a href="#t_integer"><tt>i1</tt></a>).</p>
6001 <p>If the operands are floating point vectors, then the result type is a vector
6002 of boolean with the same number of elements as the operands being
6006 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
6007 the condition code indicating the kind of comparison to perform. It is not a
6008 value, just a keyword. The possible condition code are:</p>
6011 <li><tt>false</tt>: no comparison, always returns false</li>
6012 <li><tt>oeq</tt>: ordered and equal</li>
6013 <li><tt>ogt</tt>: ordered and greater than </li>
6014 <li><tt>oge</tt>: ordered and greater than or equal</li>
6015 <li><tt>olt</tt>: ordered and less than </li>
6016 <li><tt>ole</tt>: ordered and less than or equal</li>
6017 <li><tt>one</tt>: ordered and not equal</li>
6018 <li><tt>ord</tt>: ordered (no nans)</li>
6019 <li><tt>ueq</tt>: unordered or equal</li>
6020 <li><tt>ugt</tt>: unordered or greater than </li>
6021 <li><tt>uge</tt>: unordered or greater than or equal</li>
6022 <li><tt>ult</tt>: unordered or less than </li>
6023 <li><tt>ule</tt>: unordered or less than or equal</li>
6024 <li><tt>une</tt>: unordered or not equal</li>
6025 <li><tt>uno</tt>: unordered (either nans)</li>
6026 <li><tt>true</tt>: no comparison, always returns true</li>
6029 <p><i>Ordered</i> means that neither operand is a QNAN while
6030 <i>unordered</i> means that either operand may be a QNAN.</p>
6032 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
6033 a <a href="#t_floating">floating point</a> type or
6034 a <a href="#t_vector">vector</a> of floating point type. They must have
6035 identical types.</p>
6038 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
6039 according to the condition code given as <tt>cond</tt>. If the operands are
6040 vectors, then the vectors are compared element by element. Each comparison
6041 performed always yields an <a href="#t_integer">i1</a> result, as
6045 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
6047 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6048 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6050 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6051 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6053 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6054 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6056 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6057 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6059 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6060 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6062 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6063 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6065 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
6067 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
6068 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6070 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
6071 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6073 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
6074 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6076 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
6077 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6079 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
6080 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6082 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
6083 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6085 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
6087 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
6092 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
6093 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
6094 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
6095 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
6098 <p>Note that the code generator does not yet support vector types with
6099 the <tt>fcmp</tt> instruction.</p>
6103 <!-- _______________________________________________________________________ -->
6105 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
6112 <result> = phi <ty> [ <val0>, <label0>], ...
6116 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
6117 SSA graph representing the function.</p>
6120 <p>The type of the incoming values is specified with the first type field. After
6121 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
6122 one pair for each predecessor basic block of the current block. Only values
6123 of <a href="#t_firstclass">first class</a> type may be used as the value
6124 arguments to the PHI node. Only labels may be used as the label
6127 <p>There must be no non-phi instructions between the start of a basic block and
6128 the PHI instructions: i.e. PHI instructions must be first in a basic
6131 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
6132 occur on the edge from the corresponding predecessor block to the current
6133 block (but after any definition of an '<tt>invoke</tt>' instruction's return
6134 value on the same edge).</p>
6137 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
6138 specified by the pair corresponding to the predecessor basic block that
6139 executed just prior to the current block.</p>
6143 Loop: ; Infinite loop that counts from 0 on up...
6144 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6145 %nextindvar = add i32 %indvar, 1
6151 <!-- _______________________________________________________________________ -->
6153 <a name="i_select">'<tt>select</tt>' Instruction</a>
6160 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
6162 <i>selty</i> is either i1 or {<N x i1>}
6166 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
6167 condition, without branching.</p>
6171 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
6172 values indicating the condition, and two values of the
6173 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
6174 vectors and the condition is a scalar, then entire vectors are selected, not
6175 individual elements.</p>
6178 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
6179 first value argument; otherwise, it returns the second value argument.</p>
6181 <p>If the condition is a vector of i1, then the value arguments must be vectors
6182 of the same size, and the selection is done element by element.</p>
6186 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
6191 <!-- _______________________________________________________________________ -->
6193 <a name="i_call">'<tt>call</tt>' Instruction</a>
6200 <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>]
6204 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
6207 <p>This instruction requires several arguments:</p>
6210 <li>The optional "tail" marker indicates that the callee function does not
6211 access any allocas or varargs in the caller. Note that calls may be
6212 marked "tail" even if they do not occur before
6213 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
6214 present, the function call is eligible for tail call optimization,
6215 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
6216 optimized into a jump</a>. The code generator may optimize calls marked
6217 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
6218 sibling call optimization</a> when the caller and callee have
6219 matching signatures, or 2) forced tail call optimization when the
6220 following extra requirements are met:
6222 <li>Caller and callee both have the calling
6223 convention <tt>fastcc</tt>.</li>
6224 <li>The call is in tail position (ret immediately follows call and ret
6225 uses value of call or is void).</li>
6226 <li>Option <tt>-tailcallopt</tt> is enabled,
6227 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
6228 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
6229 constraints are met.</a></li>
6233 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
6234 convention</a> the call should use. If none is specified, the call
6235 defaults to using C calling conventions. The calling convention of the
6236 call must match the calling convention of the target function, or else the
6237 behavior is undefined.</li>
6239 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
6240 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
6241 '<tt>inreg</tt>' attributes are valid here.</li>
6243 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
6244 type of the return value. Functions that return no value are marked
6245 <tt><a href="#t_void">void</a></tt>.</li>
6247 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
6248 being invoked. The argument types must match the types implied by this
6249 signature. This type can be omitted if the function is not varargs and if
6250 the function type does not return a pointer to a function.</li>
6252 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
6253 be invoked. In most cases, this is a direct function invocation, but
6254 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
6255 to function value.</li>
6257 <li>'<tt>function args</tt>': argument list whose types match the function
6258 signature argument types and parameter attributes. All arguments must be
6259 of <a href="#t_firstclass">first class</a> type. If the function
6260 signature indicates the function accepts a variable number of arguments,
6261 the extra arguments can be specified.</li>
6263 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
6264 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
6265 '<tt>readnone</tt>' attributes are valid here.</li>
6269 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
6270 a specified function, with its incoming arguments bound to the specified
6271 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
6272 function, control flow continues with the instruction after the function
6273 call, and the return value of the function is bound to the result
6278 %retval = call i32 @test(i32 %argc)
6279 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
6280 %X = tail call i32 @foo() <i>; yields i32</i>
6281 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
6282 call void %foo(i8 97 signext)
6284 %struct.A = type { i32, i8 }
6285 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
6286 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
6287 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
6288 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
6289 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
6292 <p>llvm treats calls to some functions with names and arguments that match the
6293 standard C99 library as being the C99 library functions, and may perform
6294 optimizations or generate code for them under that assumption. This is
6295 something we'd like to change in the future to provide better support for
6296 freestanding environments and non-C-based languages.</p>
6300 <!-- _______________________________________________________________________ -->
6302 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
6309 <resultval> = va_arg <va_list*> <arglist>, <argty>
6313 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6314 the "variable argument" area of a function call. It is used to implement the
6315 <tt>va_arg</tt> macro in C.</p>
6318 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6319 argument. It returns a value of the specified argument type and increments
6320 the <tt>va_list</tt> to point to the next argument. The actual type
6321 of <tt>va_list</tt> is target specific.</p>
6324 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6325 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6326 to the next argument. For more information, see the variable argument
6327 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6329 <p>It is legal for this instruction to be called in a function which does not
6330 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6333 <p><tt>va_arg</tt> is an LLVM instruction instead of
6334 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6338 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6340 <p>Note that the code generator does not yet fully support va_arg on many
6341 targets. Also, it does not currently support va_arg with aggregate types on
6346 <!-- _______________________________________________________________________ -->
6348 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6355 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6356 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6358 <clause> := catch <type> <value>
6359 <clause> := filter <array constant type> <array constant>
6363 <p>The '<tt>landingpad</tt>' instruction is used by
6364 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6365 system</a> to specify that a basic block is a landing pad — one where
6366 the exception lands, and corresponds to the code found in the
6367 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6368 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6369 re-entry to the function. The <tt>resultval</tt> has the
6370 type <tt>resultty</tt>.</p>
6373 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6374 function associated with the unwinding mechanism. The optional
6375 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6377 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6378 or <tt>filter</tt> — and contains the global variable representing the
6379 "type" that may be caught or filtered respectively. Unlike the
6380 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6381 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6382 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6383 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6386 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6387 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6388 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6389 calling conventions, how the personality function results are represented in
6390 LLVM IR is target specific.</p>
6392 <p>The clauses are applied in order from top to bottom. If two
6393 <tt>landingpad</tt> instructions are merged together through inlining, the
6394 clauses from the calling function are appended to the list of clauses.
6395 When the call stack is being unwound due to an exception being thrown, the
6396 exception is compared against each <tt>clause</tt> in turn. If it doesn't
6397 match any of the clauses, and the <tt>cleanup</tt> flag is not set, then
6398 unwinding continues further up the call stack.</p>
6400 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6403 <li>A landing pad block is a basic block which is the unwind destination of an
6404 '<tt>invoke</tt>' instruction.</li>
6405 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6406 first non-PHI instruction.</li>
6407 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6409 <li>A basic block that is not a landing pad block may not include a
6410 '<tt>landingpad</tt>' instruction.</li>
6411 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6412 personality function.</li>
6417 ;; A landing pad which can catch an integer.
6418 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6420 ;; A landing pad that is a cleanup.
6421 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6423 ;; A landing pad which can catch an integer and can only throw a double.
6424 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6426 filter [1 x i8**] [@_ZTId]
6435 <!-- *********************************************************************** -->
6436 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6437 <!-- *********************************************************************** -->
6441 <p>LLVM supports the notion of an "intrinsic function". These functions have
6442 well known names and semantics and are required to follow certain
6443 restrictions. Overall, these intrinsics represent an extension mechanism for
6444 the LLVM language that does not require changing all of the transformations
6445 in LLVM when adding to the language (or the bitcode reader/writer, the
6446 parser, etc...).</p>
6448 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6449 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6450 begin with this prefix. Intrinsic functions must always be external
6451 functions: you cannot define the body of intrinsic functions. Intrinsic
6452 functions may only be used in call or invoke instructions: it is illegal to
6453 take the address of an intrinsic function. Additionally, because intrinsic
6454 functions are part of the LLVM language, it is required if any are added that
6455 they be documented here.</p>
6457 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6458 family of functions that perform the same operation but on different data
6459 types. Because LLVM can represent over 8 million different integer types,
6460 overloading is used commonly to allow an intrinsic function to operate on any
6461 integer type. One or more of the argument types or the result type can be
6462 overloaded to accept any integer type. Argument types may also be defined as
6463 exactly matching a previous argument's type or the result type. This allows
6464 an intrinsic function which accepts multiple arguments, but needs all of them
6465 to be of the same type, to only be overloaded with respect to a single
6466 argument or the result.</p>
6468 <p>Overloaded intrinsics will have the names of its overloaded argument types
6469 encoded into its function name, each preceded by a period. Only those types
6470 which are overloaded result in a name suffix. Arguments whose type is matched
6471 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6472 can take an integer of any width and returns an integer of exactly the same
6473 integer width. This leads to a family of functions such as
6474 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6475 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6476 suffix is required. Because the argument's type is matched against the return
6477 type, it does not require its own name suffix.</p>
6479 <p>To learn how to add an intrinsic function, please see the
6480 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6482 <!-- ======================================================================= -->
6484 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6489 <p>Variable argument support is defined in LLVM with
6490 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6491 intrinsic functions. These functions are related to the similarly named
6492 macros defined in the <tt><stdarg.h></tt> header file.</p>
6494 <p>All of these functions operate on arguments that use a target-specific value
6495 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6496 not define what this type is, so all transformations should be prepared to
6497 handle these functions regardless of the type used.</p>
6499 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6500 instruction and the variable argument handling intrinsic functions are
6503 <pre class="doc_code">
6504 define i32 @test(i32 %X, ...) {
6505 ; Initialize variable argument processing
6507 %ap2 = bitcast i8** %ap to i8*
6508 call void @llvm.va_start(i8* %ap2)
6510 ; Read a single integer argument
6511 %tmp = va_arg i8** %ap, i32
6513 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6515 %aq2 = bitcast i8** %aq to i8*
6516 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6517 call void @llvm.va_end(i8* %aq2)
6519 ; Stop processing of arguments.
6520 call void @llvm.va_end(i8* %ap2)
6524 declare void @llvm.va_start(i8*)
6525 declare void @llvm.va_copy(i8*, i8*)
6526 declare void @llvm.va_end(i8*)
6529 <!-- _______________________________________________________________________ -->
6531 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6539 declare void %llvm.va_start(i8* <arglist>)
6543 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6544 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6547 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6550 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6551 macro available in C. In a target-dependent way, it initializes
6552 the <tt>va_list</tt> element to which the argument points, so that the next
6553 call to <tt>va_arg</tt> will produce the first variable argument passed to
6554 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6555 need to know the last argument of the function as the compiler can figure
6560 <!-- _______________________________________________________________________ -->
6562 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6569 declare void @llvm.va_end(i8* <arglist>)
6573 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6574 which has been initialized previously
6575 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6576 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6579 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6582 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6583 macro available in C. In a target-dependent way, it destroys
6584 the <tt>va_list</tt> element to which the argument points. Calls
6585 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6586 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6587 with calls to <tt>llvm.va_end</tt>.</p>
6591 <!-- _______________________________________________________________________ -->
6593 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6600 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6604 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6605 from the source argument list to the destination argument list.</p>
6608 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6609 The second argument is a pointer to a <tt>va_list</tt> element to copy
6613 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6614 macro available in C. In a target-dependent way, it copies the
6615 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6616 element. This intrinsic is necessary because
6617 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6618 arbitrarily complex and require, for example, memory allocation.</p>
6624 <!-- ======================================================================= -->
6626 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6631 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6632 Collection</a> (GC) requires the implementation and generation of these
6633 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6634 roots on the stack</a>, as well as garbage collector implementations that
6635 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6636 barriers. Front-ends for type-safe garbage collected languages should generate
6637 these intrinsics to make use of the LLVM garbage collectors. For more details,
6638 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6641 <p>The garbage collection intrinsics only operate on objects in the generic
6642 address space (address space zero).</p>
6644 <!-- _______________________________________________________________________ -->
6646 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6653 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6657 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6658 the code generator, and allows some metadata to be associated with it.</p>
6661 <p>The first argument specifies the address of a stack object that contains the
6662 root pointer. The second pointer (which must be either a constant or a
6663 global value address) contains the meta-data to be associated with the
6667 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6668 location. At compile-time, the code generator generates information to allow
6669 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6670 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6675 <!-- _______________________________________________________________________ -->
6677 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6684 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6688 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6689 locations, allowing garbage collector implementations that require read
6693 <p>The second argument is the address to read from, which should be an address
6694 allocated from the garbage collector. The first object is a pointer to the
6695 start of the referenced object, if needed by the language runtime (otherwise
6699 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6700 instruction, but may be replaced with substantially more complex code by the
6701 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6702 may only be used in a function which <a href="#gc">specifies a GC
6707 <!-- _______________________________________________________________________ -->
6709 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6716 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6720 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6721 locations, allowing garbage collector implementations that require write
6722 barriers (such as generational or reference counting collectors).</p>
6725 <p>The first argument is the reference to store, the second is the start of the
6726 object to store it to, and the third is the address of the field of Obj to
6727 store to. If the runtime does not require a pointer to the object, Obj may
6731 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6732 instruction, but may be replaced with substantially more complex code by the
6733 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6734 may only be used in a function which <a href="#gc">specifies a GC
6741 <!-- ======================================================================= -->
6743 <a name="int_codegen">Code Generator Intrinsics</a>
6748 <p>These intrinsics are provided by LLVM to expose special features that may
6749 only be implemented with code generator support.</p>
6751 <!-- _______________________________________________________________________ -->
6753 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6760 declare i8 *@llvm.returnaddress(i32 <level>)
6764 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6765 target-specific value indicating the return address of the current function
6766 or one of its callers.</p>
6769 <p>The argument to this intrinsic indicates which function to return the address
6770 for. Zero indicates the calling function, one indicates its caller, etc.
6771 The argument is <b>required</b> to be a constant integer value.</p>
6774 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6775 indicating the return address of the specified call frame, or zero if it
6776 cannot be identified. The value returned by this intrinsic is likely to be
6777 incorrect or 0 for arguments other than zero, so it should only be used for
6778 debugging purposes.</p>
6780 <p>Note that calling this intrinsic does not prevent function inlining or other
6781 aggressive transformations, so the value returned may not be that of the
6782 obvious source-language caller.</p>
6786 <!-- _______________________________________________________________________ -->
6788 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6795 declare i8* @llvm.frameaddress(i32 <level>)
6799 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6800 target-specific frame pointer value for the specified stack frame.</p>
6803 <p>The argument to this intrinsic indicates which function to return the frame
6804 pointer for. Zero indicates the calling function, one indicates its caller,
6805 etc. The argument is <b>required</b> to be a constant integer value.</p>
6808 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6809 indicating the frame address of the specified call frame, or zero if it
6810 cannot be identified. The value returned by this intrinsic is likely to be
6811 incorrect or 0 for arguments other than zero, so it should only be used for
6812 debugging purposes.</p>
6814 <p>Note that calling this intrinsic does not prevent function inlining or other
6815 aggressive transformations, so the value returned may not be that of the
6816 obvious source-language caller.</p>
6820 <!-- _______________________________________________________________________ -->
6822 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6829 declare i8* @llvm.stacksave()
6833 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6834 of the function stack, for use
6835 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6836 useful for implementing language features like scoped automatic variable
6837 sized arrays in C99.</p>
6840 <p>This intrinsic returns a opaque pointer value that can be passed
6841 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6842 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6843 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6844 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6845 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6846 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6850 <!-- _______________________________________________________________________ -->
6852 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6859 declare void @llvm.stackrestore(i8* %ptr)
6863 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6864 the function stack to the state it was in when the
6865 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6866 executed. This is useful for implementing language features like scoped
6867 automatic variable sized arrays in C99.</p>
6870 <p>See the description
6871 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6875 <!-- _______________________________________________________________________ -->
6877 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6884 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6888 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6889 insert a prefetch instruction if supported; otherwise, it is a noop.
6890 Prefetches have no effect on the behavior of the program but can change its
6891 performance characteristics.</p>
6894 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6895 specifier determining if the fetch should be for a read (0) or write (1),
6896 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
6897 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6898 specifies whether the prefetch is performed on the data (1) or instruction (0)
6899 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
6900 must be constant integers.</p>
6903 <p>This intrinsic does not modify the behavior of the program. In particular,
6904 prefetches cannot trap and do not produce a value. On targets that support
6905 this intrinsic, the prefetch can provide hints to the processor cache for
6906 better performance.</p>
6910 <!-- _______________________________________________________________________ -->
6912 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
6919 declare void @llvm.pcmarker(i32 <id>)
6923 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
6924 Counter (PC) in a region of code to simulators and other tools. The method
6925 is target specific, but it is expected that the marker will use exported
6926 symbols to transmit the PC of the marker. The marker makes no guarantees
6927 that it will remain with any specific instruction after optimizations. It is
6928 possible that the presence of a marker will inhibit optimizations. The
6929 intended use is to be inserted after optimizations to allow correlations of
6930 simulation runs.</p>
6933 <p><tt>id</tt> is a numerical id identifying the marker.</p>
6936 <p>This intrinsic does not modify the behavior of the program. Backends that do
6937 not support this intrinsic may ignore it.</p>
6941 <!-- _______________________________________________________________________ -->
6943 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
6950 declare i64 @llvm.readcyclecounter()
6954 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
6955 counter register (or similar low latency, high accuracy clocks) on those
6956 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6957 should map to RPCC. As the backing counters overflow quickly (on the order
6958 of 9 seconds on alpha), this should only be used for small timings.</p>
6961 <p>When directly supported, reading the cycle counter should not modify any
6962 memory. Implementations are allowed to either return a application specific
6963 value or a system wide value. On backends without support, this is lowered
6964 to a constant 0.</p>
6970 <!-- ======================================================================= -->
6972 <a name="int_libc">Standard C Library Intrinsics</a>
6977 <p>LLVM provides intrinsics for a few important standard C library functions.
6978 These intrinsics allow source-language front-ends to pass information about
6979 the alignment of the pointer arguments to the code generator, providing
6980 opportunity for more efficient code generation.</p>
6982 <!-- _______________________________________________________________________ -->
6984 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
6990 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
6991 integer bit width and for different address spaces. Not all targets support
6992 all bit widths however.</p>
6995 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6996 i32 <len>, i32 <align>, i1 <isvolatile>)
6997 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6998 i64 <len>, i32 <align>, i1 <isvolatile>)
7002 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7003 source location to the destination location.</p>
7005 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
7006 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7007 and the pointers can be in specified address spaces.</p>
7011 <p>The first argument is a pointer to the destination, the second is a pointer
7012 to the source. The third argument is an integer argument specifying the
7013 number of bytes to copy, the fourth argument is the alignment of the
7014 source and destination locations, and the fifth is a boolean indicating a
7015 volatile access.</p>
7017 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7018 then the caller guarantees that both the source and destination pointers are
7019 aligned to that boundary.</p>
7021 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7022 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
7023 The detailed access behavior is not very cleanly specified and it is unwise
7024 to depend on it.</p>
7028 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7029 source location to the destination location, which are not allowed to
7030 overlap. It copies "len" bytes of memory over. If the argument is known to
7031 be aligned to some boundary, this can be specified as the fourth argument,
7032 otherwise it should be set to 0 or 1.</p>
7036 <!-- _______________________________________________________________________ -->
7038 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
7044 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
7045 width and for different address space. Not all targets support all bit
7049 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7050 i32 <len>, i32 <align>, i1 <isvolatile>)
7051 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7052 i64 <len>, i32 <align>, i1 <isvolatile>)
7056 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
7057 source location to the destination location. It is similar to the
7058 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
7061 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
7062 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7063 and the pointers can be in specified address spaces.</p>
7067 <p>The first argument is a pointer to the destination, the second is a pointer
7068 to the source. The third argument is an integer argument specifying the
7069 number of bytes to copy, the fourth argument is the alignment of the
7070 source and destination locations, and the fifth is a boolean indicating a
7071 volatile access.</p>
7073 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7074 then the caller guarantees that the source and destination pointers are
7075 aligned to that boundary.</p>
7077 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7078 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
7079 The detailed access behavior is not very cleanly specified and it is unwise
7080 to depend on it.</p>
7084 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
7085 source location to the destination location, which may overlap. It copies
7086 "len" bytes of memory over. If the argument is known to be aligned to some
7087 boundary, this can be specified as the fourth argument, otherwise it should
7088 be set to 0 or 1.</p>
7092 <!-- _______________________________________________________________________ -->
7094 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
7100 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
7101 width and for different address spaces. However, not all targets support all
7105 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7106 i32 <len>, i32 <align>, i1 <isvolatile>)
7107 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7108 i64 <len>, i32 <align>, i1 <isvolatile>)
7112 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
7113 particular byte value.</p>
7115 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
7116 intrinsic does not return a value and takes extra alignment/volatile
7117 arguments. Also, the destination can be in an arbitrary address space.</p>
7120 <p>The first argument is a pointer to the destination to fill, the second is the
7121 byte value with which to fill it, the third argument is an integer argument
7122 specifying the number of bytes to fill, and the fourth argument is the known
7123 alignment of the destination location.</p>
7125 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7126 then the caller guarantees that the destination pointer is aligned to that
7129 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7130 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
7131 The detailed access behavior is not very cleanly specified and it is unwise
7132 to depend on it.</p>
7135 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
7136 at the destination location. If the argument is known to be aligned to some
7137 boundary, this can be specified as the fourth argument, otherwise it should
7138 be set to 0 or 1.</p>
7142 <!-- _______________________________________________________________________ -->
7144 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
7150 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
7151 floating point or vector of floating point type. Not all targets support all
7155 declare float @llvm.sqrt.f32(float %Val)
7156 declare double @llvm.sqrt.f64(double %Val)
7157 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7158 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7159 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7163 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
7164 returning the same value as the libm '<tt>sqrt</tt>' functions would.
7165 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
7166 behavior for negative numbers other than -0.0 (which allows for better
7167 optimization, because there is no need to worry about errno being
7168 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
7171 <p>The argument and return value are floating point numbers of the same
7175 <p>This function returns the sqrt of the specified operand if it is a
7176 nonnegative floating point number.</p>
7180 <!-- _______________________________________________________________________ -->
7182 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
7188 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
7189 floating point or vector of floating point type. Not all targets support all
7193 declare float @llvm.powi.f32(float %Val, i32 %power)
7194 declare double @llvm.powi.f64(double %Val, i32 %power)
7195 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7196 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7197 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7201 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
7202 specified (positive or negative) power. The order of evaluation of
7203 multiplications is not defined. When a vector of floating point type is
7204 used, the second argument remains a scalar integer value.</p>
7207 <p>The second argument is an integer power, and the first is a value to raise to
7211 <p>This function returns the first value raised to the second power with an
7212 unspecified sequence of rounding operations.</p>
7216 <!-- _______________________________________________________________________ -->
7218 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
7224 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
7225 floating point or vector of floating point type. Not all targets support all
7229 declare float @llvm.sin.f32(float %Val)
7230 declare double @llvm.sin.f64(double %Val)
7231 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7232 declare fp128 @llvm.sin.f128(fp128 %Val)
7233 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7237 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
7240 <p>The argument and return value are floating point numbers of the same
7244 <p>This function returns the sine of the specified operand, returning the same
7245 values as the libm <tt>sin</tt> functions would, and handles error conditions
7246 in the same way.</p>
7250 <!-- _______________________________________________________________________ -->
7252 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
7258 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
7259 floating point or vector of floating point type. Not all targets support all
7263 declare float @llvm.cos.f32(float %Val)
7264 declare double @llvm.cos.f64(double %Val)
7265 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7266 declare fp128 @llvm.cos.f128(fp128 %Val)
7267 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7271 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
7274 <p>The argument and return value are floating point numbers of the same
7278 <p>This function returns the cosine of the specified operand, returning the same
7279 values as the libm <tt>cos</tt> functions would, and handles error conditions
7280 in the same way.</p>
7284 <!-- _______________________________________________________________________ -->
7286 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
7292 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
7293 floating point or vector of floating point type. Not all targets support all
7297 declare float @llvm.pow.f32(float %Val, float %Power)
7298 declare double @llvm.pow.f64(double %Val, double %Power)
7299 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7300 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7301 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7305 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
7306 specified (positive or negative) power.</p>
7309 <p>The second argument is a floating point power, and the first is a value to
7310 raise to that power.</p>
7313 <p>This function returns the first value raised to the second power, returning
7314 the same values as the libm <tt>pow</tt> functions would, and handles error
7315 conditions in the same way.</p>
7319 <!-- _______________________________________________________________________ -->
7321 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7327 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7328 floating point or vector of floating point type. Not all targets support all
7332 declare float @llvm.exp.f32(float %Val)
7333 declare double @llvm.exp.f64(double %Val)
7334 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7335 declare fp128 @llvm.exp.f128(fp128 %Val)
7336 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7340 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7343 <p>The argument and return value are floating point numbers of the same
7347 <p>This function returns the same values as the libm <tt>exp</tt> functions
7348 would, and handles error conditions in the same way.</p>
7352 <!-- _______________________________________________________________________ -->
7354 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7360 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7361 floating point or vector of floating point type. Not all targets support all
7365 declare float @llvm.log.f32(float %Val)
7366 declare double @llvm.log.f64(double %Val)
7367 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7368 declare fp128 @llvm.log.f128(fp128 %Val)
7369 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7373 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7376 <p>The argument and return value are floating point numbers of the same
7380 <p>This function returns the same values as the libm <tt>log</tt> functions
7381 would, and handles error conditions in the same way.</p>
7385 <!-- _______________________________________________________________________ -->
7387 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7393 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7394 floating point or vector of floating point type. Not all targets support all
7398 declare float @llvm.fma.f32(float %a, float %b, float %c)
7399 declare double @llvm.fma.f64(double %a, double %b, double %c)
7400 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7401 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7402 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7406 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7410 <p>The argument and return value are floating point numbers of the same
7414 <p>This function returns the same values as the libm <tt>fma</tt> functions
7421 <!-- ======================================================================= -->
7423 <a name="int_manip">Bit Manipulation Intrinsics</a>
7428 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7429 These allow efficient code generation for some algorithms.</p>
7431 <!-- _______________________________________________________________________ -->
7433 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7439 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7440 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7443 declare i16 @llvm.bswap.i16(i16 <id>)
7444 declare i32 @llvm.bswap.i32(i32 <id>)
7445 declare i64 @llvm.bswap.i64(i64 <id>)
7449 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7450 values with an even number of bytes (positive multiple of 16 bits). These
7451 are useful for performing operations on data that is not in the target's
7452 native byte order.</p>
7455 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7456 and low byte of the input i16 swapped. Similarly,
7457 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7458 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7459 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7460 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7461 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7462 more, respectively).</p>
7466 <!-- _______________________________________________________________________ -->
7468 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7474 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7475 width, or on any vector with integer elements. Not all targets support all
7476 bit widths or vector types, however.</p>
7479 declare i8 @llvm.ctpop.i8(i8 <src>)
7480 declare i16 @llvm.ctpop.i16(i16 <src>)
7481 declare i32 @llvm.ctpop.i32(i32 <src>)
7482 declare i64 @llvm.ctpop.i64(i64 <src>)
7483 declare i256 @llvm.ctpop.i256(i256 <src>)
7484 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7488 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7492 <p>The only argument is the value to be counted. The argument may be of any
7493 integer type, or a vector with integer elements.
7494 The return type must match the argument type.</p>
7497 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7498 element of a vector.</p>
7502 <!-- _______________________________________________________________________ -->
7504 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7510 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7511 integer bit width, or any vector whose elements are integers. Not all
7512 targets support all bit widths or vector types, however.</p>
7515 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7516 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7517 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7518 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7519 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7520 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7524 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7525 leading zeros in a variable.</p>
7528 <p>The first argument is the value to be counted. This argument may be of any
7529 integer type, or a vectory with integer element type. The return type
7530 must match the first argument type.</p>
7532 <p>The second argument must be a constant and is a flag to indicate whether the
7533 intrinsic should ensure that a zero as the first argument produces a defined
7534 result. Historically some architectures did not provide a defined result for
7535 zero values as efficiently, and many algorithms are now predicated on
7536 avoiding zero-value inputs.</p>
7539 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7540 zeros in a variable, or within each element of the vector.
7541 If <tt>src == 0</tt> then the result is the size in bits of the type of
7542 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7543 For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7547 <!-- _______________________________________________________________________ -->
7549 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7555 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7556 integer bit width, or any vector of integer elements. Not all targets
7557 support all bit widths or vector types, however.</p>
7560 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7561 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7562 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7563 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7564 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7565 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7569 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7573 <p>The first argument is the value to be counted. This argument may be of any
7574 integer type, or a vectory with integer element type. The return type
7575 must match the first argument type.</p>
7577 <p>The second argument must be a constant and is a flag to indicate whether the
7578 intrinsic should ensure that a zero as the first argument produces a defined
7579 result. Historically some architectures did not provide a defined result for
7580 zero values as efficiently, and many algorithms are now predicated on
7581 avoiding zero-value inputs.</p>
7584 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7585 zeros in a variable, or within each element of a vector.
7586 If <tt>src == 0</tt> then the result is the size in bits of the type of
7587 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7588 For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7594 <!-- ======================================================================= -->
7596 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7601 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7603 <!-- _______________________________________________________________________ -->
7605 <a name="int_sadd_overflow">
7606 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7613 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7614 on any integer bit width.</p>
7617 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7618 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7619 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7623 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7624 a signed addition of the two arguments, and indicate whether an overflow
7625 occurred during the signed summation.</p>
7628 <p>The arguments (%a and %b) and the first element of the result structure may
7629 be of integer types of any bit width, but they must have the same bit
7630 width. The second element of the result structure must be of
7631 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7632 undergo signed addition.</p>
7635 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7636 a signed addition of the two variables. They return a structure — the
7637 first element of which is the signed summation, and the second element of
7638 which is a bit specifying if the signed summation resulted in an
7643 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7644 %sum = extractvalue {i32, i1} %res, 0
7645 %obit = extractvalue {i32, i1} %res, 1
7646 br i1 %obit, label %overflow, label %normal
7651 <!-- _______________________________________________________________________ -->
7653 <a name="int_uadd_overflow">
7654 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7661 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7662 on any integer bit width.</p>
7665 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7666 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7667 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7671 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7672 an unsigned addition of the two arguments, and indicate whether a carry
7673 occurred during the unsigned summation.</p>
7676 <p>The arguments (%a and %b) and the first element of the result structure may
7677 be of integer types of any bit width, but they must have the same bit
7678 width. The second element of the result structure must be of
7679 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7680 undergo unsigned addition.</p>
7683 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7684 an unsigned addition of the two arguments. They return a structure —
7685 the first element of which is the sum, and the second element of which is a
7686 bit specifying if the unsigned summation resulted in a carry.</p>
7690 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7691 %sum = extractvalue {i32, i1} %res, 0
7692 %obit = extractvalue {i32, i1} %res, 1
7693 br i1 %obit, label %carry, label %normal
7698 <!-- _______________________________________________________________________ -->
7700 <a name="int_ssub_overflow">
7701 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7708 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7709 on any integer bit width.</p>
7712 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7713 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7714 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7718 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7719 a signed subtraction of the two arguments, and indicate whether an overflow
7720 occurred during the signed subtraction.</p>
7723 <p>The arguments (%a and %b) and the first element of the result structure may
7724 be of integer types of any bit width, but they must have the same bit
7725 width. The second element of the result structure must be of
7726 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7727 undergo signed subtraction.</p>
7730 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7731 a signed subtraction of the two arguments. They return a structure —
7732 the first element of which is the subtraction, and the second element of
7733 which is a bit specifying if the signed subtraction resulted in an
7738 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7739 %sum = extractvalue {i32, i1} %res, 0
7740 %obit = extractvalue {i32, i1} %res, 1
7741 br i1 %obit, label %overflow, label %normal
7746 <!-- _______________________________________________________________________ -->
7748 <a name="int_usub_overflow">
7749 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7756 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7757 on any integer bit width.</p>
7760 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7761 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7762 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7766 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7767 an unsigned subtraction of the two arguments, and indicate whether an
7768 overflow occurred during the unsigned subtraction.</p>
7771 <p>The arguments (%a and %b) and the first element of the result structure may
7772 be of integer types of any bit width, but they must have the same bit
7773 width. The second element of the result structure must be of
7774 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7775 undergo unsigned subtraction.</p>
7778 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7779 an unsigned subtraction of the two arguments. They return a structure —
7780 the first element of which is the subtraction, and the second element of
7781 which is a bit specifying if the unsigned subtraction resulted in an
7786 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7787 %sum = extractvalue {i32, i1} %res, 0
7788 %obit = extractvalue {i32, i1} %res, 1
7789 br i1 %obit, label %overflow, label %normal
7794 <!-- _______________________________________________________________________ -->
7796 <a name="int_smul_overflow">
7797 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7804 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7805 on any integer bit width.</p>
7808 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7809 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7810 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7815 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7816 a signed multiplication of the two arguments, and indicate whether an
7817 overflow occurred during the signed multiplication.</p>
7820 <p>The arguments (%a and %b) and the first element of the result structure may
7821 be of integer types of any bit width, but they must have the same bit
7822 width. The second element of the result structure must be of
7823 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7824 undergo signed multiplication.</p>
7827 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7828 a signed multiplication of the two arguments. They return a structure —
7829 the first element of which is the multiplication, and the second element of
7830 which is a bit specifying if the signed multiplication resulted in an
7835 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7836 %sum = extractvalue {i32, i1} %res, 0
7837 %obit = extractvalue {i32, i1} %res, 1
7838 br i1 %obit, label %overflow, label %normal
7843 <!-- _______________________________________________________________________ -->
7845 <a name="int_umul_overflow">
7846 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
7853 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
7854 on any integer bit width.</p>
7857 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7858 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7859 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7863 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7864 a unsigned multiplication of the two arguments, and indicate whether an
7865 overflow occurred during the unsigned multiplication.</p>
7868 <p>The arguments (%a and %b) and the first element of the result structure may
7869 be of integer types of any bit width, but they must have the same bit
7870 width. The second element of the result structure must be of
7871 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7872 undergo unsigned multiplication.</p>
7875 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7876 an unsigned multiplication of the two arguments. They return a structure
7877 — the first element of which is the multiplication, and the second
7878 element of which is a bit specifying if the unsigned multiplication resulted
7883 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7884 %sum = extractvalue {i32, i1} %res, 0
7885 %obit = extractvalue {i32, i1} %res, 1
7886 br i1 %obit, label %overflow, label %normal
7893 <!-- ======================================================================= -->
7895 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
7900 <p>Half precision floating point is a storage-only format. This means that it is
7901 a dense encoding (in memory) but does not support computation in the
7904 <p>This means that code must first load the half-precision floating point
7905 value as an i16, then convert it to float with <a
7906 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
7907 Computation can then be performed on the float value (including extending to
7908 double etc). To store the value back to memory, it is first converted to
7909 float if needed, then converted to i16 with
7910 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
7911 storing as an i16 value.</p>
7913 <!-- _______________________________________________________________________ -->
7915 <a name="int_convert_to_fp16">
7916 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
7924 declare i16 @llvm.convert.to.fp16(f32 %a)
7928 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7929 a conversion from single precision floating point format to half precision
7930 floating point format.</p>
7933 <p>The intrinsic function contains single argument - the value to be
7937 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7938 a conversion from single precision floating point format to half precision
7939 floating point format. The return value is an <tt>i16</tt> which
7940 contains the converted number.</p>
7944 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7945 store i16 %res, i16* @x, align 2
7950 <!-- _______________________________________________________________________ -->
7952 <a name="int_convert_from_fp16">
7953 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
7961 declare f32 @llvm.convert.from.fp16(i16 %a)
7965 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
7966 a conversion from half precision floating point format to single precision
7967 floating point format.</p>
7970 <p>The intrinsic function contains single argument - the value to be
7974 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
7975 conversion from half single precision floating point format to single
7976 precision floating point format. The input half-float value is represented by
7977 an <tt>i16</tt> value.</p>
7981 %a = load i16* @x, align 2
7982 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7989 <!-- ======================================================================= -->
7991 <a name="int_debugger">Debugger Intrinsics</a>
7996 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
7997 prefix), are described in
7998 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
7999 Level Debugging</a> document.</p>
8003 <!-- ======================================================================= -->
8005 <a name="int_eh">Exception Handling Intrinsics</a>
8010 <p>The LLVM exception handling intrinsics (which all start with
8011 <tt>llvm.eh.</tt> prefix), are described in
8012 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
8013 Handling</a> document.</p>
8017 <!-- ======================================================================= -->
8019 <a name="int_trampoline">Trampoline Intrinsics</a>
8024 <p>These intrinsics make it possible to excise one parameter, marked with
8025 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
8026 The result is a callable
8027 function pointer lacking the nest parameter - the caller does not need to
8028 provide a value for it. Instead, the value to use is stored in advance in a
8029 "trampoline", a block of memory usually allocated on the stack, which also
8030 contains code to splice the nest value into the argument list. This is used
8031 to implement the GCC nested function address extension.</p>
8033 <p>For example, if the function is
8034 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
8035 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
8038 <pre class="doc_code">
8039 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8040 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8041 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8042 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8043 %fp = bitcast i8* %p to i32 (i32, i32)*
8046 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
8047 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
8049 <!-- _______________________________________________________________________ -->
8052 '<tt>llvm.init.trampoline</tt>' Intrinsic
8060 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8064 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
8065 turning it into a trampoline.</p>
8068 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
8069 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
8070 sufficiently aligned block of memory; this memory is written to by the
8071 intrinsic. Note that the size and the alignment are target-specific - LLVM
8072 currently provides no portable way of determining them, so a front-end that
8073 generates this intrinsic needs to have some target-specific knowledge.
8074 The <tt>func</tt> argument must hold a function bitcast to
8075 an <tt>i8*</tt>.</p>
8078 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
8079 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
8080 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
8081 which can be <a href="#int_trampoline">bitcast (to a new function) and
8082 called</a>. The new function's signature is the same as that of
8083 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
8084 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
8085 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
8086 with the same argument list, but with <tt>nval</tt> used for the missing
8087 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
8088 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
8089 to the returned function pointer is undefined.</p>
8092 <!-- _______________________________________________________________________ -->
8095 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
8103 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8107 <p>This performs any required machine-specific adjustment to the address of a
8108 trampoline (passed as <tt>tramp</tt>).</p>
8111 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
8112 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
8116 <p>On some architectures the address of the code to be executed needs to be
8117 different to the address where the trampoline is actually stored. This
8118 intrinsic returns the executable address corresponding to <tt>tramp</tt>
8119 after performing the required machine specific adjustments.
8120 The pointer returned can then be <a href="#int_trampoline"> bitcast and
8128 <!-- ======================================================================= -->
8130 <a name="int_memorymarkers">Memory Use Markers</a>
8135 <p>This class of intrinsics exists to information about the lifetime of memory
8136 objects and ranges where variables are immutable.</p>
8138 <!-- _______________________________________________________________________ -->
8140 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8147 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8151 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8152 object's lifetime.</p>
8155 <p>The first argument is a constant integer representing the size of the
8156 object, or -1 if it is variable sized. The second argument is a pointer to
8160 <p>This intrinsic indicates that before this point in the code, the value of the
8161 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8162 never be used and has an undefined value. A load from the pointer that
8163 precedes this intrinsic can be replaced with
8164 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8168 <!-- _______________________________________________________________________ -->
8170 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8177 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8181 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8182 object's lifetime.</p>
8185 <p>The first argument is a constant integer representing the size of the
8186 object, or -1 if it is variable sized. The second argument is a pointer to
8190 <p>This intrinsic indicates that after this point in the code, the value of the
8191 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8192 never be used and has an undefined value. Any stores into the memory object
8193 following this intrinsic may be removed as dead.
8197 <!-- _______________________________________________________________________ -->
8199 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8206 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8210 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8211 a memory object will not change.</p>
8214 <p>The first argument is a constant integer representing the size of the
8215 object, or -1 if it is variable sized. The second argument is a pointer to
8219 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8220 the return value, the referenced memory location is constant and
8225 <!-- _______________________________________________________________________ -->
8227 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8234 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8238 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8239 a memory object are mutable.</p>
8242 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8243 The second argument is a constant integer representing the size of the
8244 object, or -1 if it is variable sized and the third argument is a pointer
8248 <p>This intrinsic indicates that the memory is mutable again.</p>
8254 <!-- ======================================================================= -->
8256 <a name="int_general">General Intrinsics</a>
8261 <p>This class of intrinsics is designed to be generic and has no specific
8264 <!-- _______________________________________________________________________ -->
8266 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8273 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8277 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8280 <p>The first argument is a pointer to a value, the second is a pointer to a
8281 global string, the third is a pointer to a global string which is the source
8282 file name, and the last argument is the line number.</p>
8285 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8286 This can be useful for special purpose optimizations that want to look for
8287 these annotations. These have no other defined use; they are ignored by code
8288 generation and optimization.</p>
8292 <!-- _______________________________________________________________________ -->
8294 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8300 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8301 any integer bit width.</p>
8304 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8305 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8306 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8307 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8308 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8312 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8315 <p>The first argument is an integer value (result of some expression), the
8316 second is a pointer to a global string, the third is a pointer to a global
8317 string which is the source file name, and the last argument is the line
8318 number. It returns the value of the first argument.</p>
8321 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8322 arbitrary strings. This can be useful for special purpose optimizations that
8323 want to look for these annotations. These have no other defined use; they
8324 are ignored by code generation and optimization.</p>
8328 <!-- _______________________________________________________________________ -->
8330 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8337 declare void @llvm.trap()
8341 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8347 <p>This intrinsics is lowered to the target dependent trap instruction. If the
8348 target does not have a trap instruction, this intrinsic will be lowered to
8349 the call of the <tt>abort()</tt> function.</p>
8353 <!-- _______________________________________________________________________ -->
8355 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8362 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8366 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8367 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8368 ensure that it is placed on the stack before local variables.</p>
8371 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8372 arguments. The first argument is the value loaded from the stack
8373 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8374 that has enough space to hold the value of the guard.</p>
8377 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8378 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8379 stack. This is to ensure that if a local variable on the stack is
8380 overwritten, it will destroy the value of the guard. When the function exits,
8381 the guard on the stack is checked against the original guard. If they are
8382 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8387 <!-- _______________________________________________________________________ -->
8389 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8396 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <type>)
8397 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <type>)
8401 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8402 the optimizers to determine at compile time whether a) an operation (like
8403 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8404 runtime check for overflow isn't necessary. An object in this context means
8405 an allocation of a specific class, structure, array, or other object.</p>
8408 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8409 argument is a pointer to or into the <tt>object</tt>. The second argument
8410 is a boolean 0 or 1. This argument determines whether you want the
8411 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
8412 1, variables are not allowed.</p>
8415 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
8416 representing the size of the object concerned, or <tt>i32/i64 -1 or 0</tt>,
8417 depending on the <tt>type</tt> argument, if the size cannot be determined at
8421 <!-- _______________________________________________________________________ -->
8423 <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a>
8430 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8431 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8435 <p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the
8436 most probable) value of <tt>val</tt>, which can be used by optimizers.</p>
8439 <p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first
8440 argument is a value. The second argument is an expected value, this needs to
8441 be a constant value, variables are not allowed.</p>
8444 <p>This intrinsic is lowered to the <tt>val</tt>.</p>
8450 <!-- *********************************************************************** -->
8453 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8454 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8455 <a href="http://validator.w3.org/check/referer"><img
8456 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8458 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8459 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8460 Last modified: $Date$