1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <h1>LLVM Language Reference Manual</h1>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a>
25 <li><a href="#linkage_private">'<tt>private</tt>' Linkage</a></li>
26 <li><a href="#linkage_linker_private">'<tt>linker_private</tt>' Linkage</a></li>
27 <li><a href="#linkage_linker_private_weak">'<tt>linker_private_weak</tt>' Linkage</a></li>
28 <li><a href="#linkage_linker_private_weak_def_auto">'<tt>linker_private_weak_def_auto</tt>' Linkage</a></li>
29 <li><a href="#linkage_internal">'<tt>internal</tt>' Linkage</a></li>
30 <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
31 <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
32 <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
33 <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
34 <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
35 <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
36 <li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
37 <li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
38 <li><a href="#linkage_external">'<tt>external</tt>' Linkage</a></li>
39 <li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
40 <li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
43 <li><a href="#callingconv">Calling Conventions</a></li>
44 <li><a href="#namedtypes">Named Types</a></li>
45 <li><a href="#globalvars">Global Variables</a></li>
46 <li><a href="#functionstructure">Functions</a></li>
47 <li><a href="#aliasstructure">Aliases</a></li>
48 <li><a href="#namedmetadatastructure">Named Metadata</a></li>
49 <li><a href="#paramattrs">Parameter Attributes</a></li>
50 <li><a href="#fnattrs">Function Attributes</a></li>
51 <li><a href="#gc">Garbage Collector Names</a></li>
52 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
53 <li><a href="#datalayout">Data Layout</a></li>
54 <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
55 <li><a href="#volatile">Volatile Memory Accesses</a></li>
56 <li><a href="#memmodel">Memory Model for Concurrent Operations</a></li>
57 <li><a href="#ordering">Atomic Memory Ordering Constraints</a></li>
60 <li><a href="#typesystem">Type System</a>
62 <li><a href="#t_classifications">Type Classifications</a></li>
63 <li><a href="#t_primitive">Primitive Types</a>
65 <li><a href="#t_integer">Integer Type</a></li>
66 <li><a href="#t_floating">Floating Point Types</a></li>
67 <li><a href="#t_x86mmx">X86mmx Type</a></li>
68 <li><a href="#t_void">Void Type</a></li>
69 <li><a href="#t_label">Label Type</a></li>
70 <li><a href="#t_metadata">Metadata Type</a></li>
73 <li><a href="#t_derived">Derived Types</a>
75 <li><a href="#t_aggregate">Aggregate Types</a>
77 <li><a href="#t_array">Array Type</a></li>
78 <li><a href="#t_struct">Structure Type</a></li>
79 <li><a href="#t_opaque">Opaque Structure Types</a></li>
80 <li><a href="#t_vector">Vector Type</a></li>
83 <li><a href="#t_function">Function Type</a></li>
84 <li><a href="#t_pointer">Pointer Type</a></li>
89 <li><a href="#constants">Constants</a>
91 <li><a href="#simpleconstants">Simple Constants</a></li>
92 <li><a href="#complexconstants">Complex Constants</a></li>
93 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
94 <li><a href="#undefvalues">Undefined Values</a></li>
95 <li><a href="#poisonvalues">Poison Values</a></li>
96 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
97 <li><a href="#constantexprs">Constant Expressions</a></li>
100 <li><a href="#othervalues">Other Values</a>
102 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
103 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a>
105 <li><a href="#tbaa">'<tt>tbaa</tt>' Metadata</a></li>
106 <li><a href="#fpaccuracy">'<tt>fpaccuracy</tt>' Metadata</a></li>
111 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
113 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
114 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
115 Global Variable</a></li>
116 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
117 Global Variable</a></li>
118 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
119 Global Variable</a></li>
122 <li><a href="#instref">Instruction Reference</a>
124 <li><a href="#terminators">Terminator Instructions</a>
126 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
127 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
128 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
129 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
130 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
131 <li><a href="#i_resume">'<tt>resume</tt>' Instruction</a></li>
132 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
135 <li><a href="#binaryops">Binary Operations</a>
137 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
138 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
139 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
140 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
141 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
142 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
143 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
144 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
145 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
146 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
147 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
148 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
151 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
153 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
154 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
155 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
156 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
157 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
158 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
161 <li><a href="#vectorops">Vector Operations</a>
163 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
164 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
165 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
168 <li><a href="#aggregateops">Aggregate Operations</a>
170 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
171 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
174 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
176 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
177 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
178 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
179 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
180 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
181 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
182 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
185 <li><a href="#convertops">Conversion Operations</a>
187 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
188 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
189 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
190 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
191 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
192 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
193 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
194 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
195 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
196 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
197 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
198 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
201 <li><a href="#otherops">Other Operations</a>
203 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
204 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
205 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
206 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
207 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
208 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
209 <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
214 <li><a href="#intrinsics">Intrinsic Functions</a>
216 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
218 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
219 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
220 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
223 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
225 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
226 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
227 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
230 <li><a href="#int_codegen">Code Generator Intrinsics</a>
232 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
233 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
234 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
235 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
236 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
237 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
238 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
241 <li><a href="#int_libc">Standard C Library Intrinsics</a>
243 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
244 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
245 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
246 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
247 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
248 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
249 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
250 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
251 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
252 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
253 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
256 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
258 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
259 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
260 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
261 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
264 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
266 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
267 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
268 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
269 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
270 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
271 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
274 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
276 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
277 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
280 <li><a href="#int_debugger">Debugger intrinsics</a></li>
281 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
282 <li><a href="#int_trampoline">Trampoline Intrinsics</a>
284 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
285 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
288 <li><a href="#int_memorymarkers">Memory Use Markers</a>
290 <li><a href="#int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a></li>
291 <li><a href="#int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a></li>
292 <li><a href="#int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a></li>
293 <li><a href="#int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a></li>
296 <li><a href="#int_general">General intrinsics</a>
298 <li><a href="#int_var_annotation">
299 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
300 <li><a href="#int_annotation">
301 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
302 <li><a href="#int_trap">
303 '<tt>llvm.trap</tt>' Intrinsic</a></li>
304 <li><a href="#int_stackprotector">
305 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
306 <li><a href="#int_objectsize">
307 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
308 <li><a href="#int_expect">
309 '<tt>llvm.expect</tt>' Intrinsic</a></li>
316 <div class="doc_author">
317 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
318 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
321 <!-- *********************************************************************** -->
322 <h2><a name="abstract">Abstract</a></h2>
323 <!-- *********************************************************************** -->
327 <p>This document is a reference manual for the LLVM assembly language. LLVM is
328 a Static Single Assignment (SSA) based representation that provides type
329 safety, low-level operations, flexibility, and the capability of representing
330 'all' high-level languages cleanly. It is the common code representation
331 used throughout all phases of the LLVM compilation strategy.</p>
335 <!-- *********************************************************************** -->
336 <h2><a name="introduction">Introduction</a></h2>
337 <!-- *********************************************************************** -->
341 <p>The LLVM code representation is designed to be used in three different forms:
342 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
343 for fast loading by a Just-In-Time compiler), and as a human readable
344 assembly language representation. This allows LLVM to provide a powerful
345 intermediate representation for efficient compiler transformations and
346 analysis, while providing a natural means to debug and visualize the
347 transformations. The three different forms of LLVM are all equivalent. This
348 document describes the human readable representation and notation.</p>
350 <p>The LLVM representation aims to be light-weight and low-level while being
351 expressive, typed, and extensible at the same time. It aims to be a
352 "universal IR" of sorts, by being at a low enough level that high-level ideas
353 may be cleanly mapped to it (similar to how microprocessors are "universal
354 IR's", allowing many source languages to be mapped to them). By providing
355 type information, LLVM can be used as the target of optimizations: for
356 example, through pointer analysis, it can be proven that a C automatic
357 variable is never accessed outside of the current function, allowing it to
358 be promoted to a simple SSA value instead of a memory location.</p>
360 <!-- _______________________________________________________________________ -->
362 <a name="wellformed">Well-Formedness</a>
367 <p>It is important to note that this document describes 'well formed' LLVM
368 assembly language. There is a difference between what the parser accepts and
369 what is considered 'well formed'. For example, the following instruction is
370 syntactically okay, but not well formed:</p>
372 <pre class="doc_code">
373 %x = <a href="#i_add">add</a> i32 1, %x
376 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
377 LLVM infrastructure provides a verification pass that may be used to verify
378 that an LLVM module is well formed. This pass is automatically run by the
379 parser after parsing input assembly and by the optimizer before it outputs
380 bitcode. The violations pointed out by the verifier pass indicate bugs in
381 transformation passes or input to the parser.</p>
387 <!-- Describe the typesetting conventions here. -->
389 <!-- *********************************************************************** -->
390 <h2><a name="identifiers">Identifiers</a></h2>
391 <!-- *********************************************************************** -->
395 <p>LLVM identifiers come in two basic types: global and local. Global
396 identifiers (functions, global variables) begin with the <tt>'@'</tt>
397 character. Local identifiers (register names, types) begin with
398 the <tt>'%'</tt> character. Additionally, there are three different formats
399 for identifiers, for different purposes:</p>
402 <li>Named values are represented as a string of characters with their prefix.
403 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
404 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
405 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
406 other characters in their names can be surrounded with quotes. Special
407 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
408 ASCII code for the character in hexadecimal. In this way, any character
409 can be used in a name value, even quotes themselves.</li>
411 <li>Unnamed values are represented as an unsigned numeric value with their
412 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
414 <li>Constants, which are described in a <a href="#constants">section about
415 constants</a>, below.</li>
418 <p>LLVM requires that values start with a prefix for two reasons: Compilers
419 don't need to worry about name clashes with reserved words, and the set of
420 reserved words may be expanded in the future without penalty. Additionally,
421 unnamed identifiers allow a compiler to quickly come up with a temporary
422 variable without having to avoid symbol table conflicts.</p>
424 <p>Reserved words in LLVM are very similar to reserved words in other
425 languages. There are keywords for different opcodes
426 ('<tt><a href="#i_add">add</a></tt>',
427 '<tt><a href="#i_bitcast">bitcast</a></tt>',
428 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
429 ('<tt><a href="#t_void">void</a></tt>',
430 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
431 reserved words cannot conflict with variable names, because none of them
432 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
434 <p>Here is an example of LLVM code to multiply the integer variable
435 '<tt>%X</tt>' by 8:</p>
439 <pre class="doc_code">
440 %result = <a href="#i_mul">mul</a> i32 %X, 8
443 <p>After strength reduction:</p>
445 <pre class="doc_code">
446 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
449 <p>And the hard way:</p>
451 <pre class="doc_code">
452 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
453 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
454 %result = <a href="#i_add">add</a> i32 %1, %1
457 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
458 lexical features of LLVM:</p>
461 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
464 <li>Unnamed temporaries are created when the result of a computation is not
465 assigned to a named value.</li>
467 <li>Unnamed temporaries are numbered sequentially</li>
470 <p>It also shows a convention that we follow in this document. When
471 demonstrating instructions, we will follow an instruction with a comment that
472 defines the type and name of value produced. Comments are shown in italic
477 <!-- *********************************************************************** -->
478 <h2><a name="highlevel">High Level Structure</a></h2>
479 <!-- *********************************************************************** -->
481 <!-- ======================================================================= -->
483 <a name="modulestructure">Module Structure</a>
488 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
489 of the input programs. Each module consists of functions, global variables,
490 and symbol table entries. Modules may be combined together with the LLVM
491 linker, which merges function (and global variable) definitions, resolves
492 forward declarations, and merges symbol table entries. Here is an example of
493 the "hello world" module:</p>
495 <pre class="doc_code">
496 <i>; Declare the string constant as a global constant.</i>
497 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
499 <i>; External declaration of the puts function</i>
500 <a href="#functionstructure">declare</a> i32 @puts(i8*) <i>; i32 (i8*)* </i>
502 <i>; Definition of main function</i>
503 define i32 @main() { <i>; i32()* </i>
504 <i>; Convert [13 x i8]* to i8 *...</i>
505 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8*</i>
507 <i>; Call puts function to write out the string to stdout.</i>
508 <a href="#i_call">call</a> i32 @puts(i8* %cast210) <i>; i32</i>
509 <a href="#i_ret">ret</a> i32 0
512 <i>; Named metadata</i>
513 !1 = metadata !{i32 41}
517 <p>This example is made up of a <a href="#globalvars">global variable</a> named
518 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
519 a <a href="#functionstructure">function definition</a> for
520 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
523 <p>In general, a module is made up of a list of global values, where both
524 functions and global variables are global values. Global values are
525 represented by a pointer to a memory location (in this case, a pointer to an
526 array of char, and a pointer to a function), and have one of the
527 following <a href="#linkage">linkage types</a>.</p>
531 <!-- ======================================================================= -->
533 <a name="linkage">Linkage Types</a>
538 <p>All Global Variables and Functions have one of the following types of
542 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
543 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
544 by objects in the current module. In particular, linking code into a
545 module with an private global value may cause the private to be renamed as
546 necessary to avoid collisions. Because the symbol is private to the
547 module, all references can be updated. This doesn't show up in any symbol
548 table in the object file.</dd>
550 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
551 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
552 assembler and evaluated by the linker. Unlike normal strong symbols, they
553 are removed by the linker from the final linked image (executable or
554 dynamic library).</dd>
556 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
557 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
558 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
559 linker. The symbols are removed by the linker from the final linked image
560 (executable or dynamic library).</dd>
562 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
563 <dd>Similar to "<tt>linker_private_weak</tt>", but it's known that the address
564 of the object is not taken. For instance, functions that had an inline
565 definition, but the compiler decided not to inline it. Note,
566 unlike <tt>linker_private</tt> and <tt>linker_private_weak</tt>,
567 <tt>linker_private_weak_def_auto</tt> may have only <tt>default</tt>
568 visibility. The symbols are removed by the linker from the final linked
569 image (executable or dynamic library).</dd>
571 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
572 <dd>Similar to private, but the value shows as a local symbol
573 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
574 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
576 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
577 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
578 into the object file corresponding to the LLVM module. They exist to
579 allow inlining and other optimizations to take place given knowledge of
580 the definition of the global, which is known to be somewhere outside the
581 module. Globals with <tt>available_externally</tt> linkage are allowed to
582 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
583 This linkage type is only allowed on definitions, not declarations.</dd>
585 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
586 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
587 the same name when linkage occurs. This can be used to implement
588 some forms of inline functions, templates, or other code which must be
589 generated in each translation unit that uses it, but where the body may
590 be overridden with a more definitive definition later. Unreferenced
591 <tt>linkonce</tt> globals are allowed to be discarded. Note that
592 <tt>linkonce</tt> linkage does not actually allow the optimizer to
593 inline the body of this function into callers because it doesn't know if
594 this definition of the function is the definitive definition within the
595 program or whether it will be overridden by a stronger definition.
596 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
599 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
600 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
601 <tt>linkonce</tt> linkage, except that unreferenced globals with
602 <tt>weak</tt> linkage may not be discarded. This is used for globals that
603 are declared "weak" in C source code.</dd>
605 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
606 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
607 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
609 Symbols with "<tt>common</tt>" linkage are merged in the same way as
610 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
611 <tt>common</tt> symbols may not have an explicit section,
612 must have a zero initializer, and may not be marked '<a
613 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
614 have common linkage.</dd>
617 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
618 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
619 pointer to array type. When two global variables with appending linkage
620 are linked together, the two global arrays are appended together. This is
621 the LLVM, typesafe, equivalent of having the system linker append together
622 "sections" with identical names when .o files are linked.</dd>
624 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
625 <dd>The semantics of this linkage follow the ELF object file model: the symbol
626 is weak until linked, if not linked, the symbol becomes null instead of
627 being an undefined reference.</dd>
629 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
630 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
631 <dd>Some languages allow differing globals to be merged, such as two functions
632 with different semantics. Other languages, such as <tt>C++</tt>, ensure
633 that only equivalent globals are ever merged (the "one definition rule"
634 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
635 and <tt>weak_odr</tt> linkage types to indicate that the global will only
636 be merged with equivalent globals. These linkage types are otherwise the
637 same as their non-<tt>odr</tt> versions.</dd>
639 <dt><tt><b><a name="linkage_external">external</a></b></tt></dt>
640 <dd>If none of the above identifiers are used, the global is externally
641 visible, meaning that it participates in linkage and can be used to
642 resolve external symbol references.</dd>
645 <p>The next two types of linkage are targeted for Microsoft Windows platform
646 only. They are designed to support importing (exporting) symbols from (to)
647 DLLs (Dynamic Link Libraries).</p>
650 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
651 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
652 or variable via a global pointer to a pointer that is set up by the DLL
653 exporting the symbol. On Microsoft Windows targets, the pointer name is
654 formed by combining <code>__imp_</code> and the function or variable
657 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
658 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
659 pointer to a pointer in a DLL, so that it can be referenced with the
660 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
661 name is formed by combining <code>__imp_</code> and the function or
665 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
666 another module defined a "<tt>.LC0</tt>" variable and was linked with this
667 one, one of the two would be renamed, preventing a collision. Since
668 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
669 declarations), they are accessible outside of the current module.</p>
671 <p>It is illegal for a function <i>declaration</i> to have any linkage type
672 other than <tt>external</tt>, <tt>dllimport</tt>
673 or <tt>extern_weak</tt>.</p>
675 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
676 or <tt>weak_odr</tt> linkages.</p>
680 <!-- ======================================================================= -->
682 <a name="callingconv">Calling Conventions</a>
687 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
688 and <a href="#i_invoke">invokes</a> can all have an optional calling
689 convention specified for the call. The calling convention of any pair of
690 dynamic caller/callee must match, or the behavior of the program is
691 undefined. The following calling conventions are supported by LLVM, and more
692 may be added in the future:</p>
695 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
696 <dd>This calling convention (the default if no other calling convention is
697 specified) matches the target C calling conventions. This calling
698 convention supports varargs function calls and tolerates some mismatch in
699 the declared prototype and implemented declaration of the function (as
702 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
703 <dd>This calling convention attempts to make calls as fast as possible
704 (e.g. by passing things in registers). This calling convention allows the
705 target to use whatever tricks it wants to produce fast code for the
706 target, without having to conform to an externally specified ABI
707 (Application Binary Interface).
708 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
709 when this or the GHC convention is used.</a> This calling convention
710 does not support varargs and requires the prototype of all callees to
711 exactly match the prototype of the function definition.</dd>
713 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
714 <dd>This calling convention attempts to make code in the caller as efficient
715 as possible under the assumption that the call is not commonly executed.
716 As such, these calls often preserve all registers so that the call does
717 not break any live ranges in the caller side. This calling convention
718 does not support varargs and requires the prototype of all callees to
719 exactly match the prototype of the function definition.</dd>
721 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
722 <dd>This calling convention has been implemented specifically for use by the
723 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
724 It passes everything in registers, going to extremes to achieve this by
725 disabling callee save registers. This calling convention should not be
726 used lightly but only for specific situations such as an alternative to
727 the <em>register pinning</em> performance technique often used when
728 implementing functional programming languages.At the moment only X86
729 supports this convention and it has the following limitations:
731 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
732 floating point types are supported.</li>
733 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
734 6 floating point parameters.</li>
736 This calling convention supports
737 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
738 requires both the caller and callee are using it.
741 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
742 <dd>Any calling convention may be specified by number, allowing
743 target-specific calling conventions to be used. Target specific calling
744 conventions start at 64.</dd>
747 <p>More calling conventions can be added/defined on an as-needed basis, to
748 support Pascal conventions or any other well-known target-independent
753 <!-- ======================================================================= -->
755 <a name="visibility">Visibility Styles</a>
760 <p>All Global Variables and Functions have one of the following visibility
764 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
765 <dd>On targets that use the ELF object file format, default visibility means
766 that the declaration is visible to other modules and, in shared libraries,
767 means that the declared entity may be overridden. On Darwin, default
768 visibility means that the declaration is visible to other modules. Default
769 visibility corresponds to "external linkage" in the language.</dd>
771 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
772 <dd>Two declarations of an object with hidden visibility refer to the same
773 object if they are in the same shared object. Usually, hidden visibility
774 indicates that the symbol will not be placed into the dynamic symbol
775 table, so no other module (executable or shared library) can reference it
778 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
779 <dd>On ELF, protected visibility indicates that the symbol will be placed in
780 the dynamic symbol table, but that references within the defining module
781 will bind to the local symbol. That is, the symbol cannot be overridden by
787 <!-- ======================================================================= -->
789 <a name="namedtypes">Named Types</a>
794 <p>LLVM IR allows you to specify name aliases for certain types. This can make
795 it easier to read the IR and make the IR more condensed (particularly when
796 recursive types are involved). An example of a name specification is:</p>
798 <pre class="doc_code">
799 %mytype = type { %mytype*, i32 }
802 <p>You may give a name to any <a href="#typesystem">type</a> except
803 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
804 is expected with the syntax "%mytype".</p>
806 <p>Note that type names are aliases for the structural type that they indicate,
807 and that you can therefore specify multiple names for the same type. This
808 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
809 uses structural typing, the name is not part of the type. When printing out
810 LLVM IR, the printer will pick <em>one name</em> to render all types of a
811 particular shape. This means that if you have code where two different
812 source types end up having the same LLVM type, that the dumper will sometimes
813 print the "wrong" or unexpected type. This is an important design point and
814 isn't going to change.</p>
818 <!-- ======================================================================= -->
820 <a name="globalvars">Global Variables</a>
825 <p>Global variables define regions of memory allocated at compilation time
826 instead of run-time. Global variables may optionally be initialized, may
827 have an explicit section to be placed in, and may have an optional explicit
828 alignment specified. A variable may be defined as "thread_local", which
829 means that it will not be shared by threads (each thread will have a
830 separated copy of the variable). A variable may be defined as a global
831 "constant," which indicates that the contents of the variable
832 will <b>never</b> be modified (enabling better optimization, allowing the
833 global data to be placed in the read-only section of an executable, etc).
834 Note that variables that need runtime initialization cannot be marked
835 "constant" as there is a store to the variable.</p>
837 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
838 constant, even if the final definition of the global is not. This capability
839 can be used to enable slightly better optimization of the program, but
840 requires the language definition to guarantee that optimizations based on the
841 'constantness' are valid for the translation units that do not include the
844 <p>As SSA values, global variables define pointer values that are in scope
845 (i.e. they dominate) all basic blocks in the program. Global variables
846 always define a pointer to their "content" type because they describe a
847 region of memory, and all memory objects in LLVM are accessed through
850 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
851 that the address is not significant, only the content. Constants marked
852 like this can be merged with other constants if they have the same
853 initializer. Note that a constant with significant address <em>can</em>
854 be merged with a <tt>unnamed_addr</tt> constant, the result being a
855 constant whose address is significant.</p>
857 <p>A global variable may be declared to reside in a target-specific numbered
858 address space. For targets that support them, address spaces may affect how
859 optimizations are performed and/or what target instructions are used to
860 access the variable. The default address space is zero. The address space
861 qualifier must precede any other attributes.</p>
863 <p>LLVM allows an explicit section to be specified for globals. If the target
864 supports it, it will emit globals to the section specified.</p>
866 <p>An explicit alignment may be specified for a global, which must be a power
867 of 2. If not present, or if the alignment is set to zero, the alignment of
868 the global is set by the target to whatever it feels convenient. If an
869 explicit alignment is specified, the global is forced to have exactly that
870 alignment. Targets and optimizers are not allowed to over-align the global
871 if the global has an assigned section. In this case, the extra alignment
872 could be observable: for example, code could assume that the globals are
873 densely packed in their section and try to iterate over them as an array,
874 alignment padding would break this iteration.</p>
876 <p>For example, the following defines a global in a numbered address space with
877 an initializer, section, and alignment:</p>
879 <pre class="doc_code">
880 @G = addrspace(5) constant float 1.0, section "foo", align 4
886 <!-- ======================================================================= -->
888 <a name="functionstructure">Functions</a>
893 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
894 optional <a href="#linkage">linkage type</a>, an optional
895 <a href="#visibility">visibility style</a>, an optional
896 <a href="#callingconv">calling convention</a>,
897 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
898 <a href="#paramattrs">parameter attribute</a> for the return type, a function
899 name, a (possibly empty) argument list (each with optional
900 <a href="#paramattrs">parameter attributes</a>), optional
901 <a href="#fnattrs">function attributes</a>, an optional section, an optional
902 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
903 curly brace, a list of basic blocks, and a closing curly brace.</p>
905 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
906 optional <a href="#linkage">linkage type</a>, an optional
907 <a href="#visibility">visibility style</a>, an optional
908 <a href="#callingconv">calling convention</a>,
909 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
910 <a href="#paramattrs">parameter attribute</a> for the return type, a function
911 name, a possibly empty list of arguments, an optional alignment, and an
912 optional <a href="#gc">garbage collector name</a>.</p>
914 <p>A function definition contains a list of basic blocks, forming the CFG
915 (Control Flow Graph) for the function. Each basic block may optionally start
916 with a label (giving the basic block a symbol table entry), contains a list
917 of instructions, and ends with a <a href="#terminators">terminator</a>
918 instruction (such as a branch or function return).</p>
920 <p>The first basic block in a function is special in two ways: it is immediately
921 executed on entrance to the function, and it is not allowed to have
922 predecessor basic blocks (i.e. there can not be any branches to the entry
923 block of a function). Because the block can have no predecessors, it also
924 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
926 <p>LLVM allows an explicit section to be specified for functions. If the target
927 supports it, it will emit functions to the section specified.</p>
929 <p>An explicit alignment may be specified for a function. If not present, or if
930 the alignment is set to zero, the alignment of the function is set by the
931 target to whatever it feels convenient. If an explicit alignment is
932 specified, the function is forced to have at least that much alignment. All
933 alignments must be a power of 2.</p>
935 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
936 be significant and two identical functions can be merged.</p>
939 <pre class="doc_code">
940 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
941 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
942 <ResultType> @<FunctionName> ([argument list])
943 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
944 [<a href="#gc">gc</a>] { ... }
949 <!-- ======================================================================= -->
951 <a name="aliasstructure">Aliases</a>
956 <p>Aliases act as "second name" for the aliasee value (which can be either
957 function, global variable, another alias or bitcast of global value). Aliases
958 may have an optional <a href="#linkage">linkage type</a>, and an
959 optional <a href="#visibility">visibility style</a>.</p>
962 <pre class="doc_code">
963 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
968 <!-- ======================================================================= -->
970 <a name="namedmetadatastructure">Named Metadata</a>
975 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
976 nodes</a> (but not metadata strings) are the only valid operands for
977 a named metadata.</p>
980 <pre class="doc_code">
981 ; Some unnamed metadata nodes, which are referenced by the named metadata.
982 !0 = metadata !{metadata !"zero"}
983 !1 = metadata !{metadata !"one"}
984 !2 = metadata !{metadata !"two"}
986 !name = !{!0, !1, !2}
991 <!-- ======================================================================= -->
993 <a name="paramattrs">Parameter Attributes</a>
998 <p>The return type and each parameter of a function type may have a set of
999 <i>parameter attributes</i> associated with them. Parameter attributes are
1000 used to communicate additional information about the result or parameters of
1001 a function. Parameter attributes are considered to be part of the function,
1002 not of the function type, so functions with different parameter attributes
1003 can have the same function type.</p>
1005 <p>Parameter attributes are simple keywords that follow the type specified. If
1006 multiple parameter attributes are needed, they are space separated. For
1009 <pre class="doc_code">
1010 declare i32 @printf(i8* noalias nocapture, ...)
1011 declare i32 @atoi(i8 zeroext)
1012 declare signext i8 @returns_signed_char()
1015 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1016 <tt>readonly</tt>) come immediately after the argument list.</p>
1018 <p>Currently, only the following parameter attributes are defined:</p>
1021 <dt><tt><b>zeroext</b></tt></dt>
1022 <dd>This indicates to the code generator that the parameter or return value
1023 should be zero-extended to the extent required by the target's ABI (which
1024 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1025 parameter) or the callee (for a return value).</dd>
1027 <dt><tt><b>signext</b></tt></dt>
1028 <dd>This indicates to the code generator that the parameter or return value
1029 should be sign-extended to the extent required by the target's ABI (which
1030 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1033 <dt><tt><b>inreg</b></tt></dt>
1034 <dd>This indicates that this parameter or return value should be treated in a
1035 special target-dependent fashion during while emitting code for a function
1036 call or return (usually, by putting it in a register as opposed to memory,
1037 though some targets use it to distinguish between two different kinds of
1038 registers). Use of this attribute is target-specific.</dd>
1040 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1041 <dd><p>This indicates that the pointer parameter should really be passed by
1042 value to the function. The attribute implies that a hidden copy of the
1044 is made between the caller and the callee, so the callee is unable to
1045 modify the value in the callee. This attribute is only valid on LLVM
1046 pointer arguments. It is generally used to pass structs and arrays by
1047 value, but is also valid on pointers to scalars. The copy is considered
1048 to belong to the caller not the callee (for example,
1049 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1050 <tt>byval</tt> parameters). This is not a valid attribute for return
1053 <p>The byval attribute also supports specifying an alignment with
1054 the align attribute. It indicates the alignment of the stack slot to
1055 form and the known alignment of the pointer specified to the call site. If
1056 the alignment is not specified, then the code generator makes a
1057 target-specific assumption.</p></dd>
1059 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1060 <dd>This indicates that the pointer parameter specifies the address of a
1061 structure that is the return value of the function in the source program.
1062 This pointer must be guaranteed by the caller to be valid: loads and
1063 stores to the structure may be assumed by the callee to not to trap. This
1064 may only be applied to the first parameter. This is not a valid attribute
1065 for return values. </dd>
1067 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1068 <dd>This indicates that pointer values
1069 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1070 value do not alias pointer values which are not <i>based</i> on it,
1071 ignoring certain "irrelevant" dependencies.
1072 For a call to the parent function, dependencies between memory
1073 references from before or after the call and from those during the call
1074 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1075 return value used in that call.
1076 The caller shares the responsibility with the callee for ensuring that
1077 these requirements are met.
1078 For further details, please see the discussion of the NoAlias response in
1079 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1081 Note that this definition of <tt>noalias</tt> is intentionally
1082 similar to the definition of <tt>restrict</tt> in C99 for function
1083 arguments, though it is slightly weaker.
1085 For function return values, C99's <tt>restrict</tt> is not meaningful,
1086 while LLVM's <tt>noalias</tt> is.
1089 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1090 <dd>This indicates that the callee does not make any copies of the pointer
1091 that outlive the callee itself. This is not a valid attribute for return
1094 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1095 <dd>This indicates that the pointer parameter can be excised using the
1096 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1097 attribute for return values.</dd>
1102 <!-- ======================================================================= -->
1104 <a name="gc">Garbage Collector Names</a>
1109 <p>Each function may specify a garbage collector name, which is simply a
1112 <pre class="doc_code">
1113 define void @f() gc "name" { ... }
1116 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1117 collector which will cause the compiler to alter its output in order to
1118 support the named garbage collection algorithm.</p>
1122 <!-- ======================================================================= -->
1124 <a name="fnattrs">Function Attributes</a>
1129 <p>Function attributes are set to communicate additional information about a
1130 function. Function attributes are considered to be part of the function, not
1131 of the function type, so functions with different parameter attributes can
1132 have the same function type.</p>
1134 <p>Function attributes are simple keywords that follow the type specified. If
1135 multiple attributes are needed, they are space separated. For example:</p>
1137 <pre class="doc_code">
1138 define void @f() noinline { ... }
1139 define void @f() alwaysinline { ... }
1140 define void @f() alwaysinline optsize { ... }
1141 define void @f() optsize { ... }
1145 <dt><tt><b>address_safety</b></tt></dt>
1146 <dd>This attribute indicates that the address safety analysis
1147 is enabled for this function. </dd>
1149 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1150 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1151 the backend should forcibly align the stack pointer. Specify the
1152 desired alignment, which must be a power of two, in parentheses.
1154 <dt><tt><b>alwaysinline</b></tt></dt>
1155 <dd>This attribute indicates that the inliner should attempt to inline this
1156 function into callers whenever possible, ignoring any active inlining size
1157 threshold for this caller.</dd>
1159 <dt><tt><b>nonlazybind</b></tt></dt>
1160 <dd>This attribute suppresses lazy symbol binding for the function. This
1161 may make calls to the function faster, at the cost of extra program
1162 startup time if the function is not called during program startup.</dd>
1164 <dt><tt><b>inlinehint</b></tt></dt>
1165 <dd>This attribute indicates that the source code contained a hint that inlining
1166 this function is desirable (such as the "inline" keyword in C/C++). It
1167 is just a hint; it imposes no requirements on the inliner.</dd>
1169 <dt><tt><b>naked</b></tt></dt>
1170 <dd>This attribute disables prologue / epilogue emission for the function.
1171 This can have very system-specific consequences.</dd>
1173 <dt><tt><b>noimplicitfloat</b></tt></dt>
1174 <dd>This attributes disables implicit floating point instructions.</dd>
1176 <dt><tt><b>noinline</b></tt></dt>
1177 <dd>This attribute indicates that the inliner should never inline this
1178 function in any situation. This attribute may not be used together with
1179 the <tt>alwaysinline</tt> attribute.</dd>
1181 <dt><tt><b>noredzone</b></tt></dt>
1182 <dd>This attribute indicates that the code generator should not use a red
1183 zone, even if the target-specific ABI normally permits it.</dd>
1185 <dt><tt><b>noreturn</b></tt></dt>
1186 <dd>This function attribute indicates that the function never returns
1187 normally. This produces undefined behavior at runtime if the function
1188 ever does dynamically return.</dd>
1190 <dt><tt><b>nounwind</b></tt></dt>
1191 <dd>This function attribute indicates that the function never returns with an
1192 unwind or exceptional control flow. If the function does unwind, its
1193 runtime behavior is undefined.</dd>
1195 <dt><tt><b>optsize</b></tt></dt>
1196 <dd>This attribute suggests that optimization passes and code generator passes
1197 make choices that keep the code size of this function low, and otherwise
1198 do optimizations specifically to reduce code size.</dd>
1200 <dt><tt><b>readnone</b></tt></dt>
1201 <dd>This attribute indicates that the function computes its result (or decides
1202 to unwind an exception) based strictly on its arguments, without
1203 dereferencing any pointer arguments or otherwise accessing any mutable
1204 state (e.g. memory, control registers, etc) visible to caller functions.
1205 It does not write through any pointer arguments
1206 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1207 changes any state visible to callers. This means that it cannot unwind
1208 exceptions by calling the <tt>C++</tt> exception throwing methods.</dd>
1210 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1211 <dd>This attribute indicates that the function does not write through any
1212 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1213 arguments) or otherwise modify any state (e.g. memory, control registers,
1214 etc) visible to caller functions. It may dereference pointer arguments
1215 and read state that may be set in the caller. A readonly function always
1216 returns the same value (or unwinds an exception identically) when called
1217 with the same set of arguments and global state. It cannot unwind an
1218 exception by calling the <tt>C++</tt> exception throwing methods.</dd>
1220 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1221 <dd>This attribute indicates that this function can return twice. The
1222 C <code>setjmp</code> is an example of such a function. The compiler
1223 disables some optimizations (like tail calls) in the caller of these
1226 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1227 <dd>This attribute indicates that the function should emit a stack smashing
1228 protector. It is in the form of a "canary"—a random value placed on
1229 the stack before the local variables that's checked upon return from the
1230 function to see if it has been overwritten. A heuristic is used to
1231 determine if a function needs stack protectors or not.<br>
1233 If a function that has an <tt>ssp</tt> attribute is inlined into a
1234 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1235 function will have an <tt>ssp</tt> attribute.</dd>
1237 <dt><tt><b>sspreq</b></tt></dt>
1238 <dd>This attribute indicates that the function should <em>always</em> emit a
1239 stack smashing protector. This overrides
1240 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1242 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1243 function that doesn't have an <tt>sspreq</tt> attribute or which has
1244 an <tt>ssp</tt> attribute, then the resulting function will have
1245 an <tt>sspreq</tt> attribute.</dd>
1247 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1248 <dd>This attribute indicates that the ABI being targeted requires that
1249 an unwind table entry be produce for this function even if we can
1250 show that no exceptions passes by it. This is normally the case for
1251 the ELF x86-64 abi, but it can be disabled for some compilation
1257 <!-- ======================================================================= -->
1259 <a name="moduleasm">Module-Level Inline Assembly</a>
1264 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1265 the GCC "file scope inline asm" blocks. These blocks are internally
1266 concatenated by LLVM and treated as a single unit, but may be separated in
1267 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1269 <pre class="doc_code">
1270 module asm "inline asm code goes here"
1271 module asm "more can go here"
1274 <p>The strings can contain any character by escaping non-printable characters.
1275 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1278 <p>The inline asm code is simply printed to the machine code .s file when
1279 assembly code is generated.</p>
1283 <!-- ======================================================================= -->
1285 <a name="datalayout">Data Layout</a>
1290 <p>A module may specify a target specific data layout string that specifies how
1291 data is to be laid out in memory. The syntax for the data layout is
1294 <pre class="doc_code">
1295 target datalayout = "<i>layout specification</i>"
1298 <p>The <i>layout specification</i> consists of a list of specifications
1299 separated by the minus sign character ('-'). Each specification starts with
1300 a letter and may include other information after the letter to define some
1301 aspect of the data layout. The specifications accepted are as follows:</p>
1305 <dd>Specifies that the target lays out data in big-endian form. That is, the
1306 bits with the most significance have the lowest address location.</dd>
1309 <dd>Specifies that the target lays out data in little-endian form. That is,
1310 the bits with the least significance have the lowest address
1313 <dt><tt>S<i>size</i></tt></dt>
1314 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1315 of stack variables is limited to the natural stack alignment to avoid
1316 dynamic stack realignment. The stack alignment must be a multiple of
1317 8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1318 which does not prevent any alignment promotions.</dd>
1320 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1321 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1322 <i>preferred</i> alignments. All sizes are in bits. Specifying
1323 the <i>pref</i> alignment is optional. If omitted, the
1324 preceding <tt>:</tt> should be omitted too.</dd>
1326 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1327 <dd>This specifies the alignment for an integer type of a given bit
1328 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1330 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1331 <dd>This specifies the alignment for a vector type of a given bit
1334 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1335 <dd>This specifies the alignment for a floating point type of a given bit
1336 <i>size</i>. Only values of <i>size</i> that are supported by the target
1337 will work. 32 (float) and 64 (double) are supported on all targets;
1338 80 or 128 (different flavors of long double) are also supported on some
1341 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1342 <dd>This specifies the alignment for an aggregate type of a given bit
1345 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1346 <dd>This specifies the alignment for a stack object of a given bit
1349 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1350 <dd>This specifies a set of native integer widths for the target CPU
1351 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1352 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1353 this set are considered to support most general arithmetic
1354 operations efficiently.</dd>
1357 <p>When constructing the data layout for a given target, LLVM starts with a
1358 default set of specifications which are then (possibly) overridden by the
1359 specifications in the <tt>datalayout</tt> keyword. The default specifications
1360 are given in this list:</p>
1363 <li><tt>E</tt> - big endian</li>
1364 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1365 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1366 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1367 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1368 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1369 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1370 alignment of 64-bits</li>
1371 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1372 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1373 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1374 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1375 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1376 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1379 <p>When LLVM is determining the alignment for a given type, it uses the
1380 following rules:</p>
1383 <li>If the type sought is an exact match for one of the specifications, that
1384 specification is used.</li>
1386 <li>If no match is found, and the type sought is an integer type, then the
1387 smallest integer type that is larger than the bitwidth of the sought type
1388 is used. If none of the specifications are larger than the bitwidth then
1389 the the largest integer type is used. For example, given the default
1390 specifications above, the i7 type will use the alignment of i8 (next
1391 largest) while both i65 and i256 will use the alignment of i64 (largest
1394 <li>If no match is found, and the type sought is a vector type, then the
1395 largest vector type that is smaller than the sought vector type will be
1396 used as a fall back. This happens because <128 x double> can be
1397 implemented in terms of 64 <2 x double>, for example.</li>
1400 <p>The function of the data layout string may not be what you expect. Notably,
1401 this is not a specification from the frontend of what alignment the code
1402 generator should use.</p>
1404 <p>Instead, if specified, the target data layout is required to match what the
1405 ultimate <em>code generator</em> expects. This string is used by the
1406 mid-level optimizers to
1407 improve code, and this only works if it matches what the ultimate code
1408 generator uses. If you would like to generate IR that does not embed this
1409 target-specific detail into the IR, then you don't have to specify the
1410 string. This will disable some optimizations that require precise layout
1411 information, but this also prevents those optimizations from introducing
1412 target specificity into the IR.</p>
1418 <!-- ======================================================================= -->
1420 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1425 <p>Any memory access must be done through a pointer value associated
1426 with an address range of the memory access, otherwise the behavior
1427 is undefined. Pointer values are associated with address ranges
1428 according to the following rules:</p>
1431 <li>A pointer value is associated with the addresses associated with
1432 any value it is <i>based</i> on.
1433 <li>An address of a global variable is associated with the address
1434 range of the variable's storage.</li>
1435 <li>The result value of an allocation instruction is associated with
1436 the address range of the allocated storage.</li>
1437 <li>A null pointer in the default address-space is associated with
1439 <li>An integer constant other than zero or a pointer value returned
1440 from a function not defined within LLVM may be associated with address
1441 ranges allocated through mechanisms other than those provided by
1442 LLVM. Such ranges shall not overlap with any ranges of addresses
1443 allocated by mechanisms provided by LLVM.</li>
1446 <p>A pointer value is <i>based</i> on another pointer value according
1447 to the following rules:</p>
1450 <li>A pointer value formed from a
1451 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1452 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1453 <li>The result value of a
1454 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1455 of the <tt>bitcast</tt>.</li>
1456 <li>A pointer value formed by an
1457 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1458 pointer values that contribute (directly or indirectly) to the
1459 computation of the pointer's value.</li>
1460 <li>The "<i>based</i> on" relationship is transitive.</li>
1463 <p>Note that this definition of <i>"based"</i> is intentionally
1464 similar to the definition of <i>"based"</i> in C99, though it is
1465 slightly weaker.</p>
1467 <p>LLVM IR does not associate types with memory. The result type of a
1468 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1469 alignment of the memory from which to load, as well as the
1470 interpretation of the value. The first operand type of a
1471 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1472 and alignment of the store.</p>
1474 <p>Consequently, type-based alias analysis, aka TBAA, aka
1475 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1476 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1477 additional information which specialized optimization passes may use
1478 to implement type-based alias analysis.</p>
1482 <!-- ======================================================================= -->
1484 <a name="volatile">Volatile Memory Accesses</a>
1489 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1490 href="#i_store"><tt>store</tt></a>s, and <a
1491 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1492 The optimizers must not change the number of volatile operations or change their
1493 order of execution relative to other volatile operations. The optimizers
1494 <i>may</i> change the order of volatile operations relative to non-volatile
1495 operations. This is not Java's "volatile" and has no cross-thread
1496 synchronization behavior.</p>
1500 <!-- ======================================================================= -->
1502 <a name="memmodel">Memory Model for Concurrent Operations</a>
1507 <p>The LLVM IR does not define any way to start parallel threads of execution
1508 or to register signal handlers. Nonetheless, there are platform-specific
1509 ways to create them, and we define LLVM IR's behavior in their presence. This
1510 model is inspired by the C++0x memory model.</p>
1512 <p>For a more informal introduction to this model, see the
1513 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1515 <p>We define a <i>happens-before</i> partial order as the least partial order
1518 <li>Is a superset of single-thread program order, and</li>
1519 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1520 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1521 by platform-specific techniques, like pthread locks, thread
1522 creation, thread joining, etc., and by atomic instructions.
1523 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1527 <p>Note that program order does not introduce <i>happens-before</i> edges
1528 between a thread and signals executing inside that thread.</p>
1530 <p>Every (defined) read operation (load instructions, memcpy, atomic
1531 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1532 (defined) write operations (store instructions, atomic
1533 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1534 initialized globals are considered to have a write of the initializer which is
1535 atomic and happens before any other read or write of the memory in question.
1536 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1537 any write to the same byte, except:</p>
1540 <li>If <var>write<sub>1</sub></var> happens before
1541 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1542 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1543 does not see <var>write<sub>1</sub></var>.
1544 <li>If <var>R<sub>byte</sub></var> happens before
1545 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1546 see <var>write<sub>3</sub></var>.
1549 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1551 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1552 is supposed to give guarantees which can support
1553 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1554 addresses which do not behave like normal memory. It does not generally
1555 provide cross-thread synchronization.)
1556 <li>Otherwise, if there is no write to the same byte that happens before
1557 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1558 <tt>undef</tt> for that byte.
1559 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1560 <var>R<sub>byte</sub></var> returns the value written by that
1562 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1563 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1564 values written. See the <a href="#ordering">Atomic Memory Ordering
1565 Constraints</a> section for additional constraints on how the choice
1567 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1570 <p><var>R</var> returns the value composed of the series of bytes it read.
1571 This implies that some bytes within the value may be <tt>undef</tt>
1572 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1573 defines the semantics of the operation; it doesn't mean that targets will
1574 emit more than one instruction to read the series of bytes.</p>
1576 <p>Note that in cases where none of the atomic intrinsics are used, this model
1577 places only one restriction on IR transformations on top of what is required
1578 for single-threaded execution: introducing a store to a byte which might not
1579 otherwise be stored is not allowed in general. (Specifically, in the case
1580 where another thread might write to and read from an address, introducing a
1581 store can change a load that may see exactly one write into a load that may
1582 see multiple writes.)</p>
1584 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1585 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1586 none of the backends currently in the tree fall into this category; however,
1587 there might be targets which care. If there are, we want a paragraph
1590 Targets may specify that stores narrower than a certain width are not
1591 available; on such a target, for the purposes of this model, treat any
1592 non-atomic write with an alignment or width less than the minimum width
1593 as if it writes to the relevant surrounding bytes.
1598 <!-- ======================================================================= -->
1600 <a name="ordering">Atomic Memory Ordering Constraints</a>
1605 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1606 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1607 <a href="#i_fence"><code>fence</code></a>,
1608 <a href="#i_load"><code>atomic load</code></a>, and
1609 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1610 that determines which other atomic instructions on the same address they
1611 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1612 but are somewhat more colloquial. If these descriptions aren't precise enough,
1613 check those specs (see spec references in the
1614 <a href="Atomics.html#introduction">atomics guide</a>).
1615 <a href="#i_fence"><code>fence</code></a> instructions
1616 treat these orderings somewhat differently since they don't take an address.
1617 See that instruction's documentation for details.</p>
1619 <p>For a simpler introduction to the ordering constraints, see the
1620 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1623 <dt><code>unordered</code></dt>
1624 <dd>The set of values that can be read is governed by the happens-before
1625 partial order. A value cannot be read unless some operation wrote it.
1626 This is intended to provide a guarantee strong enough to model Java's
1627 non-volatile shared variables. This ordering cannot be specified for
1628 read-modify-write operations; it is not strong enough to make them atomic
1629 in any interesting way.</dd>
1630 <dt><code>monotonic</code></dt>
1631 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1632 total order for modifications by <code>monotonic</code> operations on each
1633 address. All modification orders must be compatible with the happens-before
1634 order. There is no guarantee that the modification orders can be combined to
1635 a global total order for the whole program (and this often will not be
1636 possible). The read in an atomic read-modify-write operation
1637 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1638 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1639 reads the value in the modification order immediately before the value it
1640 writes. If one atomic read happens before another atomic read of the same
1641 address, the later read must see the same value or a later value in the
1642 address's modification order. This disallows reordering of
1643 <code>monotonic</code> (or stronger) operations on the same address. If an
1644 address is written <code>monotonic</code>ally by one thread, and other threads
1645 <code>monotonic</code>ally read that address repeatedly, the other threads must
1646 eventually see the write. This corresponds to the C++0x/C1x
1647 <code>memory_order_relaxed</code>.</dd>
1648 <dt><code>acquire</code></dt>
1649 <dd>In addition to the guarantees of <code>monotonic</code>,
1650 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1651 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1652 <dt><code>release</code></dt>
1653 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1654 writes a value which is subsequently read by an <code>acquire</code> operation,
1655 it <i>synchronizes-with</i> that operation. (This isn't a complete
1656 description; see the C++0x definition of a release sequence.) This corresponds
1657 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1658 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1659 <code>acquire</code> and <code>release</code> operation on its address.
1660 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1661 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1662 <dd>In addition to the guarantees of <code>acq_rel</code>
1663 (<code>acquire</code> for an operation which only reads, <code>release</code>
1664 for an operation which only writes), there is a global total order on all
1665 sequentially-consistent operations on all addresses, which is consistent with
1666 the <i>happens-before</i> partial order and with the modification orders of
1667 all the affected addresses. Each sequentially-consistent read sees the last
1668 preceding write to the same address in this global order. This corresponds
1669 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1672 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1673 it only <i>synchronizes with</i> or participates in modification and seq_cst
1674 total orderings with other operations running in the same thread (for example,
1675 in signal handlers).</p>
1681 <!-- *********************************************************************** -->
1682 <h2><a name="typesystem">Type System</a></h2>
1683 <!-- *********************************************************************** -->
1687 <p>The LLVM type system is one of the most important features of the
1688 intermediate representation. Being typed enables a number of optimizations
1689 to be performed on the intermediate representation directly, without having
1690 to do extra analyses on the side before the transformation. A strong type
1691 system makes it easier to read the generated code and enables novel analyses
1692 and transformations that are not feasible to perform on normal three address
1693 code representations.</p>
1695 <!-- ======================================================================= -->
1697 <a name="t_classifications">Type Classifications</a>
1702 <p>The types fall into a few useful classifications:</p>
1704 <table border="1" cellspacing="0" cellpadding="4">
1706 <tr><th>Classification</th><th>Types</th></tr>
1708 <td><a href="#t_integer">integer</a></td>
1709 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1712 <td><a href="#t_floating">floating point</a></td>
1713 <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1716 <td><a name="t_firstclass">first class</a></td>
1717 <td><a href="#t_integer">integer</a>,
1718 <a href="#t_floating">floating point</a>,
1719 <a href="#t_pointer">pointer</a>,
1720 <a href="#t_vector">vector</a>,
1721 <a href="#t_struct">structure</a>,
1722 <a href="#t_array">array</a>,
1723 <a href="#t_label">label</a>,
1724 <a href="#t_metadata">metadata</a>.
1728 <td><a href="#t_primitive">primitive</a></td>
1729 <td><a href="#t_label">label</a>,
1730 <a href="#t_void">void</a>,
1731 <a href="#t_integer">integer</a>,
1732 <a href="#t_floating">floating point</a>,
1733 <a href="#t_x86mmx">x86mmx</a>,
1734 <a href="#t_metadata">metadata</a>.</td>
1737 <td><a href="#t_derived">derived</a></td>
1738 <td><a href="#t_array">array</a>,
1739 <a href="#t_function">function</a>,
1740 <a href="#t_pointer">pointer</a>,
1741 <a href="#t_struct">structure</a>,
1742 <a href="#t_vector">vector</a>,
1743 <a href="#t_opaque">opaque</a>.
1749 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1750 important. Values of these types are the only ones which can be produced by
1755 <!-- ======================================================================= -->
1757 <a name="t_primitive">Primitive Types</a>
1762 <p>The primitive types are the fundamental building blocks of the LLVM
1765 <!-- _______________________________________________________________________ -->
1767 <a name="t_integer">Integer Type</a>
1773 <p>The integer type is a very simple type that simply specifies an arbitrary
1774 bit width for the integer type desired. Any bit width from 1 bit to
1775 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1782 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1786 <table class="layout">
1788 <td class="left"><tt>i1</tt></td>
1789 <td class="left">a single-bit integer.</td>
1792 <td class="left"><tt>i32</tt></td>
1793 <td class="left">a 32-bit integer.</td>
1796 <td class="left"><tt>i1942652</tt></td>
1797 <td class="left">a really big integer of over 1 million bits.</td>
1803 <!-- _______________________________________________________________________ -->
1805 <a name="t_floating">Floating Point Types</a>
1812 <tr><th>Type</th><th>Description</th></tr>
1813 <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr>
1814 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1815 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1816 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1817 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1818 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1824 <!-- _______________________________________________________________________ -->
1826 <a name="t_x86mmx">X86mmx Type</a>
1832 <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>
1841 <!-- _______________________________________________________________________ -->
1843 <a name="t_void">Void Type</a>
1849 <p>The void type does not represent any value and has no size.</p>
1858 <!-- _______________________________________________________________________ -->
1860 <a name="t_label">Label Type</a>
1866 <p>The label type represents code labels.</p>
1875 <!-- _______________________________________________________________________ -->
1877 <a name="t_metadata">Metadata Type</a>
1883 <p>The metadata type represents embedded metadata. No derived types may be
1884 created from metadata except for <a href="#t_function">function</a>
1896 <!-- ======================================================================= -->
1898 <a name="t_derived">Derived Types</a>
1903 <p>The real power in LLVM comes from the derived types in the system. This is
1904 what allows a programmer to represent arrays, functions, pointers, and other
1905 useful types. Each of these types contain one or more element types which
1906 may be a primitive type, or another derived type. For example, it is
1907 possible to have a two dimensional array, using an array as the element type
1908 of another array.</p>
1910 <!-- _______________________________________________________________________ -->
1912 <a name="t_aggregate">Aggregate Types</a>
1917 <p>Aggregate Types are a subset of derived types that can contain multiple
1918 member types. <a href="#t_array">Arrays</a> and
1919 <a href="#t_struct">structs</a> are aggregate types.
1920 <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p>
1924 <!-- _______________________________________________________________________ -->
1926 <a name="t_array">Array Type</a>
1932 <p>The array type is a very simple derived type that arranges elements
1933 sequentially in memory. The array type requires a size (number of elements)
1934 and an underlying data type.</p>
1938 [<# elements> x <elementtype>]
1941 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1942 be any type with a size.</p>
1945 <table class="layout">
1947 <td class="left"><tt>[40 x i32]</tt></td>
1948 <td class="left">Array of 40 32-bit integer values.</td>
1951 <td class="left"><tt>[41 x i32]</tt></td>
1952 <td class="left">Array of 41 32-bit integer values.</td>
1955 <td class="left"><tt>[4 x i8]</tt></td>
1956 <td class="left">Array of 4 8-bit integer values.</td>
1959 <p>Here are some examples of multidimensional arrays:</p>
1960 <table class="layout">
1962 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1963 <td class="left">3x4 array of 32-bit integer values.</td>
1966 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1967 <td class="left">12x10 array of single precision floating point values.</td>
1970 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1971 <td class="left">2x3x4 array of 16-bit integer values.</td>
1975 <p>There is no restriction on indexing beyond the end of the array implied by
1976 a static type (though there are restrictions on indexing beyond the bounds
1977 of an allocated object in some cases). This means that single-dimension
1978 'variable sized array' addressing can be implemented in LLVM with a zero
1979 length array type. An implementation of 'pascal style arrays' in LLVM could
1980 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1984 <!-- _______________________________________________________________________ -->
1986 <a name="t_function">Function Type</a>
1992 <p>The function type can be thought of as a function signature. It consists of
1993 a return type and a list of formal parameter types. The return type of a
1994 function type is a first class type or a void type.</p>
1998 <returntype> (<parameter list>)
2001 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
2002 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
2003 which indicates that the function takes a variable number of arguments.
2004 Variable argument functions can access their arguments with
2005 the <a href="#int_varargs">variable argument handling intrinsic</a>
2006 functions. '<tt><returntype></tt>' is any type except
2007 <a href="#t_label">label</a>.</p>
2010 <table class="layout">
2012 <td class="left"><tt>i32 (i32)</tt></td>
2013 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2015 </tr><tr class="layout">
2016 <td class="left"><tt>float (i16, i32 *) *
2018 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2019 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2020 returning <tt>float</tt>.
2022 </tr><tr class="layout">
2023 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2024 <td class="left">A vararg function that takes at least one
2025 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2026 which returns an integer. This is the signature for <tt>printf</tt> in
2029 </tr><tr class="layout">
2030 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2031 <td class="left">A function taking an <tt>i32</tt>, returning a
2032 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2039 <!-- _______________________________________________________________________ -->
2041 <a name="t_struct">Structure Type</a>
2047 <p>The structure type is used to represent a collection of data members together
2048 in memory. The elements of a structure may be any type that has a size.</p>
2050 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2051 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2052 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2053 Structures in registers are accessed using the
2054 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2055 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2057 <p>Structures may optionally be "packed" structures, which indicate that the
2058 alignment of the struct is one byte, and that there is no padding between
2059 the elements. In non-packed structs, padding between field types is inserted
2060 as defined by the TargetData string in the module, which is required to match
2061 what the underlying code generator expects.</p>
2063 <p>Structures can either be "literal" or "identified". A literal structure is
2064 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2065 types are always defined at the top level with a name. Literal types are
2066 uniqued by their contents and can never be recursive or opaque since there is
2067 no way to write one. Identified types can be recursive, can be opaqued, and are
2073 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2074 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2078 <table class="layout">
2080 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2081 <td class="left">A triple of three <tt>i32</tt> values</td>
2084 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2085 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2086 second element is a <a href="#t_pointer">pointer</a> to a
2087 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2088 an <tt>i32</tt>.</td>
2091 <td class="left"><tt><{ i8, i32 }></tt></td>
2092 <td class="left">A packed struct known to be 5 bytes in size.</td>
2098 <!-- _______________________________________________________________________ -->
2100 <a name="t_opaque">Opaque Structure Types</a>
2106 <p>Opaque structure types are used to represent named structure types that do
2107 not have a body specified. This corresponds (for example) to the C notion of
2108 a forward declared structure.</p>
2117 <table class="layout">
2119 <td class="left"><tt>opaque</tt></td>
2120 <td class="left">An opaque type.</td>
2128 <!-- _______________________________________________________________________ -->
2130 <a name="t_pointer">Pointer Type</a>
2136 <p>The pointer type is used to specify memory locations.
2137 Pointers are commonly used to reference objects in memory.</p>
2139 <p>Pointer types may have an optional address space attribute defining the
2140 numbered address space where the pointed-to object resides. The default
2141 address space is number zero. The semantics of non-zero address
2142 spaces are target-specific.</p>
2144 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2145 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2153 <table class="layout">
2155 <td class="left"><tt>[4 x i32]*</tt></td>
2156 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2157 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2160 <td class="left"><tt>i32 (i32*) *</tt></td>
2161 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2162 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2166 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2167 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2168 that resides in address space #5.</td>
2174 <!-- _______________________________________________________________________ -->
2176 <a name="t_vector">Vector Type</a>
2182 <p>A vector type is a simple derived type that represents a vector of elements.
2183 Vector types are used when multiple primitive data are operated in parallel
2184 using a single instruction (SIMD). A vector type requires a size (number of
2185 elements) and an underlying primitive data type. Vector types are considered
2186 <a href="#t_firstclass">first class</a>.</p>
2190 < <# elements> x <elementtype> >
2193 <p>The number of elements is a constant integer value larger than 0; elementtype
2194 may be any integer or floating point type, or a pointer to these types.
2195 Vectors of size zero are not allowed. </p>
2198 <table class="layout">
2200 <td class="left"><tt><4 x i32></tt></td>
2201 <td class="left">Vector of 4 32-bit integer values.</td>
2204 <td class="left"><tt><8 x float></tt></td>
2205 <td class="left">Vector of 8 32-bit floating-point values.</td>
2208 <td class="left"><tt><2 x i64></tt></td>
2209 <td class="left">Vector of 2 64-bit integer values.</td>
2212 <td class="left"><tt><4 x i64*></tt></td>
2213 <td class="left">Vector of 4 pointers to 64-bit integer values.</td>
2223 <!-- *********************************************************************** -->
2224 <h2><a name="constants">Constants</a></h2>
2225 <!-- *********************************************************************** -->
2229 <p>LLVM has several different basic types of constants. This section describes
2230 them all and their syntax.</p>
2232 <!-- ======================================================================= -->
2234 <a name="simpleconstants">Simple Constants</a>
2240 <dt><b>Boolean constants</b></dt>
2241 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2242 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2244 <dt><b>Integer constants</b></dt>
2245 <dd>Standard integers (such as '4') are constants of
2246 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2247 with integer types.</dd>
2249 <dt><b>Floating point constants</b></dt>
2250 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2251 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2252 notation (see below). The assembler requires the exact decimal value of a
2253 floating-point constant. For example, the assembler accepts 1.25 but
2254 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2255 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2257 <dt><b>Null pointer constants</b></dt>
2258 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2259 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2262 <p>The one non-intuitive notation for constants is the hexadecimal form of
2263 floating point constants. For example, the form '<tt>double
2264 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2265 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2266 constants are required (and the only time that they are generated by the
2267 disassembler) is when a floating point constant must be emitted but it cannot
2268 be represented as a decimal floating point number in a reasonable number of
2269 digits. For example, NaN's, infinities, and other special values are
2270 represented in their IEEE hexadecimal format so that assembly and disassembly
2271 do not cause any bits to change in the constants.</p>
2273 <p>When using the hexadecimal form, constants of types half, float, and double are
2274 represented using the 16-digit form shown above (which matches the IEEE754
2275 representation for double); half and float values must, however, be exactly
2276 representable as IEE754 half and single precision, respectively.
2277 Hexadecimal format is always used
2278 for long double, and there are three forms of long double. The 80-bit format
2279 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2280 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2281 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2282 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2283 currently supported target uses this format. Long doubles will only work if
2284 they match the long double format on your target. All hexadecimal formats
2285 are big-endian (sign bit at the left).</p>
2287 <p>There are no constants of type x86mmx.</p>
2290 <!-- ======================================================================= -->
2292 <a name="aggregateconstants"></a> <!-- old anchor -->
2293 <a name="complexconstants">Complex Constants</a>
2298 <p>Complex constants are a (potentially recursive) combination of simple
2299 constants and smaller complex constants.</p>
2302 <dt><b>Structure constants</b></dt>
2303 <dd>Structure constants are represented with notation similar to structure
2304 type definitions (a comma separated list of elements, surrounded by braces
2305 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2306 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2307 Structure constants must have <a href="#t_struct">structure type</a>, and
2308 the number and types of elements must match those specified by the
2311 <dt><b>Array constants</b></dt>
2312 <dd>Array constants are represented with notation similar to array type
2313 definitions (a comma separated list of elements, surrounded by square
2314 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2315 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2316 the number and types of elements must match those specified by the
2319 <dt><b>Vector constants</b></dt>
2320 <dd>Vector constants are represented with notation similar to vector type
2321 definitions (a comma separated list of elements, surrounded by
2322 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2323 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2324 have <a href="#t_vector">vector type</a>, and the number and types of
2325 elements must match those specified by the type.</dd>
2327 <dt><b>Zero initialization</b></dt>
2328 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2329 value to zero of <em>any</em> type, including scalar and
2330 <a href="#t_aggregate">aggregate</a> types.
2331 This is often used to avoid having to print large zero initializers
2332 (e.g. for large arrays) and is always exactly equivalent to using explicit
2333 zero initializers.</dd>
2335 <dt><b>Metadata node</b></dt>
2336 <dd>A metadata node is a structure-like constant with
2337 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2338 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2339 be interpreted as part of the instruction stream, metadata is a place to
2340 attach additional information such as debug info.</dd>
2345 <!-- ======================================================================= -->
2347 <a name="globalconstants">Global Variable and Function Addresses</a>
2352 <p>The addresses of <a href="#globalvars">global variables</a>
2353 and <a href="#functionstructure">functions</a> are always implicitly valid
2354 (link-time) constants. These constants are explicitly referenced when
2355 the <a href="#identifiers">identifier for the global</a> is used and always
2356 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2357 legal LLVM file:</p>
2359 <pre class="doc_code">
2362 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2367 <!-- ======================================================================= -->
2369 <a name="undefvalues">Undefined Values</a>
2374 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2375 indicates that the user of the value may receive an unspecified bit-pattern.
2376 Undefined values may be of any type (other than '<tt>label</tt>'
2377 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2379 <p>Undefined values are useful because they indicate to the compiler that the
2380 program is well defined no matter what value is used. This gives the
2381 compiler more freedom to optimize. Here are some examples of (potentially
2382 surprising) transformations that are valid (in pseudo IR):</p>
2385 <pre class="doc_code">
2395 <p>This is safe because all of the output bits are affected by the undef bits.
2396 Any output bit can have a zero or one depending on the input bits.</p>
2398 <pre class="doc_code">
2409 <p>These logical operations have bits that are not always affected by the input.
2410 For example, if <tt>%X</tt> has a zero bit, then the output of the
2411 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2412 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2413 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2414 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2415 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2416 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2417 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2419 <pre class="doc_code">
2420 %A = select undef, %X, %Y
2421 %B = select undef, 42, %Y
2422 %C = select %X, %Y, undef
2433 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2434 branch) conditions can go <em>either way</em>, but they have to come from one
2435 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2436 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2437 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2438 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2439 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2442 <pre class="doc_code">
2443 %A = xor undef, undef
2461 <p>This example points out that two '<tt>undef</tt>' operands are not
2462 necessarily the same. This can be surprising to people (and also matches C
2463 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2464 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2465 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2466 its value over its "live range". This is true because the variable doesn't
2467 actually <em>have a live range</em>. Instead, the value is logically read
2468 from arbitrary registers that happen to be around when needed, so the value
2469 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2470 need to have the same semantics or the core LLVM "replace all uses with"
2471 concept would not hold.</p>
2473 <pre class="doc_code">
2481 <p>These examples show the crucial difference between an <em>undefined
2482 value</em> and <em>undefined behavior</em>. An undefined value (like
2483 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2484 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2485 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2486 defined on SNaN's. However, in the second example, we can make a more
2487 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2488 arbitrary value, we are allowed to assume that it could be zero. Since a
2489 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2490 the operation does not execute at all. This allows us to delete the divide and
2491 all code after it. Because the undefined operation "can't happen", the
2492 optimizer can assume that it occurs in dead code.</p>
2494 <pre class="doc_code">
2495 a: store undef -> %X
2496 b: store %X -> undef
2502 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2503 undefined value can be assumed to not have any effect; we can assume that the
2504 value is overwritten with bits that happen to match what was already there.
2505 However, a store <em>to</em> an undefined location could clobber arbitrary
2506 memory, therefore, it has undefined behavior.</p>
2510 <!-- ======================================================================= -->
2512 <a name="poisonvalues">Poison Values</a>
2517 <p>Poison values are similar to <a href="#undefvalues">undef values</a>, however
2518 they also represent the fact that an instruction or constant expression which
2519 cannot evoke side effects has nevertheless detected a condition which results
2520 in undefined behavior.</p>
2522 <p>There is currently no way of representing a poison value in the IR; they
2523 only exist when produced by operations such as
2524 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2526 <p>Poison value behavior is defined in terms of value <i>dependence</i>:</p>
2529 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2530 their operands.</li>
2532 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2533 to their dynamic predecessor basic block.</li>
2535 <li>Function arguments depend on the corresponding actual argument values in
2536 the dynamic callers of their functions.</li>
2538 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2539 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2540 control back to them.</li>
2542 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2543 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>,
2544 or exception-throwing call instructions that dynamically transfer control
2547 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2548 referenced memory addresses, following the order in the IR
2549 (including loads and stores implied by intrinsics such as
2550 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2552 <!-- TODO: In the case of multiple threads, this only applies if the store
2553 "happens-before" the load or store. -->
2555 <!-- TODO: floating-point exception state -->
2557 <li>An instruction with externally visible side effects depends on the most
2558 recent preceding instruction with externally visible side effects, following
2559 the order in the IR. (This includes
2560 <a href="#volatile">volatile operations</a>.)</li>
2562 <li>An instruction <i>control-depends</i> on a
2563 <a href="#terminators">terminator instruction</a>
2564 if the terminator instruction has multiple successors and the instruction
2565 is always executed when control transfers to one of the successors, and
2566 may not be executed when control is transferred to another.</li>
2568 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2569 instruction if the set of instructions it otherwise depends on would be
2570 different if the terminator had transferred control to a different
2573 <li>Dependence is transitive.</li>
2577 <p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>,
2578 with the additional affect that any instruction which has a <i>dependence</i>
2579 on a poison value has undefined behavior.</p>
2581 <p>Here are some examples:</p>
2583 <pre class="doc_code">
2585 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2586 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2587 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2588 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2590 store i32 %poison, i32* @g ; Poison value stored to memory.
2591 %poison2 = load i32* @g ; Poison value loaded back from memory.
2593 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2595 %narrowaddr = bitcast i32* @g to i16*
2596 %wideaddr = bitcast i32* @g to i64*
2597 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2598 %poison4 = load i64* %wideaddr ; Returns a poison value.
2600 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2601 br i1 %cmp, label %true, label %end ; Branch to either destination.
2604 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2605 ; it has undefined behavior.
2609 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2610 ; Both edges into this PHI are
2611 ; control-dependent on %cmp, so this
2612 ; always results in a poison value.
2614 store volatile i32 0, i32* @g ; This would depend on the store in %true
2615 ; if %cmp is true, or the store in %entry
2616 ; otherwise, so this is undefined behavior.
2618 br i1 %cmp, label %second_true, label %second_end
2619 ; The same branch again, but this time the
2620 ; true block doesn't have side effects.
2627 store volatile i32 0, i32* @g ; This time, the instruction always depends
2628 ; on the store in %end. Also, it is
2629 ; control-equivalent to %end, so this is
2630 ; well-defined (ignoring earlier undefined
2631 ; behavior in this example).
2636 <!-- ======================================================================= -->
2638 <a name="blockaddress">Addresses of Basic Blocks</a>
2643 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2645 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2646 basic block in the specified function, and always has an i8* type. Taking
2647 the address of the entry block is illegal.</p>
2649 <p>This value only has defined behavior when used as an operand to the
2650 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2651 comparisons against null. Pointer equality tests between labels addresses
2652 results in undefined behavior — though, again, comparison against null
2653 is ok, and no label is equal to the null pointer. This may be passed around
2654 as an opaque pointer sized value as long as the bits are not inspected. This
2655 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2656 long as the original value is reconstituted before the <tt>indirectbr</tt>
2659 <p>Finally, some targets may provide defined semantics when using the value as
2660 the operand to an inline assembly, but that is target specific.</p>
2665 <!-- ======================================================================= -->
2667 <a name="constantexprs">Constant Expressions</a>
2672 <p>Constant expressions are used to allow expressions involving other constants
2673 to be used as constants. Constant expressions may be of
2674 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2675 operation that does not have side effects (e.g. load and call are not
2676 supported). The following is the syntax for constant expressions:</p>
2679 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2680 <dd>Truncate a constant to another type. The bit size of CST must be larger
2681 than the bit size of TYPE. Both types must be integers.</dd>
2683 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2684 <dd>Zero extend a constant to another type. The bit size of CST must be
2685 smaller than the bit size of TYPE. Both types must be integers.</dd>
2687 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2688 <dd>Sign extend a constant to another type. The bit size of CST must be
2689 smaller than the bit size of TYPE. Both types must be integers.</dd>
2691 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2692 <dd>Truncate a floating point constant to another floating point type. The
2693 size of CST must be larger than the size of TYPE. Both types must be
2694 floating point.</dd>
2696 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2697 <dd>Floating point extend a constant to another type. The size of CST must be
2698 smaller or equal to the size of TYPE. Both types must be floating
2701 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2702 <dd>Convert a floating point constant to the corresponding unsigned integer
2703 constant. TYPE must be a scalar or vector integer type. CST must be of
2704 scalar or vector floating point type. Both CST and TYPE must be scalars,
2705 or vectors of the same number of elements. If the value won't fit in the
2706 integer type, the results are undefined.</dd>
2708 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2709 <dd>Convert a floating point constant to the corresponding signed integer
2710 constant. TYPE must be a scalar or vector integer type. CST must be of
2711 scalar or vector floating point type. Both CST and TYPE must be scalars,
2712 or vectors of the same number of elements. If the value won't fit in the
2713 integer type, the results are undefined.</dd>
2715 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2716 <dd>Convert an unsigned integer constant to the corresponding floating point
2717 constant. TYPE must be a scalar or vector floating point type. CST must be
2718 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2719 vectors of the same number of elements. If the value won't fit in the
2720 floating point type, the results are undefined.</dd>
2722 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2723 <dd>Convert a signed integer constant to the corresponding floating point
2724 constant. TYPE must be a scalar or vector floating point type. CST must be
2725 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2726 vectors of the same number of elements. If the value won't fit in the
2727 floating point type, the results are undefined.</dd>
2729 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2730 <dd>Convert a pointer typed constant to the corresponding integer constant
2731 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2732 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2733 make it fit in <tt>TYPE</tt>.</dd>
2735 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2736 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2737 type. CST must be of integer type. The CST value is zero extended,
2738 truncated, or unchanged to make it fit in a pointer size. This one is
2739 <i>really</i> dangerous!</dd>
2741 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2742 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2743 are the same as those for the <a href="#i_bitcast">bitcast
2744 instruction</a>.</dd>
2746 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2747 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2748 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2749 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2750 instruction, the index list may have zero or more indexes, which are
2751 required to make sense for the type of "CSTPTR".</dd>
2753 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2754 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2756 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2757 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2759 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2760 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2762 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2763 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2766 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2767 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2770 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2771 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2774 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2775 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2776 constants. The index list is interpreted in a similar manner as indices in
2777 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2778 index value must be specified.</dd>
2780 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2781 <dd>Perform the <a href="#i_insertvalue">insertvalue 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>OPCODE (LHS, RHS)</tt></b></dt>
2787 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2788 be any of the <a href="#binaryops">binary</a>
2789 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2790 on operands are the same as those for the corresponding instruction
2791 (e.g. no bitwise operations on floating point values are allowed).</dd>
2798 <!-- *********************************************************************** -->
2799 <h2><a name="othervalues">Other Values</a></h2>
2800 <!-- *********************************************************************** -->
2802 <!-- ======================================================================= -->
2804 <a name="inlineasm">Inline Assembler Expressions</a>
2809 <p>LLVM supports inline assembler expressions (as opposed
2810 to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of
2811 a special value. This value represents the inline assembler as a string
2812 (containing the instructions to emit), a list of operand constraints (stored
2813 as a string), a flag that indicates whether or not the inline asm
2814 expression has side effects, and a flag indicating whether the function
2815 containing the asm needs to align its stack conservatively. An example
2816 inline assembler expression is:</p>
2818 <pre class="doc_code">
2819 i32 (i32) asm "bswap $0", "=r,r"
2822 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2823 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2826 <pre class="doc_code">
2827 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2830 <p>Inline asms with side effects not visible in the constraint list must be
2831 marked as having side effects. This is done through the use of the
2832 '<tt>sideeffect</tt>' keyword, like so:</p>
2834 <pre class="doc_code">
2835 call void asm sideeffect "eieio", ""()
2838 <p>In some cases inline asms will contain code that will not work unless the
2839 stack is aligned in some way, such as calls or SSE instructions on x86,
2840 yet will not contain code that does that alignment within the asm.
2841 The compiler should make conservative assumptions about what the asm might
2842 contain and should generate its usual stack alignment code in the prologue
2843 if the '<tt>alignstack</tt>' keyword is present:</p>
2845 <pre class="doc_code">
2846 call void asm alignstack "eieio", ""()
2849 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2853 <p>TODO: The format of the asm and constraints string still need to be
2854 documented here. Constraints on what can be done (e.g. duplication, moving,
2855 etc need to be documented). This is probably best done by reference to
2856 another document that covers inline asm from a holistic perspective.</p>
2859 <!-- _______________________________________________________________________ -->
2861 <a name="inlineasm_md">Inline Asm Metadata</a>
2866 <p>The call instructions that wrap inline asm nodes may have a
2867 "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant
2868 integers. If present, the code generator will use the integer as the
2869 location cookie value when report errors through the <tt>LLVMContext</tt>
2870 error reporting mechanisms. This allows a front-end to correlate backend
2871 errors that occur with inline asm back to the source code that produced it.
2874 <pre class="doc_code">
2875 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2877 !42 = !{ i32 1234567 }
2880 <p>It is up to the front-end to make sense of the magic numbers it places in the
2881 IR. If the MDNode contains multiple constants, the code generator will use
2882 the one that corresponds to the line of the asm that the error occurs on.</p>
2888 <!-- ======================================================================= -->
2890 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2895 <p>LLVM IR allows metadata to be attached to instructions in the program that
2896 can convey extra information about the code to the optimizers and code
2897 generator. One example application of metadata is source-level debug
2898 information. There are two metadata primitives: strings and nodes. All
2899 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2900 preceding exclamation point ('<tt>!</tt>').</p>
2902 <p>A metadata string is a string surrounded by double quotes. It can contain
2903 any character by escaping non-printable characters with "<tt>\xx</tt>" where
2904 "<tt>xx</tt>" is the two digit hex code. For example:
2905 "<tt>!"test\00"</tt>".</p>
2907 <p>Metadata nodes are represented with notation similar to structure constants
2908 (a comma separated list of elements, surrounded by braces and preceded by an
2909 exclamation point). Metadata nodes can have any values as their operand. For
2912 <div class="doc_code">
2914 !{ metadata !"test\00", i32 10}
2918 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2919 metadata nodes, which can be looked up in the module symbol table. For
2922 <div class="doc_code">
2924 !foo = metadata !{!4, !3}
2928 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2929 function is using two metadata arguments:</p>
2931 <div class="doc_code">
2933 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2937 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2938 attached to the <tt>add</tt> instruction using the <tt>!dbg</tt>
2941 <div class="doc_code">
2943 %indvar.next = add i64 %indvar, 1, !dbg !21
2947 <p>More information about specific metadata nodes recognized by the optimizers
2948 and code generator is found below.</p>
2950 <!-- _______________________________________________________________________ -->
2952 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
2957 <p>In LLVM IR, memory does not have types, so LLVM's own type system is not
2958 suitable for doing TBAA. Instead, metadata is added to the IR to describe
2959 a type system of a higher level language. This can be used to implement
2960 typical C/C++ TBAA, but it can also be used to implement custom alias
2961 analysis behavior for other languages.</p>
2963 <p>The current metadata format is very simple. TBAA metadata nodes have up to
2964 three fields, e.g.:</p>
2966 <div class="doc_code">
2968 !0 = metadata !{ metadata !"an example type tree" }
2969 !1 = metadata !{ metadata !"int", metadata !0 }
2970 !2 = metadata !{ metadata !"float", metadata !0 }
2971 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2975 <p>The first field is an identity field. It can be any value, usually
2976 a metadata string, which uniquely identifies the type. The most important
2977 name in the tree is the name of the root node. Two trees with
2978 different root node names are entirely disjoint, even if they
2979 have leaves with common names.</p>
2981 <p>The second field identifies the type's parent node in the tree, or
2982 is null or omitted for a root node. A type is considered to alias
2983 all of its descendants and all of its ancestors in the tree. Also,
2984 a type is considered to alias all types in other trees, so that
2985 bitcode produced from multiple front-ends is handled conservatively.</p>
2987 <p>If the third field is present, it's an integer which if equal to 1
2988 indicates that the type is "constant" (meaning
2989 <tt>pointsToConstantMemory</tt> should return true; see
2990 <a href="AliasAnalysis.html#OtherItfs">other useful
2991 <tt>AliasAnalysis</tt> methods</a>).</p>
2995 <!-- _______________________________________________________________________ -->
2997 <a name="fpaccuracy">'<tt>fpaccuracy</tt>' Metadata</a>
3002 <p><tt>fpaccuracy</tt> metadata may be attached to any instruction of floating
3003 point type. It expresses the maximum relative error of the result of
3004 that instruction, in ULPs. ULP is defined as follows:</p>
3008 <p>If <tt>x</tt> is a real number that lies between two finite consecutive
3009 floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one
3010 of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the
3011 distance between the two non-equal finite floating-point numbers nearest
3012 <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p>
3016 <p>The maximum relative error may be any rational number. The metadata node
3017 shall consist of a pair of unsigned integers respectively representing
3018 the numerator and denominator. For example, 2.5 ULP:</p>
3020 <div class="doc_code">
3022 !0 = metadata !{ i32 5, i32 2 }
3032 <!-- *********************************************************************** -->
3034 <a name="intrinsic_globals">Intrinsic Global Variables</a>
3036 <!-- *********************************************************************** -->
3038 <p>LLVM has a number of "magic" global variables that contain data that affect
3039 code generation or other IR semantics. These are documented here. All globals
3040 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
3041 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
3044 <!-- ======================================================================= -->
3046 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
3051 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
3052 href="#linkage_appending">appending linkage</a>. This array contains a list of
3053 pointers to global variables and functions which may optionally have a pointer
3054 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
3056 <div class="doc_code">
3061 @llvm.used = appending global [2 x i8*] [
3063 i8* bitcast (i32* @Y to i8*)
3064 ], section "llvm.metadata"
3068 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
3069 compiler, assembler, and linker are required to treat the symbol as if there
3070 is a reference to the global that it cannot see. For example, if a variable
3071 has internal linkage and no references other than that from
3072 the <tt>@llvm.used</tt> list, it cannot be deleted. This is commonly used to
3073 represent references from inline asms and other things the compiler cannot
3074 "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p>
3076 <p>On some targets, the code generator must emit a directive to the assembler or
3077 object file to prevent the assembler and linker from molesting the
3082 <!-- ======================================================================= -->
3084 <a name="intg_compiler_used">
3085 The '<tt>llvm.compiler.used</tt>' Global Variable
3091 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
3092 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
3093 touching the symbol. On targets that support it, this allows an intelligent
3094 linker to optimize references to the symbol without being impeded as it would
3095 be by <tt>@llvm.used</tt>.</p>
3097 <p>This is a rare construct that should only be used in rare circumstances, and
3098 should not be exposed to source languages.</p>
3102 <!-- ======================================================================= -->
3104 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
3109 <div class="doc_code">
3111 %0 = type { i32, void ()* }
3112 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3116 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor
3117 functions and associated priorities. The functions referenced by this array
3118 will be called in ascending order of priority (i.e. lowest first) when the
3119 module is loaded. The order of functions with the same priority is not
3124 <!-- ======================================================================= -->
3126 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3131 <div class="doc_code">
3133 %0 = type { i32, void ()* }
3134 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3138 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions
3139 and associated priorities. The functions referenced by this array will be
3140 called in descending order of priority (i.e. highest first) when the module
3141 is loaded. The order of functions with the same priority is not defined.</p>
3147 <!-- *********************************************************************** -->
3148 <h2><a name="instref">Instruction Reference</a></h2>
3149 <!-- *********************************************************************** -->
3153 <p>The LLVM instruction set consists of several different classifications of
3154 instructions: <a href="#terminators">terminator
3155 instructions</a>, <a href="#binaryops">binary instructions</a>,
3156 <a href="#bitwiseops">bitwise binary instructions</a>,
3157 <a href="#memoryops">memory instructions</a>, and
3158 <a href="#otherops">other instructions</a>.</p>
3160 <!-- ======================================================================= -->
3162 <a name="terminators">Terminator Instructions</a>
3167 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3168 in a program ends with a "Terminator" instruction, which indicates which
3169 block should be executed after the current block is finished. These
3170 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3171 control flow, not values (the one exception being the
3172 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3174 <p>The terminator instructions are:
3175 '<a href="#i_ret"><tt>ret</tt></a>',
3176 '<a href="#i_br"><tt>br</tt></a>',
3177 '<a href="#i_switch"><tt>switch</tt></a>',
3178 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3179 '<a href="#i_invoke"><tt>invoke</tt></a>',
3180 '<a href="#i_resume"><tt>resume</tt></a>', and
3181 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3183 <!-- _______________________________________________________________________ -->
3185 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3192 ret <type> <value> <i>; Return a value from a non-void function</i>
3193 ret void <i>; Return from void function</i>
3197 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3198 a value) from a function back to the caller.</p>
3200 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3201 value and then causes control flow, and one that just causes control flow to
3205 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3206 return value. The type of the return value must be a
3207 '<a href="#t_firstclass">first class</a>' type.</p>
3209 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3210 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3211 value or a return value with a type that does not match its type, or if it
3212 has a void return type and contains a '<tt>ret</tt>' instruction with a
3216 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3217 the calling function's context. If the caller is a
3218 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3219 instruction after the call. If the caller was an
3220 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3221 the beginning of the "normal" destination block. If the instruction returns
3222 a value, that value shall set the call or invoke instruction's return
3227 ret i32 5 <i>; Return an integer value of 5</i>
3228 ret void <i>; Return from a void function</i>
3229 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3233 <!-- _______________________________________________________________________ -->
3235 <a name="i_br">'<tt>br</tt>' Instruction</a>
3242 br i1 <cond>, label <iftrue>, label <iffalse>
3243 br label <dest> <i>; Unconditional branch</i>
3247 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3248 different basic block in the current function. There are two forms of this
3249 instruction, corresponding to a conditional branch and an unconditional
3253 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3254 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3255 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3259 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3260 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3261 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3262 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3267 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3268 br i1 %cond, label %IfEqual, label %IfUnequal
3270 <a href="#i_ret">ret</a> i32 1
3272 <a href="#i_ret">ret</a> i32 0
3277 <!-- _______________________________________________________________________ -->
3279 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3286 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3290 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3291 several different places. It is a generalization of the '<tt>br</tt>'
3292 instruction, allowing a branch to occur to one of many possible
3296 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3297 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3298 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3299 The table is not allowed to contain duplicate constant entries.</p>
3302 <p>The <tt>switch</tt> instruction specifies a table of values and
3303 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3304 is searched for the given value. If the value is found, control flow is
3305 transferred to the corresponding destination; otherwise, control flow is
3306 transferred to the default destination.</p>
3308 <h5>Implementation:</h5>
3309 <p>Depending on properties of the target machine and the particular
3310 <tt>switch</tt> instruction, this instruction may be code generated in
3311 different ways. For example, it could be generated as a series of chained
3312 conditional branches or with a lookup table.</p>
3316 <i>; Emulate a conditional br instruction</i>
3317 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3318 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3320 <i>; Emulate an unconditional br instruction</i>
3321 switch i32 0, label %dest [ ]
3323 <i>; Implement a jump table:</i>
3324 switch i32 %val, label %otherwise [ i32 0, label %onzero
3326 i32 2, label %ontwo ]
3332 <!-- _______________________________________________________________________ -->
3334 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3341 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3346 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3347 within the current function, whose address is specified by
3348 "<tt>address</tt>". Address must be derived from a <a
3349 href="#blockaddress">blockaddress</a> constant.</p>
3353 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3354 rest of the arguments indicate the full set of possible destinations that the
3355 address may point to. Blocks are allowed to occur multiple times in the
3356 destination list, though this isn't particularly useful.</p>
3358 <p>This destination list is required so that dataflow analysis has an accurate
3359 understanding of the CFG.</p>
3363 <p>Control transfers to the block specified in the address argument. All
3364 possible destination blocks must be listed in the label list, otherwise this
3365 instruction has undefined behavior. This implies that jumps to labels
3366 defined in other functions have undefined behavior as well.</p>
3368 <h5>Implementation:</h5>
3370 <p>This is typically implemented with a jump through a register.</p>
3374 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3380 <!-- _______________________________________________________________________ -->
3382 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3389 <result> = invoke [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] <ptr to function ty> <function ptr val>(<function args>) [<a href="#fnattrs">fn attrs</a>]
3390 to label <normal label> unwind label <exception label>
3394 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3395 function, with the possibility of control flow transfer to either the
3396 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3397 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3398 control flow will return to the "normal" label. If the callee (or any
3399 indirect callees) returns via the "<a href="#i_resume"><tt>resume</tt></a>"
3400 instruction or other exception handling mechanism, control is interrupted and
3401 continued at the dynamically nearest "exception" label.</p>
3403 <p>The '<tt>exception</tt>' label is a
3404 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3405 exception. As such, '<tt>exception</tt>' label is required to have the
3406 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3407 the information about the behavior of the program after unwinding
3408 happens, as its first non-PHI instruction. The restrictions on the
3409 "<tt>landingpad</tt>" instruction's tightly couples it to the
3410 "<tt>invoke</tt>" instruction, so that the important information contained
3411 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3415 <p>This instruction requires several arguments:</p>
3418 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3419 convention</a> the call should use. If none is specified, the call
3420 defaults to using C calling conventions.</li>
3422 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3423 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3424 '<tt>inreg</tt>' attributes are valid here.</li>
3426 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3427 function value being invoked. In most cases, this is a direct function
3428 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3429 off an arbitrary pointer to function value.</li>
3431 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3432 function to be invoked. </li>
3434 <li>'<tt>function args</tt>': argument list whose types match the function
3435 signature argument types and parameter attributes. All arguments must be
3436 of <a href="#t_firstclass">first class</a> type. If the function
3437 signature indicates the function accepts a variable number of arguments,
3438 the extra arguments can be specified.</li>
3440 <li>'<tt>normal label</tt>': the label reached when the called function
3441 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3443 <li>'<tt>exception label</tt>': the label reached when a callee returns via
3444 the <a href="#i_resume"><tt>resume</tt></a> instruction or other exception
3445 handling mechanism.</li>
3447 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3448 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3449 '<tt>readnone</tt>' attributes are valid here.</li>
3453 <p>This instruction is designed to operate as a standard
3454 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3455 primary difference is that it establishes an association with a label, which
3456 is used by the runtime library to unwind the stack.</p>
3458 <p>This instruction is used in languages with destructors to ensure that proper
3459 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3460 exception. Additionally, this is important for implementation of
3461 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3463 <p>For the purposes of the SSA form, the definition of the value returned by the
3464 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3465 block to the "normal" label. If the callee unwinds then no return value is
3470 %retval = invoke i32 @Test(i32 15) to label %Continue
3471 unwind label %TestCleanup <i>; {i32}:retval set</i>
3472 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3473 unwind label %TestCleanup <i>; {i32}:retval set</i>
3478 <!-- _______________________________________________________________________ -->
3481 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3488 resume <type> <value>
3492 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3496 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3497 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3501 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3502 (in-flight) exception whose unwinding was interrupted with
3503 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3507 resume { i8*, i32 } %exn
3512 <!-- _______________________________________________________________________ -->
3515 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3526 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3527 instruction is used to inform the optimizer that a particular portion of the
3528 code is not reachable. This can be used to indicate that the code after a
3529 no-return function cannot be reached, and other facts.</p>
3532 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3538 <!-- ======================================================================= -->
3540 <a name="binaryops">Binary Operations</a>
3545 <p>Binary operators are used to do most of the computation in a program. They
3546 require two operands of the same type, execute an operation on them, and
3547 produce a single value. The operands might represent multiple data, as is
3548 the case with the <a href="#t_vector">vector</a> data type. The result value
3549 has the same type as its operands.</p>
3551 <p>There are several different binary operators:</p>
3553 <!-- _______________________________________________________________________ -->
3555 <a name="i_add">'<tt>add</tt>' Instruction</a>
3562 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3563 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3564 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3565 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3569 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3572 <p>The two arguments to the '<tt>add</tt>' instruction must
3573 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3574 integer values. Both arguments must have identical types.</p>
3577 <p>The value produced is the integer sum of the two operands.</p>
3579 <p>If the sum has unsigned overflow, the result returned is the mathematical
3580 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3582 <p>Because LLVM integers use a two's complement representation, this instruction
3583 is appropriate for both signed and unsigned integers.</p>
3585 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3586 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3587 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3588 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3589 respectively, occurs.</p>
3593 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3598 <!-- _______________________________________________________________________ -->
3600 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3607 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3611 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3614 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3615 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3616 floating point values. Both arguments must have identical types.</p>
3619 <p>The value produced is the floating point sum of the two operands.</p>
3623 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3628 <!-- _______________________________________________________________________ -->
3630 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3637 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3638 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3639 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3640 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3644 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3647 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3648 '<tt>neg</tt>' instruction present in most other intermediate
3649 representations.</p>
3652 <p>The two arguments to the '<tt>sub</tt>' instruction must
3653 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3654 integer values. Both arguments must have identical types.</p>
3657 <p>The value produced is the integer difference of the two operands.</p>
3659 <p>If the difference has unsigned overflow, the result returned is the
3660 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3663 <p>Because LLVM integers use a two's complement representation, this instruction
3664 is appropriate for both signed and unsigned integers.</p>
3666 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3667 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3668 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3669 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3670 respectively, occurs.</p>
3674 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3675 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3680 <!-- _______________________________________________________________________ -->
3682 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3689 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3693 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3696 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3697 '<tt>fneg</tt>' instruction present in most other intermediate
3698 representations.</p>
3701 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3702 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3703 floating point values. Both arguments must have identical types.</p>
3706 <p>The value produced is the floating point difference of the two operands.</p>
3710 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3711 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3716 <!-- _______________________________________________________________________ -->
3718 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3725 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3726 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3727 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3728 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3732 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3735 <p>The two arguments to the '<tt>mul</tt>' instruction must
3736 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3737 integer values. Both arguments must have identical types.</p>
3740 <p>The value produced is the integer product of the two operands.</p>
3742 <p>If the result of the multiplication has unsigned overflow, the result
3743 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3744 width of the result.</p>
3746 <p>Because LLVM integers use a two's complement representation, and the result
3747 is the same width as the operands, this instruction returns the correct
3748 result for both signed and unsigned integers. If a full product
3749 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3750 be sign-extended or zero-extended as appropriate to the width of the full
3753 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3754 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3755 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3756 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3757 respectively, occurs.</p>
3761 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3766 <!-- _______________________________________________________________________ -->
3768 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3775 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3779 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3782 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3783 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3784 floating point values. Both arguments must have identical types.</p>
3787 <p>The value produced is the floating point product of the two operands.</p>
3791 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3796 <!-- _______________________________________________________________________ -->
3798 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
3805 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3806 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3810 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3813 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3814 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3815 values. Both arguments must have identical types.</p>
3818 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3820 <p>Note that unsigned integer division and signed integer division are distinct
3821 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3823 <p>Division by zero leads to undefined behavior.</p>
3825 <p>If the <tt>exact</tt> keyword is present, the result value of the
3826 <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a
3827 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
3832 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3837 <!-- _______________________________________________________________________ -->
3839 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
3846 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3847 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3851 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3854 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3855 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3856 values. Both arguments must have identical types.</p>
3859 <p>The value produced is the signed integer quotient of the two operands rounded
3862 <p>Note that signed integer division and unsigned integer division are distinct
3863 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3865 <p>Division by zero leads to undefined behavior. Overflow also leads to
3866 undefined behavior; this is a rare case, but can occur, for example, by doing
3867 a 32-bit division of -2147483648 by -1.</p>
3869 <p>If the <tt>exact</tt> keyword is present, the result value of the
3870 <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would
3875 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3880 <!-- _______________________________________________________________________ -->
3882 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
3889 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3893 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3896 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3897 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3898 floating point values. Both arguments must have identical types.</p>
3901 <p>The value produced is the floating point quotient of the two operands.</p>
3905 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3910 <!-- _______________________________________________________________________ -->
3912 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3919 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3923 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3924 division of its two arguments.</p>
3927 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3928 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3929 values. Both arguments must have identical types.</p>
3932 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3933 This instruction always performs an unsigned division to get the
3936 <p>Note that unsigned integer remainder and signed integer remainder are
3937 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3939 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3943 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3948 <!-- _______________________________________________________________________ -->
3950 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3957 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3961 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3962 division of its two operands. This instruction can also take
3963 <a href="#t_vector">vector</a> versions of the values in which case the
3964 elements must be integers.</p>
3967 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3968 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3969 values. Both arguments must have identical types.</p>
3972 <p>This instruction returns the <i>remainder</i> of a division (where the result
3973 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
3974 <i>modulo</i> operator (where the result is either zero or has the same sign
3975 as the divisor, <tt>op2</tt>) of a value.
3976 For more information about the difference,
3977 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3978 Math Forum</a>. For a table of how this is implemented in various languages,
3979 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3980 Wikipedia: modulo operation</a>.</p>
3982 <p>Note that signed integer remainder and unsigned integer remainder are
3983 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3985 <p>Taking the remainder of a division by zero leads to undefined behavior.
3986 Overflow also leads to undefined behavior; this is a rare case, but can
3987 occur, for example, by taking the remainder of a 32-bit division of
3988 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3989 lets srem be implemented using instructions that return both the result of
3990 the division and the remainder.)</p>
3994 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3999 <!-- _______________________________________________________________________ -->
4001 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
4008 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4012 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
4013 its two operands.</p>
4016 <p>The two arguments to the '<tt>frem</tt>' instruction must be
4017 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4018 floating point values. Both arguments must have identical types.</p>
4021 <p>This instruction returns the <i>remainder</i> of a division. The remainder
4022 has the same sign as the dividend.</p>
4026 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
4033 <!-- ======================================================================= -->
4035 <a name="bitwiseops">Bitwise Binary Operations</a>
4040 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
4041 program. They are generally very efficient instructions and can commonly be
4042 strength reduced from other instructions. They require two operands of the
4043 same type, execute an operation on them, and produce a single value. The
4044 resulting value is the same type as its operands.</p>
4046 <!-- _______________________________________________________________________ -->
4048 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
4055 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4056 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4057 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4058 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4062 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
4063 a specified number of bits.</p>
4066 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
4067 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4068 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4071 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
4072 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
4073 is (statically or dynamically) negative or equal to or larger than the number
4074 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4075 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4076 shift amount in <tt>op2</tt>.</p>
4078 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
4079 <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits. If
4080 the <tt>nsw</tt> keyword is present, then the shift produces a
4081 <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree
4082 with the resultant sign bit. As such, NUW/NSW have the same semantics as
4083 they would if the shift were expressed as a mul instruction with the same
4084 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
4088 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
4089 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
4090 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
4091 <result> = shl i32 1, 32 <i>; undefined</i>
4092 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4097 <!-- _______________________________________________________________________ -->
4099 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4106 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4107 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4111 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4112 operand shifted to the right a specified number of bits with zero fill.</p>
4115 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4116 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4117 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4120 <p>This instruction always performs a logical shift right operation. The most
4121 significant bits of the result will be filled with zero bits after the shift.
4122 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4123 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4124 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4125 shift amount in <tt>op2</tt>.</p>
4127 <p>If the <tt>exact</tt> keyword is present, the result value of the
4128 <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4129 shifted out are non-zero.</p>
4134 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4135 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4136 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4137 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4138 <result> = lshr i32 1, 32 <i>; undefined</i>
4139 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4144 <!-- _______________________________________________________________________ -->
4146 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4153 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4154 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4158 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4159 operand shifted to the right a specified number of bits with sign
4163 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4164 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4165 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4168 <p>This instruction always performs an arithmetic shift right operation, The
4169 most significant bits of the result will be filled with the sign bit
4170 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4171 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4172 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4173 the corresponding shift amount in <tt>op2</tt>.</p>
4175 <p>If the <tt>exact</tt> keyword is present, the result value of the
4176 <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4177 shifted out are non-zero.</p>
4181 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4182 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4183 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4184 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4185 <result> = ashr i32 1, 32 <i>; undefined</i>
4186 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4191 <!-- _______________________________________________________________________ -->
4193 <a name="i_and">'<tt>and</tt>' Instruction</a>
4200 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4204 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4208 <p>The two arguments to the '<tt>and</tt>' instruction must be
4209 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4210 values. Both arguments must have identical types.</p>
4213 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4215 <table border="1" cellspacing="0" cellpadding="4">
4247 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4248 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4249 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4252 <!-- _______________________________________________________________________ -->
4254 <a name="i_or">'<tt>or</tt>' Instruction</a>
4261 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4265 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4269 <p>The two arguments to the '<tt>or</tt>' instruction must be
4270 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4271 values. Both arguments must have identical types.</p>
4274 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4276 <table border="1" cellspacing="0" cellpadding="4">
4308 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4309 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4310 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4315 <!-- _______________________________________________________________________ -->
4317 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4324 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4328 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4329 its two operands. The <tt>xor</tt> is used to implement the "one's
4330 complement" operation, which is the "~" operator in C.</p>
4333 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4334 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4335 values. Both arguments must have identical types.</p>
4338 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4340 <table border="1" cellspacing="0" cellpadding="4">
4372 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4373 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4374 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4375 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4382 <!-- ======================================================================= -->
4384 <a name="vectorops">Vector Operations</a>
4389 <p>LLVM supports several instructions to represent vector operations in a
4390 target-independent manner. These instructions cover the element-access and
4391 vector-specific operations needed to process vectors effectively. While LLVM
4392 does directly support these vector operations, many sophisticated algorithms
4393 will want to use target-specific intrinsics to take full advantage of a
4394 specific target.</p>
4396 <!-- _______________________________________________________________________ -->
4398 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4405 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4409 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4410 from a vector at a specified index.</p>
4414 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4415 of <a href="#t_vector">vector</a> type. The second operand is an index
4416 indicating the position from which to extract the element. The index may be
4420 <p>The result is a scalar of the same type as the element type of
4421 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4422 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4423 results are undefined.</p>
4427 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4432 <!-- _______________________________________________________________________ -->
4434 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4441 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4445 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4446 vector at a specified index.</p>
4449 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4450 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4451 whose type must equal the element type of the first operand. The third
4452 operand is an index indicating the position at which to insert the value.
4453 The index may be a variable.</p>
4456 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4457 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4458 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4459 results are undefined.</p>
4463 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4468 <!-- _______________________________________________________________________ -->
4470 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4477 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4481 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4482 from two input vectors, returning a vector with the same element type as the
4483 input and length that is the same as the shuffle mask.</p>
4486 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4487 with types that match each other. The third argument is a shuffle mask whose
4488 element type is always 'i32'. The result of the instruction is a vector
4489 whose length is the same as the shuffle mask and whose element type is the
4490 same as the element type of the first two operands.</p>
4492 <p>The shuffle mask operand is required to be a constant vector with either
4493 constant integer or undef values.</p>
4496 <p>The elements of the two input vectors are numbered from left to right across
4497 both of the vectors. The shuffle mask operand specifies, for each element of
4498 the result vector, which element of the two input vectors the result element
4499 gets. The element selector may be undef (meaning "don't care") and the
4500 second operand may be undef if performing a shuffle from only one vector.</p>
4504 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4505 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4506 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4507 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4508 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4509 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4510 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4511 <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>
4518 <!-- ======================================================================= -->
4520 <a name="aggregateops">Aggregate Operations</a>
4525 <p>LLVM supports several instructions for working with
4526 <a href="#t_aggregate">aggregate</a> values.</p>
4528 <!-- _______________________________________________________________________ -->
4530 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4537 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4541 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4542 from an <a href="#t_aggregate">aggregate</a> value.</p>
4545 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4546 of <a href="#t_struct">struct</a> or
4547 <a href="#t_array">array</a> type. The operands are constant indices to
4548 specify which value to extract in a similar manner as indices in a
4549 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4550 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4552 <li>Since the value being indexed is not a pointer, the first index is
4553 omitted and assumed to be zero.</li>
4554 <li>At least one index must be specified.</li>
4555 <li>Not only struct indices but also array indices must be in
4560 <p>The result is the value at the position in the aggregate specified by the
4565 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4570 <!-- _______________________________________________________________________ -->
4572 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4579 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4583 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4584 in an <a href="#t_aggregate">aggregate</a> value.</p>
4587 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4588 of <a href="#t_struct">struct</a> or
4589 <a href="#t_array">array</a> type. The second operand is a first-class
4590 value to insert. The following operands are constant indices indicating
4591 the position at which to insert the value in a similar manner as indices in a
4592 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4593 value to insert must have the same type as the value identified by the
4597 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4598 that of <tt>val</tt> except that the value at the position specified by the
4599 indices is that of <tt>elt</tt>.</p>
4603 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4604 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4605 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4612 <!-- ======================================================================= -->
4614 <a name="memoryops">Memory Access and Addressing Operations</a>
4619 <p>A key design point of an SSA-based representation is how it represents
4620 memory. In LLVM, no memory locations are in SSA form, which makes things
4621 very simple. This section describes how to read, write, and allocate
4624 <!-- _______________________________________________________________________ -->
4626 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4633 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4637 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4638 currently executing function, to be automatically released when this function
4639 returns to its caller. The object is always allocated in the generic address
4640 space (address space zero).</p>
4643 <p>The '<tt>alloca</tt>' instruction
4644 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4645 runtime stack, returning a pointer of the appropriate type to the program.
4646 If "NumElements" is specified, it is the number of elements allocated,
4647 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4648 specified, the value result of the allocation is guaranteed to be aligned to
4649 at least that boundary. If not specified, or if zero, the target can choose
4650 to align the allocation on any convenient boundary compatible with the
4653 <p>'<tt>type</tt>' may be any sized type.</p>
4656 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4657 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4658 memory is automatically released when the function returns. The
4659 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4660 variables that must have an address available. When the function returns
4661 (either with the <tt><a href="#i_ret">ret</a></tt>
4662 or <tt><a href="#i_resume">resume</a></tt> instructions), the memory is
4663 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4667 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4668 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4669 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4670 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4675 <!-- _______________________________________________________________________ -->
4677 <a name="i_load">'<tt>load</tt>' Instruction</a>
4684 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4685 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4686 !<index> = !{ i32 1 }
4690 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4693 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4694 from which to load. The pointer must point to
4695 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4696 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4697 number or order of execution of this <tt>load</tt> with other <a
4698 href="#volatile">volatile operations</a>.</p>
4700 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
4701 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4702 argument. The <code>release</code> and <code>acq_rel</code> orderings are
4703 not valid on <code>load</code> instructions. Atomic loads produce <a
4704 href="#memorymodel">defined</a> results when they may see multiple atomic
4705 stores. The type of the pointee must be an integer type whose bit width
4706 is a power of two greater than or equal to eight and less than or equal
4707 to a target-specific size limit. <code>align</code> must be explicitly
4708 specified on atomic loads, and the load has undefined behavior if the
4709 alignment is not set to a value which is at least the size in bytes of
4710 the pointee. <code>!nontemporal</code> does not have any defined semantics
4711 for atomic loads.</p>
4713 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4714 operation (that is, the alignment of the memory address). A value of 0 or an
4715 omitted <tt>align</tt> argument means that the operation has the preferential
4716 alignment for the target. It is the responsibility of the code emitter to
4717 ensure that the alignment information is correct. Overestimating the
4718 alignment results in undefined behavior. Underestimating the alignment may
4719 produce less efficient code. An alignment of 1 is always safe.</p>
4721 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4722 metatadata name <index> corresponding to a metadata node with
4723 one <tt>i32</tt> entry of value 1. The existence of
4724 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4725 and code generator that this load is not expected to be reused in the cache.
4726 The code generator may select special instructions to save cache bandwidth,
4727 such as the <tt>MOVNT</tt> instruction on x86.</p>
4729 <p>The optional <tt>!invariant.load</tt> metadata must reference a single
4730 metatadata name <index> corresponding to a metadata node with no
4731 entries. The existence of the <tt>!invariant.load</tt> metatadata on the
4732 instruction tells the optimizer and code generator that this load address
4733 points to memory which does not change value during program execution.
4734 The optimizer may then move this load around, for example, by hoisting it
4735 out of loops using loop invariant code motion.</p>
4738 <p>The location of memory pointed to is loaded. If the value being loaded is of
4739 scalar type then the number of bytes read does not exceed the minimum number
4740 of bytes needed to hold all bits of the type. For example, loading an
4741 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4742 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4743 is undefined if the value was not originally written using a store of the
4748 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4749 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4750 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4755 <!-- _______________________________________________________________________ -->
4757 <a name="i_store">'<tt>store</tt>' Instruction</a>
4764 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4765 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
4769 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4772 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4773 and an address at which to store it. The type of the
4774 '<tt><pointer></tt>' operand must be a pointer to
4775 the <a href="#t_firstclass">first class</a> type of the
4776 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
4777 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
4778 order of execution of this <tt>store</tt> with other <a
4779 href="#volatile">volatile operations</a>.</p>
4781 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
4782 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4783 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
4784 valid on <code>store</code> instructions. Atomic loads produce <a
4785 href="#memorymodel">defined</a> results when they may see multiple atomic
4786 stores. The type of the pointee must be an integer type whose bit width
4787 is a power of two greater than or equal to eight and less than or equal
4788 to a target-specific size limit. <code>align</code> must be explicitly
4789 specified on atomic stores, and the store has undefined behavior if the
4790 alignment is not set to a value which is at least the size in bytes of
4791 the pointee. <code>!nontemporal</code> does not have any defined semantics
4792 for atomic stores.</p>
4794 <p>The optional constant "align" argument specifies the alignment of the
4795 operation (that is, the alignment of the memory address). A value of 0 or an
4796 omitted "align" argument means that the operation has the preferential
4797 alignment for the target. It is the responsibility of the code emitter to
4798 ensure that the alignment information is correct. Overestimating the
4799 alignment results in an undefined behavior. Underestimating the alignment may
4800 produce less efficient code. An alignment of 1 is always safe.</p>
4802 <p>The optional !nontemporal metadata must reference a single metatadata
4803 name <index> corresponding to a metadata node with one i32 entry of
4804 value 1. The existence of the !nontemporal metatadata on the
4805 instruction tells the optimizer and code generator that this load is
4806 not expected to be reused in the cache. The code generator may
4807 select special instructions to save cache bandwidth, such as the
4808 MOVNT instruction on x86.</p>
4812 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4813 location specified by the '<tt><pointer></tt>' operand. If
4814 '<tt><value></tt>' is of scalar type then the number of bytes written
4815 does not exceed the minimum number of bytes needed to hold all bits of the
4816 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4817 writing a value of a type like <tt>i20</tt> with a size that is not an
4818 integral number of bytes, it is unspecified what happens to the extra bits
4819 that do not belong to the type, but they will typically be overwritten.</p>
4823 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4824 store i32 3, i32* %ptr <i>; yields {void}</i>
4825 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4830 <!-- _______________________________________________________________________ -->
4832 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
4839 fence [singlethread] <ordering> <i>; yields {void}</i>
4843 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
4844 between operations.</p>
4846 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
4847 href="#ordering">ordering</a> argument which defines what
4848 <i>synchronizes-with</i> edges they add. They can only be given
4849 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
4850 <code>seq_cst</code> orderings.</p>
4853 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
4854 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
4855 <code>acquire</code> ordering semantics if and only if there exist atomic
4856 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
4857 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
4858 <var>X</var> modifies <var>M</var> (either directly or through some side effect
4859 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
4860 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
4861 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
4862 than an explicit <code>fence</code>, one (but not both) of the atomic operations
4863 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
4864 <code>acquire</code> (resp.) ordering constraint and still
4865 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
4866 <i>happens-before</i> edge.</p>
4868 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
4869 having both <code>acquire</code> and <code>release</code> semantics specified
4870 above, participates in the global program order of other <code>seq_cst</code>
4871 operations and/or fences.</p>
4873 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
4874 specifies that the fence only synchronizes with other fences in the same
4875 thread. (This is useful for interacting with signal handlers.)</p>
4879 fence acquire <i>; yields {void}</i>
4880 fence singlethread seq_cst <i>; yields {void}</i>
4885 <!-- _______________________________________________________________________ -->
4887 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
4894 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
4898 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
4899 It loads a value in memory and compares it to a given value. If they are
4900 equal, it stores a new value into the memory.</p>
4903 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
4904 address to operate on, a value to compare to the value currently be at that
4905 address, and a new value to place at that address if the compared values are
4906 equal. The type of '<var><cmp></var>' must be an integer type whose
4907 bit width is a power of two greater than or equal to eight and less than
4908 or equal to a target-specific size limit. '<var><cmp></var>' and
4909 '<var><new></var>' must have the same type, and the type of
4910 '<var><pointer></var>' must be a pointer to that type. If the
4911 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
4912 optimizer is not allowed to modify the number or order of execution
4913 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
4916 <!-- FIXME: Extend allowed types. -->
4918 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
4919 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
4921 <p>The optional "<code>singlethread</code>" argument declares that the
4922 <code>cmpxchg</code> is only atomic with respect to code (usually signal
4923 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
4924 cmpxchg is atomic with respect to all other code in the system.</p>
4926 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
4927 the size in memory of the operand.
4930 <p>The contents of memory at the location specified by the
4931 '<tt><pointer></tt>' operand is read and compared to
4932 '<tt><cmp></tt>'; if the read value is the equal,
4933 '<tt><new></tt>' is written. The original value at the location
4936 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
4937 purpose of identifying <a href="#release_sequence">release sequences</a>. A
4938 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
4939 parameter determined by dropping any <code>release</code> part of the
4940 <code>cmpxchg</code>'s ordering.</p>
4943 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
4944 optimization work on ARM.)
4946 FIXME: Is a weaker ordering constraint on failure helpful in practice?
4952 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
4953 <a href="#i_br">br</a> label %loop
4956 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
4957 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
4958 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
4959 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
4960 <a href="#i_br">br</a> i1 %success, label %done, label %loop
4968 <!-- _______________________________________________________________________ -->
4970 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
4977 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
4981 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
4984 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
4985 operation to apply, an address whose value to modify, an argument to the
4986 operation. The operation must be one of the following keywords:</p>
5001 <p>The type of '<var><value></var>' must be an integer type whose
5002 bit width is a power of two greater than or equal to eight and less than
5003 or equal to a target-specific size limit. The type of the
5004 '<code><pointer></code>' operand must be a pointer to that type.
5005 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
5006 optimizer is not allowed to modify the number or order of execution of this
5007 <code>atomicrmw</code> with other <a href="#volatile">volatile
5010 <!-- FIXME: Extend allowed types. -->
5013 <p>The contents of memory at the location specified by the
5014 '<tt><pointer></tt>' operand are atomically read, modified, and written
5015 back. The original value at the location is returned. The modification is
5016 specified by the <var>operation</var> argument:</p>
5019 <li>xchg: <code>*ptr = val</code></li>
5020 <li>add: <code>*ptr = *ptr + val</code></li>
5021 <li>sub: <code>*ptr = *ptr - val</code></li>
5022 <li>and: <code>*ptr = *ptr & val</code></li>
5023 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
5024 <li>or: <code>*ptr = *ptr | val</code></li>
5025 <li>xor: <code>*ptr = *ptr ^ val</code></li>
5026 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
5027 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
5028 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
5029 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
5034 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
5039 <!-- _______________________________________________________________________ -->
5041 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
5048 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5049 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5050 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5054 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
5055 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
5056 It performs address calculation only and does not access memory.</p>
5059 <p>The first argument is always a pointer or a vector of pointers,
5060 and forms the basis of the
5061 calculation. The remaining arguments are indices that indicate which of the
5062 elements of the aggregate object are indexed. The interpretation of each
5063 index is dependent on the type being indexed into. The first index always
5064 indexes the pointer value given as the first argument, the second index
5065 indexes a value of the type pointed to (not necessarily the value directly
5066 pointed to, since the first index can be non-zero), etc. The first type
5067 indexed into must be a pointer value, subsequent types can be arrays,
5068 vectors, and structs. Note that subsequent types being indexed into
5069 can never be pointers, since that would require loading the pointer before
5070 continuing calculation.</p>
5072 <p>The type of each index argument depends on the type it is indexing into.
5073 When indexing into a (optionally packed) structure, only <tt>i32</tt>
5074 integer <b>constants</b> are allowed. When indexing into an array, pointer
5075 or vector, integers of any width are allowed, and they are not required to be
5076 constant. These integers are treated as signed values where relevant.</p>
5078 <p>For example, let's consider a C code fragment and how it gets compiled to
5081 <pre class="doc_code">
5093 int *foo(struct ST *s) {
5094 return &s[1].Z.B[5][13];
5098 <p>The LLVM code generated by Clang is:</p>
5100 <pre class="doc_code">
5101 %struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 }
5102 %struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT }
5104 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5106 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5112 <p>In the example above, the first index is indexing into the
5113 '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a
5114 '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a
5115 structure. The second index indexes into the third element of the structure,
5116 yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>'
5117 type, another structure. The third index indexes into the second element of
5118 the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The
5119 two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>'
5120 type. The '<tt>getelementptr</tt>' instruction returns a pointer to this
5121 element, thus computing a value of '<tt>i32*</tt>' type.</p>
5123 <p>Note that it is perfectly legal to index partially through a structure,
5124 returning a pointer to an inner element. Because of this, the LLVM code for
5125 the given testcase is equivalent to:</p>
5127 <pre class="doc_code">
5128 define i32* @foo(%struct.ST* %s) {
5129 %t1 = getelementptr %struct.ST* %s, i32 1 <i>; yields %struct.ST*:%t1</i>
5130 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 <i>; yields %struct.RT*:%t2</i>
5131 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5132 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5133 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5138 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5139 <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the
5140 base pointer is not an <i>in bounds</i> address of an allocated object,
5141 or if any of the addresses that would be formed by successive addition of
5142 the offsets implied by the indices to the base address with infinitely
5143 precise signed arithmetic are not an <i>in bounds</i> address of that
5144 allocated object. The <i>in bounds</i> addresses for an allocated object
5145 are all the addresses that point into the object, plus the address one
5147 In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword
5148 applies to each of the computations element-wise. </p>
5150 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5151 the base address with silently-wrapping two's complement arithmetic. If the
5152 offsets have a different width from the pointer, they are sign-extended or
5153 truncated to the width of the pointer. The result value of the
5154 <tt>getelementptr</tt> may be outside the object pointed to by the base
5155 pointer. The result value may not necessarily be used to access memory
5156 though, even if it happens to point into allocated storage. See the
5157 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5160 <p>The getelementptr instruction is often confusing. For some more insight into
5161 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5165 <i>; yields [12 x i8]*:aptr</i>
5166 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5167 <i>; yields i8*:vptr</i>
5168 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5169 <i>; yields i8*:eptr</i>
5170 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5171 <i>; yields i32*:iptr</i>
5172 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5175 <p>In cases where the pointer argument is a vector of pointers, only a
5176 single index may be used, and the number of vector elements has to be
5177 the same. For example: </p>
5178 <pre class="doc_code">
5179 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5186 <!-- ======================================================================= -->
5188 <a name="convertops">Conversion Operations</a>
5193 <p>The instructions in this category are the conversion instructions (casting)
5194 which all take a single operand and a type. They perform various bit
5195 conversions on the operand.</p>
5197 <!-- _______________________________________________________________________ -->
5199 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5206 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5210 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5211 type <tt>ty2</tt>.</p>
5214 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5215 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5216 of the same number of integers.
5217 The bit size of the <tt>value</tt> must be larger than
5218 the bit size of the destination type, <tt>ty2</tt>.
5219 Equal sized types are not allowed.</p>
5222 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5223 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5224 source size must be larger than the destination size, <tt>trunc</tt> cannot
5225 be a <i>no-op cast</i>. It will always truncate bits.</p>
5229 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5230 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5231 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5232 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5237 <!-- _______________________________________________________________________ -->
5239 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5246 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5250 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5255 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5256 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5257 of the same number of integers.
5258 The bit size of the <tt>value</tt> must be smaller than
5259 the bit size of the destination type,
5263 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5264 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5266 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5270 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5271 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5272 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5277 <!-- _______________________________________________________________________ -->
5279 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5286 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5290 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5293 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5294 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5295 of the same number of integers.
5296 The bit size of the <tt>value</tt> must be smaller than
5297 the bit size of the destination type,
5301 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5302 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5303 of the type <tt>ty2</tt>.</p>
5305 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5309 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5310 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5311 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5316 <!-- _______________________________________________________________________ -->
5318 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5325 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5329 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5333 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5334 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5335 to cast it to. The size of <tt>value</tt> must be larger than the size of
5336 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5337 <i>no-op cast</i>.</p>
5340 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5341 <a href="#t_floating">floating point</a> type to a smaller
5342 <a href="#t_floating">floating point</a> type. If the value cannot fit
5343 within the destination type, <tt>ty2</tt>, then the results are
5348 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5349 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5354 <!-- _______________________________________________________________________ -->
5356 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5363 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5367 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5368 floating point value.</p>
5371 <p>The '<tt>fpext</tt>' instruction takes a
5372 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5373 a <a href="#t_floating">floating point</a> type to cast it to. The source
5374 type must be smaller than the destination type.</p>
5377 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5378 <a href="#t_floating">floating point</a> type to a larger
5379 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5380 used to make a <i>no-op cast</i> because it always changes bits. Use
5381 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5385 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5386 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5391 <!-- _______________________________________________________________________ -->
5393 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5400 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5404 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5405 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5408 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5409 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5410 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5411 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5412 vector integer type with the same number of elements as <tt>ty</tt></p>
5415 <p>The '<tt>fptoui</tt>' instruction converts its
5416 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5417 towards zero) unsigned integer value. If the value cannot fit
5418 in <tt>ty2</tt>, the results are undefined.</p>
5422 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5423 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5424 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5429 <!-- _______________________________________________________________________ -->
5431 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5438 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5442 <p>The '<tt>fptosi</tt>' instruction converts
5443 <a href="#t_floating">floating point</a> <tt>value</tt> to
5444 type <tt>ty2</tt>.</p>
5447 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5448 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5449 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5450 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5451 vector integer type with the same number of elements as <tt>ty</tt></p>
5454 <p>The '<tt>fptosi</tt>' instruction converts its
5455 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5456 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5457 the results are undefined.</p>
5461 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5462 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5463 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5468 <!-- _______________________________________________________________________ -->
5470 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5477 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5481 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5482 integer and converts that value to the <tt>ty2</tt> type.</p>
5485 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5486 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5487 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5488 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5489 floating point type with the same number of elements as <tt>ty</tt></p>
5492 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5493 integer quantity and converts it to the corresponding floating point
5494 value. If the value cannot fit in the floating point value, the results are
5499 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5500 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5505 <!-- _______________________________________________________________________ -->
5507 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5514 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5518 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5519 and converts that value to the <tt>ty2</tt> type.</p>
5522 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5523 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5524 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5525 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5526 floating point type with the same number of elements as <tt>ty</tt></p>
5529 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5530 quantity and converts it to the corresponding floating point value. If the
5531 value cannot fit in the floating point value, the results are undefined.</p>
5535 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5536 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5541 <!-- _______________________________________________________________________ -->
5543 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5550 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5554 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of
5555 pointers <tt>value</tt> to
5556 the integer (or vector of integers) type <tt>ty2</tt>.</p>
5559 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5560 must be a a value of type <a href="#t_pointer">pointer</a> or a vector of
5561 pointers, and a type to cast it to
5562 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector
5563 of integers type.</p>
5566 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5567 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5568 truncating or zero extending that value to the size of the integer type. If
5569 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5570 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5571 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5576 %X = ptrtoint i32* %P to i8 <i>; yields truncation on 32-bit architecture</i>
5577 %Y = ptrtoint i32* %P to i64 <i>; yields zero extension on 32-bit architecture</i>
5578 %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>
5583 <!-- _______________________________________________________________________ -->
5585 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5592 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5596 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5597 pointer type, <tt>ty2</tt>.</p>
5600 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5601 value to cast, and a type to cast it to, which must be a
5602 <a href="#t_pointer">pointer</a> type.</p>
5605 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5606 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5607 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5608 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5609 than the size of a pointer then a zero extension is done. If they are the
5610 same size, nothing is done (<i>no-op cast</i>).</p>
5614 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5615 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5616 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5617 %Z = inttoptr <4 x i32> %G to <4 x i8*><i>; yields truncation of vector G to four pointers</i>
5622 <!-- _______________________________________________________________________ -->
5624 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5631 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5635 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5636 <tt>ty2</tt> without changing any bits.</p>
5639 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5640 non-aggregate first class value, and a type to cast it to, which must also be
5641 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5642 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5643 identical. If the source type is a pointer, the destination type must also be
5644 a pointer. This instruction supports bitwise conversion of vectors to
5645 integers and to vectors of other types (as long as they have the same
5649 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5650 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5651 this conversion. The conversion is done as if the <tt>value</tt> had been
5652 stored to memory and read back as type <tt>ty2</tt>.
5653 Pointer (or vector of pointers) types may only be converted to other pointer
5654 (or vector of pointers) types with this instruction. To convert
5655 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5656 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5660 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
5661 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
5662 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
5663 %Z = bitcast <2 x i32*> %V to <2 x i64*> <i>; yields <2 x i64*></i>
5670 <!-- ======================================================================= -->
5672 <a name="otherops">Other Operations</a>
5677 <p>The instructions in this category are the "miscellaneous" instructions, which
5678 defy better classification.</p>
5680 <!-- _______________________________________________________________________ -->
5682 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5689 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5693 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5694 boolean values based on comparison of its two integer, integer vector,
5695 pointer, or pointer vector operands.</p>
5698 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
5699 the condition code indicating the kind of comparison to perform. It is not a
5700 value, just a keyword. The possible condition code are:</p>
5703 <li><tt>eq</tt>: equal</li>
5704 <li><tt>ne</tt>: not equal </li>
5705 <li><tt>ugt</tt>: unsigned greater than</li>
5706 <li><tt>uge</tt>: unsigned greater or equal</li>
5707 <li><tt>ult</tt>: unsigned less than</li>
5708 <li><tt>ule</tt>: unsigned less or equal</li>
5709 <li><tt>sgt</tt>: signed greater than</li>
5710 <li><tt>sge</tt>: signed greater or equal</li>
5711 <li><tt>slt</tt>: signed less than</li>
5712 <li><tt>sle</tt>: signed less or equal</li>
5715 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
5716 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
5717 typed. They must also be identical types.</p>
5720 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
5721 condition code given as <tt>cond</tt>. The comparison performed always yields
5722 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
5723 result, as follows:</p>
5726 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
5727 <tt>false</tt> otherwise. No sign interpretation is necessary or
5730 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
5731 <tt>false</tt> otherwise. No sign interpretation is necessary or
5734 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
5735 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5737 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
5738 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5739 to <tt>op2</tt>.</li>
5741 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
5742 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5744 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
5745 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5747 <li><tt>sgt</tt>: interprets the operands as signed values and yields
5748 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5750 <li><tt>sge</tt>: interprets the operands as signed values and yields
5751 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5752 to <tt>op2</tt>.</li>
5754 <li><tt>slt</tt>: interprets the operands as signed values and yields
5755 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5757 <li><tt>sle</tt>: interprets the operands as signed values and yields
5758 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5761 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
5762 values are compared as if they were integers.</p>
5764 <p>If the operands are integer vectors, then they are compared element by
5765 element. The result is an <tt>i1</tt> vector with the same number of elements
5766 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
5770 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
5771 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
5772 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
5773 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
5774 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
5775 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
5778 <p>Note that the code generator does not yet support vector types with
5779 the <tt>icmp</tt> instruction.</p>
5783 <!-- _______________________________________________________________________ -->
5785 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
5792 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5796 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
5797 values based on comparison of its operands.</p>
5799 <p>If the operands are floating point scalars, then the result type is a boolean
5800 (<a href="#t_integer"><tt>i1</tt></a>).</p>
5802 <p>If the operands are floating point vectors, then the result type is a vector
5803 of boolean with the same number of elements as the operands being
5807 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
5808 the condition code indicating the kind of comparison to perform. It is not a
5809 value, just a keyword. The possible condition code are:</p>
5812 <li><tt>false</tt>: no comparison, always returns false</li>
5813 <li><tt>oeq</tt>: ordered and equal</li>
5814 <li><tt>ogt</tt>: ordered and greater than </li>
5815 <li><tt>oge</tt>: ordered and greater than or equal</li>
5816 <li><tt>olt</tt>: ordered and less than </li>
5817 <li><tt>ole</tt>: ordered and less than or equal</li>
5818 <li><tt>one</tt>: ordered and not equal</li>
5819 <li><tt>ord</tt>: ordered (no nans)</li>
5820 <li><tt>ueq</tt>: unordered or equal</li>
5821 <li><tt>ugt</tt>: unordered or greater than </li>
5822 <li><tt>uge</tt>: unordered or greater than or equal</li>
5823 <li><tt>ult</tt>: unordered or less than </li>
5824 <li><tt>ule</tt>: unordered or less than or equal</li>
5825 <li><tt>une</tt>: unordered or not equal</li>
5826 <li><tt>uno</tt>: unordered (either nans)</li>
5827 <li><tt>true</tt>: no comparison, always returns true</li>
5830 <p><i>Ordered</i> means that neither operand is a QNAN while
5831 <i>unordered</i> means that either operand may be a QNAN.</p>
5833 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
5834 a <a href="#t_floating">floating point</a> type or
5835 a <a href="#t_vector">vector</a> of floating point type. They must have
5836 identical types.</p>
5839 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
5840 according to the condition code given as <tt>cond</tt>. If the operands are
5841 vectors, then the vectors are compared element by element. Each comparison
5842 performed always yields an <a href="#t_integer">i1</a> result, as
5846 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
5848 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5849 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5851 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5852 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5854 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5855 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5857 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5858 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5860 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5861 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5863 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5864 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5866 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
5868 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
5869 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5871 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
5872 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5874 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
5875 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5877 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
5878 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5880 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
5881 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5883 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
5884 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5886 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
5888 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
5893 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
5894 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
5895 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
5896 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
5899 <p>Note that the code generator does not yet support vector types with
5900 the <tt>fcmp</tt> instruction.</p>
5904 <!-- _______________________________________________________________________ -->
5906 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
5913 <result> = phi <ty> [ <val0>, <label0>], ...
5917 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
5918 SSA graph representing the function.</p>
5921 <p>The type of the incoming values is specified with the first type field. After
5922 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
5923 one pair for each predecessor basic block of the current block. Only values
5924 of <a href="#t_firstclass">first class</a> type may be used as the value
5925 arguments to the PHI node. Only labels may be used as the label
5928 <p>There must be no non-phi instructions between the start of a basic block and
5929 the PHI instructions: i.e. PHI instructions must be first in a basic
5932 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5933 occur on the edge from the corresponding predecessor block to the current
5934 block (but after any definition of an '<tt>invoke</tt>' instruction's return
5935 value on the same edge).</p>
5938 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
5939 specified by the pair corresponding to the predecessor basic block that
5940 executed just prior to the current block.</p>
5944 Loop: ; Infinite loop that counts from 0 on up...
5945 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5946 %nextindvar = add i32 %indvar, 1
5952 <!-- _______________________________________________________________________ -->
5954 <a name="i_select">'<tt>select</tt>' Instruction</a>
5961 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5963 <i>selty</i> is either i1 or {<N x i1>}
5967 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
5968 condition, without branching.</p>
5972 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
5973 values indicating the condition, and two values of the
5974 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
5975 vectors and the condition is a scalar, then entire vectors are selected, not
5976 individual elements.</p>
5979 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
5980 first value argument; otherwise, it returns the second value argument.</p>
5982 <p>If the condition is a vector of i1, then the value arguments must be vectors
5983 of the same size, and the selection is done element by element.</p>
5987 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
5992 <!-- _______________________________________________________________________ -->
5994 <a name="i_call">'<tt>call</tt>' Instruction</a>
6001 <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>]
6005 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
6008 <p>This instruction requires several arguments:</p>
6011 <li>The optional "tail" marker indicates that the callee function does not
6012 access any allocas or varargs in the caller. Note that calls may be
6013 marked "tail" even if they do not occur before
6014 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
6015 present, the function call is eligible for tail call optimization,
6016 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
6017 optimized into a jump</a>. The code generator may optimize calls marked
6018 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
6019 sibling call optimization</a> when the caller and callee have
6020 matching signatures, or 2) forced tail call optimization when the
6021 following extra requirements are met:
6023 <li>Caller and callee both have the calling
6024 convention <tt>fastcc</tt>.</li>
6025 <li>The call is in tail position (ret immediately follows call and ret
6026 uses value of call or is void).</li>
6027 <li>Option <tt>-tailcallopt</tt> is enabled,
6028 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
6029 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
6030 constraints are met.</a></li>
6034 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
6035 convention</a> the call should use. If none is specified, the call
6036 defaults to using C calling conventions. The calling convention of the
6037 call must match the calling convention of the target function, or else the
6038 behavior is undefined.</li>
6040 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
6041 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
6042 '<tt>inreg</tt>' attributes are valid here.</li>
6044 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
6045 type of the return value. Functions that return no value are marked
6046 <tt><a href="#t_void">void</a></tt>.</li>
6048 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
6049 being invoked. The argument types must match the types implied by this
6050 signature. This type can be omitted if the function is not varargs and if
6051 the function type does not return a pointer to a function.</li>
6053 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
6054 be invoked. In most cases, this is a direct function invocation, but
6055 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
6056 to function value.</li>
6058 <li>'<tt>function args</tt>': argument list whose types match the function
6059 signature argument types and parameter attributes. All arguments must be
6060 of <a href="#t_firstclass">first class</a> type. If the function
6061 signature indicates the function accepts a variable number of arguments,
6062 the extra arguments can be specified.</li>
6064 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
6065 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
6066 '<tt>readnone</tt>' attributes are valid here.</li>
6070 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
6071 a specified function, with its incoming arguments bound to the specified
6072 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
6073 function, control flow continues with the instruction after the function
6074 call, and the return value of the function is bound to the result
6079 %retval = call i32 @test(i32 %argc)
6080 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
6081 %X = tail call i32 @foo() <i>; yields i32</i>
6082 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
6083 call void %foo(i8 97 signext)
6085 %struct.A = type { i32, i8 }
6086 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
6087 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
6088 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
6089 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
6090 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
6093 <p>llvm treats calls to some functions with names and arguments that match the
6094 standard C99 library as being the C99 library functions, and may perform
6095 optimizations or generate code for them under that assumption. This is
6096 something we'd like to change in the future to provide better support for
6097 freestanding environments and non-C-based languages.</p>
6101 <!-- _______________________________________________________________________ -->
6103 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
6110 <resultval> = va_arg <va_list*> <arglist>, <argty>
6114 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6115 the "variable argument" area of a function call. It is used to implement the
6116 <tt>va_arg</tt> macro in C.</p>
6119 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6120 argument. It returns a value of the specified argument type and increments
6121 the <tt>va_list</tt> to point to the next argument. The actual type
6122 of <tt>va_list</tt> is target specific.</p>
6125 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6126 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6127 to the next argument. For more information, see the variable argument
6128 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6130 <p>It is legal for this instruction to be called in a function which does not
6131 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6134 <p><tt>va_arg</tt> is an LLVM instruction instead of
6135 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6139 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6141 <p>Note that the code generator does not yet fully support va_arg on many
6142 targets. Also, it does not currently support va_arg with aggregate types on
6147 <!-- _______________________________________________________________________ -->
6149 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6156 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6157 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6159 <clause> := catch <type> <value>
6160 <clause> := filter <array constant type> <array constant>
6164 <p>The '<tt>landingpad</tt>' instruction is used by
6165 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6166 system</a> to specify that a basic block is a landing pad — one where
6167 the exception lands, and corresponds to the code found in the
6168 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6169 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6170 re-entry to the function. The <tt>resultval</tt> has the
6171 type <tt>resultty</tt>.</p>
6174 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6175 function associated with the unwinding mechanism. The optional
6176 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6178 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6179 or <tt>filter</tt> — and contains the global variable representing the
6180 "type" that may be caught or filtered respectively. Unlike the
6181 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6182 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6183 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6184 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6187 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6188 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6189 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6190 calling conventions, how the personality function results are represented in
6191 LLVM IR is target specific.</p>
6193 <p>The clauses are applied in order from top to bottom. If two
6194 <tt>landingpad</tt> instructions are merged together through inlining, the
6195 clauses from the calling function are appended to the list of clauses.
6196 When the call stack is being unwound due to an exception being thrown, the
6197 exception is compared against each <tt>clause</tt> in turn. If it doesn't
6198 match any of the clauses, and the <tt>cleanup</tt> flag is not set, then
6199 unwinding continues further up the call stack.</p>
6201 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6204 <li>A landing pad block is a basic block which is the unwind destination of an
6205 '<tt>invoke</tt>' instruction.</li>
6206 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6207 first non-PHI instruction.</li>
6208 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6210 <li>A basic block that is not a landing pad block may not include a
6211 '<tt>landingpad</tt>' instruction.</li>
6212 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6213 personality function.</li>
6218 ;; A landing pad which can catch an integer.
6219 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6221 ;; A landing pad that is a cleanup.
6222 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6224 ;; A landing pad which can catch an integer and can only throw a double.
6225 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6227 filter [1 x i8**] [@_ZTId]
6236 <!-- *********************************************************************** -->
6237 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6238 <!-- *********************************************************************** -->
6242 <p>LLVM supports the notion of an "intrinsic function". These functions have
6243 well known names and semantics and are required to follow certain
6244 restrictions. Overall, these intrinsics represent an extension mechanism for
6245 the LLVM language that does not require changing all of the transformations
6246 in LLVM when adding to the language (or the bitcode reader/writer, the
6247 parser, etc...).</p>
6249 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6250 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6251 begin with this prefix. Intrinsic functions must always be external
6252 functions: you cannot define the body of intrinsic functions. Intrinsic
6253 functions may only be used in call or invoke instructions: it is illegal to
6254 take the address of an intrinsic function. Additionally, because intrinsic
6255 functions are part of the LLVM language, it is required if any are added that
6256 they be documented here.</p>
6258 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6259 family of functions that perform the same operation but on different data
6260 types. Because LLVM can represent over 8 million different integer types,
6261 overloading is used commonly to allow an intrinsic function to operate on any
6262 integer type. One or more of the argument types or the result type can be
6263 overloaded to accept any integer type. Argument types may also be defined as
6264 exactly matching a previous argument's type or the result type. This allows
6265 an intrinsic function which accepts multiple arguments, but needs all of them
6266 to be of the same type, to only be overloaded with respect to a single
6267 argument or the result.</p>
6269 <p>Overloaded intrinsics will have the names of its overloaded argument types
6270 encoded into its function name, each preceded by a period. Only those types
6271 which are overloaded result in a name suffix. Arguments whose type is matched
6272 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6273 can take an integer of any width and returns an integer of exactly the same
6274 integer width. This leads to a family of functions such as
6275 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6276 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6277 suffix is required. Because the argument's type is matched against the return
6278 type, it does not require its own name suffix.</p>
6280 <p>To learn how to add an intrinsic function, please see the
6281 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6283 <!-- ======================================================================= -->
6285 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6290 <p>Variable argument support is defined in LLVM with
6291 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6292 intrinsic functions. These functions are related to the similarly named
6293 macros defined in the <tt><stdarg.h></tt> header file.</p>
6295 <p>All of these functions operate on arguments that use a target-specific value
6296 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6297 not define what this type is, so all transformations should be prepared to
6298 handle these functions regardless of the type used.</p>
6300 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6301 instruction and the variable argument handling intrinsic functions are
6304 <pre class="doc_code">
6305 define i32 @test(i32 %X, ...) {
6306 ; Initialize variable argument processing
6308 %ap2 = bitcast i8** %ap to i8*
6309 call void @llvm.va_start(i8* %ap2)
6311 ; Read a single integer argument
6312 %tmp = va_arg i8** %ap, i32
6314 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6316 %aq2 = bitcast i8** %aq to i8*
6317 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6318 call void @llvm.va_end(i8* %aq2)
6320 ; Stop processing of arguments.
6321 call void @llvm.va_end(i8* %ap2)
6325 declare void @llvm.va_start(i8*)
6326 declare void @llvm.va_copy(i8*, i8*)
6327 declare void @llvm.va_end(i8*)
6330 <!-- _______________________________________________________________________ -->
6332 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6340 declare void %llvm.va_start(i8* <arglist>)
6344 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6345 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6348 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6351 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6352 macro available in C. In a target-dependent way, it initializes
6353 the <tt>va_list</tt> element to which the argument points, so that the next
6354 call to <tt>va_arg</tt> will produce the first variable argument passed to
6355 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6356 need to know the last argument of the function as the compiler can figure
6361 <!-- _______________________________________________________________________ -->
6363 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6370 declare void @llvm.va_end(i8* <arglist>)
6374 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6375 which has been initialized previously
6376 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6377 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6380 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6383 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6384 macro available in C. In a target-dependent way, it destroys
6385 the <tt>va_list</tt> element to which the argument points. Calls
6386 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6387 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6388 with calls to <tt>llvm.va_end</tt>.</p>
6392 <!-- _______________________________________________________________________ -->
6394 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6401 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6405 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6406 from the source argument list to the destination argument list.</p>
6409 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6410 The second argument is a pointer to a <tt>va_list</tt> element to copy
6414 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6415 macro available in C. In a target-dependent way, it copies the
6416 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6417 element. This intrinsic is necessary because
6418 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6419 arbitrarily complex and require, for example, memory allocation.</p>
6425 <!-- ======================================================================= -->
6427 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6432 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6433 Collection</a> (GC) requires the implementation and generation of these
6434 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6435 roots on the stack</a>, as well as garbage collector implementations that
6436 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6437 barriers. Front-ends for type-safe garbage collected languages should generate
6438 these intrinsics to make use of the LLVM garbage collectors. For more details,
6439 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6442 <p>The garbage collection intrinsics only operate on objects in the generic
6443 address space (address space zero).</p>
6445 <!-- _______________________________________________________________________ -->
6447 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6454 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6458 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6459 the code generator, and allows some metadata to be associated with it.</p>
6462 <p>The first argument specifies the address of a stack object that contains the
6463 root pointer. The second pointer (which must be either a constant or a
6464 global value address) contains the meta-data to be associated with the
6468 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6469 location. At compile-time, the code generator generates information to allow
6470 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6471 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6476 <!-- _______________________________________________________________________ -->
6478 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6485 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6489 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6490 locations, allowing garbage collector implementations that require read
6494 <p>The second argument is the address to read from, which should be an address
6495 allocated from the garbage collector. The first object is a pointer to the
6496 start of the referenced object, if needed by the language runtime (otherwise
6500 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6501 instruction, but may be replaced with substantially more complex code by the
6502 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6503 may only be used in a function which <a href="#gc">specifies a GC
6508 <!-- _______________________________________________________________________ -->
6510 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6517 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6521 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6522 locations, allowing garbage collector implementations that require write
6523 barriers (such as generational or reference counting collectors).</p>
6526 <p>The first argument is the reference to store, the second is the start of the
6527 object to store it to, and the third is the address of the field of Obj to
6528 store to. If the runtime does not require a pointer to the object, Obj may
6532 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6533 instruction, but may be replaced with substantially more complex code by the
6534 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6535 may only be used in a function which <a href="#gc">specifies a GC
6542 <!-- ======================================================================= -->
6544 <a name="int_codegen">Code Generator Intrinsics</a>
6549 <p>These intrinsics are provided by LLVM to expose special features that may
6550 only be implemented with code generator support.</p>
6552 <!-- _______________________________________________________________________ -->
6554 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6561 declare i8 *@llvm.returnaddress(i32 <level>)
6565 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6566 target-specific value indicating the return address of the current function
6567 or one of its callers.</p>
6570 <p>The argument to this intrinsic indicates which function to return the address
6571 for. Zero indicates the calling function, one indicates its caller, etc.
6572 The argument is <b>required</b> to be a constant integer value.</p>
6575 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6576 indicating the return address of the specified call frame, or zero if it
6577 cannot be identified. The value returned by this intrinsic is likely to be
6578 incorrect or 0 for arguments other than zero, so it should only be used for
6579 debugging purposes.</p>
6581 <p>Note that calling this intrinsic does not prevent function inlining or other
6582 aggressive transformations, so the value returned may not be that of the
6583 obvious source-language caller.</p>
6587 <!-- _______________________________________________________________________ -->
6589 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6596 declare i8* @llvm.frameaddress(i32 <level>)
6600 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6601 target-specific frame pointer value for the specified stack frame.</p>
6604 <p>The argument to this intrinsic indicates which function to return the frame
6605 pointer for. Zero indicates the calling function, one indicates its caller,
6606 etc. The argument is <b>required</b> to be a constant integer value.</p>
6609 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6610 indicating the frame address of the specified call frame, or zero if it
6611 cannot be identified. The value returned by this intrinsic is likely to be
6612 incorrect or 0 for arguments other than zero, so it should only be used for
6613 debugging purposes.</p>
6615 <p>Note that calling this intrinsic does not prevent function inlining or other
6616 aggressive transformations, so the value returned may not be that of the
6617 obvious source-language caller.</p>
6621 <!-- _______________________________________________________________________ -->
6623 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6630 declare i8* @llvm.stacksave()
6634 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6635 of the function stack, for use
6636 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6637 useful for implementing language features like scoped automatic variable
6638 sized arrays in C99.</p>
6641 <p>This intrinsic returns a opaque pointer value that can be passed
6642 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6643 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6644 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6645 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6646 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6647 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6651 <!-- _______________________________________________________________________ -->
6653 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6660 declare void @llvm.stackrestore(i8* %ptr)
6664 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6665 the function stack to the state it was in when the
6666 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6667 executed. This is useful for implementing language features like scoped
6668 automatic variable sized arrays in C99.</p>
6671 <p>See the description
6672 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6676 <!-- _______________________________________________________________________ -->
6678 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6685 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6689 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6690 insert a prefetch instruction if supported; otherwise, it is a noop.
6691 Prefetches have no effect on the behavior of the program but can change its
6692 performance characteristics.</p>
6695 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6696 specifier determining if the fetch should be for a read (0) or write (1),
6697 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
6698 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6699 specifies whether the prefetch is performed on the data (1) or instruction (0)
6700 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
6701 must be constant integers.</p>
6704 <p>This intrinsic does not modify the behavior of the program. In particular,
6705 prefetches cannot trap and do not produce a value. On targets that support
6706 this intrinsic, the prefetch can provide hints to the processor cache for
6707 better performance.</p>
6711 <!-- _______________________________________________________________________ -->
6713 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
6720 declare void @llvm.pcmarker(i32 <id>)
6724 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
6725 Counter (PC) in a region of code to simulators and other tools. The method
6726 is target specific, but it is expected that the marker will use exported
6727 symbols to transmit the PC of the marker. The marker makes no guarantees
6728 that it will remain with any specific instruction after optimizations. It is
6729 possible that the presence of a marker will inhibit optimizations. The
6730 intended use is to be inserted after optimizations to allow correlations of
6731 simulation runs.</p>
6734 <p><tt>id</tt> is a numerical id identifying the marker.</p>
6737 <p>This intrinsic does not modify the behavior of the program. Backends that do
6738 not support this intrinsic may ignore it.</p>
6742 <!-- _______________________________________________________________________ -->
6744 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
6751 declare i64 @llvm.readcyclecounter()
6755 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
6756 counter register (or similar low latency, high accuracy clocks) on those
6757 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6758 should map to RPCC. As the backing counters overflow quickly (on the order
6759 of 9 seconds on alpha), this should only be used for small timings.</p>
6762 <p>When directly supported, reading the cycle counter should not modify any
6763 memory. Implementations are allowed to either return a application specific
6764 value or a system wide value. On backends without support, this is lowered
6765 to a constant 0.</p>
6771 <!-- ======================================================================= -->
6773 <a name="int_libc">Standard C Library Intrinsics</a>
6778 <p>LLVM provides intrinsics for a few important standard C library functions.
6779 These intrinsics allow source-language front-ends to pass information about
6780 the alignment of the pointer arguments to the code generator, providing
6781 opportunity for more efficient code generation.</p>
6783 <!-- _______________________________________________________________________ -->
6785 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
6791 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
6792 integer bit width and for different address spaces. Not all targets support
6793 all bit widths however.</p>
6796 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6797 i32 <len>, i32 <align>, i1 <isvolatile>)
6798 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6799 i64 <len>, i32 <align>, i1 <isvolatile>)
6803 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6804 source location to the destination location.</p>
6806 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
6807 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6808 and the pointers can be in specified address spaces.</p>
6812 <p>The first argument is a pointer to the destination, the second is a pointer
6813 to the source. The third argument is an integer argument specifying the
6814 number of bytes to copy, the fourth argument is the alignment of the
6815 source and destination locations, and the fifth is a boolean indicating a
6816 volatile access.</p>
6818 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6819 then the caller guarantees that both the source and destination pointers are
6820 aligned to that boundary.</p>
6822 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6823 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
6824 The detailed access behavior is not very cleanly specified and it is unwise
6825 to depend on it.</p>
6829 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6830 source location to the destination location, which are not allowed to
6831 overlap. It copies "len" bytes of memory over. If the argument is known to
6832 be aligned to some boundary, this can be specified as the fourth argument,
6833 otherwise it should be set to 0 or 1.</p>
6837 <!-- _______________________________________________________________________ -->
6839 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
6845 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
6846 width and for different address space. Not all targets support all bit
6850 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6851 i32 <len>, i32 <align>, i1 <isvolatile>)
6852 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6853 i64 <len>, i32 <align>, i1 <isvolatile>)
6857 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
6858 source location to the destination location. It is similar to the
6859 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
6862 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
6863 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6864 and the pointers can be in specified address spaces.</p>
6868 <p>The first argument is a pointer to the destination, the second is a pointer
6869 to the source. The third argument is an integer argument specifying the
6870 number of bytes to copy, the fourth argument is the alignment of the
6871 source and destination locations, and the fifth is a boolean indicating a
6872 volatile access.</p>
6874 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6875 then the caller guarantees that the source and destination pointers are
6876 aligned to that boundary.</p>
6878 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6879 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
6880 The detailed access behavior is not very cleanly specified and it is unwise
6881 to depend on it.</p>
6885 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
6886 source location to the destination location, which may overlap. It copies
6887 "len" bytes of memory over. If the argument is known to be aligned to some
6888 boundary, this can be specified as the fourth argument, otherwise it should
6889 be set to 0 or 1.</p>
6893 <!-- _______________________________________________________________________ -->
6895 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
6901 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6902 width and for different address spaces. However, not all targets support all
6906 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6907 i32 <len>, i32 <align>, i1 <isvolatile>)
6908 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6909 i64 <len>, i32 <align>, i1 <isvolatile>)
6913 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
6914 particular byte value.</p>
6916 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
6917 intrinsic does not return a value and takes extra alignment/volatile
6918 arguments. Also, the destination can be in an arbitrary address space.</p>
6921 <p>The first argument is a pointer to the destination to fill, the second is the
6922 byte value with which to fill it, the third argument is an integer argument
6923 specifying the number of bytes to fill, and the fourth argument is the known
6924 alignment of the destination location.</p>
6926 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6927 then the caller guarantees that the destination pointer is aligned to that
6930 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6931 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
6932 The detailed access behavior is not very cleanly specified and it is unwise
6933 to depend on it.</p>
6936 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
6937 at the destination location. If the argument is known to be aligned to some
6938 boundary, this can be specified as the fourth argument, otherwise it should
6939 be set to 0 or 1.</p>
6943 <!-- _______________________________________________________________________ -->
6945 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
6951 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
6952 floating point or vector of floating point type. Not all targets support all
6956 declare float @llvm.sqrt.f32(float %Val)
6957 declare double @llvm.sqrt.f64(double %Val)
6958 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6959 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6960 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6964 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
6965 returning the same value as the libm '<tt>sqrt</tt>' functions would.
6966 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
6967 behavior for negative numbers other than -0.0 (which allows for better
6968 optimization, because there is no need to worry about errno being
6969 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
6972 <p>The argument and return value are floating point numbers of the same
6976 <p>This function returns the sqrt of the specified operand if it is a
6977 nonnegative floating point number.</p>
6981 <!-- _______________________________________________________________________ -->
6983 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
6989 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
6990 floating point or vector of floating point type. Not all targets support all
6994 declare float @llvm.powi.f32(float %Val, i32 %power)
6995 declare double @llvm.powi.f64(double %Val, i32 %power)
6996 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6997 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6998 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7002 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
7003 specified (positive or negative) power. The order of evaluation of
7004 multiplications is not defined. When a vector of floating point type is
7005 used, the second argument remains a scalar integer value.</p>
7008 <p>The second argument is an integer power, and the first is a value to raise to
7012 <p>This function returns the first value raised to the second power with an
7013 unspecified sequence of rounding operations.</p>
7017 <!-- _______________________________________________________________________ -->
7019 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
7025 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
7026 floating point or vector of floating point type. Not all targets support all
7030 declare float @llvm.sin.f32(float %Val)
7031 declare double @llvm.sin.f64(double %Val)
7032 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7033 declare fp128 @llvm.sin.f128(fp128 %Val)
7034 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7038 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
7041 <p>The argument and return value are floating point numbers of the same
7045 <p>This function returns the sine of the specified operand, returning the same
7046 values as the libm <tt>sin</tt> functions would, and handles error conditions
7047 in the same way.</p>
7051 <!-- _______________________________________________________________________ -->
7053 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
7059 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
7060 floating point or vector of floating point type. Not all targets support all
7064 declare float @llvm.cos.f32(float %Val)
7065 declare double @llvm.cos.f64(double %Val)
7066 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7067 declare fp128 @llvm.cos.f128(fp128 %Val)
7068 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7072 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
7075 <p>The argument and return value are floating point numbers of the same
7079 <p>This function returns the cosine of the specified operand, returning the same
7080 values as the libm <tt>cos</tt> functions would, and handles error conditions
7081 in the same way.</p>
7085 <!-- _______________________________________________________________________ -->
7087 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
7093 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
7094 floating point or vector of floating point type. Not all targets support all
7098 declare float @llvm.pow.f32(float %Val, float %Power)
7099 declare double @llvm.pow.f64(double %Val, double %Power)
7100 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7101 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7102 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7106 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
7107 specified (positive or negative) power.</p>
7110 <p>The second argument is a floating point power, and the first is a value to
7111 raise to that power.</p>
7114 <p>This function returns the first value raised to the second power, returning
7115 the same values as the libm <tt>pow</tt> functions would, and handles error
7116 conditions in the same way.</p>
7120 <!-- _______________________________________________________________________ -->
7122 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7128 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7129 floating point or vector of floating point type. Not all targets support all
7133 declare float @llvm.exp.f32(float %Val)
7134 declare double @llvm.exp.f64(double %Val)
7135 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7136 declare fp128 @llvm.exp.f128(fp128 %Val)
7137 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7141 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7144 <p>The argument and return value are floating point numbers of the same
7148 <p>This function returns the same values as the libm <tt>exp</tt> functions
7149 would, and handles error conditions in the same way.</p>
7153 <!-- _______________________________________________________________________ -->
7155 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7161 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7162 floating point or vector of floating point type. Not all targets support all
7166 declare float @llvm.log.f32(float %Val)
7167 declare double @llvm.log.f64(double %Val)
7168 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7169 declare fp128 @llvm.log.f128(fp128 %Val)
7170 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7174 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7177 <p>The argument and return value are floating point numbers of the same
7181 <p>This function returns the same values as the libm <tt>log</tt> functions
7182 would, and handles error conditions in the same way.</p>
7186 <!-- _______________________________________________________________________ -->
7188 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7194 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7195 floating point or vector of floating point type. Not all targets support all
7199 declare float @llvm.fma.f32(float %a, float %b, float %c)
7200 declare double @llvm.fma.f64(double %a, double %b, double %c)
7201 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7202 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7203 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7207 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7211 <p>The argument and return value are floating point numbers of the same
7215 <p>This function returns the same values as the libm <tt>fma</tt> functions
7222 <!-- ======================================================================= -->
7224 <a name="int_manip">Bit Manipulation Intrinsics</a>
7229 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7230 These allow efficient code generation for some algorithms.</p>
7232 <!-- _______________________________________________________________________ -->
7234 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7240 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7241 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7244 declare i16 @llvm.bswap.i16(i16 <id>)
7245 declare i32 @llvm.bswap.i32(i32 <id>)
7246 declare i64 @llvm.bswap.i64(i64 <id>)
7250 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7251 values with an even number of bytes (positive multiple of 16 bits). These
7252 are useful for performing operations on data that is not in the target's
7253 native byte order.</p>
7256 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7257 and low byte of the input i16 swapped. Similarly,
7258 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7259 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7260 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7261 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7262 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7263 more, respectively).</p>
7267 <!-- _______________________________________________________________________ -->
7269 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7275 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7276 width, or on any vector with integer elements. Not all targets support all
7277 bit widths or vector types, however.</p>
7280 declare i8 @llvm.ctpop.i8(i8 <src>)
7281 declare i16 @llvm.ctpop.i16(i16 <src>)
7282 declare i32 @llvm.ctpop.i32(i32 <src>)
7283 declare i64 @llvm.ctpop.i64(i64 <src>)
7284 declare i256 @llvm.ctpop.i256(i256 <src>)
7285 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7289 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7293 <p>The only argument is the value to be counted. The argument may be of any
7294 integer type, or a vector with integer elements.
7295 The return type must match the argument type.</p>
7298 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7299 element of a vector.</p>
7303 <!-- _______________________________________________________________________ -->
7305 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7311 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7312 integer bit width, or any vector whose elements are integers. Not all
7313 targets support all bit widths or vector types, however.</p>
7316 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7317 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7318 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7319 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7320 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7321 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7325 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7326 leading zeros in a variable.</p>
7329 <p>The first argument is the value to be counted. This argument may be of any
7330 integer type, or a vectory with integer element type. The return type
7331 must match the first argument type.</p>
7333 <p>The second argument must be a constant and is a flag to indicate whether the
7334 intrinsic should ensure that a zero as the first argument produces a defined
7335 result. Historically some architectures did not provide a defined result for
7336 zero values as efficiently, and many algorithms are now predicated on
7337 avoiding zero-value inputs.</p>
7340 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7341 zeros in a variable, or within each element of the vector.
7342 If <tt>src == 0</tt> then the result is the size in bits of the type of
7343 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7344 For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7348 <!-- _______________________________________________________________________ -->
7350 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7356 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7357 integer bit width, or any vector of integer elements. Not all targets
7358 support all bit widths or vector types, however.</p>
7361 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7362 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7363 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7364 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7365 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7366 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7370 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7374 <p>The first argument is the value to be counted. This argument may be of any
7375 integer type, or a vectory with integer element type. The return type
7376 must match the first argument type.</p>
7378 <p>The second argument must be a constant and is a flag to indicate whether the
7379 intrinsic should ensure that a zero as the first argument produces a defined
7380 result. Historically some architectures did not provide a defined result for
7381 zero values as efficiently, and many algorithms are now predicated on
7382 avoiding zero-value inputs.</p>
7385 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7386 zeros in a variable, or within each element of a vector.
7387 If <tt>src == 0</tt> then the result is the size in bits of the type of
7388 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7389 For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7395 <!-- ======================================================================= -->
7397 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7402 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7404 <!-- _______________________________________________________________________ -->
7406 <a name="int_sadd_overflow">
7407 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7414 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7415 on any integer bit width.</p>
7418 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7419 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7420 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7424 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7425 a signed addition of the two arguments, and indicate whether an overflow
7426 occurred during the signed summation.</p>
7429 <p>The arguments (%a and %b) and the first element of the result structure may
7430 be of integer types of any bit width, but they must have the same bit
7431 width. The second element of the result structure must be of
7432 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7433 undergo signed addition.</p>
7436 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7437 a signed addition of the two variables. They return a structure — the
7438 first element of which is the signed summation, and the second element of
7439 which is a bit specifying if the signed summation resulted in an
7444 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7445 %sum = extractvalue {i32, i1} %res, 0
7446 %obit = extractvalue {i32, i1} %res, 1
7447 br i1 %obit, label %overflow, label %normal
7452 <!-- _______________________________________________________________________ -->
7454 <a name="int_uadd_overflow">
7455 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7462 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7463 on any integer bit width.</p>
7466 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7467 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7468 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7472 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7473 an unsigned addition of the two arguments, and indicate whether a carry
7474 occurred during the unsigned summation.</p>
7477 <p>The arguments (%a and %b) and the first element of the result structure may
7478 be of integer types of any bit width, but they must have the same bit
7479 width. The second element of the result structure must be of
7480 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7481 undergo unsigned addition.</p>
7484 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7485 an unsigned addition of the two arguments. They return a structure —
7486 the first element of which is the sum, and the second element of which is a
7487 bit specifying if the unsigned summation resulted in a carry.</p>
7491 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7492 %sum = extractvalue {i32, i1} %res, 0
7493 %obit = extractvalue {i32, i1} %res, 1
7494 br i1 %obit, label %carry, label %normal
7499 <!-- _______________________________________________________________________ -->
7501 <a name="int_ssub_overflow">
7502 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7509 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7510 on any integer bit width.</p>
7513 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7514 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7515 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7519 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7520 a signed subtraction of the two arguments, and indicate whether an overflow
7521 occurred during the signed subtraction.</p>
7524 <p>The arguments (%a and %b) and the first element of the result structure may
7525 be of integer types of any bit width, but they must have the same bit
7526 width. The second element of the result structure must be of
7527 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7528 undergo signed subtraction.</p>
7531 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7532 a signed subtraction of the two arguments. They return a structure —
7533 the first element of which is the subtraction, and the second element of
7534 which is a bit specifying if the signed subtraction resulted in an
7539 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7540 %sum = extractvalue {i32, i1} %res, 0
7541 %obit = extractvalue {i32, i1} %res, 1
7542 br i1 %obit, label %overflow, label %normal
7547 <!-- _______________________________________________________________________ -->
7549 <a name="int_usub_overflow">
7550 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7557 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7558 on any integer bit width.</p>
7561 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7562 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7563 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7567 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7568 an unsigned subtraction of the two arguments, and indicate whether an
7569 overflow occurred during the unsigned subtraction.</p>
7572 <p>The arguments (%a and %b) and the first element of the result structure may
7573 be of integer types of any bit width, but they must have the same bit
7574 width. The second element of the result structure must be of
7575 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7576 undergo unsigned subtraction.</p>
7579 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7580 an unsigned subtraction of the two arguments. They return a structure —
7581 the first element of which is the subtraction, and the second element of
7582 which is a bit specifying if the unsigned subtraction resulted in an
7587 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7588 %sum = extractvalue {i32, i1} %res, 0
7589 %obit = extractvalue {i32, i1} %res, 1
7590 br i1 %obit, label %overflow, label %normal
7595 <!-- _______________________________________________________________________ -->
7597 <a name="int_smul_overflow">
7598 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7605 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7606 on any integer bit width.</p>
7609 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7610 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7611 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7616 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7617 a signed multiplication of the two arguments, and indicate whether an
7618 overflow occurred during the signed multiplication.</p>
7621 <p>The arguments (%a and %b) and the first element of the result structure may
7622 be of integer types of any bit width, but they must have the same bit
7623 width. The second element of the result structure must be of
7624 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7625 undergo signed multiplication.</p>
7628 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7629 a signed multiplication of the two arguments. They return a structure —
7630 the first element of which is the multiplication, and the second element of
7631 which is a bit specifying if the signed multiplication resulted in an
7636 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7637 %sum = extractvalue {i32, i1} %res, 0
7638 %obit = extractvalue {i32, i1} %res, 1
7639 br i1 %obit, label %overflow, label %normal
7644 <!-- _______________________________________________________________________ -->
7646 <a name="int_umul_overflow">
7647 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
7654 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
7655 on any integer bit width.</p>
7658 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7659 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7660 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7664 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7665 a unsigned multiplication of the two arguments, and indicate whether an
7666 overflow occurred during the unsigned multiplication.</p>
7669 <p>The arguments (%a and %b) and the first element of the result structure may
7670 be of integer types of any bit width, but they must have the same bit
7671 width. The second element of the result structure must be of
7672 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7673 undergo unsigned multiplication.</p>
7676 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7677 an unsigned multiplication of the two arguments. They return a structure
7678 — the first element of which is the multiplication, and the second
7679 element of which is a bit specifying if the unsigned multiplication resulted
7684 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7685 %sum = extractvalue {i32, i1} %res, 0
7686 %obit = extractvalue {i32, i1} %res, 1
7687 br i1 %obit, label %overflow, label %normal
7694 <!-- ======================================================================= -->
7696 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
7701 <p>Half precision floating point is a storage-only format. This means that it is
7702 a dense encoding (in memory) but does not support computation in the
7705 <p>This means that code must first load the half-precision floating point
7706 value as an i16, then convert it to float with <a
7707 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
7708 Computation can then be performed on the float value (including extending to
7709 double etc). To store the value back to memory, it is first converted to
7710 float if needed, then converted to i16 with
7711 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
7712 storing as an i16 value.</p>
7714 <!-- _______________________________________________________________________ -->
7716 <a name="int_convert_to_fp16">
7717 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
7725 declare i16 @llvm.convert.to.fp16(f32 %a)
7729 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7730 a conversion from single precision floating point format to half precision
7731 floating point format.</p>
7734 <p>The intrinsic function contains single argument - the value to be
7738 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7739 a conversion from single precision floating point format to half precision
7740 floating point format. The return value is an <tt>i16</tt> which
7741 contains the converted number.</p>
7745 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7746 store i16 %res, i16* @x, align 2
7751 <!-- _______________________________________________________________________ -->
7753 <a name="int_convert_from_fp16">
7754 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
7762 declare f32 @llvm.convert.from.fp16(i16 %a)
7766 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
7767 a conversion from half precision floating point format to single precision
7768 floating point format.</p>
7771 <p>The intrinsic function contains single argument - the value to be
7775 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
7776 conversion from half single precision floating point format to single
7777 precision floating point format. The input half-float value is represented by
7778 an <tt>i16</tt> value.</p>
7782 %a = load i16* @x, align 2
7783 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7790 <!-- ======================================================================= -->
7792 <a name="int_debugger">Debugger Intrinsics</a>
7797 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
7798 prefix), are described in
7799 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
7800 Level Debugging</a> document.</p>
7804 <!-- ======================================================================= -->
7806 <a name="int_eh">Exception Handling Intrinsics</a>
7811 <p>The LLVM exception handling intrinsics (which all start with
7812 <tt>llvm.eh.</tt> prefix), are described in
7813 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
7814 Handling</a> document.</p>
7818 <!-- ======================================================================= -->
7820 <a name="int_trampoline">Trampoline Intrinsics</a>
7825 <p>These intrinsics make it possible to excise one parameter, marked with
7826 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
7827 The result is a callable
7828 function pointer lacking the nest parameter - the caller does not need to
7829 provide a value for it. Instead, the value to use is stored in advance in a
7830 "trampoline", a block of memory usually allocated on the stack, which also
7831 contains code to splice the nest value into the argument list. This is used
7832 to implement the GCC nested function address extension.</p>
7834 <p>For example, if the function is
7835 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
7836 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
7839 <pre class="doc_code">
7840 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
7841 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
7842 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
7843 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
7844 %fp = bitcast i8* %p to i32 (i32, i32)*
7847 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
7848 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
7850 <!-- _______________________________________________________________________ -->
7853 '<tt>llvm.init.trampoline</tt>' Intrinsic
7861 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
7865 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
7866 turning it into a trampoline.</p>
7869 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
7870 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
7871 sufficiently aligned block of memory; this memory is written to by the
7872 intrinsic. Note that the size and the alignment are target-specific - LLVM
7873 currently provides no portable way of determining them, so a front-end that
7874 generates this intrinsic needs to have some target-specific knowledge.
7875 The <tt>func</tt> argument must hold a function bitcast to
7876 an <tt>i8*</tt>.</p>
7879 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
7880 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
7881 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
7882 which can be <a href="#int_trampoline">bitcast (to a new function) and
7883 called</a>. The new function's signature is the same as that of
7884 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
7885 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
7886 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
7887 with the same argument list, but with <tt>nval</tt> used for the missing
7888 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
7889 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
7890 to the returned function pointer is undefined.</p>
7893 <!-- _______________________________________________________________________ -->
7896 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
7904 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
7908 <p>This performs any required machine-specific adjustment to the address of a
7909 trampoline (passed as <tt>tramp</tt>).</p>
7912 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
7913 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
7917 <p>On some architectures the address of the code to be executed needs to be
7918 different to the address where the trampoline is actually stored. This
7919 intrinsic returns the executable address corresponding to <tt>tramp</tt>
7920 after performing the required machine specific adjustments.
7921 The pointer returned can then be <a href="#int_trampoline"> bitcast and
7929 <!-- ======================================================================= -->
7931 <a name="int_memorymarkers">Memory Use Markers</a>
7936 <p>This class of intrinsics exists to information about the lifetime of memory
7937 objects and ranges where variables are immutable.</p>
7939 <!-- _______________________________________________________________________ -->
7941 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
7948 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
7952 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
7953 object's lifetime.</p>
7956 <p>The first argument is a constant integer representing the size of the
7957 object, or -1 if it is variable sized. The second argument is a pointer to
7961 <p>This intrinsic indicates that before this point in the code, the value of the
7962 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7963 never be used and has an undefined value. A load from the pointer that
7964 precedes this intrinsic can be replaced with
7965 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
7969 <!-- _______________________________________________________________________ -->
7971 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
7978 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
7982 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
7983 object's lifetime.</p>
7986 <p>The first argument is a constant integer representing the size of the
7987 object, or -1 if it is variable sized. The second argument is a pointer to
7991 <p>This intrinsic indicates that after this point in the code, the value of the
7992 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
7993 never be used and has an undefined value. Any stores into the memory object
7994 following this intrinsic may be removed as dead.
7998 <!-- _______________________________________________________________________ -->
8000 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8007 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8011 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8012 a memory object will not change.</p>
8015 <p>The first argument is a constant integer representing the size of the
8016 object, or -1 if it is variable sized. The second argument is a pointer to
8020 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8021 the return value, the referenced memory location is constant and
8026 <!-- _______________________________________________________________________ -->
8028 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8035 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8039 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8040 a memory object are mutable.</p>
8043 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8044 The second argument is a constant integer representing the size of the
8045 object, or -1 if it is variable sized and the third argument is a pointer
8049 <p>This intrinsic indicates that the memory is mutable again.</p>
8055 <!-- ======================================================================= -->
8057 <a name="int_general">General Intrinsics</a>
8062 <p>This class of intrinsics is designed to be generic and has no specific
8065 <!-- _______________________________________________________________________ -->
8067 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8074 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8078 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8081 <p>The first argument is a pointer to a value, the second is a pointer to a
8082 global string, the third is a pointer to a global string which is the source
8083 file name, and the last argument is the line number.</p>
8086 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8087 This can be useful for special purpose optimizations that want to look for
8088 these annotations. These have no other defined use; they are ignored by code
8089 generation and optimization.</p>
8093 <!-- _______________________________________________________________________ -->
8095 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8101 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8102 any integer bit width.</p>
8105 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8106 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8107 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8108 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8109 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8113 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8116 <p>The first argument is an integer value (result of some expression), the
8117 second is a pointer to a global string, the third is a pointer to a global
8118 string which is the source file name, and the last argument is the line
8119 number. It returns the value of the first argument.</p>
8122 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8123 arbitrary strings. This can be useful for special purpose optimizations that
8124 want to look for these annotations. These have no other defined use; they
8125 are ignored by code generation and optimization.</p>
8129 <!-- _______________________________________________________________________ -->
8131 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8138 declare void @llvm.trap()
8142 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8148 <p>This intrinsics is lowered to the target dependent trap instruction. If the
8149 target does not have a trap instruction, this intrinsic will be lowered to
8150 the call of the <tt>abort()</tt> function.</p>
8154 <!-- _______________________________________________________________________ -->
8156 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8163 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8167 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8168 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8169 ensure that it is placed on the stack before local variables.</p>
8172 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8173 arguments. The first argument is the value loaded from the stack
8174 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8175 that has enough space to hold the value of the guard.</p>
8178 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8179 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8180 stack. This is to ensure that if a local variable on the stack is
8181 overwritten, it will destroy the value of the guard. When the function exits,
8182 the guard on the stack is checked against the original guard. If they are
8183 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8188 <!-- _______________________________________________________________________ -->
8190 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8197 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <type>)
8198 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <type>)
8202 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8203 the optimizers to determine at compile time whether a) an operation (like
8204 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8205 runtime check for overflow isn't necessary. An object in this context means
8206 an allocation of a specific class, structure, array, or other object.</p>
8209 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8210 argument is a pointer to or into the <tt>object</tt>. The second argument
8211 is a boolean 0 or 1. This argument determines whether you want the
8212 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
8213 1, variables are not allowed.</p>
8216 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
8217 representing the size of the object concerned, or <tt>i32/i64 -1 or 0</tt>,
8218 depending on the <tt>type</tt> argument, if the size cannot be determined at
8222 <!-- _______________________________________________________________________ -->
8224 <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a>
8231 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8232 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8236 <p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the
8237 most probable) value of <tt>val</tt>, which can be used by optimizers.</p>
8240 <p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first
8241 argument is a value. The second argument is an expected value, this needs to
8242 be a constant value, variables are not allowed.</p>
8245 <p>This intrinsic is lowered to the <tt>val</tt>.</p>
8251 <!-- *********************************************************************** -->
8254 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8255 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8256 <a href="http://validator.w3.org/check/referer"><img
8257 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8259 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8260 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8261 Last modified: $Date$