1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="_static/llvm.css" type="text/css">
15 <h1>LLVM Language Reference Manual</h1>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a>
25 <li><a href="#linkage_private">'<tt>private</tt>' Linkage</a></li>
26 <li><a href="#linkage_linker_private">'<tt>linker_private</tt>' Linkage</a></li>
27 <li><a href="#linkage_linker_private_weak">'<tt>linker_private_weak</tt>' Linkage</a></li>
28 <li><a href="#linkage_linker_private_weak_def_auto">'<tt>linker_private_weak_def_auto</tt>' Linkage</a></li>
29 <li><a href="#linkage_internal">'<tt>internal</tt>' Linkage</a></li>
30 <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
31 <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
32 <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
33 <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
34 <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
35 <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
36 <li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
37 <li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
38 <li><a href="#linkage_external">'<tt>external</tt>' Linkage</a></li>
39 <li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
40 <li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
43 <li><a href="#callingconv">Calling Conventions</a></li>
44 <li><a href="#namedtypes">Named Types</a></li>
45 <li><a href="#globalvars">Global Variables</a></li>
46 <li><a href="#functionstructure">Functions</a></li>
47 <li><a href="#aliasstructure">Aliases</a></li>
48 <li><a href="#namedmetadatastructure">Named Metadata</a></li>
49 <li><a href="#paramattrs">Parameter Attributes</a></li>
50 <li><a href="#fnattrs">Function Attributes</a></li>
51 <li><a href="#gc">Garbage Collector Names</a></li>
52 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
53 <li><a href="#datalayout">Data Layout</a></li>
54 <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
55 <li><a href="#volatile">Volatile Memory Accesses</a></li>
56 <li><a href="#memmodel">Memory Model for Concurrent Operations</a></li>
57 <li><a href="#ordering">Atomic Memory Ordering Constraints</a></li>
60 <li><a href="#typesystem">Type System</a>
62 <li><a href="#t_classifications">Type Classifications</a></li>
63 <li><a href="#t_primitive">Primitive Types</a>
65 <li><a href="#t_integer">Integer Type</a></li>
66 <li><a href="#t_floating">Floating Point Types</a></li>
67 <li><a href="#t_x86mmx">X86mmx Type</a></li>
68 <li><a href="#t_void">Void Type</a></li>
69 <li><a href="#t_label">Label Type</a></li>
70 <li><a href="#t_metadata">Metadata Type</a></li>
73 <li><a href="#t_derived">Derived Types</a>
75 <li><a href="#t_aggregate">Aggregate Types</a>
77 <li><a href="#t_array">Array Type</a></li>
78 <li><a href="#t_struct">Structure Type</a></li>
79 <li><a href="#t_opaque">Opaque Structure Types</a></li>
80 <li><a href="#t_vector">Vector Type</a></li>
83 <li><a href="#t_function">Function Type</a></li>
84 <li><a href="#t_pointer">Pointer Type</a></li>
89 <li><a href="#constants">Constants</a>
91 <li><a href="#simpleconstants">Simple Constants</a></li>
92 <li><a href="#complexconstants">Complex Constants</a></li>
93 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
94 <li><a href="#undefvalues">Undefined Values</a></li>
95 <li><a href="#poisonvalues">Poison Values</a></li>
96 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
97 <li><a href="#constantexprs">Constant Expressions</a></li>
100 <li><a href="#othervalues">Other Values</a>
102 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
103 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a>
105 <li><a href="#tbaa">'<tt>tbaa</tt>' Metadata</a></li>
106 <li><a href="#fpmath">'<tt>fpmath</tt>' Metadata</a></li>
107 <li><a href="#range">'<tt>range</tt>' Metadata</a></li>
112 <li><a href="#module_flags">Module Flags Metadata</a>
114 <li><a href="#objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a></li>
117 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
119 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
120 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
121 Global Variable</a></li>
122 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
123 Global Variable</a></li>
124 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
125 Global Variable</a></li>
128 <li><a href="#instref">Instruction Reference</a>
130 <li><a href="#terminators">Terminator Instructions</a>
132 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
133 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
134 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
135 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
136 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
137 <li><a href="#i_resume">'<tt>resume</tt>' Instruction</a></li>
138 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
141 <li><a href="#binaryops">Binary Operations</a>
143 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
144 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
145 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
146 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
147 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
148 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
149 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
150 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
151 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
152 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
153 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
154 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
157 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
159 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
160 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
161 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
162 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
163 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
164 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
167 <li><a href="#vectorops">Vector Operations</a>
169 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
170 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
171 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
174 <li><a href="#aggregateops">Aggregate Operations</a>
176 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
177 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
180 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
182 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
183 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
184 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
185 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
186 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
187 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
188 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
191 <li><a href="#convertops">Conversion Operations</a>
193 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
194 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
195 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
196 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
197 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
198 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
199 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
200 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
201 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
202 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
203 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
204 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
207 <li><a href="#otherops">Other Operations</a>
209 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
210 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
211 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
212 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
213 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
214 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
215 <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
220 <li><a href="#intrinsics">Intrinsic Functions</a>
222 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
224 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
225 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
226 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
229 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
231 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
232 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
233 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
236 <li><a href="#int_codegen">Code Generator Intrinsics</a>
238 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
239 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
240 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
241 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
242 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
243 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
244 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
247 <li><a href="#int_libc">Standard C Library Intrinsics</a>
249 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
250 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
251 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
252 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
253 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
254 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
255 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
256 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
257 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
258 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
259 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
262 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
264 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
265 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
266 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
267 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
270 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
272 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
273 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
274 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
275 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
276 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
277 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
280 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
282 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
283 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
286 <li><a href="#int_debugger">Debugger intrinsics</a></li>
287 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
288 <li><a href="#int_trampoline">Trampoline Intrinsics</a>
290 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
291 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
294 <li><a href="#int_memorymarkers">Memory Use Markers</a>
296 <li><a href="#int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a></li>
297 <li><a href="#int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a></li>
298 <li><a href="#int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a></li>
299 <li><a href="#int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a></li>
302 <li><a href="#int_general">General intrinsics</a>
304 <li><a href="#int_var_annotation">
305 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
306 <li><a href="#int_annotation">
307 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
308 <li><a href="#int_trap">
309 '<tt>llvm.trap</tt>' Intrinsic</a></li>
310 <li><a href="#int_debugtrap">
311 '<tt>llvm.debugtrap</tt>' Intrinsic</a></li>
312 <li><a href="#int_stackprotector">
313 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
314 <li><a href="#int_objectsize">
315 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
316 <li><a href="#int_expect">
317 '<tt>llvm.expect</tt>' Intrinsic</a></li>
324 <div class="doc_author">
325 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
326 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
329 <!-- *********************************************************************** -->
330 <h2><a name="abstract">Abstract</a></h2>
331 <!-- *********************************************************************** -->
335 <p>This document is a reference manual for the LLVM assembly language. LLVM is
336 a Static Single Assignment (SSA) based representation that provides type
337 safety, low-level operations, flexibility, and the capability of representing
338 'all' high-level languages cleanly. It is the common code representation
339 used throughout all phases of the LLVM compilation strategy.</p>
343 <!-- *********************************************************************** -->
344 <h2><a name="introduction">Introduction</a></h2>
345 <!-- *********************************************************************** -->
349 <p>The LLVM code representation is designed to be used in three different forms:
350 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
351 for fast loading by a Just-In-Time compiler), and as a human readable
352 assembly language representation. This allows LLVM to provide a powerful
353 intermediate representation for efficient compiler transformations and
354 analysis, while providing a natural means to debug and visualize the
355 transformations. The three different forms of LLVM are all equivalent. This
356 document describes the human readable representation and notation.</p>
358 <p>The LLVM representation aims to be light-weight and low-level while being
359 expressive, typed, and extensible at the same time. It aims to be a
360 "universal IR" of sorts, by being at a low enough level that high-level ideas
361 may be cleanly mapped to it (similar to how microprocessors are "universal
362 IR's", allowing many source languages to be mapped to them). By providing
363 type information, LLVM can be used as the target of optimizations: for
364 example, through pointer analysis, it can be proven that a C automatic
365 variable is never accessed outside of the current function, allowing it to
366 be promoted to a simple SSA value instead of a memory location.</p>
368 <!-- _______________________________________________________________________ -->
370 <a name="wellformed">Well-Formedness</a>
375 <p>It is important to note that this document describes 'well formed' LLVM
376 assembly language. There is a difference between what the parser accepts and
377 what is considered 'well formed'. For example, the following instruction is
378 syntactically okay, but not well formed:</p>
380 <pre class="doc_code">
381 %x = <a href="#i_add">add</a> i32 1, %x
384 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
385 LLVM infrastructure provides a verification pass that may be used to verify
386 that an LLVM module is well formed. This pass is automatically run by the
387 parser after parsing input assembly and by the optimizer before it outputs
388 bitcode. The violations pointed out by the verifier pass indicate bugs in
389 transformation passes or input to the parser.</p>
395 <!-- Describe the typesetting conventions here. -->
397 <!-- *********************************************************************** -->
398 <h2><a name="identifiers">Identifiers</a></h2>
399 <!-- *********************************************************************** -->
403 <p>LLVM identifiers come in two basic types: global and local. Global
404 identifiers (functions, global variables) begin with the <tt>'@'</tt>
405 character. Local identifiers (register names, types) begin with
406 the <tt>'%'</tt> character. Additionally, there are three different formats
407 for identifiers, for different purposes:</p>
410 <li>Named values are represented as a string of characters with their prefix.
411 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
412 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
413 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
414 other characters in their names can be surrounded with quotes. Special
415 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
416 ASCII code for the character in hexadecimal. In this way, any character
417 can be used in a name value, even quotes themselves.</li>
419 <li>Unnamed values are represented as an unsigned numeric value with their
420 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
422 <li>Constants, which are described in a <a href="#constants">section about
423 constants</a>, below.</li>
426 <p>LLVM requires that values start with a prefix for two reasons: Compilers
427 don't need to worry about name clashes with reserved words, and the set of
428 reserved words may be expanded in the future without penalty. Additionally,
429 unnamed identifiers allow a compiler to quickly come up with a temporary
430 variable without having to avoid symbol table conflicts.</p>
432 <p>Reserved words in LLVM are very similar to reserved words in other
433 languages. There are keywords for different opcodes
434 ('<tt><a href="#i_add">add</a></tt>',
435 '<tt><a href="#i_bitcast">bitcast</a></tt>',
436 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
437 ('<tt><a href="#t_void">void</a></tt>',
438 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
439 reserved words cannot conflict with variable names, because none of them
440 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
442 <p>Here is an example of LLVM code to multiply the integer variable
443 '<tt>%X</tt>' by 8:</p>
447 <pre class="doc_code">
448 %result = <a href="#i_mul">mul</a> i32 %X, 8
451 <p>After strength reduction:</p>
453 <pre class="doc_code">
454 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
457 <p>And the hard way:</p>
459 <pre class="doc_code">
460 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
461 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
462 %result = <a href="#i_add">add</a> i32 %1, %1
465 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
466 lexical features of LLVM:</p>
469 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
472 <li>Unnamed temporaries are created when the result of a computation is not
473 assigned to a named value.</li>
475 <li>Unnamed temporaries are numbered sequentially</li>
478 <p>It also shows a convention that we follow in this document. When
479 demonstrating instructions, we will follow an instruction with a comment that
480 defines the type and name of value produced. Comments are shown in italic
485 <!-- *********************************************************************** -->
486 <h2><a name="highlevel">High Level Structure</a></h2>
487 <!-- *********************************************************************** -->
489 <!-- ======================================================================= -->
491 <a name="modulestructure">Module Structure</a>
496 <p>LLVM programs are composed of <tt>Module</tt>s, each of which is a
497 translation unit of the input programs. Each module consists of functions,
498 global variables, and symbol table entries. Modules may be combined together
499 with the LLVM linker, which merges function (and global variable)
500 definitions, resolves forward declarations, and merges symbol table
501 entries. Here is an example of the "hello world" module:</p>
503 <pre class="doc_code">
504 <i>; Declare the string constant as a global constant.</i>
505 <a href="#identifiers">@.str</a> = <a href="#linkage_private">private</a> <a href="#globalvars">unnamed_addr</a> <a href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00"
507 <i>; External declaration of the puts function</i>
508 <a href="#functionstructure">declare</a> i32 @puts(i8* <a href="#nocapture">nocapture</a>) <a href="#fnattrs">nounwind</a>
510 <i>; Definition of main function</i>
511 define i32 @main() { <i>; i32()* </i>
512 <i>; Convert [13 x i8]* to i8 *...</i>
513 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.str, i64 0, i64 0
515 <i>; Call puts function to write out the string to stdout.</i>
516 <a href="#i_call">call</a> i32 @puts(i8* %cast210)
517 <a href="#i_ret">ret</a> i32 0
520 <i>; Named metadata</i>
521 !1 = metadata !{i32 42}
525 <p>This example is made up of a <a href="#globalvars">global variable</a> named
526 "<tt>.str</tt>", an external declaration of the "<tt>puts</tt>" function,
527 a <a href="#functionstructure">function definition</a> for
528 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
531 <p>In general, a module is made up of a list of global values (where both
532 functions and global variables are global values). Global values are
533 represented by a pointer to a memory location (in this case, a pointer to an
534 array of char, and a pointer to a function), and have one of the
535 following <a href="#linkage">linkage types</a>.</p>
539 <!-- ======================================================================= -->
541 <a name="linkage">Linkage Types</a>
546 <p>All Global Variables and Functions have one of the following types of
550 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
551 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
552 by objects in the current module. In particular, linking code into a
553 module with an private global value may cause the private to be renamed as
554 necessary to avoid collisions. Because the symbol is private to the
555 module, all references can be updated. This doesn't show up in any symbol
556 table in the object file.</dd>
558 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
559 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
560 assembler and evaluated by the linker. Unlike normal strong symbols, they
561 are removed by the linker from the final linked image (executable or
562 dynamic library).</dd>
564 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
565 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
566 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
567 linker. The symbols are removed by the linker from the final linked image
568 (executable or dynamic library).</dd>
570 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
571 <dd>Similar to "<tt>linker_private_weak</tt>", but it's known that the address
572 of the object is not taken. For instance, functions that had an inline
573 definition, but the compiler decided not to inline it. Note,
574 unlike <tt>linker_private</tt> and <tt>linker_private_weak</tt>,
575 <tt>linker_private_weak_def_auto</tt> may have only <tt>default</tt>
576 visibility. The symbols are removed by the linker from the final linked
577 image (executable or dynamic library).</dd>
579 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
580 <dd>Similar to private, but the value shows as a local symbol
581 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
582 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
584 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
585 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
586 into the object file corresponding to the LLVM module. They exist to
587 allow inlining and other optimizations to take place given knowledge of
588 the definition of the global, which is known to be somewhere outside the
589 module. Globals with <tt>available_externally</tt> linkage are allowed to
590 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
591 This linkage type is only allowed on definitions, not declarations.</dd>
593 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
594 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
595 the same name when linkage occurs. This can be used to implement
596 some forms of inline functions, templates, or other code which must be
597 generated in each translation unit that uses it, but where the body may
598 be overridden with a more definitive definition later. Unreferenced
599 <tt>linkonce</tt> globals are allowed to be discarded. Note that
600 <tt>linkonce</tt> linkage does not actually allow the optimizer to
601 inline the body of this function into callers because it doesn't know if
602 this definition of the function is the definitive definition within the
603 program or whether it will be overridden by a stronger definition.
604 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
607 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
608 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
609 <tt>linkonce</tt> linkage, except that unreferenced globals with
610 <tt>weak</tt> linkage may not be discarded. This is used for globals that
611 are declared "weak" in C source code.</dd>
613 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
614 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
615 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
617 Symbols with "<tt>common</tt>" linkage are merged in the same way as
618 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
619 <tt>common</tt> symbols may not have an explicit section,
620 must have a zero initializer, and may not be marked '<a
621 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
622 have common linkage.</dd>
625 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
626 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
627 pointer to array type. When two global variables with appending linkage
628 are linked together, the two global arrays are appended together. This is
629 the LLVM, typesafe, equivalent of having the system linker append together
630 "sections" with identical names when .o files are linked.</dd>
632 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
633 <dd>The semantics of this linkage follow the ELF object file model: the symbol
634 is weak until linked, if not linked, the symbol becomes null instead of
635 being an undefined reference.</dd>
637 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
638 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
639 <dd>Some languages allow differing globals to be merged, such as two functions
640 with different semantics. Other languages, such as <tt>C++</tt>, ensure
641 that only equivalent globals are ever merged (the "one definition rule"
642 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
643 and <tt>weak_odr</tt> linkage types to indicate that the global will only
644 be merged with equivalent globals. These linkage types are otherwise the
645 same as their non-<tt>odr</tt> versions.</dd>
647 <dt><tt><b><a name="linkage_external">external</a></b></tt></dt>
648 <dd>If none of the above identifiers are used, the global is externally
649 visible, meaning that it participates in linkage and can be used to
650 resolve external symbol references.</dd>
653 <p>The next two types of linkage are targeted for Microsoft Windows platform
654 only. They are designed to support importing (exporting) symbols from (to)
655 DLLs (Dynamic Link Libraries).</p>
658 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
659 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
660 or variable via a global pointer to a pointer that is set up by the DLL
661 exporting the symbol. On Microsoft Windows targets, the pointer name is
662 formed by combining <code>__imp_</code> and the function or variable
665 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
666 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
667 pointer to a pointer in a DLL, so that it can be referenced with the
668 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
669 name is formed by combining <code>__imp_</code> and the function or
673 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
674 another module defined a "<tt>.LC0</tt>" variable and was linked with this
675 one, one of the two would be renamed, preventing a collision. Since
676 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
677 declarations), they are accessible outside of the current module.</p>
679 <p>It is illegal for a function <i>declaration</i> to have any linkage type
680 other than <tt>external</tt>, <tt>dllimport</tt>
681 or <tt>extern_weak</tt>.</p>
683 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
684 or <tt>weak_odr</tt> linkages.</p>
688 <!-- ======================================================================= -->
690 <a name="callingconv">Calling Conventions</a>
695 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
696 and <a href="#i_invoke">invokes</a> can all have an optional calling
697 convention specified for the call. The calling convention of any pair of
698 dynamic caller/callee must match, or the behavior of the program is
699 undefined. The following calling conventions are supported by LLVM, and more
700 may be added in the future:</p>
703 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
704 <dd>This calling convention (the default if no other calling convention is
705 specified) matches the target C calling conventions. This calling
706 convention supports varargs function calls and tolerates some mismatch in
707 the declared prototype and implemented declaration of the function (as
710 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
711 <dd>This calling convention attempts to make calls as fast as possible
712 (e.g. by passing things in registers). This calling convention allows the
713 target to use whatever tricks it wants to produce fast code for the
714 target, without having to conform to an externally specified ABI
715 (Application Binary Interface).
716 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
717 when this or the GHC convention is used.</a> 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>coldcc</tt>" - The cold calling convention</b>:</dt>
722 <dd>This calling convention attempts to make code in the caller as efficient
723 as possible under the assumption that the call is not commonly executed.
724 As such, these calls often preserve all registers so that the call does
725 not break any live ranges in the caller side. This calling convention
726 does not support varargs and requires the prototype of all callees to
727 exactly match the prototype of the function definition.</dd>
729 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
730 <dd>This calling convention has been implemented specifically for use by the
731 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
732 It passes everything in registers, going to extremes to achieve this by
733 disabling callee save registers. This calling convention should not be
734 used lightly but only for specific situations such as an alternative to
735 the <em>register pinning</em> performance technique often used when
736 implementing functional programming languages.At the moment only X86
737 supports this convention and it has the following limitations:
739 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
740 floating point types are supported.</li>
741 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
742 6 floating point parameters.</li>
744 This calling convention supports
745 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
746 requires both the caller and callee are using it.
749 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
750 <dd>Any calling convention may be specified by number, allowing
751 target-specific calling conventions to be used. Target specific calling
752 conventions start at 64.</dd>
755 <p>More calling conventions can be added/defined on an as-needed basis, to
756 support Pascal conventions or any other well-known target-independent
761 <!-- ======================================================================= -->
763 <a name="visibility">Visibility Styles</a>
768 <p>All Global Variables and Functions have one of the following visibility
772 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
773 <dd>On targets that use the ELF object file format, default visibility means
774 that the declaration is visible to other modules and, in shared libraries,
775 means that the declared entity may be overridden. On Darwin, default
776 visibility means that the declaration is visible to other modules. Default
777 visibility corresponds to "external linkage" in the language.</dd>
779 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
780 <dd>Two declarations of an object with hidden visibility refer to the same
781 object if they are in the same shared object. Usually, hidden visibility
782 indicates that the symbol will not be placed into the dynamic symbol
783 table, so no other module (executable or shared library) can reference it
786 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
787 <dd>On ELF, protected visibility indicates that the symbol will be placed in
788 the dynamic symbol table, but that references within the defining module
789 will bind to the local symbol. That is, the symbol cannot be overridden by
795 <!-- ======================================================================= -->
797 <a name="namedtypes">Named Types</a>
802 <p>LLVM IR allows you to specify name aliases for certain types. This can make
803 it easier to read the IR and make the IR more condensed (particularly when
804 recursive types are involved). An example of a name specification is:</p>
806 <pre class="doc_code">
807 %mytype = type { %mytype*, i32 }
810 <p>You may give a name to any <a href="#typesystem">type</a> except
811 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
812 is expected with the syntax "%mytype".</p>
814 <p>Note that type names are aliases for the structural type that they indicate,
815 and that you can therefore specify multiple names for the same type. This
816 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
817 uses structural typing, the name is not part of the type. When printing out
818 LLVM IR, the printer will pick <em>one name</em> to render all types of a
819 particular shape. This means that if you have code where two different
820 source types end up having the same LLVM type, that the dumper will sometimes
821 print the "wrong" or unexpected type. This is an important design point and
822 isn't going to change.</p>
826 <!-- ======================================================================= -->
828 <a name="globalvars">Global Variables</a>
833 <p>Global variables define regions of memory allocated at compilation time
834 instead of run-time. Global variables may optionally be initialized, may
835 have an explicit section to be placed in, and may have an optional explicit
836 alignment specified. A variable may be defined as "thread_local", which
837 means that it will not be shared by threads (each thread will have a
838 separated copy of the variable). A variable may be defined as a global
839 "constant," which indicates that the contents of the variable
840 will <b>never</b> be modified (enabling better optimization, allowing the
841 global data to be placed in the read-only section of an executable, etc).
842 Note that variables that need runtime initialization cannot be marked
843 "constant" as there is a store to the variable.</p>
845 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
846 constant, even if the final definition of the global is not. This capability
847 can be used to enable slightly better optimization of the program, but
848 requires the language definition to guarantee that optimizations based on the
849 'constantness' are valid for the translation units that do not include the
852 <p>As SSA values, global variables define pointer values that are in scope
853 (i.e. they dominate) all basic blocks in the program. Global variables
854 always define a pointer to their "content" type because they describe a
855 region of memory, and all memory objects in LLVM are accessed through
858 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
859 that the address is not significant, only the content. Constants marked
860 like this can be merged with other constants if they have the same
861 initializer. Note that a constant with significant address <em>can</em>
862 be merged with a <tt>unnamed_addr</tt> constant, the result being a
863 constant whose address is significant.</p>
865 <p>A global variable may be declared to reside in a target-specific numbered
866 address space. For targets that support them, address spaces may affect how
867 optimizations are performed and/or what target instructions are used to
868 access the variable. The default address space is zero. The address space
869 qualifier must precede any other attributes.</p>
871 <p>LLVM allows an explicit section to be specified for globals. If the target
872 supports it, it will emit globals to the section specified.</p>
874 <p>An explicit alignment may be specified for a global, which must be a power
875 of 2. If not present, or if the alignment is set to zero, the alignment of
876 the global is set by the target to whatever it feels convenient. If an
877 explicit alignment is specified, the global is forced to have exactly that
878 alignment. Targets and optimizers are not allowed to over-align the global
879 if the global has an assigned section. In this case, the extra alignment
880 could be observable: for example, code could assume that the globals are
881 densely packed in their section and try to iterate over them as an array,
882 alignment padding would break this iteration.</p>
884 <p>For example, the following defines a global in a numbered address space with
885 an initializer, section, and alignment:</p>
887 <pre class="doc_code">
888 @G = addrspace(5) constant float 1.0, section "foo", align 4
894 <!-- ======================================================================= -->
896 <a name="functionstructure">Functions</a>
901 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
902 optional <a href="#linkage">linkage type</a>, an optional
903 <a href="#visibility">visibility style</a>, an optional
904 <a href="#callingconv">calling convention</a>,
905 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
906 <a href="#paramattrs">parameter attribute</a> for the return type, a function
907 name, a (possibly empty) argument list (each with optional
908 <a href="#paramattrs">parameter attributes</a>), optional
909 <a href="#fnattrs">function attributes</a>, an optional section, an optional
910 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
911 curly brace, a list of basic blocks, and a closing curly brace.</p>
913 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
914 optional <a href="#linkage">linkage type</a>, an optional
915 <a href="#visibility">visibility style</a>, an optional
916 <a href="#callingconv">calling convention</a>,
917 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
918 <a href="#paramattrs">parameter attribute</a> for the return type, a function
919 name, a possibly empty list of arguments, an optional alignment, and an
920 optional <a href="#gc">garbage collector name</a>.</p>
922 <p>A function definition contains a list of basic blocks, forming the CFG
923 (Control Flow Graph) for the function. Each basic block may optionally start
924 with a label (giving the basic block a symbol table entry), contains a list
925 of instructions, and ends with a <a href="#terminators">terminator</a>
926 instruction (such as a branch or function return).</p>
928 <p>The first basic block in a function is special in two ways: it is immediately
929 executed on entrance to the function, and it is not allowed to have
930 predecessor basic blocks (i.e. there can not be any branches to the entry
931 block of a function). Because the block can have no predecessors, it also
932 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
934 <p>LLVM allows an explicit section to be specified for functions. If the target
935 supports it, it will emit functions to the section specified.</p>
937 <p>An explicit alignment may be specified for a function. If not present, or if
938 the alignment is set to zero, the alignment of the function is set by the
939 target to whatever it feels convenient. If an explicit alignment is
940 specified, the function is forced to have at least that much alignment. All
941 alignments must be a power of 2.</p>
943 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
944 be significant and two identical functions can be merged.</p>
947 <pre class="doc_code">
948 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
949 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
950 <ResultType> @<FunctionName> ([argument list])
951 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
952 [<a href="#gc">gc</a>] { ... }
957 <!-- ======================================================================= -->
959 <a name="aliasstructure">Aliases</a>
964 <p>Aliases act as "second name" for the aliasee value (which can be either
965 function, global variable, another alias or bitcast of global value). Aliases
966 may have an optional <a href="#linkage">linkage type</a>, and an
967 optional <a href="#visibility">visibility style</a>.</p>
970 <pre class="doc_code">
971 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
976 <!-- ======================================================================= -->
978 <a name="namedmetadatastructure">Named Metadata</a>
983 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
984 nodes</a> (but not metadata strings) are the only valid operands for
985 a named metadata.</p>
988 <pre class="doc_code">
989 ; Some unnamed metadata nodes, which are referenced by the named metadata.
990 !0 = metadata !{metadata !"zero"}
991 !1 = metadata !{metadata !"one"}
992 !2 = metadata !{metadata !"two"}
994 !name = !{!0, !1, !2}
999 <!-- ======================================================================= -->
1001 <a name="paramattrs">Parameter Attributes</a>
1006 <p>The return type and each parameter of a function type may have a set of
1007 <i>parameter attributes</i> associated with them. Parameter attributes are
1008 used to communicate additional information about the result or parameters of
1009 a function. Parameter attributes are considered to be part of the function,
1010 not of the function type, so functions with different parameter attributes
1011 can have the same function type.</p>
1013 <p>Parameter attributes are simple keywords that follow the type specified. If
1014 multiple parameter attributes are needed, they are space separated. For
1017 <pre class="doc_code">
1018 declare i32 @printf(i8* noalias nocapture, ...)
1019 declare i32 @atoi(i8 zeroext)
1020 declare signext i8 @returns_signed_char()
1023 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1024 <tt>readonly</tt>) come immediately after the argument list.</p>
1026 <p>Currently, only the following parameter attributes are defined:</p>
1029 <dt><tt><b>zeroext</b></tt></dt>
1030 <dd>This indicates to the code generator that the parameter or return value
1031 should be zero-extended to the extent required by the target's ABI (which
1032 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1033 parameter) or the callee (for a return value).</dd>
1035 <dt><tt><b>signext</b></tt></dt>
1036 <dd>This indicates to the code generator that the parameter or return value
1037 should be sign-extended to the extent required by the target's ABI (which
1038 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1041 <dt><tt><b>inreg</b></tt></dt>
1042 <dd>This indicates that this parameter or return value should be treated in a
1043 special target-dependent fashion during while emitting code for a function
1044 call or return (usually, by putting it in a register as opposed to memory,
1045 though some targets use it to distinguish between two different kinds of
1046 registers). Use of this attribute is target-specific.</dd>
1048 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1049 <dd><p>This indicates that the pointer parameter should really be passed by
1050 value to the function. The attribute implies that a hidden copy of the
1052 is made between the caller and the callee, so the callee is unable to
1053 modify the value in the callee. This attribute is only valid on LLVM
1054 pointer arguments. It is generally used to pass structs and arrays by
1055 value, but is also valid on pointers to scalars. The copy is considered
1056 to belong to the caller not the callee (for example,
1057 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1058 <tt>byval</tt> parameters). This is not a valid attribute for return
1061 <p>The byval attribute also supports specifying an alignment with
1062 the align attribute. It indicates the alignment of the stack slot to
1063 form and the known alignment of the pointer specified to the call site. If
1064 the alignment is not specified, then the code generator makes a
1065 target-specific assumption.</p></dd>
1067 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1068 <dd>This indicates that the pointer parameter specifies the address of a
1069 structure that is the return value of the function in the source program.
1070 This pointer must be guaranteed by the caller to be valid: loads and
1071 stores to the structure may be assumed by the callee to not to trap. This
1072 may only be applied to the first parameter. This is not a valid attribute
1073 for return values. </dd>
1075 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1076 <dd>This indicates that pointer values
1077 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1078 value do not alias pointer values which are not <i>based</i> on it,
1079 ignoring certain "irrelevant" dependencies.
1080 For a call to the parent function, dependencies between memory
1081 references from before or after the call and from those during the call
1082 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1083 return value used in that call.
1084 The caller shares the responsibility with the callee for ensuring that
1085 these requirements are met.
1086 For further details, please see the discussion of the NoAlias response in
1087 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1089 Note that this definition of <tt>noalias</tt> is intentionally
1090 similar to the definition of <tt>restrict</tt> in C99 for function
1091 arguments, though it is slightly weaker.
1093 For function return values, C99's <tt>restrict</tt> is not meaningful,
1094 while LLVM's <tt>noalias</tt> is.
1097 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1098 <dd>This indicates that the callee does not make any copies of the pointer
1099 that outlive the callee itself. This is not a valid attribute for return
1102 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1103 <dd>This indicates that the pointer parameter can be excised using the
1104 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1105 attribute for return values.</dd>
1110 <!-- ======================================================================= -->
1112 <a name="gc">Garbage Collector Names</a>
1117 <p>Each function may specify a garbage collector name, which is simply a
1120 <pre class="doc_code">
1121 define void @f() gc "name" { ... }
1124 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1125 collector which will cause the compiler to alter its output in order to
1126 support the named garbage collection algorithm.</p>
1130 <!-- ======================================================================= -->
1132 <a name="fnattrs">Function Attributes</a>
1137 <p>Function attributes are set to communicate additional information about a
1138 function. Function attributes are considered to be part of the function, not
1139 of the function type, so functions with different parameter attributes can
1140 have the same function type.</p>
1142 <p>Function attributes are simple keywords that follow the type specified. If
1143 multiple attributes are needed, they are space separated. For example:</p>
1145 <pre class="doc_code">
1146 define void @f() noinline { ... }
1147 define void @f() alwaysinline { ... }
1148 define void @f() alwaysinline optsize { ... }
1149 define void @f() optsize { ... }
1153 <dt><tt><b>address_safety</b></tt></dt>
1154 <dd>This attribute indicates that the address safety analysis
1155 is enabled for this function. </dd>
1157 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1158 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1159 the backend should forcibly align the stack pointer. Specify the
1160 desired alignment, which must be a power of two, in parentheses.
1162 <dt><tt><b>alwaysinline</b></tt></dt>
1163 <dd>This attribute indicates that the inliner should attempt to inline this
1164 function into callers whenever possible, ignoring any active inlining size
1165 threshold for this caller.</dd>
1167 <dt><tt><b>nonlazybind</b></tt></dt>
1168 <dd>This attribute suppresses lazy symbol binding for the function. This
1169 may make calls to the function faster, at the cost of extra program
1170 startup time if the function is not called during program startup.</dd>
1172 <dt><tt><b>inlinehint</b></tt></dt>
1173 <dd>This attribute indicates that the source code contained a hint that inlining
1174 this function is desirable (such as the "inline" keyword in C/C++). It
1175 is just a hint; it imposes no requirements on the inliner.</dd>
1177 <dt><tt><b>naked</b></tt></dt>
1178 <dd>This attribute disables prologue / epilogue emission for the function.
1179 This can have very system-specific consequences.</dd>
1181 <dt><tt><b>noimplicitfloat</b></tt></dt>
1182 <dd>This attributes disables implicit floating point instructions.</dd>
1184 <dt><tt><b>noinline</b></tt></dt>
1185 <dd>This attribute indicates that the inliner should never inline this
1186 function in any situation. This attribute may not be used together with
1187 the <tt>alwaysinline</tt> attribute.</dd>
1189 <dt><tt><b>noredzone</b></tt></dt>
1190 <dd>This attribute indicates that the code generator should not use a red
1191 zone, even if the target-specific ABI normally permits it.</dd>
1193 <dt><tt><b>noreturn</b></tt></dt>
1194 <dd>This function attribute indicates that the function never returns
1195 normally. This produces undefined behavior at runtime if the function
1196 ever does dynamically return.</dd>
1198 <dt><tt><b>nounwind</b></tt></dt>
1199 <dd>This function attribute indicates that the function never returns with an
1200 unwind or exceptional control flow. If the function does unwind, its
1201 runtime behavior is undefined.</dd>
1203 <dt><tt><b>optsize</b></tt></dt>
1204 <dd>This attribute suggests that optimization passes and code generator passes
1205 make choices that keep the code size of this function low, and otherwise
1206 do optimizations specifically to reduce code size.</dd>
1208 <dt><tt><b>readnone</b></tt></dt>
1209 <dd>This attribute indicates that the function computes its result (or decides
1210 to unwind an exception) based strictly on its arguments, without
1211 dereferencing any pointer arguments or otherwise accessing any mutable
1212 state (e.g. memory, control registers, etc) visible to caller functions.
1213 It does not write through any pointer arguments
1214 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1215 changes any state visible to callers. This means that it cannot unwind
1216 exceptions by calling the <tt>C++</tt> exception throwing methods.</dd>
1218 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1219 <dd>This attribute indicates that the function does not write through any
1220 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1221 arguments) or otherwise modify any state (e.g. memory, control registers,
1222 etc) visible to caller functions. It may dereference pointer arguments
1223 and read state that may be set in the caller. A readonly function always
1224 returns the same value (or unwinds an exception identically) when called
1225 with the same set of arguments and global state. It cannot unwind an
1226 exception by calling the <tt>C++</tt> exception throwing methods.</dd>
1228 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1229 <dd>This attribute indicates that this function can return twice. The
1230 C <code>setjmp</code> is an example of such a function. The compiler
1231 disables some optimizations (like tail calls) in the caller of these
1234 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1235 <dd>This attribute indicates that the function should emit a stack smashing
1236 protector. It is in the form of a "canary"—a random value placed on
1237 the stack before the local variables that's checked upon return from the
1238 function to see if it has been overwritten. A heuristic is used to
1239 determine if a function needs stack protectors or not.<br>
1241 If a function that has an <tt>ssp</tt> attribute is inlined into a
1242 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1243 function will have an <tt>ssp</tt> attribute.</dd>
1245 <dt><tt><b>sspreq</b></tt></dt>
1246 <dd>This attribute indicates that the function should <em>always</em> emit a
1247 stack smashing protector. This overrides
1248 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1250 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1251 function that doesn't have an <tt>sspreq</tt> attribute or which has
1252 an <tt>ssp</tt> attribute, then the resulting function will have
1253 an <tt>sspreq</tt> attribute.</dd>
1255 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1256 <dd>This attribute indicates that the ABI being targeted requires that
1257 an unwind table entry be produce for this function even if we can
1258 show that no exceptions passes by it. This is normally the case for
1259 the ELF x86-64 abi, but it can be disabled for some compilation
1265 <!-- ======================================================================= -->
1267 <a name="moduleasm">Module-Level Inline Assembly</a>
1272 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1273 the GCC "file scope inline asm" blocks. These blocks are internally
1274 concatenated by LLVM and treated as a single unit, but may be separated in
1275 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1277 <pre class="doc_code">
1278 module asm "inline asm code goes here"
1279 module asm "more can go here"
1282 <p>The strings can contain any character by escaping non-printable characters.
1283 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1286 <p>The inline asm code is simply printed to the machine code .s file when
1287 assembly code is generated.</p>
1291 <!-- ======================================================================= -->
1293 <a name="datalayout">Data Layout</a>
1298 <p>A module may specify a target specific data layout string that specifies how
1299 data is to be laid out in memory. The syntax for the data layout is
1302 <pre class="doc_code">
1303 target datalayout = "<i>layout specification</i>"
1306 <p>The <i>layout specification</i> consists of a list of specifications
1307 separated by the minus sign character ('-'). Each specification starts with
1308 a letter and may include other information after the letter to define some
1309 aspect of the data layout. The specifications accepted are as follows:</p>
1313 <dd>Specifies that the target lays out data in big-endian form. That is, the
1314 bits with the most significance have the lowest address location.</dd>
1317 <dd>Specifies that the target lays out data in little-endian form. That is,
1318 the bits with the least significance have the lowest address
1321 <dt><tt>S<i>size</i></tt></dt>
1322 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1323 of stack variables is limited to the natural stack alignment to avoid
1324 dynamic stack realignment. The stack alignment must be a multiple of
1325 8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1326 which does not prevent any alignment promotions.</dd>
1328 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1329 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1330 <i>preferred</i> alignments. All sizes are in bits. Specifying
1331 the <i>pref</i> alignment is optional. If omitted, the
1332 preceding <tt>:</tt> should be omitted too.</dd>
1334 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1335 <dd>This specifies the alignment for an integer type of a given bit
1336 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1338 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1339 <dd>This specifies the alignment for a vector type of a given bit
1342 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1343 <dd>This specifies the alignment for a floating point type of a given bit
1344 <i>size</i>. Only values of <i>size</i> that are supported by the target
1345 will work. 32 (float) and 64 (double) are supported on all targets;
1346 80 or 128 (different flavors of long double) are also supported on some
1349 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1350 <dd>This specifies the alignment for an aggregate type of a given bit
1353 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1354 <dd>This specifies the alignment for a stack object of a given bit
1357 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1358 <dd>This specifies a set of native integer widths for the target CPU
1359 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1360 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1361 this set are considered to support most general arithmetic
1362 operations efficiently.</dd>
1365 <p>When constructing the data layout for a given target, LLVM starts with a
1366 default set of specifications which are then (possibly) overridden by the
1367 specifications in the <tt>datalayout</tt> keyword. The default specifications
1368 are given in this list:</p>
1371 <li><tt>E</tt> - big endian</li>
1372 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1373 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1374 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1375 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1376 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1377 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1378 alignment of 64-bits</li>
1379 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1380 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1381 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1382 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1383 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1384 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1387 <p>When LLVM is determining the alignment for a given type, it uses the
1388 following rules:</p>
1391 <li>If the type sought is an exact match for one of the specifications, that
1392 specification is used.</li>
1394 <li>If no match is found, and the type sought is an integer type, then the
1395 smallest integer type that is larger than the bitwidth of the sought type
1396 is used. If none of the specifications are larger than the bitwidth then
1397 the the largest integer type is used. For example, given the default
1398 specifications above, the i7 type will use the alignment of i8 (next
1399 largest) while both i65 and i256 will use the alignment of i64 (largest
1402 <li>If no match is found, and the type sought is a vector type, then the
1403 largest vector type that is smaller than the sought vector type will be
1404 used as a fall back. This happens because <128 x double> can be
1405 implemented in terms of 64 <2 x double>, for example.</li>
1408 <p>The function of the data layout string may not be what you expect. Notably,
1409 this is not a specification from the frontend of what alignment the code
1410 generator should use.</p>
1412 <p>Instead, if specified, the target data layout is required to match what the
1413 ultimate <em>code generator</em> expects. This string is used by the
1414 mid-level optimizers to
1415 improve code, and this only works if it matches what the ultimate code
1416 generator uses. If you would like to generate IR that does not embed this
1417 target-specific detail into the IR, then you don't have to specify the
1418 string. This will disable some optimizations that require precise layout
1419 information, but this also prevents those optimizations from introducing
1420 target specificity into the IR.</p>
1426 <!-- ======================================================================= -->
1428 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1433 <p>Any memory access must be done through a pointer value associated
1434 with an address range of the memory access, otherwise the behavior
1435 is undefined. Pointer values are associated with address ranges
1436 according to the following rules:</p>
1439 <li>A pointer value is associated with the addresses associated with
1440 any value it is <i>based</i> on.
1441 <li>An address of a global variable is associated with the address
1442 range of the variable's storage.</li>
1443 <li>The result value of an allocation instruction is associated with
1444 the address range of the allocated storage.</li>
1445 <li>A null pointer in the default address-space is associated with
1447 <li>An integer constant other than zero or a pointer value returned
1448 from a function not defined within LLVM may be associated with address
1449 ranges allocated through mechanisms other than those provided by
1450 LLVM. Such ranges shall not overlap with any ranges of addresses
1451 allocated by mechanisms provided by LLVM.</li>
1454 <p>A pointer value is <i>based</i> on another pointer value according
1455 to the following rules:</p>
1458 <li>A pointer value formed from a
1459 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1460 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1461 <li>The result value of a
1462 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1463 of the <tt>bitcast</tt>.</li>
1464 <li>A pointer value formed by an
1465 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1466 pointer values that contribute (directly or indirectly) to the
1467 computation of the pointer's value.</li>
1468 <li>The "<i>based</i> on" relationship is transitive.</li>
1471 <p>Note that this definition of <i>"based"</i> is intentionally
1472 similar to the definition of <i>"based"</i> in C99, though it is
1473 slightly weaker.</p>
1475 <p>LLVM IR does not associate types with memory. The result type of a
1476 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1477 alignment of the memory from which to load, as well as the
1478 interpretation of the value. The first operand type of a
1479 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1480 and alignment of the store.</p>
1482 <p>Consequently, type-based alias analysis, aka TBAA, aka
1483 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1484 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1485 additional information which specialized optimization passes may use
1486 to implement type-based alias analysis.</p>
1490 <!-- ======================================================================= -->
1492 <a name="volatile">Volatile Memory Accesses</a>
1497 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1498 href="#i_store"><tt>store</tt></a>s, and <a
1499 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1500 The optimizers must not change the number of volatile operations or change their
1501 order of execution relative to other volatile operations. The optimizers
1502 <i>may</i> change the order of volatile operations relative to non-volatile
1503 operations. This is not Java's "volatile" and has no cross-thread
1504 synchronization behavior.</p>
1508 <!-- ======================================================================= -->
1510 <a name="memmodel">Memory Model for Concurrent Operations</a>
1515 <p>The LLVM IR does not define any way to start parallel threads of execution
1516 or to register signal handlers. Nonetheless, there are platform-specific
1517 ways to create them, and we define LLVM IR's behavior in their presence. This
1518 model is inspired by the C++0x memory model.</p>
1520 <p>For a more informal introduction to this model, see the
1521 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1523 <p>We define a <i>happens-before</i> partial order as the least partial order
1526 <li>Is a superset of single-thread program order, and</li>
1527 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1528 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1529 by platform-specific techniques, like pthread locks, thread
1530 creation, thread joining, etc., and by atomic instructions.
1531 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1535 <p>Note that program order does not introduce <i>happens-before</i> edges
1536 between a thread and signals executing inside that thread.</p>
1538 <p>Every (defined) read operation (load instructions, memcpy, atomic
1539 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1540 (defined) write operations (store instructions, atomic
1541 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1542 initialized globals are considered to have a write of the initializer which is
1543 atomic and happens before any other read or write of the memory in question.
1544 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1545 any write to the same byte, except:</p>
1548 <li>If <var>write<sub>1</sub></var> happens before
1549 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1550 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1551 does not see <var>write<sub>1</sub></var>.
1552 <li>If <var>R<sub>byte</sub></var> happens before
1553 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1554 see <var>write<sub>3</sub></var>.
1557 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1559 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1560 is supposed to give guarantees which can support
1561 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1562 addresses which do not behave like normal memory. It does not generally
1563 provide cross-thread synchronization.)
1564 <li>Otherwise, if there is no write to the same byte that happens before
1565 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1566 <tt>undef</tt> for that byte.
1567 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1568 <var>R<sub>byte</sub></var> returns the value written by that
1570 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1571 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1572 values written. See the <a href="#ordering">Atomic Memory Ordering
1573 Constraints</a> section for additional constraints on how the choice
1575 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1578 <p><var>R</var> returns the value composed of the series of bytes it read.
1579 This implies that some bytes within the value may be <tt>undef</tt>
1580 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1581 defines the semantics of the operation; it doesn't mean that targets will
1582 emit more than one instruction to read the series of bytes.</p>
1584 <p>Note that in cases where none of the atomic intrinsics are used, this model
1585 places only one restriction on IR transformations on top of what is required
1586 for single-threaded execution: introducing a store to a byte which might not
1587 otherwise be stored is not allowed in general. (Specifically, in the case
1588 where another thread might write to and read from an address, introducing a
1589 store can change a load that may see exactly one write into a load that may
1590 see multiple writes.)</p>
1592 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1593 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1594 none of the backends currently in the tree fall into this category; however,
1595 there might be targets which care. If there are, we want a paragraph
1598 Targets may specify that stores narrower than a certain width are not
1599 available; on such a target, for the purposes of this model, treat any
1600 non-atomic write with an alignment or width less than the minimum width
1601 as if it writes to the relevant surrounding bytes.
1606 <!-- ======================================================================= -->
1608 <a name="ordering">Atomic Memory Ordering Constraints</a>
1613 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1614 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1615 <a href="#i_fence"><code>fence</code></a>,
1616 <a href="#i_load"><code>atomic load</code></a>, and
1617 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1618 that determines which other atomic instructions on the same address they
1619 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1620 but are somewhat more colloquial. If these descriptions aren't precise enough,
1621 check those specs (see spec references in the
1622 <a href="Atomics.html#introduction">atomics guide</a>).
1623 <a href="#i_fence"><code>fence</code></a> instructions
1624 treat these orderings somewhat differently since they don't take an address.
1625 See that instruction's documentation for details.</p>
1627 <p>For a simpler introduction to the ordering constraints, see the
1628 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1631 <dt><code>unordered</code></dt>
1632 <dd>The set of values that can be read is governed by the happens-before
1633 partial order. A value cannot be read unless some operation wrote it.
1634 This is intended to provide a guarantee strong enough to model Java's
1635 non-volatile shared variables. This ordering cannot be specified for
1636 read-modify-write operations; it is not strong enough to make them atomic
1637 in any interesting way.</dd>
1638 <dt><code>monotonic</code></dt>
1639 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1640 total order for modifications by <code>monotonic</code> operations on each
1641 address. All modification orders must be compatible with the happens-before
1642 order. There is no guarantee that the modification orders can be combined to
1643 a global total order for the whole program (and this often will not be
1644 possible). The read in an atomic read-modify-write operation
1645 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1646 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1647 reads the value in the modification order immediately before the value it
1648 writes. If one atomic read happens before another atomic read of the same
1649 address, the later read must see the same value or a later value in the
1650 address's modification order. This disallows reordering of
1651 <code>monotonic</code> (or stronger) operations on the same address. If an
1652 address is written <code>monotonic</code>ally by one thread, and other threads
1653 <code>monotonic</code>ally read that address repeatedly, the other threads must
1654 eventually see the write. This corresponds to the C++0x/C1x
1655 <code>memory_order_relaxed</code>.</dd>
1656 <dt><code>acquire</code></dt>
1657 <dd>In addition to the guarantees of <code>monotonic</code>,
1658 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1659 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1660 <dt><code>release</code></dt>
1661 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1662 writes a value which is subsequently read by an <code>acquire</code> operation,
1663 it <i>synchronizes-with</i> that operation. (This isn't a complete
1664 description; see the C++0x definition of a release sequence.) This corresponds
1665 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1666 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1667 <code>acquire</code> and <code>release</code> operation on its address.
1668 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1669 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1670 <dd>In addition to the guarantees of <code>acq_rel</code>
1671 (<code>acquire</code> for an operation which only reads, <code>release</code>
1672 for an operation which only writes), there is a global total order on all
1673 sequentially-consistent operations on all addresses, which is consistent with
1674 the <i>happens-before</i> partial order and with the modification orders of
1675 all the affected addresses. Each sequentially-consistent read sees the last
1676 preceding write to the same address in this global order. This corresponds
1677 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1680 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1681 it only <i>synchronizes with</i> or participates in modification and seq_cst
1682 total orderings with other operations running in the same thread (for example,
1683 in signal handlers).</p>
1689 <!-- *********************************************************************** -->
1690 <h2><a name="typesystem">Type System</a></h2>
1691 <!-- *********************************************************************** -->
1695 <p>The LLVM type system is one of the most important features of the
1696 intermediate representation. Being typed enables a number of optimizations
1697 to be performed on the intermediate representation directly, without having
1698 to do extra analyses on the side before the transformation. A strong type
1699 system makes it easier to read the generated code and enables novel analyses
1700 and transformations that are not feasible to perform on normal three address
1701 code representations.</p>
1703 <!-- ======================================================================= -->
1705 <a name="t_classifications">Type Classifications</a>
1710 <p>The types fall into a few useful classifications:</p>
1712 <table border="1" cellspacing="0" cellpadding="4">
1714 <tr><th>Classification</th><th>Types</th></tr>
1716 <td><a href="#t_integer">integer</a></td>
1717 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1720 <td><a href="#t_floating">floating point</a></td>
1721 <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1724 <td><a name="t_firstclass">first class</a></td>
1725 <td><a href="#t_integer">integer</a>,
1726 <a href="#t_floating">floating point</a>,
1727 <a href="#t_pointer">pointer</a>,
1728 <a href="#t_vector">vector</a>,
1729 <a href="#t_struct">structure</a>,
1730 <a href="#t_array">array</a>,
1731 <a href="#t_label">label</a>,
1732 <a href="#t_metadata">metadata</a>.
1736 <td><a href="#t_primitive">primitive</a></td>
1737 <td><a href="#t_label">label</a>,
1738 <a href="#t_void">void</a>,
1739 <a href="#t_integer">integer</a>,
1740 <a href="#t_floating">floating point</a>,
1741 <a href="#t_x86mmx">x86mmx</a>,
1742 <a href="#t_metadata">metadata</a>.</td>
1745 <td><a href="#t_derived">derived</a></td>
1746 <td><a href="#t_array">array</a>,
1747 <a href="#t_function">function</a>,
1748 <a href="#t_pointer">pointer</a>,
1749 <a href="#t_struct">structure</a>,
1750 <a href="#t_vector">vector</a>,
1751 <a href="#t_opaque">opaque</a>.
1757 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1758 important. Values of these types are the only ones which can be produced by
1763 <!-- ======================================================================= -->
1765 <a name="t_primitive">Primitive Types</a>
1770 <p>The primitive types are the fundamental building blocks of the LLVM
1773 <!-- _______________________________________________________________________ -->
1775 <a name="t_integer">Integer Type</a>
1781 <p>The integer type is a very simple type that simply specifies an arbitrary
1782 bit width for the integer type desired. Any bit width from 1 bit to
1783 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1790 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1794 <table class="layout">
1796 <td class="left"><tt>i1</tt></td>
1797 <td class="left">a single-bit integer.</td>
1800 <td class="left"><tt>i32</tt></td>
1801 <td class="left">a 32-bit integer.</td>
1804 <td class="left"><tt>i1942652</tt></td>
1805 <td class="left">a really big integer of over 1 million bits.</td>
1811 <!-- _______________________________________________________________________ -->
1813 <a name="t_floating">Floating Point Types</a>
1820 <tr><th>Type</th><th>Description</th></tr>
1821 <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr>
1822 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1823 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1824 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1825 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1826 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1832 <!-- _______________________________________________________________________ -->
1834 <a name="t_x86mmx">X86mmx Type</a>
1840 <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>
1849 <!-- _______________________________________________________________________ -->
1851 <a name="t_void">Void Type</a>
1857 <p>The void type does not represent any value and has no size.</p>
1866 <!-- _______________________________________________________________________ -->
1868 <a name="t_label">Label Type</a>
1874 <p>The label type represents code labels.</p>
1883 <!-- _______________________________________________________________________ -->
1885 <a name="t_metadata">Metadata Type</a>
1891 <p>The metadata type represents embedded metadata. No derived types may be
1892 created from metadata except for <a href="#t_function">function</a>
1904 <!-- ======================================================================= -->
1906 <a name="t_derived">Derived Types</a>
1911 <p>The real power in LLVM comes from the derived types in the system. This is
1912 what allows a programmer to represent arrays, functions, pointers, and other
1913 useful types. Each of these types contain one or more element types which
1914 may be a primitive type, or another derived type. For example, it is
1915 possible to have a two dimensional array, using an array as the element type
1916 of another array.</p>
1918 <!-- _______________________________________________________________________ -->
1920 <a name="t_aggregate">Aggregate Types</a>
1925 <p>Aggregate Types are a subset of derived types that can contain multiple
1926 member types. <a href="#t_array">Arrays</a> and
1927 <a href="#t_struct">structs</a> are aggregate types.
1928 <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p>
1932 <!-- _______________________________________________________________________ -->
1934 <a name="t_array">Array Type</a>
1940 <p>The array type is a very simple derived type that arranges elements
1941 sequentially in memory. The array type requires a size (number of elements)
1942 and an underlying data type.</p>
1946 [<# elements> x <elementtype>]
1949 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1950 be any type with a size.</p>
1953 <table class="layout">
1955 <td class="left"><tt>[40 x i32]</tt></td>
1956 <td class="left">Array of 40 32-bit integer values.</td>
1959 <td class="left"><tt>[41 x i32]</tt></td>
1960 <td class="left">Array of 41 32-bit integer values.</td>
1963 <td class="left"><tt>[4 x i8]</tt></td>
1964 <td class="left">Array of 4 8-bit integer values.</td>
1967 <p>Here are some examples of multidimensional arrays:</p>
1968 <table class="layout">
1970 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1971 <td class="left">3x4 array of 32-bit integer values.</td>
1974 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1975 <td class="left">12x10 array of single precision floating point values.</td>
1978 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1979 <td class="left">2x3x4 array of 16-bit integer values.</td>
1983 <p>There is no restriction on indexing beyond the end of the array implied by
1984 a static type (though there are restrictions on indexing beyond the bounds
1985 of an allocated object in some cases). This means that single-dimension
1986 'variable sized array' addressing can be implemented in LLVM with a zero
1987 length array type. An implementation of 'pascal style arrays' in LLVM could
1988 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1992 <!-- _______________________________________________________________________ -->
1994 <a name="t_function">Function Type</a>
2000 <p>The function type can be thought of as a function signature. It consists of
2001 a return type and a list of formal parameter types. The return type of a
2002 function type is a first class type or a void type.</p>
2006 <returntype> (<parameter list>)
2009 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
2010 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
2011 which indicates that the function takes a variable number of arguments.
2012 Variable argument functions can access their arguments with
2013 the <a href="#int_varargs">variable argument handling intrinsic</a>
2014 functions. '<tt><returntype></tt>' is any type except
2015 <a href="#t_label">label</a>.</p>
2018 <table class="layout">
2020 <td class="left"><tt>i32 (i32)</tt></td>
2021 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2023 </tr><tr class="layout">
2024 <td class="left"><tt>float (i16, i32 *) *
2026 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2027 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2028 returning <tt>float</tt>.
2030 </tr><tr class="layout">
2031 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2032 <td class="left">A vararg function that takes at least one
2033 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2034 which returns an integer. This is the signature for <tt>printf</tt> in
2037 </tr><tr class="layout">
2038 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2039 <td class="left">A function taking an <tt>i32</tt>, returning a
2040 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2047 <!-- _______________________________________________________________________ -->
2049 <a name="t_struct">Structure Type</a>
2055 <p>The structure type is used to represent a collection of data members together
2056 in memory. The elements of a structure may be any type that has a size.</p>
2058 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2059 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2060 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2061 Structures in registers are accessed using the
2062 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2063 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2065 <p>Structures may optionally be "packed" structures, which indicate that the
2066 alignment of the struct is one byte, and that there is no padding between
2067 the elements. In non-packed structs, padding between field types is inserted
2068 as defined by the TargetData string in the module, which is required to match
2069 what the underlying code generator expects.</p>
2071 <p>Structures can either be "literal" or "identified". A literal structure is
2072 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2073 types are always defined at the top level with a name. Literal types are
2074 uniqued by their contents and can never be recursive or opaque since there is
2075 no way to write one. Identified types can be recursive, can be opaqued, and are
2081 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2082 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2086 <table class="layout">
2088 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2089 <td class="left">A triple of three <tt>i32</tt> values</td>
2092 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2093 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2094 second element is a <a href="#t_pointer">pointer</a> to a
2095 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2096 an <tt>i32</tt>.</td>
2099 <td class="left"><tt><{ i8, i32 }></tt></td>
2100 <td class="left">A packed struct known to be 5 bytes in size.</td>
2106 <!-- _______________________________________________________________________ -->
2108 <a name="t_opaque">Opaque Structure Types</a>
2114 <p>Opaque structure types are used to represent named structure types that do
2115 not have a body specified. This corresponds (for example) to the C notion of
2116 a forward declared structure.</p>
2125 <table class="layout">
2127 <td class="left"><tt>opaque</tt></td>
2128 <td class="left">An opaque type.</td>
2136 <!-- _______________________________________________________________________ -->
2138 <a name="t_pointer">Pointer Type</a>
2144 <p>The pointer type is used to specify memory locations.
2145 Pointers are commonly used to reference objects in memory.</p>
2147 <p>Pointer types may have an optional address space attribute defining the
2148 numbered address space where the pointed-to object resides. The default
2149 address space is number zero. The semantics of non-zero address
2150 spaces are target-specific.</p>
2152 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2153 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2161 <table class="layout">
2163 <td class="left"><tt>[4 x i32]*</tt></td>
2164 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2165 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2168 <td class="left"><tt>i32 (i32*) *</tt></td>
2169 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2170 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2174 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2175 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2176 that resides in address space #5.</td>
2182 <!-- _______________________________________________________________________ -->
2184 <a name="t_vector">Vector Type</a>
2190 <p>A vector type is a simple derived type that represents a vector of elements.
2191 Vector types are used when multiple primitive data are operated in parallel
2192 using a single instruction (SIMD). A vector type requires a size (number of
2193 elements) and an underlying primitive data type. Vector types are considered
2194 <a href="#t_firstclass">first class</a>.</p>
2198 < <# elements> x <elementtype> >
2201 <p>The number of elements is a constant integer value larger than 0; elementtype
2202 may be any integer or floating point type, or a pointer to these types.
2203 Vectors of size zero are not allowed. </p>
2206 <table class="layout">
2208 <td class="left"><tt><4 x i32></tt></td>
2209 <td class="left">Vector of 4 32-bit integer values.</td>
2212 <td class="left"><tt><8 x float></tt></td>
2213 <td class="left">Vector of 8 32-bit floating-point values.</td>
2216 <td class="left"><tt><2 x i64></tt></td>
2217 <td class="left">Vector of 2 64-bit integer values.</td>
2220 <td class="left"><tt><4 x i64*></tt></td>
2221 <td class="left">Vector of 4 pointers to 64-bit integer values.</td>
2231 <!-- *********************************************************************** -->
2232 <h2><a name="constants">Constants</a></h2>
2233 <!-- *********************************************************************** -->
2237 <p>LLVM has several different basic types of constants. This section describes
2238 them all and their syntax.</p>
2240 <!-- ======================================================================= -->
2242 <a name="simpleconstants">Simple Constants</a>
2248 <dt><b>Boolean constants</b></dt>
2249 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2250 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2252 <dt><b>Integer constants</b></dt>
2253 <dd>Standard integers (such as '4') are constants of
2254 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2255 with integer types.</dd>
2257 <dt><b>Floating point constants</b></dt>
2258 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2259 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2260 notation (see below). The assembler requires the exact decimal value of a
2261 floating-point constant. For example, the assembler accepts 1.25 but
2262 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2263 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2265 <dt><b>Null pointer constants</b></dt>
2266 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2267 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2270 <p>The one non-intuitive notation for constants is the hexadecimal form of
2271 floating point constants. For example, the form '<tt>double
2272 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2273 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2274 constants are required (and the only time that they are generated by the
2275 disassembler) is when a floating point constant must be emitted but it cannot
2276 be represented as a decimal floating point number in a reasonable number of
2277 digits. For example, NaN's, infinities, and other special values are
2278 represented in their IEEE hexadecimal format so that assembly and disassembly
2279 do not cause any bits to change in the constants.</p>
2281 <p>When using the hexadecimal form, constants of types half, float, and double are
2282 represented using the 16-digit form shown above (which matches the IEEE754
2283 representation for double); half and float values must, however, be exactly
2284 representable as IEE754 half and single precision, respectively.
2285 Hexadecimal format is always used
2286 for long double, and there are three forms of long double. The 80-bit format
2287 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2288 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2289 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2290 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2291 currently supported target uses this format. Long doubles will only work if
2292 they match the long double format on your target. All hexadecimal formats
2293 are big-endian (sign bit at the left).</p>
2295 <p>There are no constants of type x86mmx.</p>
2298 <!-- ======================================================================= -->
2300 <a name="aggregateconstants"></a> <!-- old anchor -->
2301 <a name="complexconstants">Complex Constants</a>
2306 <p>Complex constants are a (potentially recursive) combination of simple
2307 constants and smaller complex constants.</p>
2310 <dt><b>Structure constants</b></dt>
2311 <dd>Structure constants are represented with notation similar to structure
2312 type definitions (a comma separated list of elements, surrounded by braces
2313 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2314 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2315 Structure constants must have <a href="#t_struct">structure type</a>, and
2316 the number and types of elements must match those specified by the
2319 <dt><b>Array constants</b></dt>
2320 <dd>Array constants are represented with notation similar to array type
2321 definitions (a comma separated list of elements, surrounded by square
2322 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2323 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2324 the number and types of elements must match those specified by the
2327 <dt><b>Vector constants</b></dt>
2328 <dd>Vector constants are represented with notation similar to vector type
2329 definitions (a comma separated list of elements, surrounded by
2330 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2331 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2332 have <a href="#t_vector">vector type</a>, and the number and types of
2333 elements must match those specified by the type.</dd>
2335 <dt><b>Zero initialization</b></dt>
2336 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2337 value to zero of <em>any</em> type, including scalar and
2338 <a href="#t_aggregate">aggregate</a> types.
2339 This is often used to avoid having to print large zero initializers
2340 (e.g. for large arrays) and is always exactly equivalent to using explicit
2341 zero initializers.</dd>
2343 <dt><b>Metadata node</b></dt>
2344 <dd>A metadata node is a structure-like constant with
2345 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2346 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2347 be interpreted as part of the instruction stream, metadata is a place to
2348 attach additional information such as debug info.</dd>
2353 <!-- ======================================================================= -->
2355 <a name="globalconstants">Global Variable and Function Addresses</a>
2360 <p>The addresses of <a href="#globalvars">global variables</a>
2361 and <a href="#functionstructure">functions</a> are always implicitly valid
2362 (link-time) constants. These constants are explicitly referenced when
2363 the <a href="#identifiers">identifier for the global</a> is used and always
2364 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2365 legal LLVM file:</p>
2367 <pre class="doc_code">
2370 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2375 <!-- ======================================================================= -->
2377 <a name="undefvalues">Undefined Values</a>
2382 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2383 indicates that the user of the value may receive an unspecified bit-pattern.
2384 Undefined values may be of any type (other than '<tt>label</tt>'
2385 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2387 <p>Undefined values are useful because they indicate to the compiler that the
2388 program is well defined no matter what value is used. This gives the
2389 compiler more freedom to optimize. Here are some examples of (potentially
2390 surprising) transformations that are valid (in pseudo IR):</p>
2393 <pre class="doc_code">
2403 <p>This is safe because all of the output bits are affected by the undef bits.
2404 Any output bit can have a zero or one depending on the input bits.</p>
2406 <pre class="doc_code">
2417 <p>These logical operations have bits that are not always affected by the input.
2418 For example, if <tt>%X</tt> has a zero bit, then the output of the
2419 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2420 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2421 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2422 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2423 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2424 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2425 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2427 <pre class="doc_code">
2428 %A = select undef, %X, %Y
2429 %B = select undef, 42, %Y
2430 %C = select %X, %Y, undef
2441 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2442 branch) conditions can go <em>either way</em>, but they have to come from one
2443 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2444 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2445 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2446 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2447 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2450 <pre class="doc_code">
2451 %A = xor undef, undef
2469 <p>This example points out that two '<tt>undef</tt>' operands are not
2470 necessarily the same. This can be surprising to people (and also matches C
2471 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2472 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2473 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2474 its value over its "live range". This is true because the variable doesn't
2475 actually <em>have a live range</em>. Instead, the value is logically read
2476 from arbitrary registers that happen to be around when needed, so the value
2477 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2478 need to have the same semantics or the core LLVM "replace all uses with"
2479 concept would not hold.</p>
2481 <pre class="doc_code">
2489 <p>These examples show the crucial difference between an <em>undefined
2490 value</em> and <em>undefined behavior</em>. An undefined value (like
2491 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2492 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2493 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2494 defined on SNaN's. However, in the second example, we can make a more
2495 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2496 arbitrary value, we are allowed to assume that it could be zero. Since a
2497 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2498 the operation does not execute at all. This allows us to delete the divide and
2499 all code after it. Because the undefined operation "can't happen", the
2500 optimizer can assume that it occurs in dead code.</p>
2502 <pre class="doc_code">
2503 a: store undef -> %X
2504 b: store %X -> undef
2510 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2511 undefined value can be assumed to not have any effect; we can assume that the
2512 value is overwritten with bits that happen to match what was already there.
2513 However, a store <em>to</em> an undefined location could clobber arbitrary
2514 memory, therefore, it has undefined behavior.</p>
2518 <!-- ======================================================================= -->
2520 <a name="poisonvalues">Poison Values</a>
2525 <p>Poison values are similar to <a href="#undefvalues">undef values</a>, however
2526 they also represent the fact that an instruction or constant expression which
2527 cannot evoke side effects has nevertheless detected a condition which results
2528 in undefined behavior.</p>
2530 <p>There is currently no way of representing a poison value in the IR; they
2531 only exist when produced by operations such as
2532 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2534 <p>Poison value behavior is defined in terms of value <i>dependence</i>:</p>
2537 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2538 their operands.</li>
2540 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2541 to their dynamic predecessor basic block.</li>
2543 <li>Function arguments depend on the corresponding actual argument values in
2544 the dynamic callers of their functions.</li>
2546 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2547 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2548 control back to them.</li>
2550 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2551 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>,
2552 or exception-throwing call instructions that dynamically transfer control
2555 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2556 referenced memory addresses, following the order in the IR
2557 (including loads and stores implied by intrinsics such as
2558 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2560 <!-- TODO: In the case of multiple threads, this only applies if the store
2561 "happens-before" the load or store. -->
2563 <!-- TODO: floating-point exception state -->
2565 <li>An instruction with externally visible side effects depends on the most
2566 recent preceding instruction with externally visible side effects, following
2567 the order in the IR. (This includes
2568 <a href="#volatile">volatile operations</a>.)</li>
2570 <li>An instruction <i>control-depends</i> on a
2571 <a href="#terminators">terminator instruction</a>
2572 if the terminator instruction has multiple successors and the instruction
2573 is always executed when control transfers to one of the successors, and
2574 may not be executed when control is transferred to another.</li>
2576 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2577 instruction if the set of instructions it otherwise depends on would be
2578 different if the terminator had transferred control to a different
2581 <li>Dependence is transitive.</li>
2585 <p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>,
2586 with the additional affect that any instruction which has a <i>dependence</i>
2587 on a poison value has undefined behavior.</p>
2589 <p>Here are some examples:</p>
2591 <pre class="doc_code">
2593 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2594 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2595 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2596 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2598 store i32 %poison, i32* @g ; Poison value stored to memory.
2599 %poison2 = load i32* @g ; Poison value loaded back from memory.
2601 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2603 %narrowaddr = bitcast i32* @g to i16*
2604 %wideaddr = bitcast i32* @g to i64*
2605 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2606 %poison4 = load i64* %wideaddr ; Returns a poison value.
2608 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2609 br i1 %cmp, label %true, label %end ; Branch to either destination.
2612 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2613 ; it has undefined behavior.
2617 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2618 ; Both edges into this PHI are
2619 ; control-dependent on %cmp, so this
2620 ; always results in a poison value.
2622 store volatile i32 0, i32* @g ; This would depend on the store in %true
2623 ; if %cmp is true, or the store in %entry
2624 ; otherwise, so this is undefined behavior.
2626 br i1 %cmp, label %second_true, label %second_end
2627 ; The same branch again, but this time the
2628 ; true block doesn't have side effects.
2635 store volatile i32 0, i32* @g ; This time, the instruction always depends
2636 ; on the store in %end. Also, it is
2637 ; control-equivalent to %end, so this is
2638 ; well-defined (ignoring earlier undefined
2639 ; behavior in this example).
2644 <!-- ======================================================================= -->
2646 <a name="blockaddress">Addresses of Basic Blocks</a>
2651 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2653 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2654 basic block in the specified function, and always has an i8* type. Taking
2655 the address of the entry block is illegal.</p>
2657 <p>This value only has defined behavior when used as an operand to the
2658 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2659 comparisons against null. Pointer equality tests between labels addresses
2660 results in undefined behavior — though, again, comparison against null
2661 is ok, and no label is equal to the null pointer. This may be passed around
2662 as an opaque pointer sized value as long as the bits are not inspected. This
2663 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2664 long as the original value is reconstituted before the <tt>indirectbr</tt>
2667 <p>Finally, some targets may provide defined semantics when using the value as
2668 the operand to an inline assembly, but that is target specific.</p>
2673 <!-- ======================================================================= -->
2675 <a name="constantexprs">Constant Expressions</a>
2680 <p>Constant expressions are used to allow expressions involving other constants
2681 to be used as constants. Constant expressions may be of
2682 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2683 operation that does not have side effects (e.g. load and call are not
2684 supported). The following is the syntax for constant expressions:</p>
2687 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2688 <dd>Truncate a constant to another type. The bit size of CST must be larger
2689 than the bit size of TYPE. Both types must be integers.</dd>
2691 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2692 <dd>Zero extend a constant to another type. The bit size of CST must be
2693 smaller than the bit size of TYPE. Both types must be integers.</dd>
2695 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2696 <dd>Sign extend a constant to another type. The bit size of CST must be
2697 smaller than the bit size of TYPE. Both types must be integers.</dd>
2699 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2700 <dd>Truncate a floating point constant to another floating point type. The
2701 size of CST must be larger than the size of TYPE. Both types must be
2702 floating point.</dd>
2704 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2705 <dd>Floating point extend a constant to another type. The size of CST must be
2706 smaller or equal to the size of TYPE. Both types must be floating
2709 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2710 <dd>Convert a floating point constant to the corresponding unsigned integer
2711 constant. TYPE must be a scalar or vector integer type. CST must be of
2712 scalar or vector floating point type. Both CST and TYPE must be scalars,
2713 or vectors of the same number of elements. If the value won't fit in the
2714 integer type, the results are undefined.</dd>
2716 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2717 <dd>Convert a floating point constant to the corresponding signed integer
2718 constant. TYPE must be a scalar or vector integer type. CST must be of
2719 scalar or vector floating point type. Both CST and TYPE must be scalars,
2720 or vectors of the same number of elements. If the value won't fit in the
2721 integer type, the results are undefined.</dd>
2723 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2724 <dd>Convert an unsigned integer constant to the corresponding floating point
2725 constant. TYPE must be a scalar or vector floating point type. CST must be
2726 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2727 vectors of the same number of elements. If the value won't fit in the
2728 floating point type, the results are undefined.</dd>
2730 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2731 <dd>Convert a signed integer constant to the corresponding floating point
2732 constant. TYPE must be a scalar or vector floating point type. CST must be
2733 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2734 vectors of the same number of elements. If the value won't fit in the
2735 floating point type, the results are undefined.</dd>
2737 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2738 <dd>Convert a pointer typed constant to the corresponding integer constant
2739 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2740 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2741 make it fit in <tt>TYPE</tt>.</dd>
2743 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2744 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2745 type. CST must be of integer type. The CST value is zero extended,
2746 truncated, or unchanged to make it fit in a pointer size. This one is
2747 <i>really</i> dangerous!</dd>
2749 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2750 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2751 are the same as those for the <a href="#i_bitcast">bitcast
2752 instruction</a>.</dd>
2754 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2755 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2756 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2757 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2758 instruction, the index list may have zero or more indexes, which are
2759 required to make sense for the type of "CSTPTR".</dd>
2761 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2762 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2764 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2765 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2767 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2768 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2770 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2771 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2774 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2775 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2778 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2779 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2782 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2783 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2784 constants. The index list is interpreted in a similar manner as indices in
2785 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2786 index value must be specified.</dd>
2788 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2789 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2790 constants. The index list is interpreted in a similar manner as indices in
2791 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2792 index value must be specified.</dd>
2794 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2795 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2796 be any of the <a href="#binaryops">binary</a>
2797 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2798 on operands are the same as those for the corresponding instruction
2799 (e.g. no bitwise operations on floating point values are allowed).</dd>
2806 <!-- *********************************************************************** -->
2807 <h2><a name="othervalues">Other Values</a></h2>
2808 <!-- *********************************************************************** -->
2810 <!-- ======================================================================= -->
2812 <a name="inlineasm">Inline Assembler Expressions</a>
2817 <p>LLVM supports inline assembler expressions (as opposed
2818 to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of
2819 a special value. This value represents the inline assembler as a string
2820 (containing the instructions to emit), a list of operand constraints (stored
2821 as a string), a flag that indicates whether or not the inline asm
2822 expression has side effects, and a flag indicating whether the function
2823 containing the asm needs to align its stack conservatively. An example
2824 inline assembler expression is:</p>
2826 <pre class="doc_code">
2827 i32 (i32) asm "bswap $0", "=r,r"
2830 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2831 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2834 <pre class="doc_code">
2835 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2838 <p>Inline asms with side effects not visible in the constraint list must be
2839 marked as having side effects. This is done through the use of the
2840 '<tt>sideeffect</tt>' keyword, like so:</p>
2842 <pre class="doc_code">
2843 call void asm sideeffect "eieio", ""()
2846 <p>In some cases inline asms will contain code that will not work unless the
2847 stack is aligned in some way, such as calls or SSE instructions on x86,
2848 yet will not contain code that does that alignment within the asm.
2849 The compiler should make conservative assumptions about what the asm might
2850 contain and should generate its usual stack alignment code in the prologue
2851 if the '<tt>alignstack</tt>' keyword is present:</p>
2853 <pre class="doc_code">
2854 call void asm alignstack "eieio", ""()
2857 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2861 <p>TODO: The format of the asm and constraints string still need to be
2862 documented here. Constraints on what can be done (e.g. duplication, moving,
2863 etc need to be documented). This is probably best done by reference to
2864 another document that covers inline asm from a holistic perspective.</p>
2867 <!-- _______________________________________________________________________ -->
2869 <a name="inlineasm_md">Inline Asm Metadata</a>
2874 <p>The call instructions that wrap inline asm nodes may have a
2875 "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant
2876 integers. If present, the code generator will use the integer as the
2877 location cookie value when report errors through the <tt>LLVMContext</tt>
2878 error reporting mechanisms. This allows a front-end to correlate backend
2879 errors that occur with inline asm back to the source code that produced it.
2882 <pre class="doc_code">
2883 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2885 !42 = !{ i32 1234567 }
2888 <p>It is up to the front-end to make sense of the magic numbers it places in the
2889 IR. If the MDNode contains multiple constants, the code generator will use
2890 the one that corresponds to the line of the asm that the error occurs on.</p>
2896 <!-- ======================================================================= -->
2898 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2903 <p>LLVM IR allows metadata to be attached to instructions in the program that
2904 can convey extra information about the code to the optimizers and code
2905 generator. One example application of metadata is source-level debug
2906 information. There are two metadata primitives: strings and nodes. All
2907 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2908 preceding exclamation point ('<tt>!</tt>').</p>
2910 <p>A metadata string is a string surrounded by double quotes. It can contain
2911 any character by escaping non-printable characters with "<tt>\xx</tt>" where
2912 "<tt>xx</tt>" is the two digit hex code. For example:
2913 "<tt>!"test\00"</tt>".</p>
2915 <p>Metadata nodes are represented with notation similar to structure constants
2916 (a comma separated list of elements, surrounded by braces and preceded by an
2917 exclamation point). Metadata nodes can have any values as their operand. For
2920 <div class="doc_code">
2922 !{ metadata !"test\00", i32 10}
2926 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2927 metadata nodes, which can be looked up in the module symbol table. For
2930 <div class="doc_code">
2932 !foo = metadata !{!4, !3}
2936 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2937 function is using two metadata arguments:</p>
2939 <div class="doc_code">
2941 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2945 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2946 attached to the <tt>add</tt> instruction using the <tt>!dbg</tt>
2949 <div class="doc_code">
2951 %indvar.next = add i64 %indvar, 1, !dbg !21
2955 <p>More information about specific metadata nodes recognized by the optimizers
2956 and code generator is found below.</p>
2958 <!-- _______________________________________________________________________ -->
2960 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
2965 <p>In LLVM IR, memory does not have types, so LLVM's own type system is not
2966 suitable for doing TBAA. Instead, metadata is added to the IR to describe
2967 a type system of a higher level language. This can be used to implement
2968 typical C/C++ TBAA, but it can also be used to implement custom alias
2969 analysis behavior for other languages.</p>
2971 <p>The current metadata format is very simple. TBAA metadata nodes have up to
2972 three fields, e.g.:</p>
2974 <div class="doc_code">
2976 !0 = metadata !{ metadata !"an example type tree" }
2977 !1 = metadata !{ metadata !"int", metadata !0 }
2978 !2 = metadata !{ metadata !"float", metadata !0 }
2979 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
2983 <p>The first field is an identity field. It can be any value, usually
2984 a metadata string, which uniquely identifies the type. The most important
2985 name in the tree is the name of the root node. Two trees with
2986 different root node names are entirely disjoint, even if they
2987 have leaves with common names.</p>
2989 <p>The second field identifies the type's parent node in the tree, or
2990 is null or omitted for a root node. A type is considered to alias
2991 all of its descendants and all of its ancestors in the tree. Also,
2992 a type is considered to alias all types in other trees, so that
2993 bitcode produced from multiple front-ends is handled conservatively.</p>
2995 <p>If the third field is present, it's an integer which if equal to 1
2996 indicates that the type is "constant" (meaning
2997 <tt>pointsToConstantMemory</tt> should return true; see
2998 <a href="AliasAnalysis.html#OtherItfs">other useful
2999 <tt>AliasAnalysis</tt> methods</a>).</p>
3003 <!-- _______________________________________________________________________ -->
3005 <a name="fpmath">'<tt>fpmath</tt>' Metadata</a>
3010 <p><tt>fpmath</tt> metadata may be attached to any instruction of floating point
3011 type. It can be used to express the maximum acceptable error in the result of
3012 that instruction, in ULPs, thus potentially allowing the compiler to use a
3013 more efficient but less accurate method of computing it. ULP is defined as
3018 <p>If <tt>x</tt> is a real number that lies between two finite consecutive
3019 floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one
3020 of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the
3021 distance between the two non-equal finite floating-point numbers nearest
3022 <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p>
3026 <p>The metadata node shall consist of a single positive floating point number
3027 representing the maximum relative error, for example:</p>
3029 <div class="doc_code">
3031 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3037 <!-- _______________________________________________________________________ -->
3039 <a name="range">'<tt>range</tt>' Metadata</a>
3043 <p><tt>range</tt> metadata may be attached only to loads of integer types. It
3044 expresses the possible ranges the loaded value is in. The ranges are
3045 represented with a flattened list of integers. The loaded value is known to
3046 be in the union of the ranges defined by each consecutive pair. Each pair
3047 has the following properties:</p>
3049 <li>The type must match the type loaded by the instruction.</li>
3050 <li>The pair <tt>a,b</tt> represents the range <tt>[a,b)</tt>.</li>
3051 <li>Both <tt>a</tt> and <tt>b</tt> are constants.</li>
3052 <li>The range is allowed to wrap.</li>
3053 <li>The range should not represent the full or empty set. That is,
3054 <tt>a!=b</tt>. </li>
3058 <div class="doc_code">
3060 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
3061 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3062 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
3064 !0 = metadata !{ i8 0, i8 2 }
3065 !1 = metadata !{ i8 255, i8 2 }
3066 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
3074 <!-- *********************************************************************** -->
3076 <a name="module_flags">Module Flags Metadata</a>
3078 <!-- *********************************************************************** -->
3082 <p>Information about the module as a whole is difficult to convey to LLVM's
3083 subsystems. The LLVM IR isn't sufficient to transmit this
3084 information. The <tt>llvm.module.flags</tt> named metadata exists in order to
3085 facilitate this. These flags are in the form of key / value pairs —
3086 much like a dictionary — making it easy for any subsystem who cares
3087 about a flag to look it up.</p>
3089 <p>The <tt>llvm.module.flags</tt> metadata contains a list of metadata
3090 triplets. Each triplet has the following form:</p>
3093 <li>The first element is a <i>behavior</i> flag, which specifies the behavior
3094 when two (or more) modules are merged together, and it encounters two (or
3095 more) metadata with the same ID. The supported behaviors are described
3098 <li>The second element is a metadata string that is a unique ID for the
3099 metadata. How each ID is interpreted is documented below.</li>
3101 <li>The third element is the value of the flag.</li>
3104 <p>When two (or more) modules are merged together, the resulting
3105 <tt>llvm.module.flags</tt> metadata is the union of the
3106 modules' <tt>llvm.module.flags</tt> metadata. The only exception being a flag
3107 with the <i>Override</i> behavior, which may override another flag's value
3110 <p>The following behaviors are supported:</p>
3112 <table border="1" cellspacing="0" cellpadding="4">
3122 <dt><b>Error</b></dt>
3123 <dd>Emits an error if two values disagree. It is an error to have an ID
3124 with both an Error and a Warning behavior.</dd>
3132 <dt><b>Warning</b></dt>
3133 <dd>Emits a warning if two values disagree.</dd>
3141 <dt><b>Require</b></dt>
3142 <dd>Emits an error when the specified value is not present or doesn't
3143 have the specified value. It is an error for two (or more)
3144 <tt>llvm.module.flags</tt> with the same ID to have the Require
3145 behavior but different values. There may be multiple Require flags
3154 <dt><b>Override</b></dt>
3155 <dd>Uses the specified value if the two values disagree. It is an
3156 error for two (or more) <tt>llvm.module.flags</tt> with the same
3157 ID to have the Override behavior but different values.</dd>
3164 <p>An example of module flags:</p>
3166 <pre class="doc_code">
3167 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3168 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3169 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3170 !3 = metadata !{ i32 3, metadata !"qux",
3172 metadata !"foo", i32 1
3175 !llvm.module.flags = !{ !0, !1, !2, !3 }
3179 <li><p>Metadata <tt>!0</tt> has the ID <tt>!"foo"</tt> and the value '1'. The
3180 behavior if two or more <tt>!"foo"</tt> flags are seen is to emit an
3181 error if their values are not equal.</p></li>
3183 <li><p>Metadata <tt>!1</tt> has the ID <tt>!"bar"</tt> and the value '37'. The
3184 behavior if two or more <tt>!"bar"</tt> flags are seen is to use the
3185 value '37' if their values are not equal.</p></li>
3187 <li><p>Metadata <tt>!2</tt> has the ID <tt>!"qux"</tt> and the value '42'. The
3188 behavior if two or more <tt>!"qux"</tt> flags are seen is to emit a
3189 warning if their values are not equal.</p></li>
3191 <li><p>Metadata <tt>!3</tt> has the ID <tt>!"qux"</tt> and the value:</p>
3193 <pre class="doc_code">
3194 metadata !{ metadata !"foo", i32 1 }
3197 <p>The behavior is to emit an error if the <tt>llvm.module.flags</tt> does
3198 not contain a flag with the ID <tt>!"foo"</tt> that has the value
3199 '1'. If two or more <tt>!"qux"</tt> flags exist, then they must have
3200 the same value or an error will be issued.</p></li>
3204 <!-- ======================================================================= -->
3206 <a name="objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a>
3211 <p>On the Mach-O platform, Objective-C stores metadata about garbage collection
3212 in a special section called "image info". The metadata consists of a version
3213 number and a bitmask specifying what types of garbage collection are
3214 supported (if any) by the file. If two or more modules are linked together
3215 their garbage collection metadata needs to be merged rather than appended
3218 <p>The Objective-C garbage collection module flags metadata consists of the
3219 following key-value pairs:</p>
3221 <table border="1" cellspacing="0" cellpadding="4">
3229 <td><tt>Objective-C Version</tt></td>
3230 <td align="left"><b>[Required]</b> — The Objective-C ABI
3231 version. Valid values are 1 and 2.</td>
3234 <td><tt>Objective-C Image Info Version</tt></td>
3235 <td align="left"><b>[Required]</b> — The version of the image info
3236 section. Currently always 0.</td>
3239 <td><tt>Objective-C Image Info Section</tt></td>
3240 <td align="left"><b>[Required]</b> — The section to place the
3241 metadata. Valid values are <tt>"__OBJC, __image_info, regular"</tt> for
3242 Objective-C ABI version 1, and <tt>"__DATA,__objc_imageinfo, regular,
3243 no_dead_strip"</tt> for Objective-C ABI version 2.</td>
3246 <td><tt>Objective-C Garbage Collection</tt></td>
3247 <td align="left"><b>[Required]</b> — Specifies whether garbage
3248 collection is supported or not. Valid values are 0, for no garbage
3249 collection, and 2, for garbage collection supported.</td>
3252 <td><tt>Objective-C GC Only</tt></td>
3253 <td align="left"><b>[Optional]</b> — Specifies that only garbage
3254 collection is supported. If present, its value must be 6. This flag
3255 requires that the <tt>Objective-C Garbage Collection</tt> flag have the
3261 <p>Some important flag interactions:</p>
3264 <li>If a module with <tt>Objective-C Garbage Collection</tt> set to 0 is
3265 merged with a module with <tt>Objective-C Garbage Collection</tt> set to
3266 2, then the resulting module has the <tt>Objective-C Garbage
3267 Collection</tt> flag set to 0.</li>
3269 <li>A module with <tt>Objective-C Garbage Collection</tt> set to 0 cannot be
3270 merged with a module with <tt>Objective-C GC Only</tt> set to 6.</li>
3277 <!-- *********************************************************************** -->
3279 <a name="intrinsic_globals">Intrinsic Global Variables</a>
3281 <!-- *********************************************************************** -->
3283 <p>LLVM has a number of "magic" global variables that contain data that affect
3284 code generation or other IR semantics. These are documented here. All globals
3285 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
3286 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
3289 <!-- ======================================================================= -->
3291 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
3296 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
3297 href="#linkage_appending">appending linkage</a>. This array contains a list of
3298 pointers to global variables and functions which may optionally have a pointer
3299 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
3301 <div class="doc_code">
3306 @llvm.used = appending global [2 x i8*] [
3308 i8* bitcast (i32* @Y to i8*)
3309 ], section "llvm.metadata"
3313 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
3314 compiler, assembler, and linker are required to treat the symbol as if there
3315 is a reference to the global that it cannot see. For example, if a variable
3316 has internal linkage and no references other than that from
3317 the <tt>@llvm.used</tt> list, it cannot be deleted. This is commonly used to
3318 represent references from inline asms and other things the compiler cannot
3319 "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p>
3321 <p>On some targets, the code generator must emit a directive to the assembler or
3322 object file to prevent the assembler and linker from molesting the
3327 <!-- ======================================================================= -->
3329 <a name="intg_compiler_used">
3330 The '<tt>llvm.compiler.used</tt>' Global Variable
3336 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
3337 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
3338 touching the symbol. On targets that support it, this allows an intelligent
3339 linker to optimize references to the symbol without being impeded as it would
3340 be by <tt>@llvm.used</tt>.</p>
3342 <p>This is a rare construct that should only be used in rare circumstances, and
3343 should not be exposed to source languages.</p>
3347 <!-- ======================================================================= -->
3349 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
3354 <div class="doc_code">
3356 %0 = type { i32, void ()* }
3357 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3361 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor
3362 functions and associated priorities. The functions referenced by this array
3363 will be called in ascending order of priority (i.e. lowest first) when the
3364 module is loaded. The order of functions with the same priority is not
3369 <!-- ======================================================================= -->
3371 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3376 <div class="doc_code">
3378 %0 = type { i32, void ()* }
3379 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3383 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions
3384 and associated priorities. The functions referenced by this array will be
3385 called in descending order of priority (i.e. highest first) when the module
3386 is loaded. The order of functions with the same priority is not defined.</p>
3392 <!-- *********************************************************************** -->
3393 <h2><a name="instref">Instruction Reference</a></h2>
3394 <!-- *********************************************************************** -->
3398 <p>The LLVM instruction set consists of several different classifications of
3399 instructions: <a href="#terminators">terminator
3400 instructions</a>, <a href="#binaryops">binary instructions</a>,
3401 <a href="#bitwiseops">bitwise binary instructions</a>,
3402 <a href="#memoryops">memory instructions</a>, and
3403 <a href="#otherops">other instructions</a>.</p>
3405 <!-- ======================================================================= -->
3407 <a name="terminators">Terminator Instructions</a>
3412 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3413 in a program ends with a "Terminator" instruction, which indicates which
3414 block should be executed after the current block is finished. These
3415 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3416 control flow, not values (the one exception being the
3417 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3419 <p>The terminator instructions are:
3420 '<a href="#i_ret"><tt>ret</tt></a>',
3421 '<a href="#i_br"><tt>br</tt></a>',
3422 '<a href="#i_switch"><tt>switch</tt></a>',
3423 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3424 '<a href="#i_invoke"><tt>invoke</tt></a>',
3425 '<a href="#i_resume"><tt>resume</tt></a>', and
3426 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3428 <!-- _______________________________________________________________________ -->
3430 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3437 ret <type> <value> <i>; Return a value from a non-void function</i>
3438 ret void <i>; Return from void function</i>
3442 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3443 a value) from a function back to the caller.</p>
3445 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3446 value and then causes control flow, and one that just causes control flow to
3450 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3451 return value. The type of the return value must be a
3452 '<a href="#t_firstclass">first class</a>' type.</p>
3454 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3455 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3456 value or a return value with a type that does not match its type, or if it
3457 has a void return type and contains a '<tt>ret</tt>' instruction with a
3461 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3462 the calling function's context. If the caller is a
3463 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3464 instruction after the call. If the caller was an
3465 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3466 the beginning of the "normal" destination block. If the instruction returns
3467 a value, that value shall set the call or invoke instruction's return
3472 ret i32 5 <i>; Return an integer value of 5</i>
3473 ret void <i>; Return from a void function</i>
3474 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3478 <!-- _______________________________________________________________________ -->
3480 <a name="i_br">'<tt>br</tt>' Instruction</a>
3487 br i1 <cond>, label <iftrue>, label <iffalse>
3488 br label <dest> <i>; Unconditional branch</i>
3492 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3493 different basic block in the current function. There are two forms of this
3494 instruction, corresponding to a conditional branch and an unconditional
3498 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3499 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3500 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3504 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3505 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3506 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3507 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3512 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3513 br i1 %cond, label %IfEqual, label %IfUnequal
3515 <a href="#i_ret">ret</a> i32 1
3517 <a href="#i_ret">ret</a> i32 0
3522 <!-- _______________________________________________________________________ -->
3524 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3531 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3535 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3536 several different places. It is a generalization of the '<tt>br</tt>'
3537 instruction, allowing a branch to occur to one of many possible
3541 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3542 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3543 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3544 The table is not allowed to contain duplicate constant entries.</p>
3547 <p>The <tt>switch</tt> instruction specifies a table of values and
3548 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3549 is searched for the given value. If the value is found, control flow is
3550 transferred to the corresponding destination; otherwise, control flow is
3551 transferred to the default destination.</p>
3553 <h5>Implementation:</h5>
3554 <p>Depending on properties of the target machine and the particular
3555 <tt>switch</tt> instruction, this instruction may be code generated in
3556 different ways. For example, it could be generated as a series of chained
3557 conditional branches or with a lookup table.</p>
3561 <i>; Emulate a conditional br instruction</i>
3562 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3563 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3565 <i>; Emulate an unconditional br instruction</i>
3566 switch i32 0, label %dest [ ]
3568 <i>; Implement a jump table:</i>
3569 switch i32 %val, label %otherwise [ i32 0, label %onzero
3571 i32 2, label %ontwo ]
3577 <!-- _______________________________________________________________________ -->
3579 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3586 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3591 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3592 within the current function, whose address is specified by
3593 "<tt>address</tt>". Address must be derived from a <a
3594 href="#blockaddress">blockaddress</a> constant.</p>
3598 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3599 rest of the arguments indicate the full set of possible destinations that the
3600 address may point to. Blocks are allowed to occur multiple times in the
3601 destination list, though this isn't particularly useful.</p>
3603 <p>This destination list is required so that dataflow analysis has an accurate
3604 understanding of the CFG.</p>
3608 <p>Control transfers to the block specified in the address argument. All
3609 possible destination blocks must be listed in the label list, otherwise this
3610 instruction has undefined behavior. This implies that jumps to labels
3611 defined in other functions have undefined behavior as well.</p>
3613 <h5>Implementation:</h5>
3615 <p>This is typically implemented with a jump through a register.</p>
3619 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3625 <!-- _______________________________________________________________________ -->
3627 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3634 <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>]
3635 to label <normal label> unwind label <exception label>
3639 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3640 function, with the possibility of control flow transfer to either the
3641 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3642 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3643 control flow will return to the "normal" label. If the callee (or any
3644 indirect callees) returns via the "<a href="#i_resume"><tt>resume</tt></a>"
3645 instruction or other exception handling mechanism, control is interrupted and
3646 continued at the dynamically nearest "exception" label.</p>
3648 <p>The '<tt>exception</tt>' label is a
3649 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3650 exception. As such, '<tt>exception</tt>' label is required to have the
3651 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3652 the information about the behavior of the program after unwinding
3653 happens, as its first non-PHI instruction. The restrictions on the
3654 "<tt>landingpad</tt>" instruction's tightly couples it to the
3655 "<tt>invoke</tt>" instruction, so that the important information contained
3656 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3660 <p>This instruction requires several arguments:</p>
3663 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3664 convention</a> the call should use. If none is specified, the call
3665 defaults to using C calling conventions.</li>
3667 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3668 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3669 '<tt>inreg</tt>' attributes are valid here.</li>
3671 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3672 function value being invoked. In most cases, this is a direct function
3673 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3674 off an arbitrary pointer to function value.</li>
3676 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3677 function to be invoked. </li>
3679 <li>'<tt>function args</tt>': argument list whose types match the function
3680 signature argument types and parameter attributes. All arguments must be
3681 of <a href="#t_firstclass">first class</a> type. If the function
3682 signature indicates the function accepts a variable number of arguments,
3683 the extra arguments can be specified.</li>
3685 <li>'<tt>normal label</tt>': the label reached when the called function
3686 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3688 <li>'<tt>exception label</tt>': the label reached when a callee returns via
3689 the <a href="#i_resume"><tt>resume</tt></a> instruction or other exception
3690 handling mechanism.</li>
3692 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3693 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3694 '<tt>readnone</tt>' attributes are valid here.</li>
3698 <p>This instruction is designed to operate as a standard
3699 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3700 primary difference is that it establishes an association with a label, which
3701 is used by the runtime library to unwind the stack.</p>
3703 <p>This instruction is used in languages with destructors to ensure that proper
3704 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3705 exception. Additionally, this is important for implementation of
3706 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3708 <p>For the purposes of the SSA form, the definition of the value returned by the
3709 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3710 block to the "normal" label. If the callee unwinds then no return value is
3715 %retval = invoke i32 @Test(i32 15) to label %Continue
3716 unwind label %TestCleanup <i>; {i32}:retval set</i>
3717 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3718 unwind label %TestCleanup <i>; {i32}:retval set</i>
3723 <!-- _______________________________________________________________________ -->
3726 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3733 resume <type> <value>
3737 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3741 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3742 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3746 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3747 (in-flight) exception whose unwinding was interrupted with
3748 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3752 resume { i8*, i32 } %exn
3757 <!-- _______________________________________________________________________ -->
3760 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3771 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3772 instruction is used to inform the optimizer that a particular portion of the
3773 code is not reachable. This can be used to indicate that the code after a
3774 no-return function cannot be reached, and other facts.</p>
3777 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3783 <!-- ======================================================================= -->
3785 <a name="binaryops">Binary Operations</a>
3790 <p>Binary operators are used to do most of the computation in a program. They
3791 require two operands of the same type, execute an operation on them, and
3792 produce a single value. The operands might represent multiple data, as is
3793 the case with the <a href="#t_vector">vector</a> data type. The result value
3794 has the same type as its operands.</p>
3796 <p>There are several different binary operators:</p>
3798 <!-- _______________________________________________________________________ -->
3800 <a name="i_add">'<tt>add</tt>' Instruction</a>
3807 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3808 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3809 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3810 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3814 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3817 <p>The two arguments to the '<tt>add</tt>' instruction must
3818 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3819 integer values. Both arguments must have identical types.</p>
3822 <p>The value produced is the integer sum of the two operands.</p>
3824 <p>If the sum has unsigned overflow, the result returned is the mathematical
3825 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3827 <p>Because LLVM integers use a two's complement representation, this instruction
3828 is appropriate for both signed and unsigned integers.</p>
3830 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3831 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3832 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3833 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3834 respectively, occurs.</p>
3838 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3843 <!-- _______________________________________________________________________ -->
3845 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3852 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3856 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3859 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3860 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3861 floating point values. Both arguments must have identical types.</p>
3864 <p>The value produced is the floating point sum of the two operands.</p>
3868 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3873 <!-- _______________________________________________________________________ -->
3875 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3882 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3883 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3884 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3885 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3889 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3892 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3893 '<tt>neg</tt>' instruction present in most other intermediate
3894 representations.</p>
3897 <p>The two arguments to the '<tt>sub</tt>' instruction must
3898 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3899 integer values. Both arguments must have identical types.</p>
3902 <p>The value produced is the integer difference of the two operands.</p>
3904 <p>If the difference has unsigned overflow, the result returned is the
3905 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3908 <p>Because LLVM integers use a two's complement representation, this instruction
3909 is appropriate for both signed and unsigned integers.</p>
3911 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3912 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3913 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3914 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3915 respectively, occurs.</p>
3919 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3920 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3925 <!-- _______________________________________________________________________ -->
3927 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3934 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3938 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3941 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3942 '<tt>fneg</tt>' instruction present in most other intermediate
3943 representations.</p>
3946 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3947 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3948 floating point values. Both arguments must have identical types.</p>
3951 <p>The value produced is the floating point difference of the two operands.</p>
3955 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3956 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3961 <!-- _______________________________________________________________________ -->
3963 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3970 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3971 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3972 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3973 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3977 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3980 <p>The two arguments to the '<tt>mul</tt>' instruction must
3981 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3982 integer values. Both arguments must have identical types.</p>
3985 <p>The value produced is the integer product of the two operands.</p>
3987 <p>If the result of the multiplication has unsigned overflow, the result
3988 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3989 width of the result.</p>
3991 <p>Because LLVM integers use a two's complement representation, and the result
3992 is the same width as the operands, this instruction returns the correct
3993 result for both signed and unsigned integers. If a full product
3994 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3995 be sign-extended or zero-extended as appropriate to the width of the full
3998 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3999 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
4000 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
4001 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4002 respectively, occurs.</p>
4006 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
4011 <!-- _______________________________________________________________________ -->
4013 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
4020 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4024 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
4027 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
4028 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4029 floating point values. Both arguments must have identical types.</p>
4032 <p>The value produced is the floating point product of the two operands.</p>
4036 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
4041 <!-- _______________________________________________________________________ -->
4043 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
4050 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4051 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4055 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
4058 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
4059 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4060 values. Both arguments must have identical types.</p>
4063 <p>The value produced is the unsigned integer quotient of the two operands.</p>
4065 <p>Note that unsigned integer division and signed integer division are distinct
4066 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
4068 <p>Division by zero leads to undefined behavior.</p>
4070 <p>If the <tt>exact</tt> keyword is present, the result value of the
4071 <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a
4072 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
4077 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4082 <!-- _______________________________________________________________________ -->
4084 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
4091 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4092 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4096 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
4099 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
4100 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4101 values. Both arguments must have identical types.</p>
4104 <p>The value produced is the signed integer quotient of the two operands rounded
4107 <p>Note that signed integer division and unsigned integer division are distinct
4108 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
4110 <p>Division by zero leads to undefined behavior. Overflow also leads to
4111 undefined behavior; this is a rare case, but can occur, for example, by doing
4112 a 32-bit division of -2147483648 by -1.</p>
4114 <p>If the <tt>exact</tt> keyword is present, the result value of the
4115 <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would
4120 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4125 <!-- _______________________________________________________________________ -->
4127 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
4134 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4138 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
4141 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
4142 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4143 floating point values. Both arguments must have identical types.</p>
4146 <p>The value produced is the floating point quotient of the two operands.</p>
4150 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
4155 <!-- _______________________________________________________________________ -->
4157 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
4164 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4168 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
4169 division of its two arguments.</p>
4172 <p>The two arguments to the '<tt>urem</tt>' instruction must be
4173 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4174 values. Both arguments must have identical types.</p>
4177 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
4178 This instruction always performs an unsigned division to get the
4181 <p>Note that unsigned integer remainder and signed integer remainder are
4182 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
4184 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
4188 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4193 <!-- _______________________________________________________________________ -->
4195 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
4202 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4206 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
4207 division of its two operands. This instruction can also take
4208 <a href="#t_vector">vector</a> versions of the values in which case the
4209 elements must be integers.</p>
4212 <p>The two arguments to the '<tt>srem</tt>' instruction must be
4213 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4214 values. Both arguments must have identical types.</p>
4217 <p>This instruction returns the <i>remainder</i> of a division (where the result
4218 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
4219 <i>modulo</i> operator (where the result is either zero or has the same sign
4220 as the divisor, <tt>op2</tt>) of a value.
4221 For more information about the difference,
4222 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
4223 Math Forum</a>. For a table of how this is implemented in various languages,
4224 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
4225 Wikipedia: modulo operation</a>.</p>
4227 <p>Note that signed integer remainder and unsigned integer remainder are
4228 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
4230 <p>Taking the remainder of a division by zero leads to undefined behavior.
4231 Overflow also leads to undefined behavior; this is a rare case, but can
4232 occur, for example, by taking the remainder of a 32-bit division of
4233 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
4234 lets srem be implemented using instructions that return both the result of
4235 the division and the remainder.)</p>
4239 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4244 <!-- _______________________________________________________________________ -->
4246 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
4253 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4257 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
4258 its two operands.</p>
4261 <p>The two arguments to the '<tt>frem</tt>' instruction must be
4262 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4263 floating point values. Both arguments must have identical types.</p>
4266 <p>This instruction returns the <i>remainder</i> of a division. The remainder
4267 has the same sign as the dividend.</p>
4271 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
4278 <!-- ======================================================================= -->
4280 <a name="bitwiseops">Bitwise Binary Operations</a>
4285 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
4286 program. They are generally very efficient instructions and can commonly be
4287 strength reduced from other instructions. They require two operands of the
4288 same type, execute an operation on them, and produce a single value. The
4289 resulting value is the same type as its operands.</p>
4291 <!-- _______________________________________________________________________ -->
4293 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
4300 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4301 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4302 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4303 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4307 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
4308 a specified number of bits.</p>
4311 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
4312 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4313 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4316 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
4317 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
4318 is (statically or dynamically) negative or equal to or larger than the number
4319 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4320 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4321 shift amount in <tt>op2</tt>.</p>
4323 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
4324 <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits. If
4325 the <tt>nsw</tt> keyword is present, then the shift produces a
4326 <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree
4327 with the resultant sign bit. As such, NUW/NSW have the same semantics as
4328 they would if the shift were expressed as a mul instruction with the same
4329 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
4333 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
4334 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
4335 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
4336 <result> = shl i32 1, 32 <i>; undefined</i>
4337 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4342 <!-- _______________________________________________________________________ -->
4344 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4351 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4352 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4356 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4357 operand shifted to the right a specified number of bits with zero fill.</p>
4360 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4361 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4362 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4365 <p>This instruction always performs a logical shift right operation. The most
4366 significant bits of the result will be filled with zero bits after the shift.
4367 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4368 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4369 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4370 shift amount in <tt>op2</tt>.</p>
4372 <p>If the <tt>exact</tt> keyword is present, the result value of the
4373 <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4374 shifted out are non-zero.</p>
4379 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4380 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4381 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4382 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4383 <result> = lshr i32 1, 32 <i>; undefined</i>
4384 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4389 <!-- _______________________________________________________________________ -->
4391 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4398 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4399 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4403 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4404 operand shifted to the right a specified number of bits with sign
4408 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4409 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4410 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4413 <p>This instruction always performs an arithmetic shift right operation, The
4414 most significant bits of the result will be filled with the sign bit
4415 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4416 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4417 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4418 the corresponding shift amount in <tt>op2</tt>.</p>
4420 <p>If the <tt>exact</tt> keyword is present, the result value of the
4421 <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4422 shifted out are non-zero.</p>
4426 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4427 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4428 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4429 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4430 <result> = ashr i32 1, 32 <i>; undefined</i>
4431 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4436 <!-- _______________________________________________________________________ -->
4438 <a name="i_and">'<tt>and</tt>' Instruction</a>
4445 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4449 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4453 <p>The two arguments to the '<tt>and</tt>' instruction must be
4454 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4455 values. Both arguments must have identical types.</p>
4458 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4460 <table border="1" cellspacing="0" cellpadding="4">
4492 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4493 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4494 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4497 <!-- _______________________________________________________________________ -->
4499 <a name="i_or">'<tt>or</tt>' Instruction</a>
4506 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4510 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4514 <p>The two arguments to the '<tt>or</tt>' instruction must be
4515 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4516 values. Both arguments must have identical types.</p>
4519 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4521 <table border="1" cellspacing="0" cellpadding="4">
4553 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4554 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4555 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4560 <!-- _______________________________________________________________________ -->
4562 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4569 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4573 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4574 its two operands. The <tt>xor</tt> is used to implement the "one's
4575 complement" operation, which is the "~" operator in C.</p>
4578 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4579 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4580 values. Both arguments must have identical types.</p>
4583 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4585 <table border="1" cellspacing="0" cellpadding="4">
4617 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4618 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4619 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4620 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4627 <!-- ======================================================================= -->
4629 <a name="vectorops">Vector Operations</a>
4634 <p>LLVM supports several instructions to represent vector operations in a
4635 target-independent manner. These instructions cover the element-access and
4636 vector-specific operations needed to process vectors effectively. While LLVM
4637 does directly support these vector operations, many sophisticated algorithms
4638 will want to use target-specific intrinsics to take full advantage of a
4639 specific target.</p>
4641 <!-- _______________________________________________________________________ -->
4643 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4650 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4654 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4655 from a vector at a specified index.</p>
4659 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4660 of <a href="#t_vector">vector</a> type. The second operand is an index
4661 indicating the position from which to extract the element. The index may be
4665 <p>The result is a scalar of the same type as the element type of
4666 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4667 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4668 results are undefined.</p>
4672 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4677 <!-- _______________________________________________________________________ -->
4679 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4686 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4690 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4691 vector at a specified index.</p>
4694 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4695 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4696 whose type must equal the element type of the first operand. The third
4697 operand is an index indicating the position at which to insert the value.
4698 The index may be a variable.</p>
4701 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4702 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4703 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4704 results are undefined.</p>
4708 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4713 <!-- _______________________________________________________________________ -->
4715 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4722 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4726 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4727 from two input vectors, returning a vector with the same element type as the
4728 input and length that is the same as the shuffle mask.</p>
4731 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4732 with types that match each other. The third argument is a shuffle mask whose
4733 element type is always 'i32'. The result of the instruction is a vector
4734 whose length is the same as the shuffle mask and whose element type is the
4735 same as the element type of the first two operands.</p>
4737 <p>The shuffle mask operand is required to be a constant vector with either
4738 constant integer or undef values.</p>
4741 <p>The elements of the two input vectors are numbered from left to right across
4742 both of the vectors. The shuffle mask operand specifies, for each element of
4743 the result vector, which element of the two input vectors the result element
4744 gets. The element selector may be undef (meaning "don't care") and the
4745 second operand may be undef if performing a shuffle from only one vector.</p>
4749 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4750 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4751 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4752 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4753 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4754 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4755 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4756 <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>
4763 <!-- ======================================================================= -->
4765 <a name="aggregateops">Aggregate Operations</a>
4770 <p>LLVM supports several instructions for working with
4771 <a href="#t_aggregate">aggregate</a> values.</p>
4773 <!-- _______________________________________________________________________ -->
4775 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4782 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4786 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4787 from an <a href="#t_aggregate">aggregate</a> value.</p>
4790 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4791 of <a href="#t_struct">struct</a> or
4792 <a href="#t_array">array</a> type. The operands are constant indices to
4793 specify which value to extract in a similar manner as indices in a
4794 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4795 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4797 <li>Since the value being indexed is not a pointer, the first index is
4798 omitted and assumed to be zero.</li>
4799 <li>At least one index must be specified.</li>
4800 <li>Not only struct indices but also array indices must be in
4805 <p>The result is the value at the position in the aggregate specified by the
4810 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4815 <!-- _______________________________________________________________________ -->
4817 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4824 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4828 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4829 in an <a href="#t_aggregate">aggregate</a> value.</p>
4832 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4833 of <a href="#t_struct">struct</a> or
4834 <a href="#t_array">array</a> type. The second operand is a first-class
4835 value to insert. The following operands are constant indices indicating
4836 the position at which to insert the value in a similar manner as indices in a
4837 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4838 value to insert must have the same type as the value identified by the
4842 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4843 that of <tt>val</tt> except that the value at the position specified by the
4844 indices is that of <tt>elt</tt>.</p>
4848 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4849 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4850 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4857 <!-- ======================================================================= -->
4859 <a name="memoryops">Memory Access and Addressing Operations</a>
4864 <p>A key design point of an SSA-based representation is how it represents
4865 memory. In LLVM, no memory locations are in SSA form, which makes things
4866 very simple. This section describes how to read, write, and allocate
4869 <!-- _______________________________________________________________________ -->
4871 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4878 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4882 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4883 currently executing function, to be automatically released when this function
4884 returns to its caller. The object is always allocated in the generic address
4885 space (address space zero).</p>
4888 <p>The '<tt>alloca</tt>' instruction
4889 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4890 runtime stack, returning a pointer of the appropriate type to the program.
4891 If "NumElements" is specified, it is the number of elements allocated,
4892 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4893 specified, the value result of the allocation is guaranteed to be aligned to
4894 at least that boundary. If not specified, or if zero, the target can choose
4895 to align the allocation on any convenient boundary compatible with the
4898 <p>'<tt>type</tt>' may be any sized type.</p>
4901 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4902 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4903 memory is automatically released when the function returns. The
4904 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4905 variables that must have an address available. When the function returns
4906 (either with the <tt><a href="#i_ret">ret</a></tt>
4907 or <tt><a href="#i_resume">resume</a></tt> instructions), the memory is
4908 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4909 The order in which memory is allocated (ie., which way the stack grows) is
4916 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4917 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4918 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4919 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4924 <!-- _______________________________________________________________________ -->
4926 <a name="i_load">'<tt>load</tt>' Instruction</a>
4933 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4934 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4935 !<index> = !{ i32 1 }
4939 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4942 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4943 from which to load. The pointer must point to
4944 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4945 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4946 number or order of execution of this <tt>load</tt> with other <a
4947 href="#volatile">volatile operations</a>.</p>
4949 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
4950 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
4951 argument. The <code>release</code> and <code>acq_rel</code> orderings are
4952 not valid on <code>load</code> instructions. Atomic loads produce <a
4953 href="#memorymodel">defined</a> results when they may see multiple atomic
4954 stores. The type of the pointee must be an integer type whose bit width
4955 is a power of two greater than or equal to eight and less than or equal
4956 to a target-specific size limit. <code>align</code> must be explicitly
4957 specified on atomic loads, and the load has undefined behavior if the
4958 alignment is not set to a value which is at least the size in bytes of
4959 the pointee. <code>!nontemporal</code> does not have any defined semantics
4960 for atomic loads.</p>
4962 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4963 operation (that is, the alignment of the memory address). A value of 0 or an
4964 omitted <tt>align</tt> argument means that the operation has the preferential
4965 alignment for the target. It is the responsibility of the code emitter to
4966 ensure that the alignment information is correct. Overestimating the
4967 alignment results in undefined behavior. Underestimating the alignment may
4968 produce less efficient code. An alignment of 1 is always safe.</p>
4970 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4971 metatadata name <index> corresponding to a metadata node with
4972 one <tt>i32</tt> entry of value 1. The existence of
4973 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4974 and code generator that this load is not expected to be reused in the cache.
4975 The code generator may select special instructions to save cache bandwidth,
4976 such as the <tt>MOVNT</tt> instruction on x86.</p>
4978 <p>The optional <tt>!invariant.load</tt> metadata must reference a single
4979 metatadata name <index> corresponding to a metadata node with no
4980 entries. The existence of the <tt>!invariant.load</tt> metatadata on the
4981 instruction tells the optimizer and code generator that this load address
4982 points to memory which does not change value during program execution.
4983 The optimizer may then move this load around, for example, by hoisting it
4984 out of loops using loop invariant code motion.</p>
4987 <p>The location of memory pointed to is loaded. If the value being loaded is of
4988 scalar type then the number of bytes read does not exceed the minimum number
4989 of bytes needed to hold all bits of the type. For example, loading an
4990 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4991 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4992 is undefined if the value was not originally written using a store of the
4997 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4998 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4999 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
5004 <!-- _______________________________________________________________________ -->
5006 <a name="i_store">'<tt>store</tt>' Instruction</a>
5013 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
5014 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
5018 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
5021 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
5022 and an address at which to store it. The type of the
5023 '<tt><pointer></tt>' operand must be a pointer to
5024 the <a href="#t_firstclass">first class</a> type of the
5025 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
5026 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
5027 order of execution of this <tt>store</tt> with other <a
5028 href="#volatile">volatile operations</a>.</p>
5030 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
5031 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5032 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
5033 valid on <code>store</code> instructions. Atomic loads produce <a
5034 href="#memorymodel">defined</a> results when they may see multiple atomic
5035 stores. The type of the pointee must be an integer type whose bit width
5036 is a power of two greater than or equal to eight and less than or equal
5037 to a target-specific size limit. <code>align</code> must be explicitly
5038 specified on atomic stores, and the store has undefined behavior if the
5039 alignment is not set to a value which is at least the size in bytes of
5040 the pointee. <code>!nontemporal</code> does not have any defined semantics
5041 for atomic stores.</p>
5043 <p>The optional constant "align" argument specifies the alignment of the
5044 operation (that is, the alignment of the memory address). A value of 0 or an
5045 omitted "align" argument means that the operation has the preferential
5046 alignment for the target. It is the responsibility of the code emitter to
5047 ensure that the alignment information is correct. Overestimating the
5048 alignment results in an undefined behavior. Underestimating the alignment may
5049 produce less efficient code. An alignment of 1 is always safe.</p>
5051 <p>The optional !nontemporal metadata must reference a single metatadata
5052 name <index> corresponding to a metadata node with one i32 entry of
5053 value 1. The existence of the !nontemporal metatadata on the
5054 instruction tells the optimizer and code generator that this load is
5055 not expected to be reused in the cache. The code generator may
5056 select special instructions to save cache bandwidth, such as the
5057 MOVNT instruction on x86.</p>
5061 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
5062 location specified by the '<tt><pointer></tt>' operand. If
5063 '<tt><value></tt>' is of scalar type then the number of bytes written
5064 does not exceed the minimum number of bytes needed to hold all bits of the
5065 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
5066 writing a value of a type like <tt>i20</tt> with a size that is not an
5067 integral number of bytes, it is unspecified what happens to the extra bits
5068 that do not belong to the type, but they will typically be overwritten.</p>
5072 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5073 store i32 3, i32* %ptr <i>; yields {void}</i>
5074 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
5079 <!-- _______________________________________________________________________ -->
5081 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
5088 fence [singlethread] <ordering> <i>; yields {void}</i>
5092 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
5093 between operations.</p>
5095 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
5096 href="#ordering">ordering</a> argument which defines what
5097 <i>synchronizes-with</i> edges they add. They can only be given
5098 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
5099 <code>seq_cst</code> orderings.</p>
5102 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
5103 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
5104 <code>acquire</code> ordering semantics if and only if there exist atomic
5105 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
5106 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
5107 <var>X</var> modifies <var>M</var> (either directly or through some side effect
5108 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
5109 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
5110 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
5111 than an explicit <code>fence</code>, one (but not both) of the atomic operations
5112 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
5113 <code>acquire</code> (resp.) ordering constraint and still
5114 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
5115 <i>happens-before</i> edge.</p>
5117 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
5118 having both <code>acquire</code> and <code>release</code> semantics specified
5119 above, participates in the global program order of other <code>seq_cst</code>
5120 operations and/or fences.</p>
5122 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
5123 specifies that the fence only synchronizes with other fences in the same
5124 thread. (This is useful for interacting with signal handlers.)</p>
5128 fence acquire <i>; yields {void}</i>
5129 fence singlethread seq_cst <i>; yields {void}</i>
5134 <!-- _______________________________________________________________________ -->
5136 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
5143 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
5147 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
5148 It loads a value in memory and compares it to a given value. If they are
5149 equal, it stores a new value into the memory.</p>
5152 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
5153 address to operate on, a value to compare to the value currently be at that
5154 address, and a new value to place at that address if the compared values are
5155 equal. The type of '<var><cmp></var>' must be an integer type whose
5156 bit width is a power of two greater than or equal to eight and less than
5157 or equal to a target-specific size limit. '<var><cmp></var>' and
5158 '<var><new></var>' must have the same type, and the type of
5159 '<var><pointer></var>' must be a pointer to that type. If the
5160 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
5161 optimizer is not allowed to modify the number or order of execution
5162 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
5165 <!-- FIXME: Extend allowed types. -->
5167 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
5168 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
5170 <p>The optional "<code>singlethread</code>" argument declares that the
5171 <code>cmpxchg</code> is only atomic with respect to code (usually signal
5172 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
5173 cmpxchg is atomic with respect to all other code in the system.</p>
5175 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
5176 the size in memory of the operand.
5179 <p>The contents of memory at the location specified by the
5180 '<tt><pointer></tt>' operand is read and compared to
5181 '<tt><cmp></tt>'; if the read value is the equal,
5182 '<tt><new></tt>' is written. The original value at the location
5185 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
5186 purpose of identifying <a href="#release_sequence">release sequences</a>. A
5187 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
5188 parameter determined by dropping any <code>release</code> part of the
5189 <code>cmpxchg</code>'s ordering.</p>
5192 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
5193 optimization work on ARM.)
5195 FIXME: Is a weaker ordering constraint on failure helpful in practice?
5201 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
5202 <a href="#i_br">br</a> label %loop
5205 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
5206 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
5207 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
5208 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
5209 <a href="#i_br">br</a> i1 %success, label %done, label %loop
5217 <!-- _______________________________________________________________________ -->
5219 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
5226 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
5230 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
5233 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
5234 operation to apply, an address whose value to modify, an argument to the
5235 operation. The operation must be one of the following keywords:</p>
5250 <p>The type of '<var><value></var>' must be an integer type whose
5251 bit width is a power of two greater than or equal to eight and less than
5252 or equal to a target-specific size limit. The type of the
5253 '<code><pointer></code>' operand must be a pointer to that type.
5254 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
5255 optimizer is not allowed to modify the number or order of execution of this
5256 <code>atomicrmw</code> with other <a href="#volatile">volatile
5259 <!-- FIXME: Extend allowed types. -->
5262 <p>The contents of memory at the location specified by the
5263 '<tt><pointer></tt>' operand are atomically read, modified, and written
5264 back. The original value at the location is returned. The modification is
5265 specified by the <var>operation</var> argument:</p>
5268 <li>xchg: <code>*ptr = val</code></li>
5269 <li>add: <code>*ptr = *ptr + val</code></li>
5270 <li>sub: <code>*ptr = *ptr - val</code></li>
5271 <li>and: <code>*ptr = *ptr & val</code></li>
5272 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
5273 <li>or: <code>*ptr = *ptr | val</code></li>
5274 <li>xor: <code>*ptr = *ptr ^ val</code></li>
5275 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
5276 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
5277 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
5278 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
5283 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
5288 <!-- _______________________________________________________________________ -->
5290 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
5297 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5298 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5299 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5303 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
5304 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
5305 It performs address calculation only and does not access memory.</p>
5308 <p>The first argument is always a pointer or a vector of pointers,
5309 and forms the basis of the
5310 calculation. The remaining arguments are indices that indicate which of the
5311 elements of the aggregate object are indexed. The interpretation of each
5312 index is dependent on the type being indexed into. The first index always
5313 indexes the pointer value given as the first argument, the second index
5314 indexes a value of the type pointed to (not necessarily the value directly
5315 pointed to, since the first index can be non-zero), etc. The first type
5316 indexed into must be a pointer value, subsequent types can be arrays,
5317 vectors, and structs. Note that subsequent types being indexed into
5318 can never be pointers, since that would require loading the pointer before
5319 continuing calculation.</p>
5321 <p>The type of each index argument depends on the type it is indexing into.
5322 When indexing into a (optionally packed) structure, only <tt>i32</tt>
5323 integer <b>constants</b> are allowed. When indexing into an array, pointer
5324 or vector, integers of any width are allowed, and they are not required to be
5325 constant. These integers are treated as signed values where relevant.</p>
5327 <p>For example, let's consider a C code fragment and how it gets compiled to
5330 <pre class="doc_code">
5342 int *foo(struct ST *s) {
5343 return &s[1].Z.B[5][13];
5347 <p>The LLVM code generated by Clang is:</p>
5349 <pre class="doc_code">
5350 %struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 }
5351 %struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT }
5353 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5355 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5361 <p>In the example above, the first index is indexing into the
5362 '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a
5363 '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a
5364 structure. The second index indexes into the third element of the structure,
5365 yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>'
5366 type, another structure. The third index indexes into the second element of
5367 the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The
5368 two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>'
5369 type. The '<tt>getelementptr</tt>' instruction returns a pointer to this
5370 element, thus computing a value of '<tt>i32*</tt>' type.</p>
5372 <p>Note that it is perfectly legal to index partially through a structure,
5373 returning a pointer to an inner element. Because of this, the LLVM code for
5374 the given testcase is equivalent to:</p>
5376 <pre class="doc_code">
5377 define i32* @foo(%struct.ST* %s) {
5378 %t1 = getelementptr %struct.ST* %s, i32 1 <i>; yields %struct.ST*:%t1</i>
5379 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 <i>; yields %struct.RT*:%t2</i>
5380 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5381 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5382 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5387 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5388 <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the
5389 base pointer is not an <i>in bounds</i> address of an allocated object,
5390 or if any of the addresses that would be formed by successive addition of
5391 the offsets implied by the indices to the base address with infinitely
5392 precise signed arithmetic are not an <i>in bounds</i> address of that
5393 allocated object. The <i>in bounds</i> addresses for an allocated object
5394 are all the addresses that point into the object, plus the address one
5396 In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword
5397 applies to each of the computations element-wise. </p>
5399 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5400 the base address with silently-wrapping two's complement arithmetic. If the
5401 offsets have a different width from the pointer, they are sign-extended or
5402 truncated to the width of the pointer. The result value of the
5403 <tt>getelementptr</tt> may be outside the object pointed to by the base
5404 pointer. The result value may not necessarily be used to access memory
5405 though, even if it happens to point into allocated storage. See the
5406 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5409 <p>The getelementptr instruction is often confusing. For some more insight into
5410 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5414 <i>; yields [12 x i8]*:aptr</i>
5415 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5416 <i>; yields i8*:vptr</i>
5417 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5418 <i>; yields i8*:eptr</i>
5419 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5420 <i>; yields i32*:iptr</i>
5421 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5424 <p>In cases where the pointer argument is a vector of pointers, only a
5425 single index may be used, and the number of vector elements has to be
5426 the same. For example: </p>
5427 <pre class="doc_code">
5428 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5435 <!-- ======================================================================= -->
5437 <a name="convertops">Conversion Operations</a>
5442 <p>The instructions in this category are the conversion instructions (casting)
5443 which all take a single operand and a type. They perform various bit
5444 conversions on the operand.</p>
5446 <!-- _______________________________________________________________________ -->
5448 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5455 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5459 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5460 type <tt>ty2</tt>.</p>
5463 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5464 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5465 of the same number of integers.
5466 The bit size of the <tt>value</tt> must be larger than
5467 the bit size of the destination type, <tt>ty2</tt>.
5468 Equal sized types are not allowed.</p>
5471 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5472 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5473 source size must be larger than the destination size, <tt>trunc</tt> cannot
5474 be a <i>no-op cast</i>. It will always truncate bits.</p>
5478 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5479 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5480 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5481 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5486 <!-- _______________________________________________________________________ -->
5488 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5495 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5499 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5504 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5505 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5506 of the same number of integers.
5507 The bit size of the <tt>value</tt> must be smaller than
5508 the bit size of the destination type,
5512 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5513 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5515 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5519 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5520 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5521 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5526 <!-- _______________________________________________________________________ -->
5528 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5535 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5539 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5542 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5543 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5544 of the same number of integers.
5545 The bit size of the <tt>value</tt> must be smaller than
5546 the bit size of the destination type,
5550 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5551 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5552 of the type <tt>ty2</tt>.</p>
5554 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5558 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5559 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5560 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5565 <!-- _______________________________________________________________________ -->
5567 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5574 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5578 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5582 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5583 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5584 to cast it to. The size of <tt>value</tt> must be larger than the size of
5585 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5586 <i>no-op cast</i>.</p>
5589 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5590 <a href="#t_floating">floating point</a> type to a smaller
5591 <a href="#t_floating">floating point</a> type. If the value cannot fit
5592 within the destination type, <tt>ty2</tt>, then the results are
5597 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5598 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5603 <!-- _______________________________________________________________________ -->
5605 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5612 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5616 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5617 floating point value.</p>
5620 <p>The '<tt>fpext</tt>' instruction takes a
5621 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5622 a <a href="#t_floating">floating point</a> type to cast it to. The source
5623 type must be smaller than the destination type.</p>
5626 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5627 <a href="#t_floating">floating point</a> type to a larger
5628 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5629 used to make a <i>no-op cast</i> because it always changes bits. Use
5630 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5634 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5635 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5640 <!-- _______________________________________________________________________ -->
5642 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5649 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5653 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5654 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5657 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5658 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5659 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5660 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5661 vector integer type with the same number of elements as <tt>ty</tt></p>
5664 <p>The '<tt>fptoui</tt>' instruction converts its
5665 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5666 towards zero) unsigned integer value. If the value cannot fit
5667 in <tt>ty2</tt>, the results are undefined.</p>
5671 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5672 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5673 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5678 <!-- _______________________________________________________________________ -->
5680 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5687 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5691 <p>The '<tt>fptosi</tt>' instruction converts
5692 <a href="#t_floating">floating point</a> <tt>value</tt> to
5693 type <tt>ty2</tt>.</p>
5696 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5697 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5698 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5699 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5700 vector integer type with the same number of elements as <tt>ty</tt></p>
5703 <p>The '<tt>fptosi</tt>' instruction converts its
5704 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5705 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5706 the results are undefined.</p>
5710 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5711 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5712 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5717 <!-- _______________________________________________________________________ -->
5719 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5726 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5730 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5731 integer and converts that value to the <tt>ty2</tt> type.</p>
5734 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5735 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5736 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5737 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5738 floating point type with the same number of elements as <tt>ty</tt></p>
5741 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5742 integer quantity and converts it to the corresponding floating point
5743 value. If the value cannot fit in the floating point value, the results are
5748 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5749 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5754 <!-- _______________________________________________________________________ -->
5756 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5763 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5767 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5768 and converts that value to the <tt>ty2</tt> type.</p>
5771 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5772 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5773 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5774 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5775 floating point type with the same number of elements as <tt>ty</tt></p>
5778 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5779 quantity and converts it to the corresponding floating point value. If the
5780 value cannot fit in the floating point value, the results are undefined.</p>
5784 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5785 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5790 <!-- _______________________________________________________________________ -->
5792 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5799 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5803 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of
5804 pointers <tt>value</tt> to
5805 the integer (or vector of integers) type <tt>ty2</tt>.</p>
5808 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5809 must be a a value of type <a href="#t_pointer">pointer</a> or a vector of
5810 pointers, and a type to cast it to
5811 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector
5812 of integers type.</p>
5815 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5816 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5817 truncating or zero extending that value to the size of the integer type. If
5818 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5819 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5820 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5825 %X = ptrtoint i32* %P to i8 <i>; yields truncation on 32-bit architecture</i>
5826 %Y = ptrtoint i32* %P to i64 <i>; yields zero extension on 32-bit architecture</i>
5827 %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>
5832 <!-- _______________________________________________________________________ -->
5834 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5841 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5845 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5846 pointer type, <tt>ty2</tt>.</p>
5849 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5850 value to cast, and a type to cast it to, which must be a
5851 <a href="#t_pointer">pointer</a> type.</p>
5854 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5855 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5856 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5857 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5858 than the size of a pointer then a zero extension is done. If they are the
5859 same size, nothing is done (<i>no-op cast</i>).</p>
5863 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5864 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5865 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5866 %Z = inttoptr <4 x i32> %G to <4 x i8*><i>; yields truncation of vector G to four pointers</i>
5871 <!-- _______________________________________________________________________ -->
5873 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5880 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5884 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5885 <tt>ty2</tt> without changing any bits.</p>
5888 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5889 non-aggregate first class value, and a type to cast it to, which must also be
5890 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5891 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5892 identical. If the source type is a pointer, the destination type must also be
5893 a pointer. This instruction supports bitwise conversion of vectors to
5894 integers and to vectors of other types (as long as they have the same
5898 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5899 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5900 this conversion. The conversion is done as if the <tt>value</tt> had been
5901 stored to memory and read back as type <tt>ty2</tt>.
5902 Pointer (or vector of pointers) types may only be converted to other pointer
5903 (or vector of pointers) types with this instruction. To convert
5904 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5905 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5909 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
5910 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
5911 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
5912 %Z = bitcast <2 x i32*> %V to <2 x i64*> <i>; yields <2 x i64*></i>
5919 <!-- ======================================================================= -->
5921 <a name="otherops">Other Operations</a>
5926 <p>The instructions in this category are the "miscellaneous" instructions, which
5927 defy better classification.</p>
5929 <!-- _______________________________________________________________________ -->
5931 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5938 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5942 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5943 boolean values based on comparison of its two integer, integer vector,
5944 pointer, or pointer vector operands.</p>
5947 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
5948 the condition code indicating the kind of comparison to perform. It is not a
5949 value, just a keyword. The possible condition code are:</p>
5952 <li><tt>eq</tt>: equal</li>
5953 <li><tt>ne</tt>: not equal </li>
5954 <li><tt>ugt</tt>: unsigned greater than</li>
5955 <li><tt>uge</tt>: unsigned greater or equal</li>
5956 <li><tt>ult</tt>: unsigned less than</li>
5957 <li><tt>ule</tt>: unsigned less or equal</li>
5958 <li><tt>sgt</tt>: signed greater than</li>
5959 <li><tt>sge</tt>: signed greater or equal</li>
5960 <li><tt>slt</tt>: signed less than</li>
5961 <li><tt>sle</tt>: signed less or equal</li>
5964 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
5965 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
5966 typed. They must also be identical types.</p>
5969 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
5970 condition code given as <tt>cond</tt>. The comparison performed always yields
5971 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
5972 result, as follows:</p>
5975 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
5976 <tt>false</tt> otherwise. No sign interpretation is necessary or
5979 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
5980 <tt>false</tt> otherwise. No sign interpretation is necessary or
5983 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
5984 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5986 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
5987 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5988 to <tt>op2</tt>.</li>
5990 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
5991 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5993 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
5994 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5996 <li><tt>sgt</tt>: interprets the operands as signed values and yields
5997 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5999 <li><tt>sge</tt>: interprets the operands as signed values and yields
6000 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6001 to <tt>op2</tt>.</li>
6003 <li><tt>slt</tt>: interprets the operands as signed values and yields
6004 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6006 <li><tt>sle</tt>: interprets the operands as signed values and yields
6007 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6010 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
6011 values are compared as if they were integers.</p>
6013 <p>If the operands are integer vectors, then they are compared element by
6014 element. The result is an <tt>i1</tt> vector with the same number of elements
6015 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
6019 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
6020 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
6021 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
6022 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
6023 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
6024 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
6027 <p>Note that the code generator does not yet support vector types with
6028 the <tt>icmp</tt> instruction.</p>
6032 <!-- _______________________________________________________________________ -->
6034 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
6041 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
6045 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
6046 values based on comparison of its operands.</p>
6048 <p>If the operands are floating point scalars, then the result type is a boolean
6049 (<a href="#t_integer"><tt>i1</tt></a>).</p>
6051 <p>If the operands are floating point vectors, then the result type is a vector
6052 of boolean with the same number of elements as the operands being
6056 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
6057 the condition code indicating the kind of comparison to perform. It is not a
6058 value, just a keyword. The possible condition code are:</p>
6061 <li><tt>false</tt>: no comparison, always returns false</li>
6062 <li><tt>oeq</tt>: ordered and equal</li>
6063 <li><tt>ogt</tt>: ordered and greater than </li>
6064 <li><tt>oge</tt>: ordered and greater than or equal</li>
6065 <li><tt>olt</tt>: ordered and less than </li>
6066 <li><tt>ole</tt>: ordered and less than or equal</li>
6067 <li><tt>one</tt>: ordered and not equal</li>
6068 <li><tt>ord</tt>: ordered (no nans)</li>
6069 <li><tt>ueq</tt>: unordered or equal</li>
6070 <li><tt>ugt</tt>: unordered or greater than </li>
6071 <li><tt>uge</tt>: unordered or greater than or equal</li>
6072 <li><tt>ult</tt>: unordered or less than </li>
6073 <li><tt>ule</tt>: unordered or less than or equal</li>
6074 <li><tt>une</tt>: unordered or not equal</li>
6075 <li><tt>uno</tt>: unordered (either nans)</li>
6076 <li><tt>true</tt>: no comparison, always returns true</li>
6079 <p><i>Ordered</i> means that neither operand is a QNAN while
6080 <i>unordered</i> means that either operand may be a QNAN.</p>
6082 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
6083 a <a href="#t_floating">floating point</a> type or
6084 a <a href="#t_vector">vector</a> of floating point type. They must have
6085 identical types.</p>
6088 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
6089 according to the condition code given as <tt>cond</tt>. If the operands are
6090 vectors, then the vectors are compared element by element. Each comparison
6091 performed always yields an <a href="#t_integer">i1</a> result, as
6095 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
6097 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6098 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6100 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6101 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6103 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6104 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6106 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6107 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6109 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6110 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6112 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6113 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6115 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
6117 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
6118 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6120 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
6121 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6123 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
6124 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6126 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
6127 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6129 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
6130 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6132 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
6133 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6135 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
6137 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
6142 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
6143 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
6144 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
6145 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
6148 <p>Note that the code generator does not yet support vector types with
6149 the <tt>fcmp</tt> instruction.</p>
6153 <!-- _______________________________________________________________________ -->
6155 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
6162 <result> = phi <ty> [ <val0>, <label0>], ...
6166 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
6167 SSA graph representing the function.</p>
6170 <p>The type of the incoming values is specified with the first type field. After
6171 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
6172 one pair for each predecessor basic block of the current block. Only values
6173 of <a href="#t_firstclass">first class</a> type may be used as the value
6174 arguments to the PHI node. Only labels may be used as the label
6177 <p>There must be no non-phi instructions between the start of a basic block and
6178 the PHI instructions: i.e. PHI instructions must be first in a basic
6181 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
6182 occur on the edge from the corresponding predecessor block to the current
6183 block (but after any definition of an '<tt>invoke</tt>' instruction's return
6184 value on the same edge).</p>
6187 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
6188 specified by the pair corresponding to the predecessor basic block that
6189 executed just prior to the current block.</p>
6193 Loop: ; Infinite loop that counts from 0 on up...
6194 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6195 %nextindvar = add i32 %indvar, 1
6201 <!-- _______________________________________________________________________ -->
6203 <a name="i_select">'<tt>select</tt>' Instruction</a>
6210 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
6212 <i>selty</i> is either i1 or {<N x i1>}
6216 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
6217 condition, without branching.</p>
6221 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
6222 values indicating the condition, and two values of the
6223 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
6224 vectors and the condition is a scalar, then entire vectors are selected, not
6225 individual elements.</p>
6228 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
6229 first value argument; otherwise, it returns the second value argument.</p>
6231 <p>If the condition is a vector of i1, then the value arguments must be vectors
6232 of the same size, and the selection is done element by element.</p>
6236 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
6241 <!-- _______________________________________________________________________ -->
6243 <a name="i_call">'<tt>call</tt>' Instruction</a>
6250 <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>]
6254 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
6257 <p>This instruction requires several arguments:</p>
6260 <li>The optional "tail" marker indicates that the callee function does not
6261 access any allocas or varargs in the caller. Note that calls may be
6262 marked "tail" even if they do not occur before
6263 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
6264 present, the function call is eligible for tail call optimization,
6265 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
6266 optimized into a jump</a>. The code generator may optimize calls marked
6267 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
6268 sibling call optimization</a> when the caller and callee have
6269 matching signatures, or 2) forced tail call optimization when the
6270 following extra requirements are met:
6272 <li>Caller and callee both have the calling
6273 convention <tt>fastcc</tt>.</li>
6274 <li>The call is in tail position (ret immediately follows call and ret
6275 uses value of call or is void).</li>
6276 <li>Option <tt>-tailcallopt</tt> is enabled,
6277 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
6278 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
6279 constraints are met.</a></li>
6283 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
6284 convention</a> the call should use. If none is specified, the call
6285 defaults to using C calling conventions. The calling convention of the
6286 call must match the calling convention of the target function, or else the
6287 behavior is undefined.</li>
6289 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
6290 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
6291 '<tt>inreg</tt>' attributes are valid here.</li>
6293 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
6294 type of the return value. Functions that return no value are marked
6295 <tt><a href="#t_void">void</a></tt>.</li>
6297 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
6298 being invoked. The argument types must match the types implied by this
6299 signature. This type can be omitted if the function is not varargs and if
6300 the function type does not return a pointer to a function.</li>
6302 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
6303 be invoked. In most cases, this is a direct function invocation, but
6304 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
6305 to function value.</li>
6307 <li>'<tt>function args</tt>': argument list whose types match the function
6308 signature argument types and parameter attributes. All arguments must be
6309 of <a href="#t_firstclass">first class</a> type. If the function
6310 signature indicates the function accepts a variable number of arguments,
6311 the extra arguments can be specified.</li>
6313 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
6314 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
6315 '<tt>readnone</tt>' attributes are valid here.</li>
6319 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
6320 a specified function, with its incoming arguments bound to the specified
6321 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
6322 function, control flow continues with the instruction after the function
6323 call, and the return value of the function is bound to the result
6328 %retval = call i32 @test(i32 %argc)
6329 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
6330 %X = tail call i32 @foo() <i>; yields i32</i>
6331 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
6332 call void %foo(i8 97 signext)
6334 %struct.A = type { i32, i8 }
6335 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
6336 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
6337 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
6338 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
6339 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
6342 <p>llvm treats calls to some functions with names and arguments that match the
6343 standard C99 library as being the C99 library functions, and may perform
6344 optimizations or generate code for them under that assumption. This is
6345 something we'd like to change in the future to provide better support for
6346 freestanding environments and non-C-based languages.</p>
6350 <!-- _______________________________________________________________________ -->
6352 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
6359 <resultval> = va_arg <va_list*> <arglist>, <argty>
6363 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6364 the "variable argument" area of a function call. It is used to implement the
6365 <tt>va_arg</tt> macro in C.</p>
6368 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6369 argument. It returns a value of the specified argument type and increments
6370 the <tt>va_list</tt> to point to the next argument. The actual type
6371 of <tt>va_list</tt> is target specific.</p>
6374 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6375 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6376 to the next argument. For more information, see the variable argument
6377 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6379 <p>It is legal for this instruction to be called in a function which does not
6380 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6383 <p><tt>va_arg</tt> is an LLVM instruction instead of
6384 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6388 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6390 <p>Note that the code generator does not yet fully support va_arg on many
6391 targets. Also, it does not currently support va_arg with aggregate types on
6396 <!-- _______________________________________________________________________ -->
6398 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6405 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6406 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6408 <clause> := catch <type> <value>
6409 <clause> := filter <array constant type> <array constant>
6413 <p>The '<tt>landingpad</tt>' instruction is used by
6414 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6415 system</a> to specify that a basic block is a landing pad — one where
6416 the exception lands, and corresponds to the code found in the
6417 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6418 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6419 re-entry to the function. The <tt>resultval</tt> has the
6420 type <tt>resultty</tt>.</p>
6423 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6424 function associated with the unwinding mechanism. The optional
6425 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6427 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6428 or <tt>filter</tt> — and contains the global variable representing the
6429 "type" that may be caught or filtered respectively. Unlike the
6430 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6431 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6432 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6433 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6436 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6437 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6438 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6439 calling conventions, how the personality function results are represented in
6440 LLVM IR is target specific.</p>
6442 <p>The clauses are applied in order from top to bottom. If two
6443 <tt>landingpad</tt> instructions are merged together through inlining, the
6444 clauses from the calling function are appended to the list of clauses.
6445 When the call stack is being unwound due to an exception being thrown, the
6446 exception is compared against each <tt>clause</tt> in turn. If it doesn't
6447 match any of the clauses, and the <tt>cleanup</tt> flag is not set, then
6448 unwinding continues further up the call stack.</p>
6450 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6453 <li>A landing pad block is a basic block which is the unwind destination of an
6454 '<tt>invoke</tt>' instruction.</li>
6455 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6456 first non-PHI instruction.</li>
6457 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6459 <li>A basic block that is not a landing pad block may not include a
6460 '<tt>landingpad</tt>' instruction.</li>
6461 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6462 personality function.</li>
6467 ;; A landing pad which can catch an integer.
6468 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6470 ;; A landing pad that is a cleanup.
6471 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6473 ;; A landing pad which can catch an integer and can only throw a double.
6474 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6476 filter [1 x i8**] [@_ZTId]
6485 <!-- *********************************************************************** -->
6486 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6487 <!-- *********************************************************************** -->
6491 <p>LLVM supports the notion of an "intrinsic function". These functions have
6492 well known names and semantics and are required to follow certain
6493 restrictions. Overall, these intrinsics represent an extension mechanism for
6494 the LLVM language that does not require changing all of the transformations
6495 in LLVM when adding to the language (or the bitcode reader/writer, the
6496 parser, etc...).</p>
6498 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6499 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6500 begin with this prefix. Intrinsic functions must always be external
6501 functions: you cannot define the body of intrinsic functions. Intrinsic
6502 functions may only be used in call or invoke instructions: it is illegal to
6503 take the address of an intrinsic function. Additionally, because intrinsic
6504 functions are part of the LLVM language, it is required if any are added that
6505 they be documented here.</p>
6507 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6508 family of functions that perform the same operation but on different data
6509 types. Because LLVM can represent over 8 million different integer types,
6510 overloading is used commonly to allow an intrinsic function to operate on any
6511 integer type. One or more of the argument types or the result type can be
6512 overloaded to accept any integer type. Argument types may also be defined as
6513 exactly matching a previous argument's type or the result type. This allows
6514 an intrinsic function which accepts multiple arguments, but needs all of them
6515 to be of the same type, to only be overloaded with respect to a single
6516 argument or the result.</p>
6518 <p>Overloaded intrinsics will have the names of its overloaded argument types
6519 encoded into its function name, each preceded by a period. Only those types
6520 which are overloaded result in a name suffix. Arguments whose type is matched
6521 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6522 can take an integer of any width and returns an integer of exactly the same
6523 integer width. This leads to a family of functions such as
6524 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6525 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6526 suffix is required. Because the argument's type is matched against the return
6527 type, it does not require its own name suffix.</p>
6529 <p>To learn how to add an intrinsic function, please see the
6530 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6532 <!-- ======================================================================= -->
6534 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6539 <p>Variable argument support is defined in LLVM with
6540 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6541 intrinsic functions. These functions are related to the similarly named
6542 macros defined in the <tt><stdarg.h></tt> header file.</p>
6544 <p>All of these functions operate on arguments that use a target-specific value
6545 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6546 not define what this type is, so all transformations should be prepared to
6547 handle these functions regardless of the type used.</p>
6549 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6550 instruction and the variable argument handling intrinsic functions are
6553 <pre class="doc_code">
6554 define i32 @test(i32 %X, ...) {
6555 ; Initialize variable argument processing
6557 %ap2 = bitcast i8** %ap to i8*
6558 call void @llvm.va_start(i8* %ap2)
6560 ; Read a single integer argument
6561 %tmp = va_arg i8** %ap, i32
6563 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6565 %aq2 = bitcast i8** %aq to i8*
6566 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6567 call void @llvm.va_end(i8* %aq2)
6569 ; Stop processing of arguments.
6570 call void @llvm.va_end(i8* %ap2)
6574 declare void @llvm.va_start(i8*)
6575 declare void @llvm.va_copy(i8*, i8*)
6576 declare void @llvm.va_end(i8*)
6579 <!-- _______________________________________________________________________ -->
6581 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6589 declare void %llvm.va_start(i8* <arglist>)
6593 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6594 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6597 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6600 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6601 macro available in C. In a target-dependent way, it initializes
6602 the <tt>va_list</tt> element to which the argument points, so that the next
6603 call to <tt>va_arg</tt> will produce the first variable argument passed to
6604 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6605 need to know the last argument of the function as the compiler can figure
6610 <!-- _______________________________________________________________________ -->
6612 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6619 declare void @llvm.va_end(i8* <arglist>)
6623 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6624 which has been initialized previously
6625 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6626 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6629 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6632 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6633 macro available in C. In a target-dependent way, it destroys
6634 the <tt>va_list</tt> element to which the argument points. Calls
6635 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6636 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6637 with calls to <tt>llvm.va_end</tt>.</p>
6641 <!-- _______________________________________________________________________ -->
6643 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6650 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6654 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6655 from the source argument list to the destination argument list.</p>
6658 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6659 The second argument is a pointer to a <tt>va_list</tt> element to copy
6663 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6664 macro available in C. In a target-dependent way, it copies the
6665 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6666 element. This intrinsic is necessary because
6667 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6668 arbitrarily complex and require, for example, memory allocation.</p>
6674 <!-- ======================================================================= -->
6676 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6681 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6682 Collection</a> (GC) requires the implementation and generation of these
6683 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6684 roots on the stack</a>, as well as garbage collector implementations that
6685 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6686 barriers. Front-ends for type-safe garbage collected languages should generate
6687 these intrinsics to make use of the LLVM garbage collectors. For more details,
6688 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6691 <p>The garbage collection intrinsics only operate on objects in the generic
6692 address space (address space zero).</p>
6694 <!-- _______________________________________________________________________ -->
6696 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6703 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6707 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6708 the code generator, and allows some metadata to be associated with it.</p>
6711 <p>The first argument specifies the address of a stack object that contains the
6712 root pointer. The second pointer (which must be either a constant or a
6713 global value address) contains the meta-data to be associated with the
6717 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6718 location. At compile-time, the code generator generates information to allow
6719 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6720 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6725 <!-- _______________________________________________________________________ -->
6727 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6734 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6738 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6739 locations, allowing garbage collector implementations that require read
6743 <p>The second argument is the address to read from, which should be an address
6744 allocated from the garbage collector. The first object is a pointer to the
6745 start of the referenced object, if needed by the language runtime (otherwise
6749 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6750 instruction, but may be replaced with substantially more complex code by the
6751 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6752 may only be used in a function which <a href="#gc">specifies a GC
6757 <!-- _______________________________________________________________________ -->
6759 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6766 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6770 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6771 locations, allowing garbage collector implementations that require write
6772 barriers (such as generational or reference counting collectors).</p>
6775 <p>The first argument is the reference to store, the second is the start of the
6776 object to store it to, and the third is the address of the field of Obj to
6777 store to. If the runtime does not require a pointer to the object, Obj may
6781 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6782 instruction, but may be replaced with substantially more complex code by the
6783 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6784 may only be used in a function which <a href="#gc">specifies a GC
6791 <!-- ======================================================================= -->
6793 <a name="int_codegen">Code Generator Intrinsics</a>
6798 <p>These intrinsics are provided by LLVM to expose special features that may
6799 only be implemented with code generator support.</p>
6801 <!-- _______________________________________________________________________ -->
6803 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6810 declare i8 *@llvm.returnaddress(i32 <level>)
6814 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6815 target-specific value indicating the return address of the current function
6816 or one of its callers.</p>
6819 <p>The argument to this intrinsic indicates which function to return the address
6820 for. Zero indicates the calling function, one indicates its caller, etc.
6821 The argument is <b>required</b> to be a constant integer value.</p>
6824 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6825 indicating the return address of the specified call frame, or zero if it
6826 cannot be identified. The value returned by this intrinsic is likely to be
6827 incorrect or 0 for arguments other than zero, so it should only be used for
6828 debugging purposes.</p>
6830 <p>Note that calling this intrinsic does not prevent function inlining or other
6831 aggressive transformations, so the value returned may not be that of the
6832 obvious source-language caller.</p>
6836 <!-- _______________________________________________________________________ -->
6838 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6845 declare i8* @llvm.frameaddress(i32 <level>)
6849 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6850 target-specific frame pointer value for the specified stack frame.</p>
6853 <p>The argument to this intrinsic indicates which function to return the frame
6854 pointer for. Zero indicates the calling function, one indicates its caller,
6855 etc. The argument is <b>required</b> to be a constant integer value.</p>
6858 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6859 indicating the frame address of the specified call frame, or zero if it
6860 cannot be identified. The value returned by this intrinsic is likely to be
6861 incorrect or 0 for arguments other than zero, so it should only be used for
6862 debugging purposes.</p>
6864 <p>Note that calling this intrinsic does not prevent function inlining or other
6865 aggressive transformations, so the value returned may not be that of the
6866 obvious source-language caller.</p>
6870 <!-- _______________________________________________________________________ -->
6872 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6879 declare i8* @llvm.stacksave()
6883 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6884 of the function stack, for use
6885 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6886 useful for implementing language features like scoped automatic variable
6887 sized arrays in C99.</p>
6890 <p>This intrinsic returns a opaque pointer value that can be passed
6891 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6892 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6893 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6894 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6895 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6896 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6900 <!-- _______________________________________________________________________ -->
6902 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6909 declare void @llvm.stackrestore(i8* %ptr)
6913 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6914 the function stack to the state it was in when the
6915 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6916 executed. This is useful for implementing language features like scoped
6917 automatic variable sized arrays in C99.</p>
6920 <p>See the description
6921 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6925 <!-- _______________________________________________________________________ -->
6927 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6934 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6938 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6939 insert a prefetch instruction if supported; otherwise, it is a noop.
6940 Prefetches have no effect on the behavior of the program but can change its
6941 performance characteristics.</p>
6944 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6945 specifier determining if the fetch should be for a read (0) or write (1),
6946 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
6947 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6948 specifies whether the prefetch is performed on the data (1) or instruction (0)
6949 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
6950 must be constant integers.</p>
6953 <p>This intrinsic does not modify the behavior of the program. In particular,
6954 prefetches cannot trap and do not produce a value. On targets that support
6955 this intrinsic, the prefetch can provide hints to the processor cache for
6956 better performance.</p>
6960 <!-- _______________________________________________________________________ -->
6962 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
6969 declare void @llvm.pcmarker(i32 <id>)
6973 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
6974 Counter (PC) in a region of code to simulators and other tools. The method
6975 is target specific, but it is expected that the marker will use exported
6976 symbols to transmit the PC of the marker. The marker makes no guarantees
6977 that it will remain with any specific instruction after optimizations. It is
6978 possible that the presence of a marker will inhibit optimizations. The
6979 intended use is to be inserted after optimizations to allow correlations of
6980 simulation runs.</p>
6983 <p><tt>id</tt> is a numerical id identifying the marker.</p>
6986 <p>This intrinsic does not modify the behavior of the program. Backends that do
6987 not support this intrinsic may ignore it.</p>
6991 <!-- _______________________________________________________________________ -->
6993 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
7000 declare i64 @llvm.readcyclecounter()
7004 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
7005 counter register (or similar low latency, high accuracy clocks) on those
7006 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7007 should map to RPCC. As the backing counters overflow quickly (on the order
7008 of 9 seconds on alpha), this should only be used for small timings.</p>
7011 <p>When directly supported, reading the cycle counter should not modify any
7012 memory. Implementations are allowed to either return a application specific
7013 value or a system wide value. On backends without support, this is lowered
7014 to a constant 0.</p>
7020 <!-- ======================================================================= -->
7022 <a name="int_libc">Standard C Library Intrinsics</a>
7027 <p>LLVM provides intrinsics for a few important standard C library functions.
7028 These intrinsics allow source-language front-ends to pass information about
7029 the alignment of the pointer arguments to the code generator, providing
7030 opportunity for more efficient code generation.</p>
7032 <!-- _______________________________________________________________________ -->
7034 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
7040 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
7041 integer bit width and for different address spaces. Not all targets support
7042 all bit widths however.</p>
7045 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7046 i32 <len>, i32 <align>, i1 <isvolatile>)
7047 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7048 i64 <len>, i32 <align>, i1 <isvolatile>)
7052 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7053 source location to the destination location.</p>
7055 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
7056 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7057 and the pointers can be in specified address spaces.</p>
7061 <p>The first argument is a pointer to the destination, the second is a pointer
7062 to the source. The third argument is an integer argument specifying the
7063 number of bytes to copy, the fourth argument is the alignment of the
7064 source and destination locations, and the fifth is a boolean indicating a
7065 volatile access.</p>
7067 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7068 then the caller guarantees that both the source and destination pointers are
7069 aligned to that boundary.</p>
7071 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7072 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
7073 The detailed access behavior is not very cleanly specified and it is unwise
7074 to depend on it.</p>
7078 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7079 source location to the destination location, which are not allowed to
7080 overlap. It copies "len" bytes of memory over. If the argument is known to
7081 be aligned to some boundary, this can be specified as the fourth argument,
7082 otherwise it should be set to 0 or 1.</p>
7086 <!-- _______________________________________________________________________ -->
7088 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
7094 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
7095 width and for different address space. Not all targets support all bit
7099 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7100 i32 <len>, i32 <align>, i1 <isvolatile>)
7101 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7102 i64 <len>, i32 <align>, i1 <isvolatile>)
7106 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
7107 source location to the destination location. It is similar to the
7108 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
7111 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
7112 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7113 and the pointers can be in specified address spaces.</p>
7117 <p>The first argument is a pointer to the destination, the second is a pointer
7118 to the source. The third argument is an integer argument specifying the
7119 number of bytes to copy, the fourth argument is the alignment of the
7120 source and destination locations, and the fifth is a boolean indicating a
7121 volatile access.</p>
7123 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7124 then the caller guarantees that the source and destination pointers are
7125 aligned to that boundary.</p>
7127 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7128 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
7129 The detailed access behavior is not very cleanly specified and it is unwise
7130 to depend on it.</p>
7134 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
7135 source location to the destination location, which may overlap. It copies
7136 "len" bytes of memory over. If the argument is known to be aligned to some
7137 boundary, this can be specified as the fourth argument, otherwise it should
7138 be set to 0 or 1.</p>
7142 <!-- _______________________________________________________________________ -->
7144 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
7150 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
7151 width and for different address spaces. However, not all targets support all
7155 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7156 i32 <len>, i32 <align>, i1 <isvolatile>)
7157 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7158 i64 <len>, i32 <align>, i1 <isvolatile>)
7162 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
7163 particular byte value.</p>
7165 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
7166 intrinsic does not return a value and takes extra alignment/volatile
7167 arguments. Also, the destination can be in an arbitrary address space.</p>
7170 <p>The first argument is a pointer to the destination to fill, the second is the
7171 byte value with which to fill it, the third argument is an integer argument
7172 specifying the number of bytes to fill, and the fourth argument is the known
7173 alignment of the destination location.</p>
7175 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7176 then the caller guarantees that the destination pointer is aligned to that
7179 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7180 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
7181 The detailed access behavior is not very cleanly specified and it is unwise
7182 to depend on it.</p>
7185 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
7186 at the destination location. If the argument is known to be aligned to some
7187 boundary, this can be specified as the fourth argument, otherwise it should
7188 be set to 0 or 1.</p>
7192 <!-- _______________________________________________________________________ -->
7194 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
7200 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
7201 floating point or vector of floating point type. Not all targets support all
7205 declare float @llvm.sqrt.f32(float %Val)
7206 declare double @llvm.sqrt.f64(double %Val)
7207 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7208 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7209 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7213 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
7214 returning the same value as the libm '<tt>sqrt</tt>' functions would.
7215 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
7216 behavior for negative numbers other than -0.0 (which allows for better
7217 optimization, because there is no need to worry about errno being
7218 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
7221 <p>The argument and return value are floating point numbers of the same
7225 <p>This function returns the sqrt of the specified operand if it is a
7226 nonnegative floating point number.</p>
7230 <!-- _______________________________________________________________________ -->
7232 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
7238 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
7239 floating point or vector of floating point type. Not all targets support all
7243 declare float @llvm.powi.f32(float %Val, i32 %power)
7244 declare double @llvm.powi.f64(double %Val, i32 %power)
7245 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7246 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7247 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7251 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
7252 specified (positive or negative) power. The order of evaluation of
7253 multiplications is not defined. When a vector of floating point type is
7254 used, the second argument remains a scalar integer value.</p>
7257 <p>The second argument is an integer power, and the first is a value to raise to
7261 <p>This function returns the first value raised to the second power with an
7262 unspecified sequence of rounding operations.</p>
7266 <!-- _______________________________________________________________________ -->
7268 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
7274 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
7275 floating point or vector of floating point type. Not all targets support all
7279 declare float @llvm.sin.f32(float %Val)
7280 declare double @llvm.sin.f64(double %Val)
7281 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7282 declare fp128 @llvm.sin.f128(fp128 %Val)
7283 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7287 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
7290 <p>The argument and return value are floating point numbers of the same
7294 <p>This function returns the sine of the specified operand, returning the same
7295 values as the libm <tt>sin</tt> functions would, and handles error conditions
7296 in the same way.</p>
7300 <!-- _______________________________________________________________________ -->
7302 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
7308 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
7309 floating point or vector of floating point type. Not all targets support all
7313 declare float @llvm.cos.f32(float %Val)
7314 declare double @llvm.cos.f64(double %Val)
7315 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7316 declare fp128 @llvm.cos.f128(fp128 %Val)
7317 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7321 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
7324 <p>The argument and return value are floating point numbers of the same
7328 <p>This function returns the cosine of the specified operand, returning the same
7329 values as the libm <tt>cos</tt> functions would, and handles error conditions
7330 in the same way.</p>
7334 <!-- _______________________________________________________________________ -->
7336 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
7342 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
7343 floating point or vector of floating point type. Not all targets support all
7347 declare float @llvm.pow.f32(float %Val, float %Power)
7348 declare double @llvm.pow.f64(double %Val, double %Power)
7349 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7350 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7351 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7355 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
7356 specified (positive or negative) power.</p>
7359 <p>The second argument is a floating point power, and the first is a value to
7360 raise to that power.</p>
7363 <p>This function returns the first value raised to the second power, returning
7364 the same values as the libm <tt>pow</tt> functions would, and handles error
7365 conditions in the same way.</p>
7369 <!-- _______________________________________________________________________ -->
7371 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7377 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7378 floating point or vector of floating point type. Not all targets support all
7382 declare float @llvm.exp.f32(float %Val)
7383 declare double @llvm.exp.f64(double %Val)
7384 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7385 declare fp128 @llvm.exp.f128(fp128 %Val)
7386 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7390 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7393 <p>The argument and return value are floating point numbers of the same
7397 <p>This function returns the same values as the libm <tt>exp</tt> functions
7398 would, and handles error conditions in the same way.</p>
7402 <!-- _______________________________________________________________________ -->
7404 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7410 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7411 floating point or vector of floating point type. Not all targets support all
7415 declare float @llvm.log.f32(float %Val)
7416 declare double @llvm.log.f64(double %Val)
7417 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7418 declare fp128 @llvm.log.f128(fp128 %Val)
7419 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7423 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7426 <p>The argument and return value are floating point numbers of the same
7430 <p>This function returns the same values as the libm <tt>log</tt> functions
7431 would, and handles error conditions in the same way.</p>
7435 <!-- _______________________________________________________________________ -->
7437 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7443 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7444 floating point or vector of floating point type. Not all targets support all
7448 declare float @llvm.fma.f32(float %a, float %b, float %c)
7449 declare double @llvm.fma.f64(double %a, double %b, double %c)
7450 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7451 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7452 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7456 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7460 <p>The argument and return value are floating point numbers of the same
7464 <p>This function returns the same values as the libm <tt>fma</tt> functions
7471 <!-- ======================================================================= -->
7473 <a name="int_manip">Bit Manipulation Intrinsics</a>
7478 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7479 These allow efficient code generation for some algorithms.</p>
7481 <!-- _______________________________________________________________________ -->
7483 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7489 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7490 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7493 declare i16 @llvm.bswap.i16(i16 <id>)
7494 declare i32 @llvm.bswap.i32(i32 <id>)
7495 declare i64 @llvm.bswap.i64(i64 <id>)
7499 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7500 values with an even number of bytes (positive multiple of 16 bits). These
7501 are useful for performing operations on data that is not in the target's
7502 native byte order.</p>
7505 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7506 and low byte of the input i16 swapped. Similarly,
7507 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7508 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7509 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7510 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7511 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7512 more, respectively).</p>
7516 <!-- _______________________________________________________________________ -->
7518 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7524 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7525 width, or on any vector with integer elements. Not all targets support all
7526 bit widths or vector types, however.</p>
7529 declare i8 @llvm.ctpop.i8(i8 <src>)
7530 declare i16 @llvm.ctpop.i16(i16 <src>)
7531 declare i32 @llvm.ctpop.i32(i32 <src>)
7532 declare i64 @llvm.ctpop.i64(i64 <src>)
7533 declare i256 @llvm.ctpop.i256(i256 <src>)
7534 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7538 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7542 <p>The only argument is the value to be counted. The argument may be of any
7543 integer type, or a vector with integer elements.
7544 The return type must match the argument type.</p>
7547 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7548 element of a vector.</p>
7552 <!-- _______________________________________________________________________ -->
7554 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7560 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7561 integer bit width, or any vector whose elements are integers. Not all
7562 targets support all bit widths or vector types, however.</p>
7565 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7566 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7567 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7568 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7569 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7570 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7574 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7575 leading zeros in a variable.</p>
7578 <p>The first argument is the value to be counted. This argument may be of any
7579 integer type, or a vectory with integer element type. The return type
7580 must match the first argument type.</p>
7582 <p>The second argument must be a constant and is a flag to indicate whether the
7583 intrinsic should ensure that a zero as the first argument produces a defined
7584 result. Historically some architectures did not provide a defined result for
7585 zero values as efficiently, and many algorithms are now predicated on
7586 avoiding zero-value inputs.</p>
7589 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7590 zeros in a variable, or within each element of the vector.
7591 If <tt>src == 0</tt> then the result is the size in bits of the type of
7592 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7593 For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7597 <!-- _______________________________________________________________________ -->
7599 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7605 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7606 integer bit width, or any vector of integer elements. Not all targets
7607 support all bit widths or vector types, however.</p>
7610 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7611 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7612 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7613 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7614 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7615 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7619 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7623 <p>The first argument is the value to be counted. This argument may be of any
7624 integer type, or a vectory with integer element type. The return type
7625 must match the first argument type.</p>
7627 <p>The second argument must be a constant and is a flag to indicate whether the
7628 intrinsic should ensure that a zero as the first argument produces a defined
7629 result. Historically some architectures did not provide a defined result for
7630 zero values as efficiently, and many algorithms are now predicated on
7631 avoiding zero-value inputs.</p>
7634 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7635 zeros in a variable, or within each element of a vector.
7636 If <tt>src == 0</tt> then the result is the size in bits of the type of
7637 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7638 For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7644 <!-- ======================================================================= -->
7646 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7651 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7653 <!-- _______________________________________________________________________ -->
7655 <a name="int_sadd_overflow">
7656 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7663 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7664 on any integer bit width.</p>
7667 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7668 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7669 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7673 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7674 a signed addition of the two arguments, and indicate whether an overflow
7675 occurred during the signed summation.</p>
7678 <p>The arguments (%a and %b) and the first element of the result structure may
7679 be of integer types of any bit width, but they must have the same bit
7680 width. The second element of the result structure must be of
7681 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7682 undergo signed addition.</p>
7685 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7686 a signed addition of the two variables. They return a structure — the
7687 first element of which is the signed summation, and the second element of
7688 which is a bit specifying if the signed summation resulted in an
7693 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7694 %sum = extractvalue {i32, i1} %res, 0
7695 %obit = extractvalue {i32, i1} %res, 1
7696 br i1 %obit, label %overflow, label %normal
7701 <!-- _______________________________________________________________________ -->
7703 <a name="int_uadd_overflow">
7704 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7711 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7712 on any integer bit width.</p>
7715 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7716 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7717 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7721 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7722 an unsigned addition of the two arguments, and indicate whether a carry
7723 occurred during the unsigned summation.</p>
7726 <p>The arguments (%a and %b) and the first element of the result structure may
7727 be of integer types of any bit width, but they must have the same bit
7728 width. The second element of the result structure must be of
7729 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7730 undergo unsigned addition.</p>
7733 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7734 an unsigned addition of the two arguments. They return a structure —
7735 the first element of which is the sum, and the second element of which is a
7736 bit specifying if the unsigned summation resulted in a carry.</p>
7740 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7741 %sum = extractvalue {i32, i1} %res, 0
7742 %obit = extractvalue {i32, i1} %res, 1
7743 br i1 %obit, label %carry, label %normal
7748 <!-- _______________________________________________________________________ -->
7750 <a name="int_ssub_overflow">
7751 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7758 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7759 on any integer bit width.</p>
7762 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7763 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7764 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7768 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7769 a signed subtraction of the two arguments, and indicate whether an overflow
7770 occurred during the signed subtraction.</p>
7773 <p>The arguments (%a and %b) and the first element of the result structure may
7774 be of integer types of any bit width, but they must have the same bit
7775 width. The second element of the result structure must be of
7776 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7777 undergo signed subtraction.</p>
7780 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7781 a signed subtraction of the two arguments. They return a structure —
7782 the first element of which is the subtraction, and the second element of
7783 which is a bit specifying if the signed subtraction resulted in an
7788 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7789 %sum = extractvalue {i32, i1} %res, 0
7790 %obit = extractvalue {i32, i1} %res, 1
7791 br i1 %obit, label %overflow, label %normal
7796 <!-- _______________________________________________________________________ -->
7798 <a name="int_usub_overflow">
7799 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7806 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7807 on any integer bit width.</p>
7810 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7811 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7812 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7816 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7817 an unsigned subtraction of the two arguments, and indicate whether an
7818 overflow occurred during the unsigned subtraction.</p>
7821 <p>The arguments (%a and %b) and the first element of the result structure may
7822 be of integer types of any bit width, but they must have the same bit
7823 width. The second element of the result structure must be of
7824 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7825 undergo unsigned subtraction.</p>
7828 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7829 an unsigned subtraction of the two arguments. They return a structure —
7830 the first element of which is the subtraction, and the second element of
7831 which is a bit specifying if the unsigned subtraction resulted in an
7836 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7837 %sum = extractvalue {i32, i1} %res, 0
7838 %obit = extractvalue {i32, i1} %res, 1
7839 br i1 %obit, label %overflow, label %normal
7844 <!-- _______________________________________________________________________ -->
7846 <a name="int_smul_overflow">
7847 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7854 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7855 on any integer bit width.</p>
7858 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7859 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7860 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7865 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7866 a signed multiplication of the two arguments, and indicate whether an
7867 overflow occurred during the signed multiplication.</p>
7870 <p>The arguments (%a and %b) and the first element of the result structure may
7871 be of integer types of any bit width, but they must have the same bit
7872 width. The second element of the result structure must be of
7873 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7874 undergo signed multiplication.</p>
7877 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7878 a signed multiplication of the two arguments. They return a structure —
7879 the first element of which is the multiplication, and the second element of
7880 which is a bit specifying if the signed multiplication resulted in an
7885 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7886 %sum = extractvalue {i32, i1} %res, 0
7887 %obit = extractvalue {i32, i1} %res, 1
7888 br i1 %obit, label %overflow, label %normal
7893 <!-- _______________________________________________________________________ -->
7895 <a name="int_umul_overflow">
7896 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
7903 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
7904 on any integer bit width.</p>
7907 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7908 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7909 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7913 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7914 a unsigned multiplication of the two arguments, and indicate whether an
7915 overflow occurred during the unsigned multiplication.</p>
7918 <p>The arguments (%a and %b) and the first element of the result structure may
7919 be of integer types of any bit width, but they must have the same bit
7920 width. The second element of the result structure must be of
7921 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7922 undergo unsigned multiplication.</p>
7925 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7926 an unsigned multiplication of the two arguments. They return a structure
7927 — the first element of which is the multiplication, and the second
7928 element of which is a bit specifying if the unsigned multiplication resulted
7933 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7934 %sum = extractvalue {i32, i1} %res, 0
7935 %obit = extractvalue {i32, i1} %res, 1
7936 br i1 %obit, label %overflow, label %normal
7943 <!-- ======================================================================= -->
7945 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
7950 <p>Half precision floating point is a storage-only format. This means that it is
7951 a dense encoding (in memory) but does not support computation in the
7954 <p>This means that code must first load the half-precision floating point
7955 value as an i16, then convert it to float with <a
7956 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
7957 Computation can then be performed on the float value (including extending to
7958 double etc). To store the value back to memory, it is first converted to
7959 float if needed, then converted to i16 with
7960 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
7961 storing as an i16 value.</p>
7963 <!-- _______________________________________________________________________ -->
7965 <a name="int_convert_to_fp16">
7966 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
7974 declare i16 @llvm.convert.to.fp16(f32 %a)
7978 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7979 a conversion from single precision floating point format to half precision
7980 floating point format.</p>
7983 <p>The intrinsic function contains single argument - the value to be
7987 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7988 a conversion from single precision floating point format to half precision
7989 floating point format. The return value is an <tt>i16</tt> which
7990 contains the converted number.</p>
7994 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7995 store i16 %res, i16* @x, align 2
8000 <!-- _______________________________________________________________________ -->
8002 <a name="int_convert_from_fp16">
8003 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
8011 declare f32 @llvm.convert.from.fp16(i16 %a)
8015 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
8016 a conversion from half precision floating point format to single precision
8017 floating point format.</p>
8020 <p>The intrinsic function contains single argument - the value to be
8024 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
8025 conversion from half single precision floating point format to single
8026 precision floating point format. The input half-float value is represented by
8027 an <tt>i16</tt> value.</p>
8031 %a = load i16* @x, align 2
8032 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8039 <!-- ======================================================================= -->
8041 <a name="int_debugger">Debugger Intrinsics</a>
8046 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
8047 prefix), are described in
8048 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
8049 Level Debugging</a> document.</p>
8053 <!-- ======================================================================= -->
8055 <a name="int_eh">Exception Handling Intrinsics</a>
8060 <p>The LLVM exception handling intrinsics (which all start with
8061 <tt>llvm.eh.</tt> prefix), are described in
8062 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
8063 Handling</a> document.</p>
8067 <!-- ======================================================================= -->
8069 <a name="int_trampoline">Trampoline Intrinsics</a>
8074 <p>These intrinsics make it possible to excise one parameter, marked with
8075 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
8076 The result is a callable
8077 function pointer lacking the nest parameter - the caller does not need to
8078 provide a value for it. Instead, the value to use is stored in advance in a
8079 "trampoline", a block of memory usually allocated on the stack, which also
8080 contains code to splice the nest value into the argument list. This is used
8081 to implement the GCC nested function address extension.</p>
8083 <p>For example, if the function is
8084 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
8085 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
8088 <pre class="doc_code">
8089 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8090 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8091 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8092 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8093 %fp = bitcast i8* %p to i32 (i32, i32)*
8096 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
8097 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
8099 <!-- _______________________________________________________________________ -->
8102 '<tt>llvm.init.trampoline</tt>' Intrinsic
8110 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8114 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
8115 turning it into a trampoline.</p>
8118 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
8119 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
8120 sufficiently aligned block of memory; this memory is written to by the
8121 intrinsic. Note that the size and the alignment are target-specific - LLVM
8122 currently provides no portable way of determining them, so a front-end that
8123 generates this intrinsic needs to have some target-specific knowledge.
8124 The <tt>func</tt> argument must hold a function bitcast to
8125 an <tt>i8*</tt>.</p>
8128 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
8129 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
8130 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
8131 which can be <a href="#int_trampoline">bitcast (to a new function) and
8132 called</a>. The new function's signature is the same as that of
8133 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
8134 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
8135 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
8136 with the same argument list, but with <tt>nval</tt> used for the missing
8137 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
8138 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
8139 to the returned function pointer is undefined.</p>
8142 <!-- _______________________________________________________________________ -->
8145 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
8153 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8157 <p>This performs any required machine-specific adjustment to the address of a
8158 trampoline (passed as <tt>tramp</tt>).</p>
8161 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
8162 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
8166 <p>On some architectures the address of the code to be executed needs to be
8167 different to the address where the trampoline is actually stored. This
8168 intrinsic returns the executable address corresponding to <tt>tramp</tt>
8169 after performing the required machine specific adjustments.
8170 The pointer returned can then be <a href="#int_trampoline"> bitcast and
8178 <!-- ======================================================================= -->
8180 <a name="int_memorymarkers">Memory Use Markers</a>
8185 <p>This class of intrinsics exists to information about the lifetime of memory
8186 objects and ranges where variables are immutable.</p>
8188 <!-- _______________________________________________________________________ -->
8190 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8197 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8201 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8202 object's lifetime.</p>
8205 <p>The first argument is a constant integer representing the size of the
8206 object, or -1 if it is variable sized. The second argument is a pointer to
8210 <p>This intrinsic indicates that before this point in the code, the value of the
8211 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8212 never be used and has an undefined value. A load from the pointer that
8213 precedes this intrinsic can be replaced with
8214 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8218 <!-- _______________________________________________________________________ -->
8220 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8227 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8231 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8232 object's lifetime.</p>
8235 <p>The first argument is a constant integer representing the size of the
8236 object, or -1 if it is variable sized. The second argument is a pointer to
8240 <p>This intrinsic indicates that after this point in the code, the value of the
8241 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8242 never be used and has an undefined value. Any stores into the memory object
8243 following this intrinsic may be removed as dead.
8247 <!-- _______________________________________________________________________ -->
8249 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8256 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8260 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8261 a memory object will not change.</p>
8264 <p>The first argument is a constant integer representing the size of the
8265 object, or -1 if it is variable sized. The second argument is a pointer to
8269 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8270 the return value, the referenced memory location is constant and
8275 <!-- _______________________________________________________________________ -->
8277 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8284 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8288 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8289 a memory object are mutable.</p>
8292 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8293 The second argument is a constant integer representing the size of the
8294 object, or -1 if it is variable sized and the third argument is a pointer
8298 <p>This intrinsic indicates that the memory is mutable again.</p>
8304 <!-- ======================================================================= -->
8306 <a name="int_general">General Intrinsics</a>
8311 <p>This class of intrinsics is designed to be generic and has no specific
8314 <!-- _______________________________________________________________________ -->
8316 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8323 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8327 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8330 <p>The first argument is a pointer to a value, the second is a pointer to a
8331 global string, the third is a pointer to a global string which is the source
8332 file name, and the last argument is the line number.</p>
8335 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8336 This can be useful for special purpose optimizations that want to look for
8337 these annotations. These have no other defined use; they are ignored by code
8338 generation and optimization.</p>
8342 <!-- _______________________________________________________________________ -->
8344 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8350 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8351 any integer bit width.</p>
8354 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8355 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8356 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8357 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8358 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8362 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8365 <p>The first argument is an integer value (result of some expression), the
8366 second is a pointer to a global string, the third is a pointer to a global
8367 string which is the source file name, and the last argument is the line
8368 number. It returns the value of the first argument.</p>
8371 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8372 arbitrary strings. This can be useful for special purpose optimizations that
8373 want to look for these annotations. These have no other defined use; they
8374 are ignored by code generation and optimization.</p>
8378 <!-- _______________________________________________________________________ -->
8380 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8387 declare void @llvm.trap()
8391 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8397 <p>This intrinsics is lowered to the target dependent trap instruction. If the
8398 target does not have a trap instruction, this intrinsic will be lowered to
8399 the call of the <tt>abort()</tt> function.</p>
8403 <!-- _______________________________________________________________________ -->
8405 <a name="int_debugtrap">'<tt>llvm.debugtrap</tt>' Intrinsic</a>
8412 declare void @llvm.debugtrap()
8416 <p>The '<tt>llvm.debugtrap</tt>' intrinsic.</p>
8422 <p>This intrinsic is lowered to code which is intended to cause an execution
8423 trap with the intention of requesting the attention of a debugger.</p>
8427 <!-- _______________________________________________________________________ -->
8429 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8436 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8440 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8441 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8442 ensure that it is placed on the stack before local variables.</p>
8445 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8446 arguments. The first argument is the value loaded from the stack
8447 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8448 that has enough space to hold the value of the guard.</p>
8451 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8452 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8453 stack. This is to ensure that if a local variable on the stack is
8454 overwritten, it will destroy the value of the guard. When the function exits,
8455 the guard on the stack is checked against the original guard. If they are
8456 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8461 <!-- _______________________________________________________________________ -->
8463 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8470 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>, i32 <runtime>)
8471 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>, i32 <runtime>)
8475 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8476 the optimizers to determine at compile time whether a) an operation (like
8477 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8478 runtime check for overflow isn't necessary. An object in this context means
8479 an allocation of a specific class, structure, array, or other object.</p>
8482 <p>The <tt>llvm.objectsize</tt> intrinsic takes three arguments. The first
8483 argument is a pointer to or into the <tt>object</tt>. The second argument
8484 is a boolean and determines whether <tt>llvm.objectsize</tt> returns 0 (if true)
8485 or -1 (if false) when the object size is unknown.
8486 The third argument, <tt>runtime</tt>, indicates whether the compiler is allowed
8487 to return a non-constant value. The higher the value, the higher the potential
8488 run-time performance impact.
8489 The second and third arguments only accepts constants.</p>
8492 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to a constant representing
8493 the size of the object concerned. If the size cannot be determined at compile
8494 time, <tt>llvm.objectsize</tt> either returns <tt>i32/i64 -1 or 0</tt>
8495 (depending on the <tt>min</tt> argument) if <tt>runtime</tt> is 0, or a run-time
8496 value (if <tt>runtime</tt> > 0 and an expression could be generated).</p>
8499 <!-- _______________________________________________________________________ -->
8501 <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a>
8508 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8509 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8513 <p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the
8514 most probable) value of <tt>val</tt>, which can be used by optimizers.</p>
8517 <p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first
8518 argument is a value. The second argument is an expected value, this needs to
8519 be a constant value, variables are not allowed.</p>
8522 <p>This intrinsic is lowered to the <tt>val</tt>.</p>
8528 <!-- *********************************************************************** -->
8531 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8532 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8533 <a href="http://validator.w3.org/check/referer"><img
8534 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8536 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8537 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8538 Last modified: $Date$