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>
107 <li><a href="#range">'<tt>range</tt>' Metadata</a></li>
112 <li><a href="#module_flags">Module Flags Metadata</a>
114 <li><a href="#objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a></li>
117 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
119 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
120 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
121 Global Variable</a></li>
122 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
123 Global Variable</a></li>
124 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
125 Global Variable</a></li>
128 <li><a href="#instref">Instruction Reference</a>
130 <li><a href="#terminators">Terminator Instructions</a>
132 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
133 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
134 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
135 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
136 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
137 <li><a href="#i_resume">'<tt>resume</tt>' Instruction</a></li>
138 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
141 <li><a href="#binaryops">Binary Operations</a>
143 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
144 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
145 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
146 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
147 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
148 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
149 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
150 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
151 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
152 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
153 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
154 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
157 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
159 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
160 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
161 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
162 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
163 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
164 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
167 <li><a href="#vectorops">Vector Operations</a>
169 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
170 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
171 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
174 <li><a href="#aggregateops">Aggregate Operations</a>
176 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
177 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
180 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
182 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
183 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
184 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
185 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
186 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
187 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
188 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
191 <li><a href="#convertops">Conversion Operations</a>
193 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
194 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
195 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
196 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
197 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
198 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
199 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
200 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
201 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
202 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
203 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
204 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
207 <li><a href="#otherops">Other Operations</a>
209 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
210 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
211 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
212 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
213 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
214 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
215 <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
220 <li><a href="#intrinsics">Intrinsic Functions</a>
222 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
224 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
225 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
226 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
229 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
231 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
232 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
233 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
236 <li><a href="#int_codegen">Code Generator Intrinsics</a>
238 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
239 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
240 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
241 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
242 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
243 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
244 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
247 <li><a href="#int_libc">Standard C Library Intrinsics</a>
249 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
250 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
251 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
252 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
253 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
254 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
255 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
256 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
257 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
258 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
259 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
262 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
264 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
265 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
266 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
267 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
270 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
272 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
273 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
274 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
275 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
276 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
277 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
280 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
282 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
283 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
286 <li><a href="#int_debugger">Debugger intrinsics</a></li>
287 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
288 <li><a href="#int_trampoline">Trampoline Intrinsics</a>
290 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
291 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
294 <li><a href="#int_memorymarkers">Memory Use Markers</a>
296 <li><a href="#int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a></li>
297 <li><a href="#int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a></li>
298 <li><a href="#int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a></li>
299 <li><a href="#int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a></li>
302 <li><a href="#int_general">General intrinsics</a>
304 <li><a href="#int_var_annotation">
305 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
306 <li><a href="#int_annotation">
307 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
308 <li><a href="#int_trap">
309 '<tt>llvm.trap</tt>' Intrinsic</a></li>
310 <li><a href="#int_stackprotector">
311 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
312 <li><a href="#int_objectsize">
313 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
314 <li><a href="#int_expect">
315 '<tt>llvm.expect</tt>' Intrinsic</a></li>
322 <div class="doc_author">
323 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
324 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
327 <!-- *********************************************************************** -->
328 <h2><a name="abstract">Abstract</a></h2>
329 <!-- *********************************************************************** -->
333 <p>This document is a reference manual for the LLVM assembly language. LLVM is
334 a Static Single Assignment (SSA) based representation that provides type
335 safety, low-level operations, flexibility, and the capability of representing
336 'all' high-level languages cleanly. It is the common code representation
337 used throughout all phases of the LLVM compilation strategy.</p>
341 <!-- *********************************************************************** -->
342 <h2><a name="introduction">Introduction</a></h2>
343 <!-- *********************************************************************** -->
347 <p>The LLVM code representation is designed to be used in three different forms:
348 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
349 for fast loading by a Just-In-Time compiler), and as a human readable
350 assembly language representation. This allows LLVM to provide a powerful
351 intermediate representation for efficient compiler transformations and
352 analysis, while providing a natural means to debug and visualize the
353 transformations. The three different forms of LLVM are all equivalent. This
354 document describes the human readable representation and notation.</p>
356 <p>The LLVM representation aims to be light-weight and low-level while being
357 expressive, typed, and extensible at the same time. It aims to be a
358 "universal IR" of sorts, by being at a low enough level that high-level ideas
359 may be cleanly mapped to it (similar to how microprocessors are "universal
360 IR's", allowing many source languages to be mapped to them). By providing
361 type information, LLVM can be used as the target of optimizations: for
362 example, through pointer analysis, it can be proven that a C automatic
363 variable is never accessed outside of the current function, allowing it to
364 be promoted to a simple SSA value instead of a memory location.</p>
366 <!-- _______________________________________________________________________ -->
368 <a name="wellformed">Well-Formedness</a>
373 <p>It is important to note that this document describes 'well formed' LLVM
374 assembly language. There is a difference between what the parser accepts and
375 what is considered 'well formed'. For example, the following instruction is
376 syntactically okay, but not well formed:</p>
378 <pre class="doc_code">
379 %x = <a href="#i_add">add</a> i32 1, %x
382 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
383 LLVM infrastructure provides a verification pass that may be used to verify
384 that an LLVM module is well formed. This pass is automatically run by the
385 parser after parsing input assembly and by the optimizer before it outputs
386 bitcode. The violations pointed out by the verifier pass indicate bugs in
387 transformation passes or input to the parser.</p>
393 <!-- Describe the typesetting conventions here. -->
395 <!-- *********************************************************************** -->
396 <h2><a name="identifiers">Identifiers</a></h2>
397 <!-- *********************************************************************** -->
401 <p>LLVM identifiers come in two basic types: global and local. Global
402 identifiers (functions, global variables) begin with the <tt>'@'</tt>
403 character. Local identifiers (register names, types) begin with
404 the <tt>'%'</tt> character. Additionally, there are three different formats
405 for identifiers, for different purposes:</p>
408 <li>Named values are represented as a string of characters with their prefix.
409 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
410 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
411 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
412 other characters in their names can be surrounded with quotes. Special
413 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
414 ASCII code for the character in hexadecimal. In this way, any character
415 can be used in a name value, even quotes themselves.</li>
417 <li>Unnamed values are represented as an unsigned numeric value with their
418 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
420 <li>Constants, which are described in a <a href="#constants">section about
421 constants</a>, below.</li>
424 <p>LLVM requires that values start with a prefix for two reasons: Compilers
425 don't need to worry about name clashes with reserved words, and the set of
426 reserved words may be expanded in the future without penalty. Additionally,
427 unnamed identifiers allow a compiler to quickly come up with a temporary
428 variable without having to avoid symbol table conflicts.</p>
430 <p>Reserved words in LLVM are very similar to reserved words in other
431 languages. There are keywords for different opcodes
432 ('<tt><a href="#i_add">add</a></tt>',
433 '<tt><a href="#i_bitcast">bitcast</a></tt>',
434 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
435 ('<tt><a href="#t_void">void</a></tt>',
436 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
437 reserved words cannot conflict with variable names, because none of them
438 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
440 <p>Here is an example of LLVM code to multiply the integer variable
441 '<tt>%X</tt>' by 8:</p>
445 <pre class="doc_code">
446 %result = <a href="#i_mul">mul</a> i32 %X, 8
449 <p>After strength reduction:</p>
451 <pre class="doc_code">
452 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
455 <p>And the hard way:</p>
457 <pre class="doc_code">
458 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
459 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
460 %result = <a href="#i_add">add</a> i32 %1, %1
463 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
464 lexical features of LLVM:</p>
467 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
470 <li>Unnamed temporaries are created when the result of a computation is not
471 assigned to a named value.</li>
473 <li>Unnamed temporaries are numbered sequentially</li>
476 <p>It also shows a convention that we follow in this document. When
477 demonstrating instructions, we will follow an instruction with a comment that
478 defines the type and name of value produced. Comments are shown in italic
483 <!-- *********************************************************************** -->
484 <h2><a name="highlevel">High Level Structure</a></h2>
485 <!-- *********************************************************************** -->
487 <!-- ======================================================================= -->
489 <a name="modulestructure">Module Structure</a>
494 <p>LLVM programs are composed of <tt>Module</tt>s, each of which is a
495 translation unit of the input programs. Each module consists of functions,
496 global variables, and symbol table entries. Modules may be combined together
497 with the LLVM linker, which merges function (and global variable)
498 definitions, resolves forward declarations, and merges symbol table
499 entries. Here is an example of the "hello world" module:</p>
501 <pre class="doc_code">
502 <i>; Declare the string constant as a global constant.</i>
503 <a href="#identifiers">@.str</a> = <a href="#linkage_private">private</a> <a href="#globalvars">unnamed_addr</a> <a href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00"
505 <i>; External declaration of the puts function</i>
506 <a href="#functionstructure">declare</a> i32 @puts(i8* <a href="#nocapture">nocapture</a>) <a href="#fnattrs">nounwind</a>
508 <i>; Definition of main function</i>
509 define i32 @main() { <i>; i32()* </i>
510 <i>; Convert [13 x i8]* to i8 *...</i>
511 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.str, i64 0, i64 0
513 <i>; Call puts function to write out the string to stdout.</i>
514 <a href="#i_call">call</a> i32 @puts(i8* %cast210)
515 <a href="#i_ret">ret</a> i32 0
518 <i>; Named metadata</i>
519 !1 = metadata !{i32 42}
523 <p>This example is made up of a <a href="#globalvars">global variable</a> named
524 "<tt>.str</tt>", an external declaration of the "<tt>puts</tt>" function,
525 a <a href="#functionstructure">function definition</a> for
526 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
529 <p>In general, a module is made up of a list of global values (where both
530 functions and global variables are global values). Global values are
531 represented by a pointer to a memory location (in this case, a pointer to an
532 array of char, and a pointer to a function), and have one of the
533 following <a href="#linkage">linkage types</a>.</p>
537 <!-- ======================================================================= -->
539 <a name="linkage">Linkage Types</a>
544 <p>All Global Variables and Functions have one of the following types of
548 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
549 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
550 by objects in the current module. In particular, linking code into a
551 module with an private global value may cause the private to be renamed as
552 necessary to avoid collisions. Because the symbol is private to the
553 module, all references can be updated. This doesn't show up in any symbol
554 table in the object file.</dd>
556 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
557 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
558 assembler and evaluated by the linker. Unlike normal strong symbols, they
559 are removed by the linker from the final linked image (executable or
560 dynamic library).</dd>
562 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
563 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
564 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
565 linker. The symbols are removed by the linker from the final linked image
566 (executable or dynamic library).</dd>
568 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
569 <dd>Similar to "<tt>linker_private_weak</tt>", but it's known that the address
570 of the object is not taken. For instance, functions that had an inline
571 definition, but the compiler decided not to inline it. Note,
572 unlike <tt>linker_private</tt> and <tt>linker_private_weak</tt>,
573 <tt>linker_private_weak_def_auto</tt> may have only <tt>default</tt>
574 visibility. The symbols are removed by the linker from the final linked
575 image (executable or dynamic library).</dd>
577 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
578 <dd>Similar to private, but the value shows as a local symbol
579 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
580 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
582 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
583 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
584 into the object file corresponding to the LLVM module. They exist to
585 allow inlining and other optimizations to take place given knowledge of
586 the definition of the global, which is known to be somewhere outside the
587 module. Globals with <tt>available_externally</tt> linkage are allowed to
588 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
589 This linkage type is only allowed on definitions, not declarations.</dd>
591 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
592 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
593 the same name when linkage occurs. This can be used to implement
594 some forms of inline functions, templates, or other code which must be
595 generated in each translation unit that uses it, but where the body may
596 be overridden with a more definitive definition later. Unreferenced
597 <tt>linkonce</tt> globals are allowed to be discarded. Note that
598 <tt>linkonce</tt> linkage does not actually allow the optimizer to
599 inline the body of this function into callers because it doesn't know if
600 this definition of the function is the definitive definition within the
601 program or whether it will be overridden by a stronger definition.
602 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
605 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
606 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
607 <tt>linkonce</tt> linkage, except that unreferenced globals with
608 <tt>weak</tt> linkage may not be discarded. This is used for globals that
609 are declared "weak" in C source code.</dd>
611 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
612 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
613 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
615 Symbols with "<tt>common</tt>" linkage are merged in the same way as
616 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
617 <tt>common</tt> symbols may not have an explicit section,
618 must have a zero initializer, and may not be marked '<a
619 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
620 have common linkage.</dd>
623 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
624 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
625 pointer to array type. When two global variables with appending linkage
626 are linked together, the two global arrays are appended together. This is
627 the LLVM, typesafe, equivalent of having the system linker append together
628 "sections" with identical names when .o files are linked.</dd>
630 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
631 <dd>The semantics of this linkage follow the ELF object file model: the symbol
632 is weak until linked, if not linked, the symbol becomes null instead of
633 being an undefined reference.</dd>
635 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
636 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
637 <dd>Some languages allow differing globals to be merged, such as two functions
638 with different semantics. Other languages, such as <tt>C++</tt>, ensure
639 that only equivalent globals are ever merged (the "one definition rule"
640 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
641 and <tt>weak_odr</tt> linkage types to indicate that the global will only
642 be merged with equivalent globals. These linkage types are otherwise the
643 same as their non-<tt>odr</tt> versions.</dd>
645 <dt><tt><b><a name="linkage_external">external</a></b></tt></dt>
646 <dd>If none of the above identifiers are used, the global is externally
647 visible, meaning that it participates in linkage and can be used to
648 resolve external symbol references.</dd>
651 <p>The next two types of linkage are targeted for Microsoft Windows platform
652 only. They are designed to support importing (exporting) symbols from (to)
653 DLLs (Dynamic Link Libraries).</p>
656 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
657 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
658 or variable via a global pointer to a pointer that is set up by the DLL
659 exporting the symbol. On Microsoft Windows targets, the pointer name is
660 formed by combining <code>__imp_</code> and the function or variable
663 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
664 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
665 pointer to a pointer in a DLL, so that it can be referenced with the
666 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
667 name is formed by combining <code>__imp_</code> and the function or
671 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
672 another module defined a "<tt>.LC0</tt>" variable and was linked with this
673 one, one of the two would be renamed, preventing a collision. Since
674 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
675 declarations), they are accessible outside of the current module.</p>
677 <p>It is illegal for a function <i>declaration</i> to have any linkage type
678 other than <tt>external</tt>, <tt>dllimport</tt>
679 or <tt>extern_weak</tt>.</p>
681 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
682 or <tt>weak_odr</tt> linkages.</p>
686 <!-- ======================================================================= -->
688 <a name="callingconv">Calling Conventions</a>
693 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
694 and <a href="#i_invoke">invokes</a> can all have an optional calling
695 convention specified for the call. The calling convention of any pair of
696 dynamic caller/callee must match, or the behavior of the program is
697 undefined. The following calling conventions are supported by LLVM, and more
698 may be added in the future:</p>
701 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
702 <dd>This calling convention (the default if no other calling convention is
703 specified) matches the target C calling conventions. This calling
704 convention supports varargs function calls and tolerates some mismatch in
705 the declared prototype and implemented declaration of the function (as
708 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
709 <dd>This calling convention attempts to make calls as fast as possible
710 (e.g. by passing things in registers). This calling convention allows the
711 target to use whatever tricks it wants to produce fast code for the
712 target, without having to conform to an externally specified ABI
713 (Application Binary Interface).
714 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
715 when this or the GHC convention is used.</a> This calling convention
716 does not support varargs and requires the prototype of all callees to
717 exactly match the prototype of the function definition.</dd>
719 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
720 <dd>This calling convention attempts to make code in the caller as efficient
721 as possible under the assumption that the call is not commonly executed.
722 As such, these calls often preserve all registers so that the call does
723 not break any live ranges in the caller side. This calling convention
724 does not support varargs and requires the prototype of all callees to
725 exactly match the prototype of the function definition.</dd>
727 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
728 <dd>This calling convention has been implemented specifically for use by the
729 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
730 It passes everything in registers, going to extremes to achieve this by
731 disabling callee save registers. This calling convention should not be
732 used lightly but only for specific situations such as an alternative to
733 the <em>register pinning</em> performance technique often used when
734 implementing functional programming languages.At the moment only X86
735 supports this convention and it has the following limitations:
737 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
738 floating point types are supported.</li>
739 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
740 6 floating point parameters.</li>
742 This calling convention supports
743 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
744 requires both the caller and callee are using it.
747 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
748 <dd>Any calling convention may be specified by number, allowing
749 target-specific calling conventions to be used. Target specific calling
750 conventions start at 64.</dd>
753 <p>More calling conventions can be added/defined on an as-needed basis, to
754 support Pascal conventions or any other well-known target-independent
759 <!-- ======================================================================= -->
761 <a name="visibility">Visibility Styles</a>
766 <p>All Global Variables and Functions have one of the following visibility
770 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
771 <dd>On targets that use the ELF object file format, default visibility means
772 that the declaration is visible to other modules and, in shared libraries,
773 means that the declared entity may be overridden. On Darwin, default
774 visibility means that the declaration is visible to other modules. Default
775 visibility corresponds to "external linkage" in the language.</dd>
777 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
778 <dd>Two declarations of an object with hidden visibility refer to the same
779 object if they are in the same shared object. Usually, hidden visibility
780 indicates that the symbol will not be placed into the dynamic symbol
781 table, so no other module (executable or shared library) can reference it
784 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
785 <dd>On ELF, protected visibility indicates that the symbol will be placed in
786 the dynamic symbol table, but that references within the defining module
787 will bind to the local symbol. That is, the symbol cannot be overridden by
793 <!-- ======================================================================= -->
795 <a name="namedtypes">Named Types</a>
800 <p>LLVM IR allows you to specify name aliases for certain types. This can make
801 it easier to read the IR and make the IR more condensed (particularly when
802 recursive types are involved). An example of a name specification is:</p>
804 <pre class="doc_code">
805 %mytype = type { %mytype*, i32 }
808 <p>You may give a name to any <a href="#typesystem">type</a> except
809 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
810 is expected with the syntax "%mytype".</p>
812 <p>Note that type names are aliases for the structural type that they indicate,
813 and that you can therefore specify multiple names for the same type. This
814 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
815 uses structural typing, the name is not part of the type. When printing out
816 LLVM IR, the printer will pick <em>one name</em> to render all types of a
817 particular shape. This means that if you have code where two different
818 source types end up having the same LLVM type, that the dumper will sometimes
819 print the "wrong" or unexpected type. This is an important design point and
820 isn't going to change.</p>
824 <!-- ======================================================================= -->
826 <a name="globalvars">Global Variables</a>
831 <p>Global variables define regions of memory allocated at compilation time
832 instead of run-time. Global variables may optionally be initialized, may
833 have an explicit section to be placed in, and may have an optional explicit
834 alignment specified. A variable may be defined as "thread_local", which
835 means that it will not be shared by threads (each thread will have a
836 separated copy of the variable). A variable may be defined as a global
837 "constant," which indicates that the contents of the variable
838 will <b>never</b> be modified (enabling better optimization, allowing the
839 global data to be placed in the read-only section of an executable, etc).
840 Note that variables that need runtime initialization cannot be marked
841 "constant" as there is a store to the variable.</p>
843 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
844 constant, even if the final definition of the global is not. This capability
845 can be used to enable slightly better optimization of the program, but
846 requires the language definition to guarantee that optimizations based on the
847 'constantness' are valid for the translation units that do not include the
850 <p>As SSA values, global variables define pointer values that are in scope
851 (i.e. they dominate) all basic blocks in the program. Global variables
852 always define a pointer to their "content" type because they describe a
853 region of memory, and all memory objects in LLVM are accessed through
856 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
857 that the address is not significant, only the content. Constants marked
858 like this can be merged with other constants if they have the same
859 initializer. Note that a constant with significant address <em>can</em>
860 be merged with a <tt>unnamed_addr</tt> constant, the result being a
861 constant whose address is significant.</p>
863 <p>A global variable may be declared to reside in a target-specific numbered
864 address space. For targets that support them, address spaces may affect how
865 optimizations are performed and/or what target instructions are used to
866 access the variable. The default address space is zero. The address space
867 qualifier must precede any other attributes.</p>
869 <p>LLVM allows an explicit section to be specified for globals. If the target
870 supports it, it will emit globals to the section specified.</p>
872 <p>An explicit alignment may be specified for a global, which must be a power
873 of 2. If not present, or if the alignment is set to zero, the alignment of
874 the global is set by the target to whatever it feels convenient. If an
875 explicit alignment is specified, the global is forced to have exactly that
876 alignment. Targets and optimizers are not allowed to over-align the global
877 if the global has an assigned section. In this case, the extra alignment
878 could be observable: for example, code could assume that the globals are
879 densely packed in their section and try to iterate over them as an array,
880 alignment padding would break this iteration.</p>
882 <p>For example, the following defines a global in a numbered address space with
883 an initializer, section, and alignment:</p>
885 <pre class="doc_code">
886 @G = addrspace(5) constant float 1.0, section "foo", align 4
892 <!-- ======================================================================= -->
894 <a name="functionstructure">Functions</a>
899 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
900 optional <a href="#linkage">linkage type</a>, an optional
901 <a href="#visibility">visibility style</a>, an optional
902 <a href="#callingconv">calling convention</a>,
903 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
904 <a href="#paramattrs">parameter attribute</a> for the return type, a function
905 name, a (possibly empty) argument list (each with optional
906 <a href="#paramattrs">parameter attributes</a>), optional
907 <a href="#fnattrs">function attributes</a>, an optional section, an optional
908 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
909 curly brace, a list of basic blocks, and a closing curly brace.</p>
911 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
912 optional <a href="#linkage">linkage type</a>, an optional
913 <a href="#visibility">visibility style</a>, an optional
914 <a href="#callingconv">calling convention</a>,
915 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
916 <a href="#paramattrs">parameter attribute</a> for the return type, a function
917 name, a possibly empty list of arguments, an optional alignment, and an
918 optional <a href="#gc">garbage collector name</a>.</p>
920 <p>A function definition contains a list of basic blocks, forming the CFG
921 (Control Flow Graph) for the function. Each basic block may optionally start
922 with a label (giving the basic block a symbol table entry), contains a list
923 of instructions, and ends with a <a href="#terminators">terminator</a>
924 instruction (such as a branch or function return).</p>
926 <p>The first basic block in a function is special in two ways: it is immediately
927 executed on entrance to the function, and it is not allowed to have
928 predecessor basic blocks (i.e. there can not be any branches to the entry
929 block of a function). Because the block can have no predecessors, it also
930 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
932 <p>LLVM allows an explicit section to be specified for functions. If the target
933 supports it, it will emit functions to the section specified.</p>
935 <p>An explicit alignment may be specified for a function. If not present, or if
936 the alignment is set to zero, the alignment of the function is set by the
937 target to whatever it feels convenient. If an explicit alignment is
938 specified, the function is forced to have at least that much alignment. All
939 alignments must be a power of 2.</p>
941 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
942 be significant and two identical functions can be merged.</p>
945 <pre class="doc_code">
946 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
947 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
948 <ResultType> @<FunctionName> ([argument list])
949 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
950 [<a href="#gc">gc</a>] { ... }
955 <!-- ======================================================================= -->
957 <a name="aliasstructure">Aliases</a>
962 <p>Aliases act as "second name" for the aliasee value (which can be either
963 function, global variable, another alias or bitcast of global value). Aliases
964 may have an optional <a href="#linkage">linkage type</a>, and an
965 optional <a href="#visibility">visibility style</a>.</p>
968 <pre class="doc_code">
969 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
974 <!-- ======================================================================= -->
976 <a name="namedmetadatastructure">Named Metadata</a>
981 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
982 nodes</a> (but not metadata strings) are the only valid operands for
983 a named metadata.</p>
986 <pre class="doc_code">
987 ; Some unnamed metadata nodes, which are referenced by the named metadata.
988 !0 = metadata !{metadata !"zero"}
989 !1 = metadata !{metadata !"one"}
990 !2 = metadata !{metadata !"two"}
992 !name = !{!0, !1, !2}
997 <!-- ======================================================================= -->
999 <a name="paramattrs">Parameter Attributes</a>
1004 <p>The return type and each parameter of a function type may have a set of
1005 <i>parameter attributes</i> associated with them. Parameter attributes are
1006 used to communicate additional information about the result or parameters of
1007 a function. Parameter attributes are considered to be part of the function,
1008 not of the function type, so functions with different parameter attributes
1009 can have the same function type.</p>
1011 <p>Parameter attributes are simple keywords that follow the type specified. If
1012 multiple parameter attributes are needed, they are space separated. For
1015 <pre class="doc_code">
1016 declare i32 @printf(i8* noalias nocapture, ...)
1017 declare i32 @atoi(i8 zeroext)
1018 declare signext i8 @returns_signed_char()
1021 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1022 <tt>readonly</tt>) come immediately after the argument list.</p>
1024 <p>Currently, only the following parameter attributes are defined:</p>
1027 <dt><tt><b>zeroext</b></tt></dt>
1028 <dd>This indicates to the code generator that the parameter or return value
1029 should be zero-extended to the extent required by the target's ABI (which
1030 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1031 parameter) or the callee (for a return value).</dd>
1033 <dt><tt><b>signext</b></tt></dt>
1034 <dd>This indicates to the code generator that the parameter or return value
1035 should be sign-extended to the extent required by the target's ABI (which
1036 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1039 <dt><tt><b>inreg</b></tt></dt>
1040 <dd>This indicates that this parameter or return value should be treated in a
1041 special target-dependent fashion during while emitting code for a function
1042 call or return (usually, by putting it in a register as opposed to memory,
1043 though some targets use it to distinguish between two different kinds of
1044 registers). Use of this attribute is target-specific.</dd>
1046 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1047 <dd><p>This indicates that the pointer parameter should really be passed by
1048 value to the function. The attribute implies that a hidden copy of the
1050 is made between the caller and the callee, so the callee is unable to
1051 modify the value in the callee. This attribute is only valid on LLVM
1052 pointer arguments. It is generally used to pass structs and arrays by
1053 value, but is also valid on pointers to scalars. The copy is considered
1054 to belong to the caller not the callee (for example,
1055 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1056 <tt>byval</tt> parameters). This is not a valid attribute for return
1059 <p>The byval attribute also supports specifying an alignment with
1060 the align attribute. It indicates the alignment of the stack slot to
1061 form and the known alignment of the pointer specified to the call site. If
1062 the alignment is not specified, then the code generator makes a
1063 target-specific assumption.</p></dd>
1065 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1066 <dd>This indicates that the pointer parameter specifies the address of a
1067 structure that is the return value of the function in the source program.
1068 This pointer must be guaranteed by the caller to be valid: loads and
1069 stores to the structure may be assumed by the callee to not to trap. This
1070 may only be applied to the first parameter. This is not a valid attribute
1071 for return values. </dd>
1073 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1074 <dd>This indicates that pointer values
1075 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1076 value do not alias pointer values which are not <i>based</i> on it,
1077 ignoring certain "irrelevant" dependencies.
1078 For a call to the parent function, dependencies between memory
1079 references from before or after the call and from those during the call
1080 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1081 return value used in that call.
1082 The caller shares the responsibility with the callee for ensuring that
1083 these requirements are met.
1084 For further details, please see the discussion of the NoAlias response in
1085 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1087 Note that this definition of <tt>noalias</tt> is intentionally
1088 similar to the definition of <tt>restrict</tt> in C99 for function
1089 arguments, though it is slightly weaker.
1091 For function return values, C99's <tt>restrict</tt> is not meaningful,
1092 while LLVM's <tt>noalias</tt> is.
1095 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1096 <dd>This indicates that the callee does not make any copies of the pointer
1097 that outlive the callee itself. This is not a valid attribute for return
1100 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1101 <dd>This indicates that the pointer parameter can be excised using the
1102 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1103 attribute for return values.</dd>
1108 <!-- ======================================================================= -->
1110 <a name="gc">Garbage Collector Names</a>
1115 <p>Each function may specify a garbage collector name, which is simply a
1118 <pre class="doc_code">
1119 define void @f() gc "name" { ... }
1122 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1123 collector which will cause the compiler to alter its output in order to
1124 support the named garbage collection algorithm.</p>
1128 <!-- ======================================================================= -->
1130 <a name="fnattrs">Function Attributes</a>
1135 <p>Function attributes are set to communicate additional information about a
1136 function. Function attributes are considered to be part of the function, not
1137 of the function type, so functions with different parameter attributes can
1138 have the same function type.</p>
1140 <p>Function attributes are simple keywords that follow the type specified. If
1141 multiple attributes are needed, they are space separated. For example:</p>
1143 <pre class="doc_code">
1144 define void @f() noinline { ... }
1145 define void @f() alwaysinline { ... }
1146 define void @f() alwaysinline optsize { ... }
1147 define void @f() optsize { ... }
1151 <dt><tt><b>address_safety</b></tt></dt>
1152 <dd>This attribute indicates that the address safety analysis
1153 is enabled for this function. </dd>
1155 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1156 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1157 the backend should forcibly align the stack pointer. Specify the
1158 desired alignment, which must be a power of two, in parentheses.
1160 <dt><tt><b>alwaysinline</b></tt></dt>
1161 <dd>This attribute indicates that the inliner should attempt to inline this
1162 function into callers whenever possible, ignoring any active inlining size
1163 threshold for this caller.</dd>
1165 <dt><tt><b>nonlazybind</b></tt></dt>
1166 <dd>This attribute suppresses lazy symbol binding for the function. This
1167 may make calls to the function faster, at the cost of extra program
1168 startup time if the function is not called during program startup.</dd>
1170 <dt><tt><b>inlinehint</b></tt></dt>
1171 <dd>This attribute indicates that the source code contained a hint that inlining
1172 this function is desirable (such as the "inline" keyword in C/C++). It
1173 is just a hint; it imposes no requirements on the inliner.</dd>
1175 <dt><tt><b>naked</b></tt></dt>
1176 <dd>This attribute disables prologue / epilogue emission for the function.
1177 This can have very system-specific consequences.</dd>
1179 <dt><tt><b>noimplicitfloat</b></tt></dt>
1180 <dd>This attributes disables implicit floating point instructions.</dd>
1182 <dt><tt><b>noinline</b></tt></dt>
1183 <dd>This attribute indicates that the inliner should never inline this
1184 function in any situation. This attribute may not be used together with
1185 the <tt>alwaysinline</tt> attribute.</dd>
1187 <dt><tt><b>noredzone</b></tt></dt>
1188 <dd>This attribute indicates that the code generator should not use a red
1189 zone, even if the target-specific ABI normally permits it.</dd>
1191 <dt><tt><b>noreturn</b></tt></dt>
1192 <dd>This function attribute indicates that the function never returns
1193 normally. This produces undefined behavior at runtime if the function
1194 ever does dynamically return.</dd>
1196 <dt><tt><b>nounwind</b></tt></dt>
1197 <dd>This function attribute indicates that the function never returns with an
1198 unwind or exceptional control flow. If the function does unwind, its
1199 runtime behavior is undefined.</dd>
1201 <dt><tt><b>optsize</b></tt></dt>
1202 <dd>This attribute suggests that optimization passes and code generator passes
1203 make choices that keep the code size of this function low, and otherwise
1204 do optimizations specifically to reduce code size.</dd>
1206 <dt><tt><b>readnone</b></tt></dt>
1207 <dd>This attribute indicates that the function computes its result (or decides
1208 to unwind an exception) based strictly on its arguments, without
1209 dereferencing any pointer arguments or otherwise accessing any mutable
1210 state (e.g. memory, control registers, etc) visible to caller functions.
1211 It does not write through any pointer arguments
1212 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1213 changes any state visible to callers. This means that it cannot unwind
1214 exceptions by calling the <tt>C++</tt> exception throwing methods.</dd>
1216 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1217 <dd>This attribute indicates that the function does not write through any
1218 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1219 arguments) or otherwise modify any state (e.g. memory, control registers,
1220 etc) visible to caller functions. It may dereference pointer arguments
1221 and read state that may be set in the caller. A readonly function always
1222 returns the same value (or unwinds an exception identically) when called
1223 with the same set of arguments and global state. It cannot unwind an
1224 exception by calling the <tt>C++</tt> exception throwing methods.</dd>
1226 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1227 <dd>This attribute indicates that this function can return twice. The
1228 C <code>setjmp</code> is an example of such a function. The compiler
1229 disables some optimizations (like tail calls) in the caller of these
1232 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1233 <dd>This attribute indicates that the function should emit a stack smashing
1234 protector. It is in the form of a "canary"—a random value placed on
1235 the stack before the local variables that's checked upon return from the
1236 function to see if it has been overwritten. A heuristic is used to
1237 determine if a function needs stack protectors or not.<br>
1239 If a function that has an <tt>ssp</tt> attribute is inlined into a
1240 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1241 function will have an <tt>ssp</tt> attribute.</dd>
1243 <dt><tt><b>sspreq</b></tt></dt>
1244 <dd>This attribute indicates that the function should <em>always</em> emit a
1245 stack smashing protector. This overrides
1246 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1248 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1249 function that doesn't have an <tt>sspreq</tt> attribute or which has
1250 an <tt>ssp</tt> attribute, then the resulting function will have
1251 an <tt>sspreq</tt> attribute.</dd>
1253 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1254 <dd>This attribute indicates that the ABI being targeted requires that
1255 an unwind table entry be produce for this function even if we can
1256 show that no exceptions passes by it. This is normally the case for
1257 the ELF x86-64 abi, but it can be disabled for some compilation
1263 <!-- ======================================================================= -->
1265 <a name="moduleasm">Module-Level Inline Assembly</a>
1270 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1271 the GCC "file scope inline asm" blocks. These blocks are internally
1272 concatenated by LLVM and treated as a single unit, but may be separated in
1273 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1275 <pre class="doc_code">
1276 module asm "inline asm code goes here"
1277 module asm "more can go here"
1280 <p>The strings can contain any character by escaping non-printable characters.
1281 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1284 <p>The inline asm code is simply printed to the machine code .s file when
1285 assembly code is generated.</p>
1289 <!-- ======================================================================= -->
1291 <a name="datalayout">Data Layout</a>
1296 <p>A module may specify a target specific data layout string that specifies how
1297 data is to be laid out in memory. The syntax for the data layout is
1300 <pre class="doc_code">
1301 target datalayout = "<i>layout specification</i>"
1304 <p>The <i>layout specification</i> consists of a list of specifications
1305 separated by the minus sign character ('-'). Each specification starts with
1306 a letter and may include other information after the letter to define some
1307 aspect of the data layout. The specifications accepted are as follows:</p>
1311 <dd>Specifies that the target lays out data in big-endian form. That is, the
1312 bits with the most significance have the lowest address location.</dd>
1315 <dd>Specifies that the target lays out data in little-endian form. That is,
1316 the bits with the least significance have the lowest address
1319 <dt><tt>S<i>size</i></tt></dt>
1320 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1321 of stack variables is limited to the natural stack alignment to avoid
1322 dynamic stack realignment. The stack alignment must be a multiple of
1323 8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1324 which does not prevent any alignment promotions.</dd>
1326 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1327 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1328 <i>preferred</i> alignments. All sizes are in bits. Specifying
1329 the <i>pref</i> alignment is optional. If omitted, the
1330 preceding <tt>:</tt> should be omitted too.</dd>
1332 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1333 <dd>This specifies the alignment for an integer type of a given bit
1334 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1336 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1337 <dd>This specifies the alignment for a vector type of a given bit
1340 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1341 <dd>This specifies the alignment for a floating point type of a given bit
1342 <i>size</i>. Only values of <i>size</i> that are supported by the target
1343 will work. 32 (float) and 64 (double) are supported on all targets;
1344 80 or 128 (different flavors of long double) are also supported on some
1347 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1348 <dd>This specifies the alignment for an aggregate type of a given bit
1351 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1352 <dd>This specifies the alignment for a stack object of a given bit
1355 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1356 <dd>This specifies a set of native integer widths for the target CPU
1357 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1358 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1359 this set are considered to support most general arithmetic
1360 operations efficiently.</dd>
1363 <p>When constructing the data layout for a given target, LLVM starts with a
1364 default set of specifications which are then (possibly) overridden by the
1365 specifications in the <tt>datalayout</tt> keyword. The default specifications
1366 are given in this list:</p>
1369 <li><tt>E</tt> - big endian</li>
1370 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1371 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1372 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1373 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1374 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1375 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1376 alignment of 64-bits</li>
1377 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1378 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1379 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1380 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1381 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1382 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1385 <p>When LLVM is determining the alignment for a given type, it uses the
1386 following rules:</p>
1389 <li>If the type sought is an exact match for one of the specifications, that
1390 specification is used.</li>
1392 <li>If no match is found, and the type sought is an integer type, then the
1393 smallest integer type that is larger than the bitwidth of the sought type
1394 is used. If none of the specifications are larger than the bitwidth then
1395 the the largest integer type is used. For example, given the default
1396 specifications above, the i7 type will use the alignment of i8 (next
1397 largest) while both i65 and i256 will use the alignment of i64 (largest
1400 <li>If no match is found, and the type sought is a vector type, then the
1401 largest vector type that is smaller than the sought vector type will be
1402 used as a fall back. This happens because <128 x double> can be
1403 implemented in terms of 64 <2 x double>, for example.</li>
1406 <p>The function of the data layout string may not be what you expect. Notably,
1407 this is not a specification from the frontend of what alignment the code
1408 generator should use.</p>
1410 <p>Instead, if specified, the target data layout is required to match what the
1411 ultimate <em>code generator</em> expects. This string is used by the
1412 mid-level optimizers to
1413 improve code, and this only works if it matches what the ultimate code
1414 generator uses. If you would like to generate IR that does not embed this
1415 target-specific detail into the IR, then you don't have to specify the
1416 string. This will disable some optimizations that require precise layout
1417 information, but this also prevents those optimizations from introducing
1418 target specificity into the IR.</p>
1424 <!-- ======================================================================= -->
1426 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1431 <p>Any memory access must be done through a pointer value associated
1432 with an address range of the memory access, otherwise the behavior
1433 is undefined. Pointer values are associated with address ranges
1434 according to the following rules:</p>
1437 <li>A pointer value is associated with the addresses associated with
1438 any value it is <i>based</i> on.
1439 <li>An address of a global variable is associated with the address
1440 range of the variable's storage.</li>
1441 <li>The result value of an allocation instruction is associated with
1442 the address range of the allocated storage.</li>
1443 <li>A null pointer in the default address-space is associated with
1445 <li>An integer constant other than zero or a pointer value returned
1446 from a function not defined within LLVM may be associated with address
1447 ranges allocated through mechanisms other than those provided by
1448 LLVM. Such ranges shall not overlap with any ranges of addresses
1449 allocated by mechanisms provided by LLVM.</li>
1452 <p>A pointer value is <i>based</i> on another pointer value according
1453 to the following rules:</p>
1456 <li>A pointer value formed from a
1457 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1458 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1459 <li>The result value of a
1460 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1461 of the <tt>bitcast</tt>.</li>
1462 <li>A pointer value formed by an
1463 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1464 pointer values that contribute (directly or indirectly) to the
1465 computation of the pointer's value.</li>
1466 <li>The "<i>based</i> on" relationship is transitive.</li>
1469 <p>Note that this definition of <i>"based"</i> is intentionally
1470 similar to the definition of <i>"based"</i> in C99, though it is
1471 slightly weaker.</p>
1473 <p>LLVM IR does not associate types with memory. The result type of a
1474 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1475 alignment of the memory from which to load, as well as the
1476 interpretation of the value. The first operand type of a
1477 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1478 and alignment of the store.</p>
1480 <p>Consequently, type-based alias analysis, aka TBAA, aka
1481 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1482 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1483 additional information which specialized optimization passes may use
1484 to implement type-based alias analysis.</p>
1488 <!-- ======================================================================= -->
1490 <a name="volatile">Volatile Memory Accesses</a>
1495 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1496 href="#i_store"><tt>store</tt></a>s, and <a
1497 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1498 The optimizers must not change the number of volatile operations or change their
1499 order of execution relative to other volatile operations. The optimizers
1500 <i>may</i> change the order of volatile operations relative to non-volatile
1501 operations. This is not Java's "volatile" and has no cross-thread
1502 synchronization behavior.</p>
1506 <!-- ======================================================================= -->
1508 <a name="memmodel">Memory Model for Concurrent Operations</a>
1513 <p>The LLVM IR does not define any way to start parallel threads of execution
1514 or to register signal handlers. Nonetheless, there are platform-specific
1515 ways to create them, and we define LLVM IR's behavior in their presence. This
1516 model is inspired by the C++0x memory model.</p>
1518 <p>For a more informal introduction to this model, see the
1519 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1521 <p>We define a <i>happens-before</i> partial order as the least partial order
1524 <li>Is a superset of single-thread program order, and</li>
1525 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1526 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1527 by platform-specific techniques, like pthread locks, thread
1528 creation, thread joining, etc., and by atomic instructions.
1529 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1533 <p>Note that program order does not introduce <i>happens-before</i> edges
1534 between a thread and signals executing inside that thread.</p>
1536 <p>Every (defined) read operation (load instructions, memcpy, atomic
1537 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1538 (defined) write operations (store instructions, atomic
1539 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1540 initialized globals are considered to have a write of the initializer which is
1541 atomic and happens before any other read or write of the memory in question.
1542 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1543 any write to the same byte, except:</p>
1546 <li>If <var>write<sub>1</sub></var> happens before
1547 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1548 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1549 does not see <var>write<sub>1</sub></var>.
1550 <li>If <var>R<sub>byte</sub></var> happens before
1551 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1552 see <var>write<sub>3</sub></var>.
1555 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1557 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1558 is supposed to give guarantees which can support
1559 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1560 addresses which do not behave like normal memory. It does not generally
1561 provide cross-thread synchronization.)
1562 <li>Otherwise, if there is no write to the same byte that happens before
1563 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1564 <tt>undef</tt> for that byte.
1565 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1566 <var>R<sub>byte</sub></var> returns the value written by that
1568 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1569 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1570 values written. See the <a href="#ordering">Atomic Memory Ordering
1571 Constraints</a> section for additional constraints on how the choice
1573 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1576 <p><var>R</var> returns the value composed of the series of bytes it read.
1577 This implies that some bytes within the value may be <tt>undef</tt>
1578 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1579 defines the semantics of the operation; it doesn't mean that targets will
1580 emit more than one instruction to read the series of bytes.</p>
1582 <p>Note that in cases where none of the atomic intrinsics are used, this model
1583 places only one restriction on IR transformations on top of what is required
1584 for single-threaded execution: introducing a store to a byte which might not
1585 otherwise be stored is not allowed in general. (Specifically, in the case
1586 where another thread might write to and read from an address, introducing a
1587 store can change a load that may see exactly one write into a load that may
1588 see multiple writes.)</p>
1590 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1591 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1592 none of the backends currently in the tree fall into this category; however,
1593 there might be targets which care. If there are, we want a paragraph
1596 Targets may specify that stores narrower than a certain width are not
1597 available; on such a target, for the purposes of this model, treat any
1598 non-atomic write with an alignment or width less than the minimum width
1599 as if it writes to the relevant surrounding bytes.
1604 <!-- ======================================================================= -->
1606 <a name="ordering">Atomic Memory Ordering Constraints</a>
1611 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1612 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1613 <a href="#i_fence"><code>fence</code></a>,
1614 <a href="#i_load"><code>atomic load</code></a>, and
1615 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1616 that determines which other atomic instructions on the same address they
1617 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1618 but are somewhat more colloquial. If these descriptions aren't precise enough,
1619 check those specs (see spec references in the
1620 <a href="Atomics.html#introduction">atomics guide</a>).
1621 <a href="#i_fence"><code>fence</code></a> instructions
1622 treat these orderings somewhat differently since they don't take an address.
1623 See that instruction's documentation for details.</p>
1625 <p>For a simpler introduction to the ordering constraints, see the
1626 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1629 <dt><code>unordered</code></dt>
1630 <dd>The set of values that can be read is governed by the happens-before
1631 partial order. A value cannot be read unless some operation wrote it.
1632 This is intended to provide a guarantee strong enough to model Java's
1633 non-volatile shared variables. This ordering cannot be specified for
1634 read-modify-write operations; it is not strong enough to make them atomic
1635 in any interesting way.</dd>
1636 <dt><code>monotonic</code></dt>
1637 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1638 total order for modifications by <code>monotonic</code> operations on each
1639 address. All modification orders must be compatible with the happens-before
1640 order. There is no guarantee that the modification orders can be combined to
1641 a global total order for the whole program (and this often will not be
1642 possible). The read in an atomic read-modify-write operation
1643 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1644 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1645 reads the value in the modification order immediately before the value it
1646 writes. If one atomic read happens before another atomic read of the same
1647 address, the later read must see the same value or a later value in the
1648 address's modification order. This disallows reordering of
1649 <code>monotonic</code> (or stronger) operations on the same address. If an
1650 address is written <code>monotonic</code>ally by one thread, and other threads
1651 <code>monotonic</code>ally read that address repeatedly, the other threads must
1652 eventually see the write. This corresponds to the C++0x/C1x
1653 <code>memory_order_relaxed</code>.</dd>
1654 <dt><code>acquire</code></dt>
1655 <dd>In addition to the guarantees of <code>monotonic</code>,
1656 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1657 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1658 <dt><code>release</code></dt>
1659 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1660 writes a value which is subsequently read by an <code>acquire</code> operation,
1661 it <i>synchronizes-with</i> that operation. (This isn't a complete
1662 description; see the C++0x definition of a release sequence.) This corresponds
1663 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1664 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1665 <code>acquire</code> and <code>release</code> operation on its address.
1666 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1667 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1668 <dd>In addition to the guarantees of <code>acq_rel</code>
1669 (<code>acquire</code> for an operation which only reads, <code>release</code>
1670 for an operation which only writes), there is a global total order on all
1671 sequentially-consistent operations on all addresses, which is consistent with
1672 the <i>happens-before</i> partial order and with the modification orders of
1673 all the affected addresses. Each sequentially-consistent read sees the last
1674 preceding write to the same address in this global order. This corresponds
1675 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1678 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1679 it only <i>synchronizes with</i> or participates in modification and seq_cst
1680 total orderings with other operations running in the same thread (for example,
1681 in signal handlers).</p>
1687 <!-- *********************************************************************** -->
1688 <h2><a name="typesystem">Type System</a></h2>
1689 <!-- *********************************************************************** -->
1693 <p>The LLVM type system is one of the most important features of the
1694 intermediate representation. Being typed enables a number of optimizations
1695 to be performed on the intermediate representation directly, without having
1696 to do extra analyses on the side before the transformation. A strong type
1697 system makes it easier to read the generated code and enables novel analyses
1698 and transformations that are not feasible to perform on normal three address
1699 code representations.</p>
1701 <!-- ======================================================================= -->
1703 <a name="t_classifications">Type Classifications</a>
1708 <p>The types fall into a few useful classifications:</p>
1710 <table border="1" cellspacing="0" cellpadding="4">
1712 <tr><th>Classification</th><th>Types</th></tr>
1714 <td><a href="#t_integer">integer</a></td>
1715 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1718 <td><a href="#t_floating">floating point</a></td>
1719 <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1722 <td><a name="t_firstclass">first class</a></td>
1723 <td><a href="#t_integer">integer</a>,
1724 <a href="#t_floating">floating point</a>,
1725 <a href="#t_pointer">pointer</a>,
1726 <a href="#t_vector">vector</a>,
1727 <a href="#t_struct">structure</a>,
1728 <a href="#t_array">array</a>,
1729 <a href="#t_label">label</a>,
1730 <a href="#t_metadata">metadata</a>.
1734 <td><a href="#t_primitive">primitive</a></td>
1735 <td><a href="#t_label">label</a>,
1736 <a href="#t_void">void</a>,
1737 <a href="#t_integer">integer</a>,
1738 <a href="#t_floating">floating point</a>,
1739 <a href="#t_x86mmx">x86mmx</a>,
1740 <a href="#t_metadata">metadata</a>.</td>
1743 <td><a href="#t_derived">derived</a></td>
1744 <td><a href="#t_array">array</a>,
1745 <a href="#t_function">function</a>,
1746 <a href="#t_pointer">pointer</a>,
1747 <a href="#t_struct">structure</a>,
1748 <a href="#t_vector">vector</a>,
1749 <a href="#t_opaque">opaque</a>.
1755 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1756 important. Values of these types are the only ones which can be produced by
1761 <!-- ======================================================================= -->
1763 <a name="t_primitive">Primitive Types</a>
1768 <p>The primitive types are the fundamental building blocks of the LLVM
1771 <!-- _______________________________________________________________________ -->
1773 <a name="t_integer">Integer Type</a>
1779 <p>The integer type is a very simple type that simply specifies an arbitrary
1780 bit width for the integer type desired. Any bit width from 1 bit to
1781 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1788 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1792 <table class="layout">
1794 <td class="left"><tt>i1</tt></td>
1795 <td class="left">a single-bit integer.</td>
1798 <td class="left"><tt>i32</tt></td>
1799 <td class="left">a 32-bit integer.</td>
1802 <td class="left"><tt>i1942652</tt></td>
1803 <td class="left">a really big integer of over 1 million bits.</td>
1809 <!-- _______________________________________________________________________ -->
1811 <a name="t_floating">Floating Point Types</a>
1818 <tr><th>Type</th><th>Description</th></tr>
1819 <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr>
1820 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1821 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1822 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1823 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1824 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1830 <!-- _______________________________________________________________________ -->
1832 <a name="t_x86mmx">X86mmx Type</a>
1838 <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>
1847 <!-- _______________________________________________________________________ -->
1849 <a name="t_void">Void Type</a>
1855 <p>The void type does not represent any value and has no size.</p>
1864 <!-- _______________________________________________________________________ -->
1866 <a name="t_label">Label Type</a>
1872 <p>The label type represents code labels.</p>
1881 <!-- _______________________________________________________________________ -->
1883 <a name="t_metadata">Metadata Type</a>
1889 <p>The metadata type represents embedded metadata. No derived types may be
1890 created from metadata except for <a href="#t_function">function</a>
1902 <!-- ======================================================================= -->
1904 <a name="t_derived">Derived Types</a>
1909 <p>The real power in LLVM comes from the derived types in the system. This is
1910 what allows a programmer to represent arrays, functions, pointers, and other
1911 useful types. Each of these types contain one or more element types which
1912 may be a primitive type, or another derived type. For example, it is
1913 possible to have a two dimensional array, using an array as the element type
1914 of another array.</p>
1916 <!-- _______________________________________________________________________ -->
1918 <a name="t_aggregate">Aggregate Types</a>
1923 <p>Aggregate Types are a subset of derived types that can contain multiple
1924 member types. <a href="#t_array">Arrays</a> and
1925 <a href="#t_struct">structs</a> are aggregate types.
1926 <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p>
1930 <!-- _______________________________________________________________________ -->
1932 <a name="t_array">Array Type</a>
1938 <p>The array type is a very simple derived type that arranges elements
1939 sequentially in memory. The array type requires a size (number of elements)
1940 and an underlying data type.</p>
1944 [<# elements> x <elementtype>]
1947 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1948 be any type with a size.</p>
1951 <table class="layout">
1953 <td class="left"><tt>[40 x i32]</tt></td>
1954 <td class="left">Array of 40 32-bit integer values.</td>
1957 <td class="left"><tt>[41 x i32]</tt></td>
1958 <td class="left">Array of 41 32-bit integer values.</td>
1961 <td class="left"><tt>[4 x i8]</tt></td>
1962 <td class="left">Array of 4 8-bit integer values.</td>
1965 <p>Here are some examples of multidimensional arrays:</p>
1966 <table class="layout">
1968 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1969 <td class="left">3x4 array of 32-bit integer values.</td>
1972 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1973 <td class="left">12x10 array of single precision floating point values.</td>
1976 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1977 <td class="left">2x3x4 array of 16-bit integer values.</td>
1981 <p>There is no restriction on indexing beyond the end of the array implied by
1982 a static type (though there are restrictions on indexing beyond the bounds
1983 of an allocated object in some cases). This means that single-dimension
1984 'variable sized array' addressing can be implemented in LLVM with a zero
1985 length array type. An implementation of 'pascal style arrays' in LLVM could
1986 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1990 <!-- _______________________________________________________________________ -->
1992 <a name="t_function">Function Type</a>
1998 <p>The function type can be thought of as a function signature. It consists of
1999 a return type and a list of formal parameter types. The return type of a
2000 function type is a first class type or a void type.</p>
2004 <returntype> (<parameter list>)
2007 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
2008 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
2009 which indicates that the function takes a variable number of arguments.
2010 Variable argument functions can access their arguments with
2011 the <a href="#int_varargs">variable argument handling intrinsic</a>
2012 functions. '<tt><returntype></tt>' is any type except
2013 <a href="#t_label">label</a>.</p>
2016 <table class="layout">
2018 <td class="left"><tt>i32 (i32)</tt></td>
2019 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2021 </tr><tr class="layout">
2022 <td class="left"><tt>float (i16, i32 *) *
2024 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2025 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2026 returning <tt>float</tt>.
2028 </tr><tr class="layout">
2029 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2030 <td class="left">A vararg function that takes at least one
2031 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2032 which returns an integer. This is the signature for <tt>printf</tt> in
2035 </tr><tr class="layout">
2036 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2037 <td class="left">A function taking an <tt>i32</tt>, returning a
2038 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2045 <!-- _______________________________________________________________________ -->
2047 <a name="t_struct">Structure Type</a>
2053 <p>The structure type is used to represent a collection of data members together
2054 in memory. The elements of a structure may be any type that has a size.</p>
2056 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2057 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2058 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2059 Structures in registers are accessed using the
2060 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2061 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2063 <p>Structures may optionally be "packed" structures, which indicate that the
2064 alignment of the struct is one byte, and that there is no padding between
2065 the elements. In non-packed structs, padding between field types is inserted
2066 as defined by the TargetData string in the module, which is required to match
2067 what the underlying code generator expects.</p>
2069 <p>Structures can either be "literal" or "identified". A literal structure is
2070 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2071 types are always defined at the top level with a name. Literal types are
2072 uniqued by their contents and can never be recursive or opaque since there is
2073 no way to write one. Identified types can be recursive, can be opaqued, and are
2079 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2080 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2084 <table class="layout">
2086 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2087 <td class="left">A triple of three <tt>i32</tt> values</td>
2090 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2091 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2092 second element is a <a href="#t_pointer">pointer</a> to a
2093 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2094 an <tt>i32</tt>.</td>
2097 <td class="left"><tt><{ i8, i32 }></tt></td>
2098 <td class="left">A packed struct known to be 5 bytes in size.</td>
2104 <!-- _______________________________________________________________________ -->
2106 <a name="t_opaque">Opaque Structure Types</a>
2112 <p>Opaque structure types are used to represent named structure types that do
2113 not have a body specified. This corresponds (for example) to the C notion of
2114 a forward declared structure.</p>
2123 <table class="layout">
2125 <td class="left"><tt>opaque</tt></td>
2126 <td class="left">An opaque type.</td>
2134 <!-- _______________________________________________________________________ -->
2136 <a name="t_pointer">Pointer Type</a>
2142 <p>The pointer type is used to specify memory locations.
2143 Pointers are commonly used to reference objects in memory.</p>
2145 <p>Pointer types may have an optional address space attribute defining the
2146 numbered address space where the pointed-to object resides. The default
2147 address space is number zero. The semantics of non-zero address
2148 spaces are target-specific.</p>
2150 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2151 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2159 <table class="layout">
2161 <td class="left"><tt>[4 x i32]*</tt></td>
2162 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2163 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2166 <td class="left"><tt>i32 (i32*) *</tt></td>
2167 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2168 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2172 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2173 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2174 that resides in address space #5.</td>
2180 <!-- _______________________________________________________________________ -->
2182 <a name="t_vector">Vector Type</a>
2188 <p>A vector type is a simple derived type that represents a vector of elements.
2189 Vector types are used when multiple primitive data are operated in parallel
2190 using a single instruction (SIMD). A vector type requires a size (number of
2191 elements) and an underlying primitive data type. Vector types are considered
2192 <a href="#t_firstclass">first class</a>.</p>
2196 < <# elements> x <elementtype> >
2199 <p>The number of elements is a constant integer value larger than 0; elementtype
2200 may be any integer or floating point type, or a pointer to these types.
2201 Vectors of size zero are not allowed. </p>
2204 <table class="layout">
2206 <td class="left"><tt><4 x i32></tt></td>
2207 <td class="left">Vector of 4 32-bit integer values.</td>
2210 <td class="left"><tt><8 x float></tt></td>
2211 <td class="left">Vector of 8 32-bit floating-point values.</td>
2214 <td class="left"><tt><2 x i64></tt></td>
2215 <td class="left">Vector of 2 64-bit integer values.</td>
2218 <td class="left"><tt><4 x i64*></tt></td>
2219 <td class="left">Vector of 4 pointers to 64-bit integer values.</td>
2229 <!-- *********************************************************************** -->
2230 <h2><a name="constants">Constants</a></h2>
2231 <!-- *********************************************************************** -->
2235 <p>LLVM has several different basic types of constants. This section describes
2236 them all and their syntax.</p>
2238 <!-- ======================================================================= -->
2240 <a name="simpleconstants">Simple Constants</a>
2246 <dt><b>Boolean constants</b></dt>
2247 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2248 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2250 <dt><b>Integer constants</b></dt>
2251 <dd>Standard integers (such as '4') are constants of
2252 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2253 with integer types.</dd>
2255 <dt><b>Floating point constants</b></dt>
2256 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2257 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2258 notation (see below). The assembler requires the exact decimal value of a
2259 floating-point constant. For example, the assembler accepts 1.25 but
2260 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2261 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2263 <dt><b>Null pointer constants</b></dt>
2264 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2265 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2268 <p>The one non-intuitive notation for constants is the hexadecimal form of
2269 floating point constants. For example, the form '<tt>double
2270 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2271 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2272 constants are required (and the only time that they are generated by the
2273 disassembler) is when a floating point constant must be emitted but it cannot
2274 be represented as a decimal floating point number in a reasonable number of
2275 digits. For example, NaN's, infinities, and other special values are
2276 represented in their IEEE hexadecimal format so that assembly and disassembly
2277 do not cause any bits to change in the constants.</p>
2279 <p>When using the hexadecimal form, constants of types half, float, and double are
2280 represented using the 16-digit form shown above (which matches the IEEE754
2281 representation for double); half and float values must, however, be exactly
2282 representable as IEE754 half and single precision, respectively.
2283 Hexadecimal format is always used
2284 for long double, and there are three forms of long double. The 80-bit format
2285 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2286 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2287 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2288 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2289 currently supported target uses this format. Long doubles will only work if
2290 they match the long double format on your target. All hexadecimal formats
2291 are big-endian (sign bit at the left).</p>
2293 <p>There are no constants of type x86mmx.</p>
2296 <!-- ======================================================================= -->
2298 <a name="aggregateconstants"></a> <!-- old anchor -->
2299 <a name="complexconstants">Complex Constants</a>
2304 <p>Complex constants are a (potentially recursive) combination of simple
2305 constants and smaller complex constants.</p>
2308 <dt><b>Structure constants</b></dt>
2309 <dd>Structure constants are represented with notation similar to structure
2310 type definitions (a comma separated list of elements, surrounded by braces
2311 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2312 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2313 Structure constants must have <a href="#t_struct">structure type</a>, and
2314 the number and types of elements must match those specified by the
2317 <dt><b>Array constants</b></dt>
2318 <dd>Array constants are represented with notation similar to array type
2319 definitions (a comma separated list of elements, surrounded by square
2320 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2321 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2322 the number and types of elements must match those specified by the
2325 <dt><b>Vector constants</b></dt>
2326 <dd>Vector constants are represented with notation similar to vector type
2327 definitions (a comma separated list of elements, surrounded by
2328 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2329 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2330 have <a href="#t_vector">vector type</a>, and the number and types of
2331 elements must match those specified by the type.</dd>
2333 <dt><b>Zero initialization</b></dt>
2334 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2335 value to zero of <em>any</em> type, including scalar and
2336 <a href="#t_aggregate">aggregate</a> types.
2337 This is often used to avoid having to print large zero initializers
2338 (e.g. for large arrays) and is always exactly equivalent to using explicit
2339 zero initializers.</dd>
2341 <dt><b>Metadata node</b></dt>
2342 <dd>A metadata node is a structure-like constant with
2343 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2344 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2345 be interpreted as part of the instruction stream, metadata is a place to
2346 attach additional information such as debug info.</dd>
2351 <!-- ======================================================================= -->
2353 <a name="globalconstants">Global Variable and Function Addresses</a>
2358 <p>The addresses of <a href="#globalvars">global variables</a>
2359 and <a href="#functionstructure">functions</a> are always implicitly valid
2360 (link-time) constants. These constants are explicitly referenced when
2361 the <a href="#identifiers">identifier for the global</a> is used and always
2362 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2363 legal LLVM file:</p>
2365 <pre class="doc_code">
2368 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2373 <!-- ======================================================================= -->
2375 <a name="undefvalues">Undefined Values</a>
2380 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2381 indicates that the user of the value may receive an unspecified bit-pattern.
2382 Undefined values may be of any type (other than '<tt>label</tt>'
2383 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2385 <p>Undefined values are useful because they indicate to the compiler that the
2386 program is well defined no matter what value is used. This gives the
2387 compiler more freedom to optimize. Here are some examples of (potentially
2388 surprising) transformations that are valid (in pseudo IR):</p>
2391 <pre class="doc_code">
2401 <p>This is safe because all of the output bits are affected by the undef bits.
2402 Any output bit can have a zero or one depending on the input bits.</p>
2404 <pre class="doc_code">
2415 <p>These logical operations have bits that are not always affected by the input.
2416 For example, if <tt>%X</tt> has a zero bit, then the output of the
2417 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2418 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2419 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2420 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2421 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2422 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2423 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2425 <pre class="doc_code">
2426 %A = select undef, %X, %Y
2427 %B = select undef, 42, %Y
2428 %C = select %X, %Y, undef
2439 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2440 branch) conditions can go <em>either way</em>, but they have to come from one
2441 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2442 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2443 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2444 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2445 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2448 <pre class="doc_code">
2449 %A = xor undef, undef
2467 <p>This example points out that two '<tt>undef</tt>' operands are not
2468 necessarily the same. This can be surprising to people (and also matches C
2469 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2470 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2471 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2472 its value over its "live range". This is true because the variable doesn't
2473 actually <em>have a live range</em>. Instead, the value is logically read
2474 from arbitrary registers that happen to be around when needed, so the value
2475 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2476 need to have the same semantics or the core LLVM "replace all uses with"
2477 concept would not hold.</p>
2479 <pre class="doc_code">
2487 <p>These examples show the crucial difference between an <em>undefined
2488 value</em> and <em>undefined behavior</em>. An undefined value (like
2489 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2490 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2491 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2492 defined on SNaN's. However, in the second example, we can make a more
2493 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2494 arbitrary value, we are allowed to assume that it could be zero. Since a
2495 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2496 the operation does not execute at all. This allows us to delete the divide and
2497 all code after it. Because the undefined operation "can't happen", the
2498 optimizer can assume that it occurs in dead code.</p>
2500 <pre class="doc_code">
2501 a: store undef -> %X
2502 b: store %X -> undef
2508 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2509 undefined value can be assumed to not have any effect; we can assume that the
2510 value is overwritten with bits that happen to match what was already there.
2511 However, a store <em>to</em> an undefined location could clobber arbitrary
2512 memory, therefore, it has undefined behavior.</p>
2516 <!-- ======================================================================= -->
2518 <a name="poisonvalues">Poison Values</a>
2523 <p>Poison values are similar to <a href="#undefvalues">undef values</a>, however
2524 they also represent the fact that an instruction or constant expression which
2525 cannot evoke side effects has nevertheless detected a condition which results
2526 in undefined behavior.</p>
2528 <p>There is currently no way of representing a poison value in the IR; they
2529 only exist when produced by operations such as
2530 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2532 <p>Poison value behavior is defined in terms of value <i>dependence</i>:</p>
2535 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2536 their operands.</li>
2538 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2539 to their dynamic predecessor basic block.</li>
2541 <li>Function arguments depend on the corresponding actual argument values in
2542 the dynamic callers of their functions.</li>
2544 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2545 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2546 control back to them.</li>
2548 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2549 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>,
2550 or exception-throwing call instructions that dynamically transfer control
2553 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2554 referenced memory addresses, following the order in the IR
2555 (including loads and stores implied by intrinsics such as
2556 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2558 <!-- TODO: In the case of multiple threads, this only applies if the store
2559 "happens-before" the load or store. -->
2561 <!-- TODO: floating-point exception state -->
2563 <li>An instruction with externally visible side effects depends on the most
2564 recent preceding instruction with externally visible side effects, following
2565 the order in the IR. (This includes
2566 <a href="#volatile">volatile operations</a>.)</li>
2568 <li>An instruction <i>control-depends</i> on a
2569 <a href="#terminators">terminator instruction</a>
2570 if the terminator instruction has multiple successors and the instruction
2571 is always executed when control transfers to one of the successors, and
2572 may not be executed when control is transferred to another.</li>
2574 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2575 instruction if the set of instructions it otherwise depends on would be
2576 different if the terminator had transferred control to a different
2579 <li>Dependence is transitive.</li>
2583 <p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>,
2584 with the additional affect that any instruction which has a <i>dependence</i>
2585 on a poison value has undefined behavior.</p>
2587 <p>Here are some examples:</p>
2589 <pre class="doc_code">
2591 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2592 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2593 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2594 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2596 store i32 %poison, i32* @g ; Poison value stored to memory.
2597 %poison2 = load i32* @g ; Poison value loaded back from memory.
2599 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2601 %narrowaddr = bitcast i32* @g to i16*
2602 %wideaddr = bitcast i32* @g to i64*
2603 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2604 %poison4 = load i64* %wideaddr ; Returns a poison value.
2606 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2607 br i1 %cmp, label %true, label %end ; Branch to either destination.
2610 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2611 ; it has undefined behavior.
2615 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2616 ; Both edges into this PHI are
2617 ; control-dependent on %cmp, so this
2618 ; always results in a poison value.
2620 store volatile i32 0, i32* @g ; This would depend on the store in %true
2621 ; if %cmp is true, or the store in %entry
2622 ; otherwise, so this is undefined behavior.
2624 br i1 %cmp, label %second_true, label %second_end
2625 ; The same branch again, but this time the
2626 ; true block doesn't have side effects.
2633 store volatile i32 0, i32* @g ; This time, the instruction always depends
2634 ; on the store in %end. Also, it is
2635 ; control-equivalent to %end, so this is
2636 ; well-defined (ignoring earlier undefined
2637 ; behavior in this example).
2642 <!-- ======================================================================= -->
2644 <a name="blockaddress">Addresses of Basic Blocks</a>
2649 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2651 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2652 basic block in the specified function, and always has an i8* type. Taking
2653 the address of the entry block is illegal.</p>
2655 <p>This value only has defined behavior when used as an operand to the
2656 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2657 comparisons against null. Pointer equality tests between labels addresses
2658 results in undefined behavior — though, again, comparison against null
2659 is ok, and no label is equal to the null pointer. This may be passed around
2660 as an opaque pointer sized value as long as the bits are not inspected. This
2661 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2662 long as the original value is reconstituted before the <tt>indirectbr</tt>
2665 <p>Finally, some targets may provide defined semantics when using the value as
2666 the operand to an inline assembly, but that is target specific.</p>
2671 <!-- ======================================================================= -->
2673 <a name="constantexprs">Constant Expressions</a>
2678 <p>Constant expressions are used to allow expressions involving other constants
2679 to be used as constants. Constant expressions may be of
2680 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2681 operation that does not have side effects (e.g. load and call are not
2682 supported). The following is the syntax for constant expressions:</p>
2685 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2686 <dd>Truncate a constant to another type. The bit size of CST must be larger
2687 than the bit size of TYPE. Both types must be integers.</dd>
2689 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2690 <dd>Zero extend a constant to another type. The bit size of CST must be
2691 smaller than the bit size of TYPE. Both types must be integers.</dd>
2693 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2694 <dd>Sign extend a constant to another type. The bit size of CST must be
2695 smaller than the bit size of TYPE. Both types must be integers.</dd>
2697 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2698 <dd>Truncate a floating point constant to another floating point type. The
2699 size of CST must be larger than the size of TYPE. Both types must be
2700 floating point.</dd>
2702 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2703 <dd>Floating point extend a constant to another type. The size of CST must be
2704 smaller or equal to the size of TYPE. Both types must be floating
2707 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2708 <dd>Convert a floating point constant to the corresponding unsigned integer
2709 constant. TYPE must be a scalar or vector integer type. CST must be of
2710 scalar or vector floating point type. Both CST and TYPE must be scalars,
2711 or vectors of the same number of elements. If the value won't fit in the
2712 integer type, the results are undefined.</dd>
2714 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2715 <dd>Convert a floating point constant to the corresponding signed integer
2716 constant. TYPE must be a scalar or vector integer type. CST must be of
2717 scalar or vector floating point type. Both CST and TYPE must be scalars,
2718 or vectors of the same number of elements. If the value won't fit in the
2719 integer type, the results are undefined.</dd>
2721 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2722 <dd>Convert an unsigned integer constant to the corresponding floating point
2723 constant. TYPE must be a scalar or vector floating point type. CST must be
2724 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2725 vectors of the same number of elements. If the value won't fit in the
2726 floating point type, the results are undefined.</dd>
2728 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2729 <dd>Convert a signed integer constant to the corresponding floating point
2730 constant. TYPE must be a scalar or vector floating point type. CST must be
2731 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2732 vectors of the same number of elements. If the value won't fit in the
2733 floating point type, the results are undefined.</dd>
2735 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2736 <dd>Convert a pointer typed constant to the corresponding integer constant
2737 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2738 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2739 make it fit in <tt>TYPE</tt>.</dd>
2741 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2742 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2743 type. CST must be of integer type. The CST value is zero extended,
2744 truncated, or unchanged to make it fit in a pointer size. This one is
2745 <i>really</i> dangerous!</dd>
2747 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2748 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2749 are the same as those for the <a href="#i_bitcast">bitcast
2750 instruction</a>.</dd>
2752 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2753 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2754 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2755 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2756 instruction, the index list may have zero or more indexes, which are
2757 required to make sense for the type of "CSTPTR".</dd>
2759 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2760 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2762 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2763 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2765 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2766 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2768 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2769 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2772 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2773 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2776 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2777 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2780 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2781 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2782 constants. The index list is interpreted in a similar manner as indices in
2783 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2784 index value must be specified.</dd>
2786 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2787 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2788 constants. The index list is interpreted in a similar manner as indices in
2789 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2790 index value must be specified.</dd>
2792 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2793 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2794 be any of the <a href="#binaryops">binary</a>
2795 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2796 on operands are the same as those for the corresponding instruction
2797 (e.g. no bitwise operations on floating point values are allowed).</dd>
2804 <!-- *********************************************************************** -->
2805 <h2><a name="othervalues">Other Values</a></h2>
2806 <!-- *********************************************************************** -->
2808 <!-- ======================================================================= -->
2810 <a name="inlineasm">Inline Assembler Expressions</a>
2815 <p>LLVM supports inline assembler expressions (as opposed
2816 to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of
2817 a special value. This value represents the inline assembler as a string
2818 (containing the instructions to emit), a list of operand constraints (stored
2819 as a string), a flag that indicates whether or not the inline asm
2820 expression has side effects, and a flag indicating whether the function
2821 containing the asm needs to align its stack conservatively. An example
2822 inline assembler expression is:</p>
2824 <pre class="doc_code">
2825 i32 (i32) asm "bswap $0", "=r,r"
2828 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2829 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2832 <pre class="doc_code">
2833 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2836 <p>Inline asms with side effects not visible in the constraint list must be
2837 marked as having side effects. This is done through the use of the
2838 '<tt>sideeffect</tt>' keyword, like so:</p>
2840 <pre class="doc_code">
2841 call void asm sideeffect "eieio", ""()
2844 <p>In some cases inline asms will contain code that will not work unless the
2845 stack is aligned in some way, such as calls or SSE instructions on x86,
2846 yet will not contain code that does that alignment within the asm.
2847 The compiler should make conservative assumptions about what the asm might
2848 contain and should generate its usual stack alignment code in the prologue
2849 if the '<tt>alignstack</tt>' keyword is present:</p>
2851 <pre class="doc_code">
2852 call void asm alignstack "eieio", ""()
2855 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2859 <p>TODO: The format of the asm and constraints string still need to be
2860 documented here. Constraints on what can be done (e.g. duplication, moving,
2861 etc need to be documented). This is probably best done by reference to
2862 another document that covers inline asm from a holistic perspective.</p>
2865 <!-- _______________________________________________________________________ -->
2867 <a name="inlineasm_md">Inline Asm Metadata</a>
2872 <p>The call instructions that wrap inline asm nodes may have a
2873 "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant
2874 integers. If present, the code generator will use the integer as the
2875 location cookie value when report errors through the <tt>LLVMContext</tt>
2876 error reporting mechanisms. This allows a front-end to correlate backend
2877 errors that occur with inline asm back to the source code that produced it.
2880 <pre class="doc_code">
2881 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2883 !42 = !{ i32 1234567 }
2886 <p>It is up to the front-end to make sense of the magic numbers it places in the
2887 IR. If the MDNode contains multiple constants, the code generator will use
2888 the one that corresponds to the line of the asm that the error occurs on.</p>
2894 <!-- ======================================================================= -->
2896 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2901 <p>LLVM IR allows metadata to be attached to instructions in the program that
2902 can convey extra information about the code to the optimizers and code
2903 generator. One example application of metadata is source-level debug
2904 information. There are two metadata primitives: strings and nodes. All
2905 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2906 preceding exclamation point ('<tt>!</tt>').</p>
2908 <p>A metadata string is a string surrounded by double quotes. It can contain
2909 any character by escaping non-printable characters with "<tt>\xx</tt>" where
2910 "<tt>xx</tt>" is the two digit hex code. For example:
2911 "<tt>!"test\00"</tt>".</p>
2913 <p>Metadata nodes are represented with notation similar to structure constants
2914 (a comma separated list of elements, surrounded by braces and preceded by an
2915 exclamation point). Metadata nodes can have any values as their operand. For
2918 <div class="doc_code">
2920 !{ metadata !"test\00", i32 10}
2924 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2925 metadata nodes, which can be looked up in the module symbol table. For
2928 <div class="doc_code">
2930 !foo = metadata !{!4, !3}
2934 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2935 function is using two metadata arguments:</p>
2937 <div class="doc_code">
2939 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2943 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2944 attached to the <tt>add</tt> instruction using the <tt>!dbg</tt>
2947 <div class="doc_code">
2949 %indvar.next = add i64 %indvar, 1, !dbg !21
2953 <p>More information about specific metadata nodes recognized by the optimizers
2954 and code generator is found below.</p>
2956 <!-- _______________________________________________________________________ -->
2958 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
2963 <p>In LLVM IR, memory does not have types, so LLVM's own type system is not
2964 suitable for doing TBAA. Instead, metadata is added to the IR to describe
2965 a type system of a higher level language. This can be used to implement
2966 typical C/C++ TBAA, but it can also be used to implement custom alias
2967 analysis behavior for other languages.</p>
2969 <p>The current metadata format is very simple. TBAA metadata nodes have up to
2970 three fields, e.g.:</p>
2972 <div class="doc_code">
2974 !0 = metadata !{ metadata !"an example type tree" }
2975 !1 = metadata !{ metadata !"int", metadata !0 }
2976 !2 = metadata !{ metadata !"float", metadata !0 }
2977 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2981 <p>The first field is an identity field. It can be any value, usually
2982 a metadata string, which uniquely identifies the type. The most important
2983 name in the tree is the name of the root node. Two trees with
2984 different root node names are entirely disjoint, even if they
2985 have leaves with common names.</p>
2987 <p>The second field identifies the type's parent node in the tree, or
2988 is null or omitted for a root node. A type is considered to alias
2989 all of its descendants and all of its ancestors in the tree. Also,
2990 a type is considered to alias all types in other trees, so that
2991 bitcode produced from multiple front-ends is handled conservatively.</p>
2993 <p>If the third field is present, it's an integer which if equal to 1
2994 indicates that the type is "constant" (meaning
2995 <tt>pointsToConstantMemory</tt> should return true; see
2996 <a href="AliasAnalysis.html#OtherItfs">other useful
2997 <tt>AliasAnalysis</tt> methods</a>).</p>
3001 <!-- _______________________________________________________________________ -->
3003 <a name="fpaccuracy">'<tt>fpaccuracy</tt>' Metadata</a>
3008 <p><tt>fpaccuracy</tt> metadata may be attached to any instruction of floating
3009 point type. It expresses the maximum relative error allowed in the result
3010 of that instruction, in ULPs, thus potentially allowing the compiler to use
3011 a more efficient but less accurate method of computing it.
3012 ULP is defined as follows:</p>
3016 <p>If <tt>x</tt> is a real number that lies between two finite consecutive
3017 floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one
3018 of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the
3019 distance between the two non-equal finite floating-point numbers nearest
3020 <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p>
3024 <p>The metadata node shall consist of a single non-negative floating
3025 point number representing the maximum relative error. For example,
3028 <div class="doc_code">
3030 !0 = metadata !{ float 2.5 }
3036 <!-- _______________________________________________________________________ -->
3038 <a name="range">'<tt>range</tt>' Metadata</a>
3042 <p><tt>range</tt> metadata may be attached only to loads of integer types. It
3043 expresses the possible ranges the loaded value is in. The ranges are
3044 represented with a flattened list of integers. The loaded value is known to
3045 be in the union of the ranges defined by each consecutive pair. Each pair
3046 has the following properties:</p>
3048 <li>The type must match the type loaded by the instruction.</li>
3049 <li>The pair <tt>a,b</tt> represents the range <tt>[a,b)</tt>.</li>
3050 <li>Both <tt>a</tt> and <tt>b</tt> are constants.</li>
3051 <li>The range is allowed to wrap.</li>
3052 <li>The range should not represent the full or empty set. That is,
3053 <tt>a!=b</tt>. </li>
3057 <div class="doc_code">
3059 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
3060 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3061 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
3063 !0 = metadata !{ i8 0, i8 2 }
3064 !1 = metadata !{ i8 255, i8 2 }
3065 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
3073 <!-- *********************************************************************** -->
3075 <a name="module_flags">Module Flags Metadata</a>
3077 <!-- *********************************************************************** -->
3081 <p>Information about the module as a whole is difficult to convey to LLVM's
3082 subsystems. The LLVM IR isn't sufficient to transmit this
3083 information. The <tt>llvm.module.flags</tt> named metadata exists in order to
3084 facilitate this. These flags are in the form of key / value pairs —
3085 much like a dictionary — making it easy for any subsystem who cares
3086 about a flag to look it up.</p>
3088 <p>The <tt>llvm.module.flags</tt> metadata contains a list of metadata
3089 triplets. Each triplet has the following form:</p>
3092 <li>The first element is a <i>behavior</i> flag, which specifies the behavior
3093 when two (or more) modules are merged together, and it encounters two (or
3094 more) metadata with the same ID. The supported behaviors are described
3097 <li>The second element is a metadata string that is a unique ID for the
3098 metadata. How each ID is interpreted is documented below.</li>
3100 <li>The third element is the value of the flag.</li>
3103 <p>When two (or more) modules are merged together, the resulting
3104 <tt>llvm.module.flags</tt> metadata is the union of the
3105 modules' <tt>llvm.module.flags</tt> metadata. The only exception being a flag
3106 with the <i>Override</i> behavior, which may override another flag's value
3109 <p>The following behaviors are supported:</p>
3111 <table border="1" cellspacing="0" cellpadding="4">
3121 <dt><b>Error</b></dt>
3122 <dd>Emits an error if two values disagree. It is an error to have an ID
3123 with both an Error and a Warning behavior.</dd>
3131 <dt><b>Warning</b></dt>
3132 <dd>Emits a warning if two values disagree.</dd>
3140 <dt><b>Require</b></dt>
3141 <dd>Emits an error when the specified value is not present or doesn't
3142 have the specified value. It is an error for two (or more)
3143 <tt>llvm.module.flags</tt> with the same ID to have the Require
3144 behavior but different values. There may be multiple Require flags
3153 <dt><b>Override</b></dt>
3154 <dd>Uses the specified value if the two values disagree. It is an
3155 error for two (or more) <tt>llvm.module.flags</tt> with the same
3156 ID to have the Override behavior but different values.</dd>
3163 <p>An example of module flags:</p>
3165 <pre class="doc_code">
3166 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3167 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3168 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3169 !3 = metadata !{ i32 3, metadata !"qux",
3171 metadata !"foo", i32 1
3174 !llvm.module.flags = !{ !0, !1, !2, !3 }
3178 <li><p>Metadata <tt>!0</tt> has the ID <tt>!"foo"</tt> and the value '1'. The
3179 behavior if two or more <tt>!"foo"</tt> flags are seen is to emit an
3180 error if their values are not equal.</p></li>
3182 <li><p>Metadata <tt>!1</tt> has the ID <tt>!"bar"</tt> and the value '37'. The
3183 behavior if two or more <tt>!"bar"</tt> flags are seen is to use the
3184 value '37' if their values are not equal.</p></li>
3186 <li><p>Metadata <tt>!2</tt> has the ID <tt>!"qux"</tt> and the value '42'. The
3187 behavior if two or more <tt>!"qux"</tt> flags are seen is to emit a
3188 warning if their values are not equal.</p></li>
3190 <li><p>Metadata <tt>!3</tt> has the ID <tt>!"qux"</tt> and the value:</p>
3192 <pre class="doc_code">
3193 metadata !{ metadata !"foo", i32 1 }
3196 <p>The behavior is to emit an error if the <tt>llvm.module.flags</tt> does
3197 not contain a flag with the ID <tt>!"foo"</tt> that has the value
3198 '1'. If two or more <tt>!"qux"</tt> flags exist, then they must have
3199 the same value or an error will be issued.</p></li>
3203 <!-- ======================================================================= -->
3205 <a name="objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a>
3210 <p>On the Mach-O platform, Objective-C stores metadata about garbage collection
3211 in a special section called "image info". The metadata consists of a version
3212 number and a bitmask specifying what types of garbage collection are
3213 supported (if any) by the file. If two or more modules are linked together
3214 their garbage collection metadata needs to be merged rather than appended
3217 <p>The Objective-C garbage collection module flags metadata consists of the
3218 following key-value pairs:</p>
3220 <table border="1" cellspacing="0" cellpadding="4">
3228 <td><tt>Objective-C Version</tt></td>
3229 <td align="left"><b>[Required]</b> — The Objective-C ABI
3230 version. Valid values are 1 and 2.</td>
3233 <td><tt>Objective-C Image Info Version</tt></td>
3234 <td align="left"><b>[Required]</b> — The version of the image info
3235 section. Currently always 0.</td>
3238 <td><tt>Objective-C Image Info Section</tt></td>
3239 <td align="left"><b>[Required]</b> — The section to place the
3240 metadata. Valid values are <tt>"__OBJC, __image_info, regular"</tt> for
3241 Objective-C ABI version 1, and <tt>"__DATA,__objc_imageinfo, regular,
3242 no_dead_strip"</tt> for Objective-C ABI version 2.</td>
3245 <td><tt>Objective-C Garbage Collection</tt></td>
3246 <td align="left"><b>[Required]</b> — Specifies whether garbage
3247 collection is supported or not. Valid values are 0, for no garbage
3248 collection, and 2, for garbage collection supported.</td>
3251 <td><tt>Objective-C GC Only</tt></td>
3252 <td align="left"><b>[Optional]</b> — Specifies that only garbage
3253 collection is supported. If present, its value must be 6. This flag
3254 requires that the <tt>Objective-C Garbage Collection</tt> flag have the
3260 <p>Some important flag interactions:</p>
3263 <li>If a module with <tt>Objective-C Garbage Collection</tt> set to 0 is
3264 merged with a module with <tt>Objective-C Garbage Collection</tt> set to
3265 2, then the resulting module has the <tt>Objective-C Garbage
3266 Collection</tt> flag set to 0.</li>
3268 <li>A module with <tt>Objective-C Garbage Collection</tt> set to 0 cannot be
3269 merged with a module with <tt>Objective-C GC Only</tt> set to 6.</li>
3276 <!-- *********************************************************************** -->
3278 <a name="intrinsic_globals">Intrinsic Global Variables</a>
3280 <!-- *********************************************************************** -->
3282 <p>LLVM has a number of "magic" global variables that contain data that affect
3283 code generation or other IR semantics. These are documented here. All globals
3284 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
3285 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
3288 <!-- ======================================================================= -->
3290 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
3295 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
3296 href="#linkage_appending">appending linkage</a>. This array contains a list of
3297 pointers to global variables and functions which may optionally have a pointer
3298 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
3300 <div class="doc_code">
3305 @llvm.used = appending global [2 x i8*] [
3307 i8* bitcast (i32* @Y to i8*)
3308 ], section "llvm.metadata"
3312 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
3313 compiler, assembler, and linker are required to treat the symbol as if there
3314 is a reference to the global that it cannot see. For example, if a variable
3315 has internal linkage and no references other than that from
3316 the <tt>@llvm.used</tt> list, it cannot be deleted. This is commonly used to
3317 represent references from inline asms and other things the compiler cannot
3318 "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p>
3320 <p>On some targets, the code generator must emit a directive to the assembler or
3321 object file to prevent the assembler and linker from molesting the
3326 <!-- ======================================================================= -->
3328 <a name="intg_compiler_used">
3329 The '<tt>llvm.compiler.used</tt>' Global Variable
3335 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
3336 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
3337 touching the symbol. On targets that support it, this allows an intelligent
3338 linker to optimize references to the symbol without being impeded as it would
3339 be by <tt>@llvm.used</tt>.</p>
3341 <p>This is a rare construct that should only be used in rare circumstances, and
3342 should not be exposed to source languages.</p>
3346 <!-- ======================================================================= -->
3348 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
3353 <div class="doc_code">
3355 %0 = type { i32, void ()* }
3356 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3360 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor
3361 functions and associated priorities. The functions referenced by this array
3362 will be called in ascending order of priority (i.e. lowest first) when the
3363 module is loaded. The order of functions with the same priority is not
3368 <!-- ======================================================================= -->
3370 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3375 <div class="doc_code">
3377 %0 = type { i32, void ()* }
3378 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3382 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions
3383 and associated priorities. The functions referenced by this array will be
3384 called in descending order of priority (i.e. highest first) when the module
3385 is loaded. The order of functions with the same priority is not defined.</p>
3391 <!-- *********************************************************************** -->
3392 <h2><a name="instref">Instruction Reference</a></h2>
3393 <!-- *********************************************************************** -->
3397 <p>The LLVM instruction set consists of several different classifications of
3398 instructions: <a href="#terminators">terminator
3399 instructions</a>, <a href="#binaryops">binary instructions</a>,
3400 <a href="#bitwiseops">bitwise binary instructions</a>,
3401 <a href="#memoryops">memory instructions</a>, and
3402 <a href="#otherops">other instructions</a>.</p>
3404 <!-- ======================================================================= -->
3406 <a name="terminators">Terminator Instructions</a>
3411 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3412 in a program ends with a "Terminator" instruction, which indicates which
3413 block should be executed after the current block is finished. These
3414 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3415 control flow, not values (the one exception being the
3416 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3418 <p>The terminator instructions are:
3419 '<a href="#i_ret"><tt>ret</tt></a>',
3420 '<a href="#i_br"><tt>br</tt></a>',
3421 '<a href="#i_switch"><tt>switch</tt></a>',
3422 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3423 '<a href="#i_invoke"><tt>invoke</tt></a>',
3424 '<a href="#i_resume"><tt>resume</tt></a>', and
3425 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3427 <!-- _______________________________________________________________________ -->
3429 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3436 ret <type> <value> <i>; Return a value from a non-void function</i>
3437 ret void <i>; Return from void function</i>
3441 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3442 a value) from a function back to the caller.</p>
3444 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3445 value and then causes control flow, and one that just causes control flow to
3449 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3450 return value. The type of the return value must be a
3451 '<a href="#t_firstclass">first class</a>' type.</p>
3453 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3454 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3455 value or a return value with a type that does not match its type, or if it
3456 has a void return type and contains a '<tt>ret</tt>' instruction with a
3460 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3461 the calling function's context. If the caller is a
3462 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3463 instruction after the call. If the caller was an
3464 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3465 the beginning of the "normal" destination block. If the instruction returns
3466 a value, that value shall set the call or invoke instruction's return
3471 ret i32 5 <i>; Return an integer value of 5</i>
3472 ret void <i>; Return from a void function</i>
3473 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3477 <!-- _______________________________________________________________________ -->
3479 <a name="i_br">'<tt>br</tt>' Instruction</a>
3486 br i1 <cond>, label <iftrue>, label <iffalse>
3487 br label <dest> <i>; Unconditional branch</i>
3491 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3492 different basic block in the current function. There are two forms of this
3493 instruction, corresponding to a conditional branch and an unconditional
3497 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3498 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3499 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3503 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3504 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3505 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3506 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3511 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3512 br i1 %cond, label %IfEqual, label %IfUnequal
3514 <a href="#i_ret">ret</a> i32 1
3516 <a href="#i_ret">ret</a> i32 0
3521 <!-- _______________________________________________________________________ -->
3523 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3530 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3534 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3535 several different places. It is a generalization of the '<tt>br</tt>'
3536 instruction, allowing a branch to occur to one of many possible
3540 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3541 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3542 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3543 The table is not allowed to contain duplicate constant entries.</p>
3546 <p>The <tt>switch</tt> instruction specifies a table of values and
3547 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3548 is searched for the given value. If the value is found, control flow is
3549 transferred to the corresponding destination; otherwise, control flow is
3550 transferred to the default destination.</p>
3552 <h5>Implementation:</h5>
3553 <p>Depending on properties of the target machine and the particular
3554 <tt>switch</tt> instruction, this instruction may be code generated in
3555 different ways. For example, it could be generated as a series of chained
3556 conditional branches or with a lookup table.</p>
3560 <i>; Emulate a conditional br instruction</i>
3561 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3562 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3564 <i>; Emulate an unconditional br instruction</i>
3565 switch i32 0, label %dest [ ]
3567 <i>; Implement a jump table:</i>
3568 switch i32 %val, label %otherwise [ i32 0, label %onzero
3570 i32 2, label %ontwo ]
3576 <!-- _______________________________________________________________________ -->
3578 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3585 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3590 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3591 within the current function, whose address is specified by
3592 "<tt>address</tt>". Address must be derived from a <a
3593 href="#blockaddress">blockaddress</a> constant.</p>
3597 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3598 rest of the arguments indicate the full set of possible destinations that the
3599 address may point to. Blocks are allowed to occur multiple times in the
3600 destination list, though this isn't particularly useful.</p>
3602 <p>This destination list is required so that dataflow analysis has an accurate
3603 understanding of the CFG.</p>
3607 <p>Control transfers to the block specified in the address argument. All
3608 possible destination blocks must be listed in the label list, otherwise this
3609 instruction has undefined behavior. This implies that jumps to labels
3610 defined in other functions have undefined behavior as well.</p>
3612 <h5>Implementation:</h5>
3614 <p>This is typically implemented with a jump through a register.</p>
3618 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3624 <!-- _______________________________________________________________________ -->
3626 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3633 <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>]
3634 to label <normal label> unwind label <exception label>
3638 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3639 function, with the possibility of control flow transfer to either the
3640 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3641 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3642 control flow will return to the "normal" label. If the callee (or any
3643 indirect callees) returns via the "<a href="#i_resume"><tt>resume</tt></a>"
3644 instruction or other exception handling mechanism, control is interrupted and
3645 continued at the dynamically nearest "exception" label.</p>
3647 <p>The '<tt>exception</tt>' label is a
3648 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3649 exception. As such, '<tt>exception</tt>' label is required to have the
3650 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3651 the information about the behavior of the program after unwinding
3652 happens, as its first non-PHI instruction. The restrictions on the
3653 "<tt>landingpad</tt>" instruction's tightly couples it to the
3654 "<tt>invoke</tt>" instruction, so that the important information contained
3655 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3659 <p>This instruction requires several arguments:</p>
3662 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3663 convention</a> the call should use. If none is specified, the call
3664 defaults to using C calling conventions.</li>
3666 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3667 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3668 '<tt>inreg</tt>' attributes are valid here.</li>
3670 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3671 function value being invoked. In most cases, this is a direct function
3672 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3673 off an arbitrary pointer to function value.</li>
3675 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3676 function to be invoked. </li>
3678 <li>'<tt>function args</tt>': argument list whose types match the function
3679 signature argument types and parameter attributes. All arguments must be
3680 of <a href="#t_firstclass">first class</a> type. If the function
3681 signature indicates the function accepts a variable number of arguments,
3682 the extra arguments can be specified.</li>
3684 <li>'<tt>normal label</tt>': the label reached when the called function
3685 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3687 <li>'<tt>exception label</tt>': the label reached when a callee returns via
3688 the <a href="#i_resume"><tt>resume</tt></a> instruction or other exception
3689 handling mechanism.</li>
3691 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3692 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3693 '<tt>readnone</tt>' attributes are valid here.</li>
3697 <p>This instruction is designed to operate as a standard
3698 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3699 primary difference is that it establishes an association with a label, which
3700 is used by the runtime library to unwind the stack.</p>
3702 <p>This instruction is used in languages with destructors to ensure that proper
3703 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3704 exception. Additionally, this is important for implementation of
3705 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3707 <p>For the purposes of the SSA form, the definition of the value returned by the
3708 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3709 block to the "normal" label. If the callee unwinds then no return value is
3714 %retval = invoke i32 @Test(i32 15) to label %Continue
3715 unwind label %TestCleanup <i>; {i32}:retval set</i>
3716 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3717 unwind label %TestCleanup <i>; {i32}:retval set</i>
3722 <!-- _______________________________________________________________________ -->
3725 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3732 resume <type> <value>
3736 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3740 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3741 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3745 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3746 (in-flight) exception whose unwinding was interrupted with
3747 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3751 resume { i8*, i32 } %exn
3756 <!-- _______________________________________________________________________ -->
3759 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3770 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3771 instruction is used to inform the optimizer that a particular portion of the
3772 code is not reachable. This can be used to indicate that the code after a
3773 no-return function cannot be reached, and other facts.</p>
3776 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3782 <!-- ======================================================================= -->
3784 <a name="binaryops">Binary Operations</a>
3789 <p>Binary operators are used to do most of the computation in a program. They
3790 require two operands of the same type, execute an operation on them, and
3791 produce a single value. The operands might represent multiple data, as is
3792 the case with the <a href="#t_vector">vector</a> data type. The result value
3793 has the same type as its operands.</p>
3795 <p>There are several different binary operators:</p>
3797 <!-- _______________________________________________________________________ -->
3799 <a name="i_add">'<tt>add</tt>' Instruction</a>
3806 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3807 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3808 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3809 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3813 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3816 <p>The two arguments to the '<tt>add</tt>' instruction must
3817 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3818 integer values. Both arguments must have identical types.</p>
3821 <p>The value produced is the integer sum of the two operands.</p>
3823 <p>If the sum has unsigned overflow, the result returned is the mathematical
3824 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3826 <p>Because LLVM integers use a two's complement representation, this instruction
3827 is appropriate for both signed and unsigned integers.</p>
3829 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3830 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3831 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3832 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3833 respectively, occurs.</p>
3837 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3842 <!-- _______________________________________________________________________ -->
3844 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3851 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3855 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3858 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3859 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3860 floating point values. Both arguments must have identical types.</p>
3863 <p>The value produced is the floating point sum of the two operands.</p>
3867 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3872 <!-- _______________________________________________________________________ -->
3874 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3881 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3882 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3883 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3884 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3888 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3891 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3892 '<tt>neg</tt>' instruction present in most other intermediate
3893 representations.</p>
3896 <p>The two arguments to the '<tt>sub</tt>' instruction must
3897 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3898 integer values. Both arguments must have identical types.</p>
3901 <p>The value produced is the integer difference of the two operands.</p>
3903 <p>If the difference has unsigned overflow, the result returned is the
3904 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3907 <p>Because LLVM integers use a two's complement representation, this instruction
3908 is appropriate for both signed and unsigned integers.</p>
3910 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3911 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3912 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3913 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3914 respectively, occurs.</p>
3918 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3919 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3924 <!-- _______________________________________________________________________ -->
3926 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3933 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3937 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3940 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3941 '<tt>fneg</tt>' instruction present in most other intermediate
3942 representations.</p>
3945 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3946 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3947 floating point values. Both arguments must have identical types.</p>
3950 <p>The value produced is the floating point difference of the two operands.</p>
3954 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3955 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3960 <!-- _______________________________________________________________________ -->
3962 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3969 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3970 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3971 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3972 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3976 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3979 <p>The two arguments to the '<tt>mul</tt>' instruction must
3980 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3981 integer values. Both arguments must have identical types.</p>
3984 <p>The value produced is the integer product of the two operands.</p>
3986 <p>If the result of the multiplication has unsigned overflow, the result
3987 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3988 width of the result.</p>
3990 <p>Because LLVM integers use a two's complement representation, and the result
3991 is the same width as the operands, this instruction returns the correct
3992 result for both signed and unsigned integers. If a full product
3993 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3994 be sign-extended or zero-extended as appropriate to the width of the full
3997 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3998 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3999 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
4000 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4001 respectively, occurs.</p>
4005 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
4010 <!-- _______________________________________________________________________ -->
4012 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
4019 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4023 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
4026 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
4027 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4028 floating point values. Both arguments must have identical types.</p>
4031 <p>The value produced is the floating point product of the two operands.</p>
4035 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
4040 <!-- _______________________________________________________________________ -->
4042 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
4049 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4050 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4054 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
4057 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
4058 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4059 values. Both arguments must have identical types.</p>
4062 <p>The value produced is the unsigned integer quotient of the two operands.</p>
4064 <p>Note that unsigned integer division and signed integer division are distinct
4065 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
4067 <p>Division by zero leads to undefined behavior.</p>
4069 <p>If the <tt>exact</tt> keyword is present, the result value of the
4070 <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a
4071 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
4076 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4081 <!-- _______________________________________________________________________ -->
4083 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
4090 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4091 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4095 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
4098 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
4099 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4100 values. Both arguments must have identical types.</p>
4103 <p>The value produced is the signed integer quotient of the two operands rounded
4106 <p>Note that signed integer division and unsigned integer division are distinct
4107 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
4109 <p>Division by zero leads to undefined behavior. Overflow also leads to
4110 undefined behavior; this is a rare case, but can occur, for example, by doing
4111 a 32-bit division of -2147483648 by -1.</p>
4113 <p>If the <tt>exact</tt> keyword is present, the result value of the
4114 <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would
4119 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4124 <!-- _______________________________________________________________________ -->
4126 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
4133 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4137 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
4140 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
4141 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4142 floating point values. Both arguments must have identical types.</p>
4145 <p>The value produced is the floating point quotient of the two operands.</p>
4149 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
4154 <!-- _______________________________________________________________________ -->
4156 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
4163 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4167 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
4168 division of its two arguments.</p>
4171 <p>The two arguments to the '<tt>urem</tt>' instruction must be
4172 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4173 values. Both arguments must have identical types.</p>
4176 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
4177 This instruction always performs an unsigned division to get the
4180 <p>Note that unsigned integer remainder and signed integer remainder are
4181 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
4183 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
4187 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4192 <!-- _______________________________________________________________________ -->
4194 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
4201 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4205 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
4206 division of its two operands. This instruction can also take
4207 <a href="#t_vector">vector</a> versions of the values in which case the
4208 elements must be integers.</p>
4211 <p>The two arguments to the '<tt>srem</tt>' instruction must be
4212 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4213 values. Both arguments must have identical types.</p>
4216 <p>This instruction returns the <i>remainder</i> of a division (where the result
4217 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
4218 <i>modulo</i> operator (where the result is either zero or has the same sign
4219 as the divisor, <tt>op2</tt>) of a value.
4220 For more information about the difference,
4221 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
4222 Math Forum</a>. For a table of how this is implemented in various languages,
4223 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
4224 Wikipedia: modulo operation</a>.</p>
4226 <p>Note that signed integer remainder and unsigned integer remainder are
4227 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
4229 <p>Taking the remainder of a division by zero leads to undefined behavior.
4230 Overflow also leads to undefined behavior; this is a rare case, but can
4231 occur, for example, by taking the remainder of a 32-bit division of
4232 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
4233 lets srem be implemented using instructions that return both the result of
4234 the division and the remainder.)</p>
4238 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4243 <!-- _______________________________________________________________________ -->
4245 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
4252 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4256 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
4257 its two operands.</p>
4260 <p>The two arguments to the '<tt>frem</tt>' instruction must be
4261 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4262 floating point values. Both arguments must have identical types.</p>
4265 <p>This instruction returns the <i>remainder</i> of a division. The remainder
4266 has the same sign as the dividend.</p>
4270 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
4277 <!-- ======================================================================= -->
4279 <a name="bitwiseops">Bitwise Binary Operations</a>
4284 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
4285 program. They are generally very efficient instructions and can commonly be
4286 strength reduced from other instructions. They require two operands of the
4287 same type, execute an operation on them, and produce a single value. The
4288 resulting value is the same type as its operands.</p>
4290 <!-- _______________________________________________________________________ -->
4292 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
4299 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4300 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4301 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4302 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4306 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
4307 a specified number of bits.</p>
4310 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
4311 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4312 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4315 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
4316 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
4317 is (statically or dynamically) negative or equal to or larger than the number
4318 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4319 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4320 shift amount in <tt>op2</tt>.</p>
4322 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
4323 <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits. If
4324 the <tt>nsw</tt> keyword is present, then the shift produces a
4325 <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree
4326 with the resultant sign bit. As such, NUW/NSW have the same semantics as
4327 they would if the shift were expressed as a mul instruction with the same
4328 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
4332 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
4333 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
4334 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
4335 <result> = shl i32 1, 32 <i>; undefined</i>
4336 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4341 <!-- _______________________________________________________________________ -->
4343 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4350 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4351 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4355 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4356 operand shifted to the right a specified number of bits with zero fill.</p>
4359 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4360 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4361 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4364 <p>This instruction always performs a logical shift right operation. The most
4365 significant bits of the result will be filled with zero bits after the shift.
4366 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4367 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4368 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4369 shift amount in <tt>op2</tt>.</p>
4371 <p>If the <tt>exact</tt> keyword is present, the result value of the
4372 <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4373 shifted out are non-zero.</p>
4378 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4379 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4380 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4381 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4382 <result> = lshr i32 1, 32 <i>; undefined</i>
4383 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4388 <!-- _______________________________________________________________________ -->
4390 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4397 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4398 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4402 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4403 operand shifted to the right a specified number of bits with sign
4407 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4408 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4409 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4412 <p>This instruction always performs an arithmetic shift right operation, The
4413 most significant bits of the result will be filled with the sign bit
4414 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4415 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4416 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4417 the corresponding shift amount in <tt>op2</tt>.</p>
4419 <p>If the <tt>exact</tt> keyword is present, the result value of the
4420 <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4421 shifted out are non-zero.</p>
4425 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4426 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4427 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4428 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4429 <result> = ashr i32 1, 32 <i>; undefined</i>
4430 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4435 <!-- _______________________________________________________________________ -->
4437 <a name="i_and">'<tt>and</tt>' Instruction</a>
4444 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4448 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4452 <p>The two arguments to the '<tt>and</tt>' instruction must be
4453 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4454 values. Both arguments must have identical types.</p>
4457 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4459 <table border="1" cellspacing="0" cellpadding="4">
4491 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4492 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4493 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4496 <!-- _______________________________________________________________________ -->
4498 <a name="i_or">'<tt>or</tt>' Instruction</a>
4505 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4509 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4513 <p>The two arguments to the '<tt>or</tt>' instruction must be
4514 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4515 values. Both arguments must have identical types.</p>
4518 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4520 <table border="1" cellspacing="0" cellpadding="4">
4552 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4553 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4554 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4559 <!-- _______________________________________________________________________ -->
4561 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4568 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4572 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4573 its two operands. The <tt>xor</tt> is used to implement the "one's
4574 complement" operation, which is the "~" operator in C.</p>
4577 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4578 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4579 values. Both arguments must have identical types.</p>
4582 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4584 <table border="1" cellspacing="0" cellpadding="4">
4616 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4617 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4618 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4619 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4626 <!-- ======================================================================= -->
4628 <a name="vectorops">Vector Operations</a>
4633 <p>LLVM supports several instructions to represent vector operations in a
4634 target-independent manner. These instructions cover the element-access and
4635 vector-specific operations needed to process vectors effectively. While LLVM
4636 does directly support these vector operations, many sophisticated algorithms
4637 will want to use target-specific intrinsics to take full advantage of a
4638 specific target.</p>
4640 <!-- _______________________________________________________________________ -->
4642 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4649 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4653 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4654 from a vector at a specified index.</p>
4658 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4659 of <a href="#t_vector">vector</a> type. The second operand is an index
4660 indicating the position from which to extract the element. The index may be
4664 <p>The result is a scalar of the same type as the element type of
4665 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4666 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4667 results are undefined.</p>
4671 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4676 <!-- _______________________________________________________________________ -->
4678 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4685 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4689 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4690 vector at a specified index.</p>
4693 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4694 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4695 whose type must equal the element type of the first operand. The third
4696 operand is an index indicating the position at which to insert the value.
4697 The index may be a variable.</p>
4700 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4701 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4702 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4703 results are undefined.</p>
4707 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4712 <!-- _______________________________________________________________________ -->
4714 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4721 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4725 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4726 from two input vectors, returning a vector with the same element type as the
4727 input and length that is the same as the shuffle mask.</p>
4730 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4731 with types that match each other. The third argument is a shuffle mask whose
4732 element type is always 'i32'. The result of the instruction is a vector
4733 whose length is the same as the shuffle mask and whose element type is the
4734 same as the element type of the first two operands.</p>
4736 <p>The shuffle mask operand is required to be a constant vector with either
4737 constant integer or undef values.</p>
4740 <p>The elements of the two input vectors are numbered from left to right across
4741 both of the vectors. The shuffle mask operand specifies, for each element of
4742 the result vector, which element of the two input vectors the result element
4743 gets. The element selector may be undef (meaning "don't care") and the
4744 second operand may be undef if performing a shuffle from only one vector.</p>
4748 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4749 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4750 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4751 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4752 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4753 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4754 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4755 <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>
4762 <!-- ======================================================================= -->
4764 <a name="aggregateops">Aggregate Operations</a>
4769 <p>LLVM supports several instructions for working with
4770 <a href="#t_aggregate">aggregate</a> values.</p>
4772 <!-- _______________________________________________________________________ -->
4774 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4781 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4785 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4786 from an <a href="#t_aggregate">aggregate</a> value.</p>
4789 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4790 of <a href="#t_struct">struct</a> or
4791 <a href="#t_array">array</a> type. The operands are constant indices to
4792 specify which value to extract in a similar manner as indices in a
4793 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4794 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4796 <li>Since the value being indexed is not a pointer, the first index is
4797 omitted and assumed to be zero.</li>
4798 <li>At least one index must be specified.</li>
4799 <li>Not only struct indices but also array indices must be in
4804 <p>The result is the value at the position in the aggregate specified by the
4809 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4814 <!-- _______________________________________________________________________ -->
4816 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4823 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4827 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4828 in an <a href="#t_aggregate">aggregate</a> value.</p>
4831 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4832 of <a href="#t_struct">struct</a> or
4833 <a href="#t_array">array</a> type. The second operand is a first-class
4834 value to insert. The following operands are constant indices indicating
4835 the position at which to insert the value in a similar manner as indices in a
4836 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4837 value to insert must have the same type as the value identified by the
4841 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4842 that of <tt>val</tt> except that the value at the position specified by the
4843 indices is that of <tt>elt</tt>.</p>
4847 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4848 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4849 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4856 <!-- ======================================================================= -->
4858 <a name="memoryops">Memory Access and Addressing Operations</a>
4863 <p>A key design point of an SSA-based representation is how it represents
4864 memory. In LLVM, no memory locations are in SSA form, which makes things
4865 very simple. This section describes how to read, write, and allocate
4868 <!-- _______________________________________________________________________ -->
4870 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4877 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4881 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4882 currently executing function, to be automatically released when this function
4883 returns to its caller. The object is always allocated in the generic address
4884 space (address space zero).</p>
4887 <p>The '<tt>alloca</tt>' instruction
4888 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4889 runtime stack, returning a pointer of the appropriate type to the program.
4890 If "NumElements" is specified, it is the number of elements allocated,
4891 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4892 specified, the value result of the allocation is guaranteed to be aligned to
4893 at least that boundary. If not specified, or if zero, the target can choose
4894 to align the allocation on any convenient boundary compatible with the
4897 <p>'<tt>type</tt>' may be any sized type.</p>
4900 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4901 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4902 memory is automatically released when the function returns. The
4903 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4904 variables that must have an address available. When the function returns
4905 (either with the <tt><a href="#i_ret">ret</a></tt>
4906 or <tt><a href="#i_resume">resume</a></tt> instructions), the memory is
4907 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4908 The order in which memory is allocated (ie., which way the stack grows) is
4915 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4916 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4917 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4918 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4923 <!-- _______________________________________________________________________ -->
4925 <a name="i_load">'<tt>load</tt>' Instruction</a>
4932 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4933 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4934 !<index> = !{ i32 1 }
4938 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4941 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4942 from which to load. The pointer must point to
4943 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4944 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4945 number or order of execution of this <tt>load</tt> with other <a
4946 href="#volatile">volatile operations</a>.</p>
4948 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
4949 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4950 argument. The <code>release</code> and <code>acq_rel</code> orderings are
4951 not valid on <code>load</code> instructions. Atomic loads produce <a
4952 href="#memorymodel">defined</a> results when they may see multiple atomic
4953 stores. The type of the pointee must be an integer type whose bit width
4954 is a power of two greater than or equal to eight and less than or equal
4955 to a target-specific size limit. <code>align</code> must be explicitly
4956 specified on atomic loads, and the load has undefined behavior if the
4957 alignment is not set to a value which is at least the size in bytes of
4958 the pointee. <code>!nontemporal</code> does not have any defined semantics
4959 for atomic loads.</p>
4961 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4962 operation (that is, the alignment of the memory address). A value of 0 or an
4963 omitted <tt>align</tt> argument means that the operation has the preferential
4964 alignment for the target. It is the responsibility of the code emitter to
4965 ensure that the alignment information is correct. Overestimating the
4966 alignment results in undefined behavior. Underestimating the alignment may
4967 produce less efficient code. An alignment of 1 is always safe.</p>
4969 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4970 metatadata name <index> corresponding to a metadata node with
4971 one <tt>i32</tt> entry of value 1. The existence of
4972 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4973 and code generator that this load is not expected to be reused in the cache.
4974 The code generator may select special instructions to save cache bandwidth,
4975 such as the <tt>MOVNT</tt> instruction on x86.</p>
4977 <p>The optional <tt>!invariant.load</tt> metadata must reference a single
4978 metatadata name <index> corresponding to a metadata node with no
4979 entries. The existence of the <tt>!invariant.load</tt> metatadata on the
4980 instruction tells the optimizer and code generator that this load address
4981 points to memory which does not change value during program execution.
4982 The optimizer may then move this load around, for example, by hoisting it
4983 out of loops using loop invariant code motion.</p>
4986 <p>The location of memory pointed to is loaded. If the value being loaded is of
4987 scalar type then the number of bytes read does not exceed the minimum number
4988 of bytes needed to hold all bits of the type. For example, loading an
4989 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4990 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4991 is undefined if the value was not originally written using a store of the
4996 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4997 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4998 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
5003 <!-- _______________________________________________________________________ -->
5005 <a name="i_store">'<tt>store</tt>' Instruction</a>
5012 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
5013 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
5017 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
5020 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
5021 and an address at which to store it. The type of the
5022 '<tt><pointer></tt>' operand must be a pointer to
5023 the <a href="#t_firstclass">first class</a> type of the
5024 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
5025 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
5026 order of execution of this <tt>store</tt> with other <a
5027 href="#volatile">volatile operations</a>.</p>
5029 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
5030 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5031 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
5032 valid on <code>store</code> instructions. Atomic loads produce <a
5033 href="#memorymodel">defined</a> results when they may see multiple atomic
5034 stores. The type of the pointee must be an integer type whose bit width
5035 is a power of two greater than or equal to eight and less than or equal
5036 to a target-specific size limit. <code>align</code> must be explicitly
5037 specified on atomic stores, and the store has undefined behavior if the
5038 alignment is not set to a value which is at least the size in bytes of
5039 the pointee. <code>!nontemporal</code> does not have any defined semantics
5040 for atomic stores.</p>
5042 <p>The optional constant "align" argument specifies the alignment of the
5043 operation (that is, the alignment of the memory address). A value of 0 or an
5044 omitted "align" argument means that the operation has the preferential
5045 alignment for the target. It is the responsibility of the code emitter to
5046 ensure that the alignment information is correct. Overestimating the
5047 alignment results in an undefined behavior. Underestimating the alignment may
5048 produce less efficient code. An alignment of 1 is always safe.</p>
5050 <p>The optional !nontemporal metadata must reference a single metatadata
5051 name <index> corresponding to a metadata node with one i32 entry of
5052 value 1. The existence of the !nontemporal metatadata on the
5053 instruction tells the optimizer and code generator that this load is
5054 not expected to be reused in the cache. The code generator may
5055 select special instructions to save cache bandwidth, such as the
5056 MOVNT instruction on x86.</p>
5060 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
5061 location specified by the '<tt><pointer></tt>' operand. If
5062 '<tt><value></tt>' is of scalar type then the number of bytes written
5063 does not exceed the minimum number of bytes needed to hold all bits of the
5064 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
5065 writing a value of a type like <tt>i20</tt> with a size that is not an
5066 integral number of bytes, it is unspecified what happens to the extra bits
5067 that do not belong to the type, but they will typically be overwritten.</p>
5071 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5072 store i32 3, i32* %ptr <i>; yields {void}</i>
5073 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
5078 <!-- _______________________________________________________________________ -->
5080 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
5087 fence [singlethread] <ordering> <i>; yields {void}</i>
5091 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
5092 between operations.</p>
5094 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
5095 href="#ordering">ordering</a> argument which defines what
5096 <i>synchronizes-with</i> edges they add. They can only be given
5097 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
5098 <code>seq_cst</code> orderings.</p>
5101 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
5102 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
5103 <code>acquire</code> ordering semantics if and only if there exist atomic
5104 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
5105 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
5106 <var>X</var> modifies <var>M</var> (either directly or through some side effect
5107 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
5108 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
5109 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
5110 than an explicit <code>fence</code>, one (but not both) of the atomic operations
5111 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
5112 <code>acquire</code> (resp.) ordering constraint and still
5113 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
5114 <i>happens-before</i> edge.</p>
5116 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
5117 having both <code>acquire</code> and <code>release</code> semantics specified
5118 above, participates in the global program order of other <code>seq_cst</code>
5119 operations and/or fences.</p>
5121 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
5122 specifies that the fence only synchronizes with other fences in the same
5123 thread. (This is useful for interacting with signal handlers.)</p>
5127 fence acquire <i>; yields {void}</i>
5128 fence singlethread seq_cst <i>; yields {void}</i>
5133 <!-- _______________________________________________________________________ -->
5135 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
5142 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
5146 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
5147 It loads a value in memory and compares it to a given value. If they are
5148 equal, it stores a new value into the memory.</p>
5151 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
5152 address to operate on, a value to compare to the value currently be at that
5153 address, and a new value to place at that address if the compared values are
5154 equal. The type of '<var><cmp></var>' must be an integer type whose
5155 bit width is a power of two greater than or equal to eight and less than
5156 or equal to a target-specific size limit. '<var><cmp></var>' and
5157 '<var><new></var>' must have the same type, and the type of
5158 '<var><pointer></var>' must be a pointer to that type. If the
5159 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
5160 optimizer is not allowed to modify the number or order of execution
5161 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
5164 <!-- FIXME: Extend allowed types. -->
5166 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
5167 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
5169 <p>The optional "<code>singlethread</code>" argument declares that the
5170 <code>cmpxchg</code> is only atomic with respect to code (usually signal
5171 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
5172 cmpxchg is atomic with respect to all other code in the system.</p>
5174 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
5175 the size in memory of the operand.
5178 <p>The contents of memory at the location specified by the
5179 '<tt><pointer></tt>' operand is read and compared to
5180 '<tt><cmp></tt>'; if the read value is the equal,
5181 '<tt><new></tt>' is written. The original value at the location
5184 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
5185 purpose of identifying <a href="#release_sequence">release sequences</a>. A
5186 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
5187 parameter determined by dropping any <code>release</code> part of the
5188 <code>cmpxchg</code>'s ordering.</p>
5191 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
5192 optimization work on ARM.)
5194 FIXME: Is a weaker ordering constraint on failure helpful in practice?
5200 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
5201 <a href="#i_br">br</a> label %loop
5204 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
5205 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
5206 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
5207 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
5208 <a href="#i_br">br</a> i1 %success, label %done, label %loop
5216 <!-- _______________________________________________________________________ -->
5218 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
5225 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
5229 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
5232 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
5233 operation to apply, an address whose value to modify, an argument to the
5234 operation. The operation must be one of the following keywords:</p>
5249 <p>The type of '<var><value></var>' must be an integer type whose
5250 bit width is a power of two greater than or equal to eight and less than
5251 or equal to a target-specific size limit. The type of the
5252 '<code><pointer></code>' operand must be a pointer to that type.
5253 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
5254 optimizer is not allowed to modify the number or order of execution of this
5255 <code>atomicrmw</code> with other <a href="#volatile">volatile
5258 <!-- FIXME: Extend allowed types. -->
5261 <p>The contents of memory at the location specified by the
5262 '<tt><pointer></tt>' operand are atomically read, modified, and written
5263 back. The original value at the location is returned. The modification is
5264 specified by the <var>operation</var> argument:</p>
5267 <li>xchg: <code>*ptr = val</code></li>
5268 <li>add: <code>*ptr = *ptr + val</code></li>
5269 <li>sub: <code>*ptr = *ptr - val</code></li>
5270 <li>and: <code>*ptr = *ptr & val</code></li>
5271 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
5272 <li>or: <code>*ptr = *ptr | val</code></li>
5273 <li>xor: <code>*ptr = *ptr ^ val</code></li>
5274 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
5275 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
5276 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
5277 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
5282 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
5287 <!-- _______________________________________________________________________ -->
5289 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
5296 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5297 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5298 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5302 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
5303 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
5304 It performs address calculation only and does not access memory.</p>
5307 <p>The first argument is always a pointer or a vector of pointers,
5308 and forms the basis of the
5309 calculation. The remaining arguments are indices that indicate which of the
5310 elements of the aggregate object are indexed. The interpretation of each
5311 index is dependent on the type being indexed into. The first index always
5312 indexes the pointer value given as the first argument, the second index
5313 indexes a value of the type pointed to (not necessarily the value directly
5314 pointed to, since the first index can be non-zero), etc. The first type
5315 indexed into must be a pointer value, subsequent types can be arrays,
5316 vectors, and structs. Note that subsequent types being indexed into
5317 can never be pointers, since that would require loading the pointer before
5318 continuing calculation.</p>
5320 <p>The type of each index argument depends on the type it is indexing into.
5321 When indexing into a (optionally packed) structure, only <tt>i32</tt>
5322 integer <b>constants</b> are allowed. When indexing into an array, pointer
5323 or vector, integers of any width are allowed, and they are not required to be
5324 constant. These integers are treated as signed values where relevant.</p>
5326 <p>For example, let's consider a C code fragment and how it gets compiled to
5329 <pre class="doc_code">
5341 int *foo(struct ST *s) {
5342 return &s[1].Z.B[5][13];
5346 <p>The LLVM code generated by Clang is:</p>
5348 <pre class="doc_code">
5349 %struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 }
5350 %struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT }
5352 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5354 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5360 <p>In the example above, the first index is indexing into the
5361 '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a
5362 '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a
5363 structure. The second index indexes into the third element of the structure,
5364 yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>'
5365 type, another structure. The third index indexes into the second element of
5366 the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The
5367 two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>'
5368 type. The '<tt>getelementptr</tt>' instruction returns a pointer to this
5369 element, thus computing a value of '<tt>i32*</tt>' type.</p>
5371 <p>Note that it is perfectly legal to index partially through a structure,
5372 returning a pointer to an inner element. Because of this, the LLVM code for
5373 the given testcase is equivalent to:</p>
5375 <pre class="doc_code">
5376 define i32* @foo(%struct.ST* %s) {
5377 %t1 = getelementptr %struct.ST* %s, i32 1 <i>; yields %struct.ST*:%t1</i>
5378 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 <i>; yields %struct.RT*:%t2</i>
5379 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5380 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5381 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5386 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5387 <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the
5388 base pointer is not an <i>in bounds</i> address of an allocated object,
5389 or if any of the addresses that would be formed by successive addition of
5390 the offsets implied by the indices to the base address with infinitely
5391 precise signed arithmetic are not an <i>in bounds</i> address of that
5392 allocated object. The <i>in bounds</i> addresses for an allocated object
5393 are all the addresses that point into the object, plus the address one
5395 In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword
5396 applies to each of the computations element-wise. </p>
5398 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5399 the base address with silently-wrapping two's complement arithmetic. If the
5400 offsets have a different width from the pointer, they are sign-extended or
5401 truncated to the width of the pointer. The result value of the
5402 <tt>getelementptr</tt> may be outside the object pointed to by the base
5403 pointer. The result value may not necessarily be used to access memory
5404 though, even if it happens to point into allocated storage. See the
5405 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5408 <p>The getelementptr instruction is often confusing. For some more insight into
5409 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5413 <i>; yields [12 x i8]*:aptr</i>
5414 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5415 <i>; yields i8*:vptr</i>
5416 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5417 <i>; yields i8*:eptr</i>
5418 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5419 <i>; yields i32*:iptr</i>
5420 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5423 <p>In cases where the pointer argument is a vector of pointers, only a
5424 single index may be used, and the number of vector elements has to be
5425 the same. For example: </p>
5426 <pre class="doc_code">
5427 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5434 <!-- ======================================================================= -->
5436 <a name="convertops">Conversion Operations</a>
5441 <p>The instructions in this category are the conversion instructions (casting)
5442 which all take a single operand and a type. They perform various bit
5443 conversions on the operand.</p>
5445 <!-- _______________________________________________________________________ -->
5447 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5454 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5458 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5459 type <tt>ty2</tt>.</p>
5462 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5463 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5464 of the same number of integers.
5465 The bit size of the <tt>value</tt> must be larger than
5466 the bit size of the destination type, <tt>ty2</tt>.
5467 Equal sized types are not allowed.</p>
5470 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5471 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5472 source size must be larger than the destination size, <tt>trunc</tt> cannot
5473 be a <i>no-op cast</i>. It will always truncate bits.</p>
5477 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5478 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5479 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5480 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5485 <!-- _______________________________________________________________________ -->
5487 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5494 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5498 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5503 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5504 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5505 of the same number of integers.
5506 The bit size of the <tt>value</tt> must be smaller than
5507 the bit size of the destination type,
5511 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5512 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5514 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5518 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5519 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5520 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5525 <!-- _______________________________________________________________________ -->
5527 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5534 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5538 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5541 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5542 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5543 of the same number of integers.
5544 The bit size of the <tt>value</tt> must be smaller than
5545 the bit size of the destination type,
5549 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5550 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5551 of the type <tt>ty2</tt>.</p>
5553 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5557 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5558 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5559 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5564 <!-- _______________________________________________________________________ -->
5566 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5573 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5577 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5581 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5582 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5583 to cast it to. The size of <tt>value</tt> must be larger than the size of
5584 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5585 <i>no-op cast</i>.</p>
5588 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5589 <a href="#t_floating">floating point</a> type to a smaller
5590 <a href="#t_floating">floating point</a> type. If the value cannot fit
5591 within the destination type, <tt>ty2</tt>, then the results are
5596 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5597 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5602 <!-- _______________________________________________________________________ -->
5604 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5611 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5615 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5616 floating point value.</p>
5619 <p>The '<tt>fpext</tt>' instruction takes a
5620 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5621 a <a href="#t_floating">floating point</a> type to cast it to. The source
5622 type must be smaller than the destination type.</p>
5625 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5626 <a href="#t_floating">floating point</a> type to a larger
5627 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5628 used to make a <i>no-op cast</i> because it always changes bits. Use
5629 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5633 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5634 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5639 <!-- _______________________________________________________________________ -->
5641 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5648 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5652 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5653 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5656 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5657 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5658 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5659 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5660 vector integer type with the same number of elements as <tt>ty</tt></p>
5663 <p>The '<tt>fptoui</tt>' instruction converts its
5664 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5665 towards zero) unsigned integer value. If the value cannot fit
5666 in <tt>ty2</tt>, the results are undefined.</p>
5670 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5671 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5672 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5677 <!-- _______________________________________________________________________ -->
5679 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5686 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5690 <p>The '<tt>fptosi</tt>' instruction converts
5691 <a href="#t_floating">floating point</a> <tt>value</tt> to
5692 type <tt>ty2</tt>.</p>
5695 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5696 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5697 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5698 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5699 vector integer type with the same number of elements as <tt>ty</tt></p>
5702 <p>The '<tt>fptosi</tt>' instruction converts its
5703 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5704 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5705 the results are undefined.</p>
5709 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5710 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5711 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5716 <!-- _______________________________________________________________________ -->
5718 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5725 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5729 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5730 integer and converts that value to the <tt>ty2</tt> type.</p>
5733 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5734 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5735 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5736 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5737 floating point type with the same number of elements as <tt>ty</tt></p>
5740 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5741 integer quantity and converts it to the corresponding floating point
5742 value. If the value cannot fit in the floating point value, the results are
5747 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5748 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5753 <!-- _______________________________________________________________________ -->
5755 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5762 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5766 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5767 and converts that value to the <tt>ty2</tt> type.</p>
5770 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5771 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5772 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5773 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5774 floating point type with the same number of elements as <tt>ty</tt></p>
5777 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5778 quantity and converts it to the corresponding floating point value. If the
5779 value cannot fit in the floating point value, the results are undefined.</p>
5783 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5784 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5789 <!-- _______________________________________________________________________ -->
5791 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5798 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5802 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of
5803 pointers <tt>value</tt> to
5804 the integer (or vector of integers) type <tt>ty2</tt>.</p>
5807 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5808 must be a a value of type <a href="#t_pointer">pointer</a> or a vector of
5809 pointers, and a type to cast it to
5810 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector
5811 of integers type.</p>
5814 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5815 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5816 truncating or zero extending that value to the size of the integer type. If
5817 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5818 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5819 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5824 %X = ptrtoint i32* %P to i8 <i>; yields truncation on 32-bit architecture</i>
5825 %Y = ptrtoint i32* %P to i64 <i>; yields zero extension on 32-bit architecture</i>
5826 %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>
5831 <!-- _______________________________________________________________________ -->
5833 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5840 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5844 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5845 pointer type, <tt>ty2</tt>.</p>
5848 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5849 value to cast, and a type to cast it to, which must be a
5850 <a href="#t_pointer">pointer</a> type.</p>
5853 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5854 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5855 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5856 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5857 than the size of a pointer then a zero extension is done. If they are the
5858 same size, nothing is done (<i>no-op cast</i>).</p>
5862 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5863 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5864 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5865 %Z = inttoptr <4 x i32> %G to <4 x i8*><i>; yields truncation of vector G to four pointers</i>
5870 <!-- _______________________________________________________________________ -->
5872 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5879 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5883 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5884 <tt>ty2</tt> without changing any bits.</p>
5887 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5888 non-aggregate first class value, and a type to cast it to, which must also be
5889 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5890 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5891 identical. If the source type is a pointer, the destination type must also be
5892 a pointer. This instruction supports bitwise conversion of vectors to
5893 integers and to vectors of other types (as long as they have the same
5897 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5898 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5899 this conversion. The conversion is done as if the <tt>value</tt> had been
5900 stored to memory and read back as type <tt>ty2</tt>.
5901 Pointer (or vector of pointers) types may only be converted to other pointer
5902 (or vector of pointers) types with this instruction. To convert
5903 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5904 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5908 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
5909 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
5910 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
5911 %Z = bitcast <2 x i32*> %V to <2 x i64*> <i>; yields <2 x i64*></i>
5918 <!-- ======================================================================= -->
5920 <a name="otherops">Other Operations</a>
5925 <p>The instructions in this category are the "miscellaneous" instructions, which
5926 defy better classification.</p>
5928 <!-- _______________________________________________________________________ -->
5930 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5937 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5941 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5942 boolean values based on comparison of its two integer, integer vector,
5943 pointer, or pointer vector operands.</p>
5946 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
5947 the condition code indicating the kind of comparison to perform. It is not a
5948 value, just a keyword. The possible condition code are:</p>
5951 <li><tt>eq</tt>: equal</li>
5952 <li><tt>ne</tt>: not equal </li>
5953 <li><tt>ugt</tt>: unsigned greater than</li>
5954 <li><tt>uge</tt>: unsigned greater or equal</li>
5955 <li><tt>ult</tt>: unsigned less than</li>
5956 <li><tt>ule</tt>: unsigned less or equal</li>
5957 <li><tt>sgt</tt>: signed greater than</li>
5958 <li><tt>sge</tt>: signed greater or equal</li>
5959 <li><tt>slt</tt>: signed less than</li>
5960 <li><tt>sle</tt>: signed less or equal</li>
5963 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
5964 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
5965 typed. They must also be identical types.</p>
5968 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
5969 condition code given as <tt>cond</tt>. The comparison performed always yields
5970 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
5971 result, as follows:</p>
5974 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
5975 <tt>false</tt> otherwise. No sign interpretation is necessary or
5978 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
5979 <tt>false</tt> otherwise. No sign interpretation is necessary or
5982 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
5983 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5985 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
5986 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5987 to <tt>op2</tt>.</li>
5989 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
5990 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5992 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
5993 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5995 <li><tt>sgt</tt>: interprets the operands as signed values and yields
5996 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5998 <li><tt>sge</tt>: interprets the operands as signed values and yields
5999 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6000 to <tt>op2</tt>.</li>
6002 <li><tt>slt</tt>: interprets the operands as signed values and yields
6003 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6005 <li><tt>sle</tt>: interprets the operands as signed values and yields
6006 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6009 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
6010 values are compared as if they were integers.</p>
6012 <p>If the operands are integer vectors, then they are compared element by
6013 element. The result is an <tt>i1</tt> vector with the same number of elements
6014 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
6018 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
6019 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
6020 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
6021 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
6022 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
6023 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
6026 <p>Note that the code generator does not yet support vector types with
6027 the <tt>icmp</tt> instruction.</p>
6031 <!-- _______________________________________________________________________ -->
6033 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
6040 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
6044 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
6045 values based on comparison of its operands.</p>
6047 <p>If the operands are floating point scalars, then the result type is a boolean
6048 (<a href="#t_integer"><tt>i1</tt></a>).</p>
6050 <p>If the operands are floating point vectors, then the result type is a vector
6051 of boolean with the same number of elements as the operands being
6055 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
6056 the condition code indicating the kind of comparison to perform. It is not a
6057 value, just a keyword. The possible condition code are:</p>
6060 <li><tt>false</tt>: no comparison, always returns false</li>
6061 <li><tt>oeq</tt>: ordered and equal</li>
6062 <li><tt>ogt</tt>: ordered and greater than </li>
6063 <li><tt>oge</tt>: ordered and greater than or equal</li>
6064 <li><tt>olt</tt>: ordered and less than </li>
6065 <li><tt>ole</tt>: ordered and less than or equal</li>
6066 <li><tt>one</tt>: ordered and not equal</li>
6067 <li><tt>ord</tt>: ordered (no nans)</li>
6068 <li><tt>ueq</tt>: unordered or equal</li>
6069 <li><tt>ugt</tt>: unordered or greater than </li>
6070 <li><tt>uge</tt>: unordered or greater than or equal</li>
6071 <li><tt>ult</tt>: unordered or less than </li>
6072 <li><tt>ule</tt>: unordered or less than or equal</li>
6073 <li><tt>une</tt>: unordered or not equal</li>
6074 <li><tt>uno</tt>: unordered (either nans)</li>
6075 <li><tt>true</tt>: no comparison, always returns true</li>
6078 <p><i>Ordered</i> means that neither operand is a QNAN while
6079 <i>unordered</i> means that either operand may be a QNAN.</p>
6081 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
6082 a <a href="#t_floating">floating point</a> type or
6083 a <a href="#t_vector">vector</a> of floating point type. They must have
6084 identical types.</p>
6087 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
6088 according to the condition code given as <tt>cond</tt>. If the operands are
6089 vectors, then the vectors are compared element by element. Each comparison
6090 performed always yields an <a href="#t_integer">i1</a> result, as
6094 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
6096 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6097 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6099 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6100 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6102 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6103 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6105 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6106 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6108 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6109 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6111 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6112 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6114 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
6116 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
6117 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6119 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
6120 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6122 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
6123 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6125 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
6126 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6128 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
6129 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6131 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
6132 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6134 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
6136 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
6141 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
6142 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
6143 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
6144 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
6147 <p>Note that the code generator does not yet support vector types with
6148 the <tt>fcmp</tt> instruction.</p>
6152 <!-- _______________________________________________________________________ -->
6154 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
6161 <result> = phi <ty> [ <val0>, <label0>], ...
6165 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
6166 SSA graph representing the function.</p>
6169 <p>The type of the incoming values is specified with the first type field. After
6170 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
6171 one pair for each predecessor basic block of the current block. Only values
6172 of <a href="#t_firstclass">first class</a> type may be used as the value
6173 arguments to the PHI node. Only labels may be used as the label
6176 <p>There must be no non-phi instructions between the start of a basic block and
6177 the PHI instructions: i.e. PHI instructions must be first in a basic
6180 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
6181 occur on the edge from the corresponding predecessor block to the current
6182 block (but after any definition of an '<tt>invoke</tt>' instruction's return
6183 value on the same edge).</p>
6186 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
6187 specified by the pair corresponding to the predecessor basic block that
6188 executed just prior to the current block.</p>
6192 Loop: ; Infinite loop that counts from 0 on up...
6193 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6194 %nextindvar = add i32 %indvar, 1
6200 <!-- _______________________________________________________________________ -->
6202 <a name="i_select">'<tt>select</tt>' Instruction</a>
6209 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
6211 <i>selty</i> is either i1 or {<N x i1>}
6215 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
6216 condition, without branching.</p>
6220 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
6221 values indicating the condition, and two values of the
6222 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
6223 vectors and the condition is a scalar, then entire vectors are selected, not
6224 individual elements.</p>
6227 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
6228 first value argument; otherwise, it returns the second value argument.</p>
6230 <p>If the condition is a vector of i1, then the value arguments must be vectors
6231 of the same size, and the selection is done element by element.</p>
6235 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
6240 <!-- _______________________________________________________________________ -->
6242 <a name="i_call">'<tt>call</tt>' Instruction</a>
6249 <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>]
6253 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
6256 <p>This instruction requires several arguments:</p>
6259 <li>The optional "tail" marker indicates that the callee function does not
6260 access any allocas or varargs in the caller. Note that calls may be
6261 marked "tail" even if they do not occur before
6262 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
6263 present, the function call is eligible for tail call optimization,
6264 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
6265 optimized into a jump</a>. The code generator may optimize calls marked
6266 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
6267 sibling call optimization</a> when the caller and callee have
6268 matching signatures, or 2) forced tail call optimization when the
6269 following extra requirements are met:
6271 <li>Caller and callee both have the calling
6272 convention <tt>fastcc</tt>.</li>
6273 <li>The call is in tail position (ret immediately follows call and ret
6274 uses value of call or is void).</li>
6275 <li>Option <tt>-tailcallopt</tt> is enabled,
6276 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
6277 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
6278 constraints are met.</a></li>
6282 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
6283 convention</a> the call should use. If none is specified, the call
6284 defaults to using C calling conventions. The calling convention of the
6285 call must match the calling convention of the target function, or else the
6286 behavior is undefined.</li>
6288 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
6289 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
6290 '<tt>inreg</tt>' attributes are valid here.</li>
6292 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
6293 type of the return value. Functions that return no value are marked
6294 <tt><a href="#t_void">void</a></tt>.</li>
6296 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
6297 being invoked. The argument types must match the types implied by this
6298 signature. This type can be omitted if the function is not varargs and if
6299 the function type does not return a pointer to a function.</li>
6301 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
6302 be invoked. In most cases, this is a direct function invocation, but
6303 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
6304 to function value.</li>
6306 <li>'<tt>function args</tt>': argument list whose types match the function
6307 signature argument types and parameter attributes. All arguments must be
6308 of <a href="#t_firstclass">first class</a> type. If the function
6309 signature indicates the function accepts a variable number of arguments,
6310 the extra arguments can be specified.</li>
6312 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
6313 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
6314 '<tt>readnone</tt>' attributes are valid here.</li>
6318 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
6319 a specified function, with its incoming arguments bound to the specified
6320 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
6321 function, control flow continues with the instruction after the function
6322 call, and the return value of the function is bound to the result
6327 %retval = call i32 @test(i32 %argc)
6328 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
6329 %X = tail call i32 @foo() <i>; yields i32</i>
6330 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
6331 call void %foo(i8 97 signext)
6333 %struct.A = type { i32, i8 }
6334 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
6335 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
6336 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
6337 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
6338 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
6341 <p>llvm treats calls to some functions with names and arguments that match the
6342 standard C99 library as being the C99 library functions, and may perform
6343 optimizations or generate code for them under that assumption. This is
6344 something we'd like to change in the future to provide better support for
6345 freestanding environments and non-C-based languages.</p>
6349 <!-- _______________________________________________________________________ -->
6351 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
6358 <resultval> = va_arg <va_list*> <arglist>, <argty>
6362 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6363 the "variable argument" area of a function call. It is used to implement the
6364 <tt>va_arg</tt> macro in C.</p>
6367 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6368 argument. It returns a value of the specified argument type and increments
6369 the <tt>va_list</tt> to point to the next argument. The actual type
6370 of <tt>va_list</tt> is target specific.</p>
6373 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6374 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6375 to the next argument. For more information, see the variable argument
6376 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6378 <p>It is legal for this instruction to be called in a function which does not
6379 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6382 <p><tt>va_arg</tt> is an LLVM instruction instead of
6383 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6387 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6389 <p>Note that the code generator does not yet fully support va_arg on many
6390 targets. Also, it does not currently support va_arg with aggregate types on
6395 <!-- _______________________________________________________________________ -->
6397 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6404 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6405 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6407 <clause> := catch <type> <value>
6408 <clause> := filter <array constant type> <array constant>
6412 <p>The '<tt>landingpad</tt>' instruction is used by
6413 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6414 system</a> to specify that a basic block is a landing pad — one where
6415 the exception lands, and corresponds to the code found in the
6416 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6417 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6418 re-entry to the function. The <tt>resultval</tt> has the
6419 type <tt>resultty</tt>.</p>
6422 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6423 function associated with the unwinding mechanism. The optional
6424 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6426 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6427 or <tt>filter</tt> — and contains the global variable representing the
6428 "type" that may be caught or filtered respectively. Unlike the
6429 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6430 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6431 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6432 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6435 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6436 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6437 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6438 calling conventions, how the personality function results are represented in
6439 LLVM IR is target specific.</p>
6441 <p>The clauses are applied in order from top to bottom. If two
6442 <tt>landingpad</tt> instructions are merged together through inlining, the
6443 clauses from the calling function are appended to the list of clauses.
6444 When the call stack is being unwound due to an exception being thrown, the
6445 exception is compared against each <tt>clause</tt> in turn. If it doesn't
6446 match any of the clauses, and the <tt>cleanup</tt> flag is not set, then
6447 unwinding continues further up the call stack.</p>
6449 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6452 <li>A landing pad block is a basic block which is the unwind destination of an
6453 '<tt>invoke</tt>' instruction.</li>
6454 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6455 first non-PHI instruction.</li>
6456 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6458 <li>A basic block that is not a landing pad block may not include a
6459 '<tt>landingpad</tt>' instruction.</li>
6460 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6461 personality function.</li>
6466 ;; A landing pad which can catch an integer.
6467 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6469 ;; A landing pad that is a cleanup.
6470 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6472 ;; A landing pad which can catch an integer and can only throw a double.
6473 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6475 filter [1 x i8**] [@_ZTId]
6484 <!-- *********************************************************************** -->
6485 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6486 <!-- *********************************************************************** -->
6490 <p>LLVM supports the notion of an "intrinsic function". These functions have
6491 well known names and semantics and are required to follow certain
6492 restrictions. Overall, these intrinsics represent an extension mechanism for
6493 the LLVM language that does not require changing all of the transformations
6494 in LLVM when adding to the language (or the bitcode reader/writer, the
6495 parser, etc...).</p>
6497 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6498 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6499 begin with this prefix. Intrinsic functions must always be external
6500 functions: you cannot define the body of intrinsic functions. Intrinsic
6501 functions may only be used in call or invoke instructions: it is illegal to
6502 take the address of an intrinsic function. Additionally, because intrinsic
6503 functions are part of the LLVM language, it is required if any are added that
6504 they be documented here.</p>
6506 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6507 family of functions that perform the same operation but on different data
6508 types. Because LLVM can represent over 8 million different integer types,
6509 overloading is used commonly to allow an intrinsic function to operate on any
6510 integer type. One or more of the argument types or the result type can be
6511 overloaded to accept any integer type. Argument types may also be defined as
6512 exactly matching a previous argument's type or the result type. This allows
6513 an intrinsic function which accepts multiple arguments, but needs all of them
6514 to be of the same type, to only be overloaded with respect to a single
6515 argument or the result.</p>
6517 <p>Overloaded intrinsics will have the names of its overloaded argument types
6518 encoded into its function name, each preceded by a period. Only those types
6519 which are overloaded result in a name suffix. Arguments whose type is matched
6520 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6521 can take an integer of any width and returns an integer of exactly the same
6522 integer width. This leads to a family of functions such as
6523 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6524 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6525 suffix is required. Because the argument's type is matched against the return
6526 type, it does not require its own name suffix.</p>
6528 <p>To learn how to add an intrinsic function, please see the
6529 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6531 <!-- ======================================================================= -->
6533 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6538 <p>Variable argument support is defined in LLVM with
6539 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6540 intrinsic functions. These functions are related to the similarly named
6541 macros defined in the <tt><stdarg.h></tt> header file.</p>
6543 <p>All of these functions operate on arguments that use a target-specific value
6544 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6545 not define what this type is, so all transformations should be prepared to
6546 handle these functions regardless of the type used.</p>
6548 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6549 instruction and the variable argument handling intrinsic functions are
6552 <pre class="doc_code">
6553 define i32 @test(i32 %X, ...) {
6554 ; Initialize variable argument processing
6556 %ap2 = bitcast i8** %ap to i8*
6557 call void @llvm.va_start(i8* %ap2)
6559 ; Read a single integer argument
6560 %tmp = va_arg i8** %ap, i32
6562 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6564 %aq2 = bitcast i8** %aq to i8*
6565 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6566 call void @llvm.va_end(i8* %aq2)
6568 ; Stop processing of arguments.
6569 call void @llvm.va_end(i8* %ap2)
6573 declare void @llvm.va_start(i8*)
6574 declare void @llvm.va_copy(i8*, i8*)
6575 declare void @llvm.va_end(i8*)
6578 <!-- _______________________________________________________________________ -->
6580 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6588 declare void %llvm.va_start(i8* <arglist>)
6592 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6593 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6596 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6599 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6600 macro available in C. In a target-dependent way, it initializes
6601 the <tt>va_list</tt> element to which the argument points, so that the next
6602 call to <tt>va_arg</tt> will produce the first variable argument passed to
6603 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6604 need to know the last argument of the function as the compiler can figure
6609 <!-- _______________________________________________________________________ -->
6611 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6618 declare void @llvm.va_end(i8* <arglist>)
6622 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6623 which has been initialized previously
6624 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6625 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6628 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6631 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6632 macro available in C. In a target-dependent way, it destroys
6633 the <tt>va_list</tt> element to which the argument points. Calls
6634 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6635 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6636 with calls to <tt>llvm.va_end</tt>.</p>
6640 <!-- _______________________________________________________________________ -->
6642 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6649 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6653 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6654 from the source argument list to the destination argument list.</p>
6657 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6658 The second argument is a pointer to a <tt>va_list</tt> element to copy
6662 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6663 macro available in C. In a target-dependent way, it copies the
6664 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6665 element. This intrinsic is necessary because
6666 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6667 arbitrarily complex and require, for example, memory allocation.</p>
6673 <!-- ======================================================================= -->
6675 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6680 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6681 Collection</a> (GC) requires the implementation and generation of these
6682 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6683 roots on the stack</a>, as well as garbage collector implementations that
6684 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6685 barriers. Front-ends for type-safe garbage collected languages should generate
6686 these intrinsics to make use of the LLVM garbage collectors. For more details,
6687 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6690 <p>The garbage collection intrinsics only operate on objects in the generic
6691 address space (address space zero).</p>
6693 <!-- _______________________________________________________________________ -->
6695 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6702 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6706 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6707 the code generator, and allows some metadata to be associated with it.</p>
6710 <p>The first argument specifies the address of a stack object that contains the
6711 root pointer. The second pointer (which must be either a constant or a
6712 global value address) contains the meta-data to be associated with the
6716 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6717 location. At compile-time, the code generator generates information to allow
6718 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6719 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6724 <!-- _______________________________________________________________________ -->
6726 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6733 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6737 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6738 locations, allowing garbage collector implementations that require read
6742 <p>The second argument is the address to read from, which should be an address
6743 allocated from the garbage collector. The first object is a pointer to the
6744 start of the referenced object, if needed by the language runtime (otherwise
6748 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6749 instruction, but may be replaced with substantially more complex code by the
6750 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6751 may only be used in a function which <a href="#gc">specifies a GC
6756 <!-- _______________________________________________________________________ -->
6758 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6765 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6769 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6770 locations, allowing garbage collector implementations that require write
6771 barriers (such as generational or reference counting collectors).</p>
6774 <p>The first argument is the reference to store, the second is the start of the
6775 object to store it to, and the third is the address of the field of Obj to
6776 store to. If the runtime does not require a pointer to the object, Obj may
6780 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6781 instruction, but may be replaced with substantially more complex code by the
6782 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6783 may only be used in a function which <a href="#gc">specifies a GC
6790 <!-- ======================================================================= -->
6792 <a name="int_codegen">Code Generator Intrinsics</a>
6797 <p>These intrinsics are provided by LLVM to expose special features that may
6798 only be implemented with code generator support.</p>
6800 <!-- _______________________________________________________________________ -->
6802 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6809 declare i8 *@llvm.returnaddress(i32 <level>)
6813 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6814 target-specific value indicating the return address of the current function
6815 or one of its callers.</p>
6818 <p>The argument to this intrinsic indicates which function to return the address
6819 for. Zero indicates the calling function, one indicates its caller, etc.
6820 The argument is <b>required</b> to be a constant integer value.</p>
6823 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6824 indicating the return address of the specified call frame, or zero if it
6825 cannot be identified. The value returned by this intrinsic is likely to be
6826 incorrect or 0 for arguments other than zero, so it should only be used for
6827 debugging purposes.</p>
6829 <p>Note that calling this intrinsic does not prevent function inlining or other
6830 aggressive transformations, so the value returned may not be that of the
6831 obvious source-language caller.</p>
6835 <!-- _______________________________________________________________________ -->
6837 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6844 declare i8* @llvm.frameaddress(i32 <level>)
6848 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6849 target-specific frame pointer value for the specified stack frame.</p>
6852 <p>The argument to this intrinsic indicates which function to return the frame
6853 pointer for. Zero indicates the calling function, one indicates its caller,
6854 etc. The argument is <b>required</b> to be a constant integer value.</p>
6857 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6858 indicating the frame address of the specified call frame, or zero if it
6859 cannot be identified. The value returned by this intrinsic is likely to be
6860 incorrect or 0 for arguments other than zero, so it should only be used for
6861 debugging purposes.</p>
6863 <p>Note that calling this intrinsic does not prevent function inlining or other
6864 aggressive transformations, so the value returned may not be that of the
6865 obvious source-language caller.</p>
6869 <!-- _______________________________________________________________________ -->
6871 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6878 declare i8* @llvm.stacksave()
6882 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6883 of the function stack, for use
6884 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6885 useful for implementing language features like scoped automatic variable
6886 sized arrays in C99.</p>
6889 <p>This intrinsic returns a opaque pointer value that can be passed
6890 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6891 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6892 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6893 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6894 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6895 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6899 <!-- _______________________________________________________________________ -->
6901 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6908 declare void @llvm.stackrestore(i8* %ptr)
6912 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6913 the function stack to the state it was in when the
6914 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6915 executed. This is useful for implementing language features like scoped
6916 automatic variable sized arrays in C99.</p>
6919 <p>See the description
6920 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6924 <!-- _______________________________________________________________________ -->
6926 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6933 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6937 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6938 insert a prefetch instruction if supported; otherwise, it is a noop.
6939 Prefetches have no effect on the behavior of the program but can change its
6940 performance characteristics.</p>
6943 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6944 specifier determining if the fetch should be for a read (0) or write (1),
6945 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
6946 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6947 specifies whether the prefetch is performed on the data (1) or instruction (0)
6948 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
6949 must be constant integers.</p>
6952 <p>This intrinsic does not modify the behavior of the program. In particular,
6953 prefetches cannot trap and do not produce a value. On targets that support
6954 this intrinsic, the prefetch can provide hints to the processor cache for
6955 better performance.</p>
6959 <!-- _______________________________________________________________________ -->
6961 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
6968 declare void @llvm.pcmarker(i32 <id>)
6972 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
6973 Counter (PC) in a region of code to simulators and other tools. The method
6974 is target specific, but it is expected that the marker will use exported
6975 symbols to transmit the PC of the marker. The marker makes no guarantees
6976 that it will remain with any specific instruction after optimizations. It is
6977 possible that the presence of a marker will inhibit optimizations. The
6978 intended use is to be inserted after optimizations to allow correlations of
6979 simulation runs.</p>
6982 <p><tt>id</tt> is a numerical id identifying the marker.</p>
6985 <p>This intrinsic does not modify the behavior of the program. Backends that do
6986 not support this intrinsic may ignore it.</p>
6990 <!-- _______________________________________________________________________ -->
6992 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
6999 declare i64 @llvm.readcyclecounter()
7003 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
7004 counter register (or similar low latency, high accuracy clocks) on those
7005 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7006 should map to RPCC. As the backing counters overflow quickly (on the order
7007 of 9 seconds on alpha), this should only be used for small timings.</p>
7010 <p>When directly supported, reading the cycle counter should not modify any
7011 memory. Implementations are allowed to either return a application specific
7012 value or a system wide value. On backends without support, this is lowered
7013 to a constant 0.</p>
7019 <!-- ======================================================================= -->
7021 <a name="int_libc">Standard C Library Intrinsics</a>
7026 <p>LLVM provides intrinsics for a few important standard C library functions.
7027 These intrinsics allow source-language front-ends to pass information about
7028 the alignment of the pointer arguments to the code generator, providing
7029 opportunity for more efficient code generation.</p>
7031 <!-- _______________________________________________________________________ -->
7033 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
7039 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
7040 integer bit width and for different address spaces. Not all targets support
7041 all bit widths however.</p>
7044 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7045 i32 <len>, i32 <align>, i1 <isvolatile>)
7046 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7047 i64 <len>, i32 <align>, i1 <isvolatile>)
7051 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7052 source location to the destination location.</p>
7054 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
7055 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7056 and the pointers can be in specified address spaces.</p>
7060 <p>The first argument is a pointer to the destination, the second is a pointer
7061 to the source. The third argument is an integer argument specifying the
7062 number of bytes to copy, the fourth argument is the alignment of the
7063 source and destination locations, and the fifth is a boolean indicating a
7064 volatile access.</p>
7066 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7067 then the caller guarantees that both the source and destination pointers are
7068 aligned to that boundary.</p>
7070 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7071 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
7072 The detailed access behavior is not very cleanly specified and it is unwise
7073 to depend on it.</p>
7077 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7078 source location to the destination location, which are not allowed to
7079 overlap. It copies "len" bytes of memory over. If the argument is known to
7080 be aligned to some boundary, this can be specified as the fourth argument,
7081 otherwise it should be set to 0 or 1.</p>
7085 <!-- _______________________________________________________________________ -->
7087 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
7093 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
7094 width and for different address space. Not all targets support all bit
7098 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7099 i32 <len>, i32 <align>, i1 <isvolatile>)
7100 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7101 i64 <len>, i32 <align>, i1 <isvolatile>)
7105 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
7106 source location to the destination location. It is similar to the
7107 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
7110 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
7111 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7112 and the pointers can be in specified address spaces.</p>
7116 <p>The first argument is a pointer to the destination, the second is a pointer
7117 to the source. The third argument is an integer argument specifying the
7118 number of bytes to copy, the fourth argument is the alignment of the
7119 source and destination locations, and the fifth is a boolean indicating a
7120 volatile access.</p>
7122 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7123 then the caller guarantees that the source and destination pointers are
7124 aligned to that boundary.</p>
7126 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7127 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
7128 The detailed access behavior is not very cleanly specified and it is unwise
7129 to depend on it.</p>
7133 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
7134 source location to the destination location, which may overlap. It copies
7135 "len" bytes of memory over. If the argument is known to be aligned to some
7136 boundary, this can be specified as the fourth argument, otherwise it should
7137 be set to 0 or 1.</p>
7141 <!-- _______________________________________________________________________ -->
7143 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
7149 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
7150 width and for different address spaces. However, not all targets support all
7154 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7155 i32 <len>, i32 <align>, i1 <isvolatile>)
7156 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7157 i64 <len>, i32 <align>, i1 <isvolatile>)
7161 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
7162 particular byte value.</p>
7164 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
7165 intrinsic does not return a value and takes extra alignment/volatile
7166 arguments. Also, the destination can be in an arbitrary address space.</p>
7169 <p>The first argument is a pointer to the destination to fill, the second is the
7170 byte value with which to fill it, the third argument is an integer argument
7171 specifying the number of bytes to fill, and the fourth argument is the known
7172 alignment of the destination location.</p>
7174 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7175 then the caller guarantees that the destination pointer is aligned to that
7178 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7179 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
7180 The detailed access behavior is not very cleanly specified and it is unwise
7181 to depend on it.</p>
7184 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
7185 at the destination location. If the argument is known to be aligned to some
7186 boundary, this can be specified as the fourth argument, otherwise it should
7187 be set to 0 or 1.</p>
7191 <!-- _______________________________________________________________________ -->
7193 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
7199 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
7200 floating point or vector of floating point type. Not all targets support all
7204 declare float @llvm.sqrt.f32(float %Val)
7205 declare double @llvm.sqrt.f64(double %Val)
7206 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7207 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7208 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7212 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
7213 returning the same value as the libm '<tt>sqrt</tt>' functions would.
7214 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
7215 behavior for negative numbers other than -0.0 (which allows for better
7216 optimization, because there is no need to worry about errno being
7217 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
7220 <p>The argument and return value are floating point numbers of the same
7224 <p>This function returns the sqrt of the specified operand if it is a
7225 nonnegative floating point number.</p>
7229 <!-- _______________________________________________________________________ -->
7231 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
7237 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
7238 floating point or vector of floating point type. Not all targets support all
7242 declare float @llvm.powi.f32(float %Val, i32 %power)
7243 declare double @llvm.powi.f64(double %Val, i32 %power)
7244 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7245 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7246 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7250 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
7251 specified (positive or negative) power. The order of evaluation of
7252 multiplications is not defined. When a vector of floating point type is
7253 used, the second argument remains a scalar integer value.</p>
7256 <p>The second argument is an integer power, and the first is a value to raise to
7260 <p>This function returns the first value raised to the second power with an
7261 unspecified sequence of rounding operations.</p>
7265 <!-- _______________________________________________________________________ -->
7267 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
7273 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
7274 floating point or vector of floating point type. Not all targets support all
7278 declare float @llvm.sin.f32(float %Val)
7279 declare double @llvm.sin.f64(double %Val)
7280 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7281 declare fp128 @llvm.sin.f128(fp128 %Val)
7282 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7286 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
7289 <p>The argument and return value are floating point numbers of the same
7293 <p>This function returns the sine of the specified operand, returning the same
7294 values as the libm <tt>sin</tt> functions would, and handles error conditions
7295 in the same way.</p>
7299 <!-- _______________________________________________________________________ -->
7301 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
7307 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
7308 floating point or vector of floating point type. Not all targets support all
7312 declare float @llvm.cos.f32(float %Val)
7313 declare double @llvm.cos.f64(double %Val)
7314 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7315 declare fp128 @llvm.cos.f128(fp128 %Val)
7316 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7320 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
7323 <p>The argument and return value are floating point numbers of the same
7327 <p>This function returns the cosine of the specified operand, returning the same
7328 values as the libm <tt>cos</tt> functions would, and handles error conditions
7329 in the same way.</p>
7333 <!-- _______________________________________________________________________ -->
7335 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
7341 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
7342 floating point or vector of floating point type. Not all targets support all
7346 declare float @llvm.pow.f32(float %Val, float %Power)
7347 declare double @llvm.pow.f64(double %Val, double %Power)
7348 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7349 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7350 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7354 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
7355 specified (positive or negative) power.</p>
7358 <p>The second argument is a floating point power, and the first is a value to
7359 raise to that power.</p>
7362 <p>This function returns the first value raised to the second power, returning
7363 the same values as the libm <tt>pow</tt> functions would, and handles error
7364 conditions in the same way.</p>
7368 <!-- _______________________________________________________________________ -->
7370 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7376 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7377 floating point or vector of floating point type. Not all targets support all
7381 declare float @llvm.exp.f32(float %Val)
7382 declare double @llvm.exp.f64(double %Val)
7383 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7384 declare fp128 @llvm.exp.f128(fp128 %Val)
7385 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7389 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7392 <p>The argument and return value are floating point numbers of the same
7396 <p>This function returns the same values as the libm <tt>exp</tt> functions
7397 would, and handles error conditions in the same way.</p>
7401 <!-- _______________________________________________________________________ -->
7403 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7409 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7410 floating point or vector of floating point type. Not all targets support all
7414 declare float @llvm.log.f32(float %Val)
7415 declare double @llvm.log.f64(double %Val)
7416 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7417 declare fp128 @llvm.log.f128(fp128 %Val)
7418 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7422 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7425 <p>The argument and return value are floating point numbers of the same
7429 <p>This function returns the same values as the libm <tt>log</tt> functions
7430 would, and handles error conditions in the same way.</p>
7434 <!-- _______________________________________________________________________ -->
7436 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7442 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7443 floating point or vector of floating point type. Not all targets support all
7447 declare float @llvm.fma.f32(float %a, float %b, float %c)
7448 declare double @llvm.fma.f64(double %a, double %b, double %c)
7449 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7450 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7451 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7455 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7459 <p>The argument and return value are floating point numbers of the same
7463 <p>This function returns the same values as the libm <tt>fma</tt> functions
7470 <!-- ======================================================================= -->
7472 <a name="int_manip">Bit Manipulation Intrinsics</a>
7477 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7478 These allow efficient code generation for some algorithms.</p>
7480 <!-- _______________________________________________________________________ -->
7482 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7488 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7489 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7492 declare i16 @llvm.bswap.i16(i16 <id>)
7493 declare i32 @llvm.bswap.i32(i32 <id>)
7494 declare i64 @llvm.bswap.i64(i64 <id>)
7498 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7499 values with an even number of bytes (positive multiple of 16 bits). These
7500 are useful for performing operations on data that is not in the target's
7501 native byte order.</p>
7504 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7505 and low byte of the input i16 swapped. Similarly,
7506 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7507 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7508 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7509 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7510 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7511 more, respectively).</p>
7515 <!-- _______________________________________________________________________ -->
7517 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7523 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7524 width, or on any vector with integer elements. Not all targets support all
7525 bit widths or vector types, however.</p>
7528 declare i8 @llvm.ctpop.i8(i8 <src>)
7529 declare i16 @llvm.ctpop.i16(i16 <src>)
7530 declare i32 @llvm.ctpop.i32(i32 <src>)
7531 declare i64 @llvm.ctpop.i64(i64 <src>)
7532 declare i256 @llvm.ctpop.i256(i256 <src>)
7533 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7537 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7541 <p>The only argument is the value to be counted. The argument may be of any
7542 integer type, or a vector with integer elements.
7543 The return type must match the argument type.</p>
7546 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7547 element of a vector.</p>
7551 <!-- _______________________________________________________________________ -->
7553 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7559 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7560 integer bit width, or any vector whose elements are integers. Not all
7561 targets support all bit widths or vector types, however.</p>
7564 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7565 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7566 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7567 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7568 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7569 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7573 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7574 leading zeros in a variable.</p>
7577 <p>The first argument is the value to be counted. This argument may be of any
7578 integer type, or a vectory with integer element type. The return type
7579 must match the first argument type.</p>
7581 <p>The second argument must be a constant and is a flag to indicate whether the
7582 intrinsic should ensure that a zero as the first argument produces a defined
7583 result. Historically some architectures did not provide a defined result for
7584 zero values as efficiently, and many algorithms are now predicated on
7585 avoiding zero-value inputs.</p>
7588 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7589 zeros in a variable, or within each element of the vector.
7590 If <tt>src == 0</tt> then the result is the size in bits of the type of
7591 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7592 For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7596 <!-- _______________________________________________________________________ -->
7598 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7604 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7605 integer bit width, or any vector of integer elements. Not all targets
7606 support all bit widths or vector types, however.</p>
7609 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7610 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7611 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7612 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7613 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7614 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7618 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7622 <p>The first argument is the value to be counted. This argument may be of any
7623 integer type, or a vectory with integer element type. The return type
7624 must match the first argument type.</p>
7626 <p>The second argument must be a constant and is a flag to indicate whether the
7627 intrinsic should ensure that a zero as the first argument produces a defined
7628 result. Historically some architectures did not provide a defined result for
7629 zero values as efficiently, and many algorithms are now predicated on
7630 avoiding zero-value inputs.</p>
7633 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7634 zeros in a variable, or within each element of a vector.
7635 If <tt>src == 0</tt> then the result is the size in bits of the type of
7636 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7637 For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7643 <!-- ======================================================================= -->
7645 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7650 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7652 <!-- _______________________________________________________________________ -->
7654 <a name="int_sadd_overflow">
7655 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7662 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7663 on any integer bit width.</p>
7666 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7667 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7668 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7672 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7673 a signed addition of the two arguments, and indicate whether an overflow
7674 occurred during the signed summation.</p>
7677 <p>The arguments (%a and %b) and the first element of the result structure may
7678 be of integer types of any bit width, but they must have the same bit
7679 width. The second element of the result structure must be of
7680 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7681 undergo signed addition.</p>
7684 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7685 a signed addition of the two variables. They return a structure — the
7686 first element of which is the signed summation, and the second element of
7687 which is a bit specifying if the signed summation resulted in an
7692 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7693 %sum = extractvalue {i32, i1} %res, 0
7694 %obit = extractvalue {i32, i1} %res, 1
7695 br i1 %obit, label %overflow, label %normal
7700 <!-- _______________________________________________________________________ -->
7702 <a name="int_uadd_overflow">
7703 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7710 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7711 on any integer bit width.</p>
7714 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7715 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7716 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7720 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7721 an unsigned addition of the two arguments, and indicate whether a carry
7722 occurred during the unsigned summation.</p>
7725 <p>The arguments (%a and %b) and the first element of the result structure may
7726 be of integer types of any bit width, but they must have the same bit
7727 width. The second element of the result structure must be of
7728 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7729 undergo unsigned addition.</p>
7732 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7733 an unsigned addition of the two arguments. They return a structure —
7734 the first element of which is the sum, and the second element of which is a
7735 bit specifying if the unsigned summation resulted in a carry.</p>
7739 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7740 %sum = extractvalue {i32, i1} %res, 0
7741 %obit = extractvalue {i32, i1} %res, 1
7742 br i1 %obit, label %carry, label %normal
7747 <!-- _______________________________________________________________________ -->
7749 <a name="int_ssub_overflow">
7750 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7757 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7758 on any integer bit width.</p>
7761 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7762 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7763 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7767 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7768 a signed subtraction of the two arguments, and indicate whether an overflow
7769 occurred during the signed subtraction.</p>
7772 <p>The arguments (%a and %b) and the first element of the result structure may
7773 be of integer types of any bit width, but they must have the same bit
7774 width. The second element of the result structure must be of
7775 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7776 undergo signed subtraction.</p>
7779 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7780 a signed subtraction of the two arguments. They return a structure —
7781 the first element of which is the subtraction, and the second element of
7782 which is a bit specifying if the signed subtraction resulted in an
7787 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7788 %sum = extractvalue {i32, i1} %res, 0
7789 %obit = extractvalue {i32, i1} %res, 1
7790 br i1 %obit, label %overflow, label %normal
7795 <!-- _______________________________________________________________________ -->
7797 <a name="int_usub_overflow">
7798 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7805 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7806 on any integer bit width.</p>
7809 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7810 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7811 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7815 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7816 an unsigned subtraction of the two arguments, and indicate whether an
7817 overflow occurred during the unsigned subtraction.</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 unsigned subtraction.</p>
7827 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7828 an unsigned subtraction of the two arguments. They return a structure —
7829 the first element of which is the subtraction, and the second element of
7830 which is a bit specifying if the unsigned subtraction resulted in an
7835 %res = call {i32, i1} @llvm.usub.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_smul_overflow">
7846 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7853 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7854 on any integer bit width.</p>
7857 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7858 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7859 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7864 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7865 a signed multiplication of the two arguments, and indicate whether an
7866 overflow occurred during the signed multiplication.</p>
7869 <p>The arguments (%a and %b) and the first element of the result structure may
7870 be of integer types of any bit width, but they must have the same bit
7871 width. The second element of the result structure must be of
7872 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7873 undergo signed multiplication.</p>
7876 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7877 a signed multiplication of the two arguments. They return a structure —
7878 the first element of which is the multiplication, and the second element of
7879 which is a bit specifying if the signed multiplication resulted in an
7884 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7885 %sum = extractvalue {i32, i1} %res, 0
7886 %obit = extractvalue {i32, i1} %res, 1
7887 br i1 %obit, label %overflow, label %normal
7892 <!-- _______________________________________________________________________ -->
7894 <a name="int_umul_overflow">
7895 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
7902 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
7903 on any integer bit width.</p>
7906 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7907 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7908 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7912 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7913 a unsigned multiplication of the two arguments, and indicate whether an
7914 overflow occurred during the unsigned multiplication.</p>
7917 <p>The arguments (%a and %b) and the first element of the result structure may
7918 be of integer types of any bit width, but they must have the same bit
7919 width. The second element of the result structure must be of
7920 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7921 undergo unsigned multiplication.</p>
7924 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7925 an unsigned multiplication of the two arguments. They return a structure
7926 — the first element of which is the multiplication, and the second
7927 element of which is a bit specifying if the unsigned multiplication resulted
7932 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7933 %sum = extractvalue {i32, i1} %res, 0
7934 %obit = extractvalue {i32, i1} %res, 1
7935 br i1 %obit, label %overflow, label %normal
7942 <!-- ======================================================================= -->
7944 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
7949 <p>Half precision floating point is a storage-only format. This means that it is
7950 a dense encoding (in memory) but does not support computation in the
7953 <p>This means that code must first load the half-precision floating point
7954 value as an i16, then convert it to float with <a
7955 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
7956 Computation can then be performed on the float value (including extending to
7957 double etc). To store the value back to memory, it is first converted to
7958 float if needed, then converted to i16 with
7959 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
7960 storing as an i16 value.</p>
7962 <!-- _______________________________________________________________________ -->
7964 <a name="int_convert_to_fp16">
7965 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
7973 declare i16 @llvm.convert.to.fp16(f32 %a)
7977 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7978 a conversion from single precision floating point format to half precision
7979 floating point format.</p>
7982 <p>The intrinsic function contains single argument - the value to be
7986 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7987 a conversion from single precision floating point format to half precision
7988 floating point format. The return value is an <tt>i16</tt> which
7989 contains the converted number.</p>
7993 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7994 store i16 %res, i16* @x, align 2
7999 <!-- _______________________________________________________________________ -->
8001 <a name="int_convert_from_fp16">
8002 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
8010 declare f32 @llvm.convert.from.fp16(i16 %a)
8014 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
8015 a conversion from half precision floating point format to single precision
8016 floating point format.</p>
8019 <p>The intrinsic function contains single argument - the value to be
8023 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
8024 conversion from half single precision floating point format to single
8025 precision floating point format. The input half-float value is represented by
8026 an <tt>i16</tt> value.</p>
8030 %a = load i16* @x, align 2
8031 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8038 <!-- ======================================================================= -->
8040 <a name="int_debugger">Debugger Intrinsics</a>
8045 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
8046 prefix), are described in
8047 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
8048 Level Debugging</a> document.</p>
8052 <!-- ======================================================================= -->
8054 <a name="int_eh">Exception Handling Intrinsics</a>
8059 <p>The LLVM exception handling intrinsics (which all start with
8060 <tt>llvm.eh.</tt> prefix), are described in
8061 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
8062 Handling</a> document.</p>
8066 <!-- ======================================================================= -->
8068 <a name="int_trampoline">Trampoline Intrinsics</a>
8073 <p>These intrinsics make it possible to excise one parameter, marked with
8074 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
8075 The result is a callable
8076 function pointer lacking the nest parameter - the caller does not need to
8077 provide a value for it. Instead, the value to use is stored in advance in a
8078 "trampoline", a block of memory usually allocated on the stack, which also
8079 contains code to splice the nest value into the argument list. This is used
8080 to implement the GCC nested function address extension.</p>
8082 <p>For example, if the function is
8083 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
8084 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
8087 <pre class="doc_code">
8088 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8089 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8090 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8091 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8092 %fp = bitcast i8* %p to i32 (i32, i32)*
8095 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
8096 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
8098 <!-- _______________________________________________________________________ -->
8101 '<tt>llvm.init.trampoline</tt>' Intrinsic
8109 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8113 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
8114 turning it into a trampoline.</p>
8117 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
8118 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
8119 sufficiently aligned block of memory; this memory is written to by the
8120 intrinsic. Note that the size and the alignment are target-specific - LLVM
8121 currently provides no portable way of determining them, so a front-end that
8122 generates this intrinsic needs to have some target-specific knowledge.
8123 The <tt>func</tt> argument must hold a function bitcast to
8124 an <tt>i8*</tt>.</p>
8127 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
8128 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
8129 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
8130 which can be <a href="#int_trampoline">bitcast (to a new function) and
8131 called</a>. The new function's signature is the same as that of
8132 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
8133 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
8134 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
8135 with the same argument list, but with <tt>nval</tt> used for the missing
8136 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
8137 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
8138 to the returned function pointer is undefined.</p>
8141 <!-- _______________________________________________________________________ -->
8144 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
8152 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8156 <p>This performs any required machine-specific adjustment to the address of a
8157 trampoline (passed as <tt>tramp</tt>).</p>
8160 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
8161 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
8165 <p>On some architectures the address of the code to be executed needs to be
8166 different to the address where the trampoline is actually stored. This
8167 intrinsic returns the executable address corresponding to <tt>tramp</tt>
8168 after performing the required machine specific adjustments.
8169 The pointer returned can then be <a href="#int_trampoline"> bitcast and
8177 <!-- ======================================================================= -->
8179 <a name="int_memorymarkers">Memory Use Markers</a>
8184 <p>This class of intrinsics exists to information about the lifetime of memory
8185 objects and ranges where variables are immutable.</p>
8187 <!-- _______________________________________________________________________ -->
8189 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8196 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8200 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8201 object's lifetime.</p>
8204 <p>The first argument is a constant integer representing the size of the
8205 object, or -1 if it is variable sized. The second argument is a pointer to
8209 <p>This intrinsic indicates that before this point in the code, the value of the
8210 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8211 never be used and has an undefined value. A load from the pointer that
8212 precedes this intrinsic can be replaced with
8213 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8217 <!-- _______________________________________________________________________ -->
8219 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8226 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8230 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8231 object's lifetime.</p>
8234 <p>The first argument is a constant integer representing the size of the
8235 object, or -1 if it is variable sized. The second argument is a pointer to
8239 <p>This intrinsic indicates that after this point in the code, the value of the
8240 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8241 never be used and has an undefined value. Any stores into the memory object
8242 following this intrinsic may be removed as dead.
8246 <!-- _______________________________________________________________________ -->
8248 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8255 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8259 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8260 a memory object will not change.</p>
8263 <p>The first argument is a constant integer representing the size of the
8264 object, or -1 if it is variable sized. The second argument is a pointer to
8268 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8269 the return value, the referenced memory location is constant and
8274 <!-- _______________________________________________________________________ -->
8276 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8283 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8287 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8288 a memory object are mutable.</p>
8291 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8292 The second argument is a constant integer representing the size of the
8293 object, or -1 if it is variable sized and the third argument is a pointer
8297 <p>This intrinsic indicates that the memory is mutable again.</p>
8303 <!-- ======================================================================= -->
8305 <a name="int_general">General Intrinsics</a>
8310 <p>This class of intrinsics is designed to be generic and has no specific
8313 <!-- _______________________________________________________________________ -->
8315 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8322 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8326 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8329 <p>The first argument is a pointer to a value, the second is a pointer to a
8330 global string, the third is a pointer to a global string which is the source
8331 file name, and the last argument is the line number.</p>
8334 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8335 This can be useful for special purpose optimizations that want to look for
8336 these annotations. These have no other defined use; they are ignored by code
8337 generation and optimization.</p>
8341 <!-- _______________________________________________________________________ -->
8343 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8349 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8350 any integer bit width.</p>
8353 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8354 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8355 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8356 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8357 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8361 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8364 <p>The first argument is an integer value (result of some expression), the
8365 second is a pointer to a global string, the third is a pointer to a global
8366 string which is the source file name, and the last argument is the line
8367 number. It returns the value of the first argument.</p>
8370 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8371 arbitrary strings. This can be useful for special purpose optimizations that
8372 want to look for these annotations. These have no other defined use; they
8373 are ignored by code generation and optimization.</p>
8377 <!-- _______________________________________________________________________ -->
8379 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8386 declare void @llvm.trap()
8390 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8396 <p>This intrinsics is lowered to the target dependent trap instruction. If the
8397 target does not have a trap instruction, this intrinsic will be lowered to
8398 the call of the <tt>abort()</tt> function.</p>
8402 <!-- _______________________________________________________________________ -->
8404 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8411 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8415 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8416 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8417 ensure that it is placed on the stack before local variables.</p>
8420 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8421 arguments. The first argument is the value loaded from the stack
8422 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8423 that has enough space to hold the value of the guard.</p>
8426 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8427 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8428 stack. This is to ensure that if a local variable on the stack is
8429 overwritten, it will destroy the value of the guard. When the function exits,
8430 the guard on the stack is checked against the original guard. If they are
8431 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8436 <!-- _______________________________________________________________________ -->
8438 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8445 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <type>)
8446 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <type>)
8450 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8451 the optimizers to determine at compile time whether a) an operation (like
8452 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8453 runtime check for overflow isn't necessary. An object in this context means
8454 an allocation of a specific class, structure, array, or other object.</p>
8457 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8458 argument is a pointer to or into the <tt>object</tt>. The second argument
8459 is a boolean 0 or 1. This argument determines whether you want the
8460 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
8461 1, variables are not allowed.</p>
8464 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
8465 representing the size of the object concerned, or <tt>i32/i64 -1 or 0</tt>,
8466 depending on the <tt>type</tt> argument, if the size cannot be determined at
8470 <!-- _______________________________________________________________________ -->
8472 <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a>
8479 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8480 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8484 <p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the
8485 most probable) value of <tt>val</tt>, which can be used by optimizers.</p>
8488 <p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first
8489 argument is a value. The second argument is an expected value, this needs to
8490 be a constant value, variables are not allowed.</p>
8493 <p>This intrinsic is lowered to the <tt>val</tt>.</p>
8499 <!-- *********************************************************************** -->
8502 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8503 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8504 <a href="http://validator.w3.org/check/referer"><img
8505 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8507 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8508 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8509 Last modified: $Date$