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_internal">'<tt>internal</tt>' Linkage</a></li>
29 <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
30 <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
31 <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
32 <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
33 <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
34 <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
35 <li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
36 <li><a href="#linkage_linkonce_odr_auto_hide">'<tt>linkonce_odr_auto_hide</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>
260 <li><a href="#int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a></li>
261 <li><a href="#int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a></li>
264 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
266 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
267 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
268 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
269 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
272 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
274 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
275 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
276 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
277 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
278 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
279 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
282 <li><a href="#spec_arithmetic">Specialised Arithmetic Intrinsics</a>
284 <li><a href="#fmuladd">'<tt>llvm.fmuladd</tt> Intrinsic</a></li>
287 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
289 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
290 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
293 <li><a href="#int_debugger">Debugger intrinsics</a></li>
294 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
295 <li><a href="#int_trampoline">Trampoline Intrinsics</a>
297 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
298 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
301 <li><a href="#int_memorymarkers">Memory Use Markers</a>
303 <li><a href="#int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a></li>
304 <li><a href="#int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a></li>
305 <li><a href="#int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a></li>
306 <li><a href="#int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a></li>
309 <li><a href="#int_general">General intrinsics</a>
311 <li><a href="#int_var_annotation">
312 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
313 <li><a href="#int_annotation">
314 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
315 <li><a href="#int_trap">
316 '<tt>llvm.trap</tt>' Intrinsic</a></li>
317 <li><a href="#int_debugtrap">
318 '<tt>llvm.debugtrap</tt>' Intrinsic</a></li>
319 <li><a href="#int_stackprotector">
320 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
321 <li><a href="#int_objectsize">
322 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
323 <li><a href="#int_expect">
324 '<tt>llvm.expect</tt>' Intrinsic</a></li>
325 <li><a href="#int_donothing">
326 '<tt>llvm.donothing</tt>' Intrinsic</a></li>
333 <div class="doc_author">
334 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
335 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
338 <!-- *********************************************************************** -->
339 <h2><a name="abstract">Abstract</a></h2>
340 <!-- *********************************************************************** -->
344 <p>This document is a reference manual for the LLVM assembly language. LLVM is
345 a Static Single Assignment (SSA) based representation that provides type
346 safety, low-level operations, flexibility, and the capability of representing
347 'all' high-level languages cleanly. It is the common code representation
348 used throughout all phases of the LLVM compilation strategy.</p>
352 <!-- *********************************************************************** -->
353 <h2><a name="introduction">Introduction</a></h2>
354 <!-- *********************************************************************** -->
358 <p>The LLVM code representation is designed to be used in three different forms:
359 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
360 for fast loading by a Just-In-Time compiler), and as a human readable
361 assembly language representation. This allows LLVM to provide a powerful
362 intermediate representation for efficient compiler transformations and
363 analysis, while providing a natural means to debug and visualize the
364 transformations. The three different forms of LLVM are all equivalent. This
365 document describes the human readable representation and notation.</p>
367 <p>The LLVM representation aims to be light-weight and low-level while being
368 expressive, typed, and extensible at the same time. It aims to be a
369 "universal IR" of sorts, by being at a low enough level that high-level ideas
370 may be cleanly mapped to it (similar to how microprocessors are "universal
371 IR's", allowing many source languages to be mapped to them). By providing
372 type information, LLVM can be used as the target of optimizations: for
373 example, through pointer analysis, it can be proven that a C automatic
374 variable is never accessed outside of the current function, allowing it to
375 be promoted to a simple SSA value instead of a memory location.</p>
377 <!-- _______________________________________________________________________ -->
379 <a name="wellformed">Well-Formedness</a>
384 <p>It is important to note that this document describes 'well formed' LLVM
385 assembly language. There is a difference between what the parser accepts and
386 what is considered 'well formed'. For example, the following instruction is
387 syntactically okay, but not well formed:</p>
389 <pre class="doc_code">
390 %x = <a href="#i_add">add</a> i32 1, %x
393 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
394 LLVM infrastructure provides a verification pass that may be used to verify
395 that an LLVM module is well formed. This pass is automatically run by the
396 parser after parsing input assembly and by the optimizer before it outputs
397 bitcode. The violations pointed out by the verifier pass indicate bugs in
398 transformation passes or input to the parser.</p>
404 <!-- Describe the typesetting conventions here. -->
406 <!-- *********************************************************************** -->
407 <h2><a name="identifiers">Identifiers</a></h2>
408 <!-- *********************************************************************** -->
412 <p>LLVM identifiers come in two basic types: global and local. Global
413 identifiers (functions, global variables) begin with the <tt>'@'</tt>
414 character. Local identifiers (register names, types) begin with
415 the <tt>'%'</tt> character. Additionally, there are three different formats
416 for identifiers, for different purposes:</p>
419 <li>Named values are represented as a string of characters with their prefix.
420 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
421 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
422 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
423 other characters in their names can be surrounded with quotes. Special
424 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
425 ASCII code for the character in hexadecimal. In this way, any character
426 can be used in a name value, even quotes themselves.</li>
428 <li>Unnamed values are represented as an unsigned numeric value with their
429 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
431 <li>Constants, which are described in a <a href="#constants">section about
432 constants</a>, below.</li>
435 <p>LLVM requires that values start with a prefix for two reasons: Compilers
436 don't need to worry about name clashes with reserved words, and the set of
437 reserved words may be expanded in the future without penalty. Additionally,
438 unnamed identifiers allow a compiler to quickly come up with a temporary
439 variable without having to avoid symbol table conflicts.</p>
441 <p>Reserved words in LLVM are very similar to reserved words in other
442 languages. There are keywords for different opcodes
443 ('<tt><a href="#i_add">add</a></tt>',
444 '<tt><a href="#i_bitcast">bitcast</a></tt>',
445 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
446 ('<tt><a href="#t_void">void</a></tt>',
447 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
448 reserved words cannot conflict with variable names, because none of them
449 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
451 <p>Here is an example of LLVM code to multiply the integer variable
452 '<tt>%X</tt>' by 8:</p>
456 <pre class="doc_code">
457 %result = <a href="#i_mul">mul</a> i32 %X, 8
460 <p>After strength reduction:</p>
462 <pre class="doc_code">
463 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
466 <p>And the hard way:</p>
468 <pre class="doc_code">
469 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
470 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
471 %result = <a href="#i_add">add</a> i32 %1, %1
474 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
475 lexical features of LLVM:</p>
478 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
481 <li>Unnamed temporaries are created when the result of a computation is not
482 assigned to a named value.</li>
484 <li>Unnamed temporaries are numbered sequentially</li>
487 <p>It also shows a convention that we follow in this document. When
488 demonstrating instructions, we will follow an instruction with a comment that
489 defines the type and name of value produced. Comments are shown in italic
494 <!-- *********************************************************************** -->
495 <h2><a name="highlevel">High Level Structure</a></h2>
496 <!-- *********************************************************************** -->
498 <!-- ======================================================================= -->
500 <a name="modulestructure">Module Structure</a>
505 <p>LLVM programs are composed of <tt>Module</tt>s, each of which is a
506 translation unit of the input programs. Each module consists of functions,
507 global variables, and symbol table entries. Modules may be combined together
508 with the LLVM linker, which merges function (and global variable)
509 definitions, resolves forward declarations, and merges symbol table
510 entries. Here is an example of the "hello world" module:</p>
512 <pre class="doc_code">
513 <i>; Declare the string constant as a global constant.</i>
514 <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"
516 <i>; External declaration of the puts function</i>
517 <a href="#functionstructure">declare</a> i32 @puts(i8* <a href="#nocapture">nocapture</a>) <a href="#fnattrs">nounwind</a>
519 <i>; Definition of main function</i>
520 define i32 @main() { <i>; i32()* </i>
521 <i>; Convert [13 x i8]* to i8 *...</i>
522 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.str, i64 0, i64 0
524 <i>; Call puts function to write out the string to stdout.</i>
525 <a href="#i_call">call</a> i32 @puts(i8* %cast210)
526 <a href="#i_ret">ret</a> i32 0
529 <i>; Named metadata</i>
530 !1 = metadata !{i32 42}
534 <p>This example is made up of a <a href="#globalvars">global variable</a> named
535 "<tt>.str</tt>", an external declaration of the "<tt>puts</tt>" function,
536 a <a href="#functionstructure">function definition</a> for
537 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
540 <p>In general, a module is made up of a list of global values (where both
541 functions and global variables are global values). Global values are
542 represented by a pointer to a memory location (in this case, a pointer to an
543 array of char, and a pointer to a function), and have one of the
544 following <a href="#linkage">linkage types</a>.</p>
548 <!-- ======================================================================= -->
550 <a name="linkage">Linkage Types</a>
555 <p>All Global Variables and Functions have one of the following types of
559 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
560 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
561 by objects in the current module. In particular, linking code into a
562 module with an private global value may cause the private to be renamed as
563 necessary to avoid collisions. Because the symbol is private to the
564 module, all references can be updated. This doesn't show up in any symbol
565 table in the object file.</dd>
567 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
568 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
569 assembler and evaluated by the linker. Unlike normal strong symbols, they
570 are removed by the linker from the final linked image (executable or
571 dynamic library).</dd>
573 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
574 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
575 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
576 linker. The symbols are removed by the linker from the final linked image
577 (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_linkonce_odr_auto_hide">linkonce_odr_auto_hide</a></b></tt></dt>
648 <dd>Similar to "<tt>linkonce_odr</tt>", but nothing in the translation unit
649 takes the address of this definition. For instance, functions that had an
650 inline definition, but the compiler decided not to inline it.
651 <tt>linkonce_odr_auto_hide</tt> may have only <tt>default</tt> visibility.
652 The symbols are removed by the linker from the final linked image
653 (executable or dynamic library).</dd>
655 <dt><tt><b><a name="linkage_external">external</a></b></tt></dt>
656 <dd>If none of the above identifiers are used, the global is externally
657 visible, meaning that it participates in linkage and can be used to
658 resolve external symbol references.</dd>
661 <p>The next two types of linkage are targeted for Microsoft Windows platform
662 only. They are designed to support importing (exporting) symbols from (to)
663 DLLs (Dynamic Link Libraries).</p>
666 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
667 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
668 or variable via a global pointer to a pointer that is set up by the DLL
669 exporting the symbol. On Microsoft Windows targets, the pointer name is
670 formed by combining <code>__imp_</code> and the function or variable
673 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
674 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
675 pointer to a pointer in a DLL, so that it can be referenced with the
676 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
677 name is formed by combining <code>__imp_</code> and the function or
681 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
682 another module defined a "<tt>.LC0</tt>" variable and was linked with this
683 one, one of the two would be renamed, preventing a collision. Since
684 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
685 declarations), they are accessible outside of the current module.</p>
687 <p>It is illegal for a function <i>declaration</i> to have any linkage type
688 other than <tt>external</tt>, <tt>dllimport</tt>
689 or <tt>extern_weak</tt>.</p>
691 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
692 or <tt>weak_odr</tt> linkages.</p>
696 <!-- ======================================================================= -->
698 <a name="callingconv">Calling Conventions</a>
703 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
704 and <a href="#i_invoke">invokes</a> can all have an optional calling
705 convention specified for the call. The calling convention of any pair of
706 dynamic caller/callee must match, or the behavior of the program is
707 undefined. The following calling conventions are supported by LLVM, and more
708 may be added in the future:</p>
711 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
712 <dd>This calling convention (the default if no other calling convention is
713 specified) matches the target C calling conventions. This calling
714 convention supports varargs function calls and tolerates some mismatch in
715 the declared prototype and implemented declaration of the function (as
718 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
719 <dd>This calling convention attempts to make calls as fast as possible
720 (e.g. by passing things in registers). This calling convention allows the
721 target to use whatever tricks it wants to produce fast code for the
722 target, without having to conform to an externally specified ABI
723 (Application Binary Interface).
724 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
725 when this or the GHC convention is used.</a> 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>coldcc</tt>" - The cold calling convention</b>:</dt>
730 <dd>This calling convention attempts to make code in the caller as efficient
731 as possible under the assumption that the call is not commonly executed.
732 As such, these calls often preserve all registers so that the call does
733 not break any live ranges in the caller side. This calling convention
734 does not support varargs and requires the prototype of all callees to
735 exactly match the prototype of the function definition.</dd>
737 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
738 <dd>This calling convention has been implemented specifically for use by the
739 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
740 It passes everything in registers, going to extremes to achieve this by
741 disabling callee save registers. This calling convention should not be
742 used lightly but only for specific situations such as an alternative to
743 the <em>register pinning</em> performance technique often used when
744 implementing functional programming languages.At the moment only X86
745 supports this convention and it has the following limitations:
747 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
748 floating point types are supported.</li>
749 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
750 6 floating point parameters.</li>
752 This calling convention supports
753 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
754 requires both the caller and callee are using it.
757 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
758 <dd>Any calling convention may be specified by number, allowing
759 target-specific calling conventions to be used. Target specific calling
760 conventions start at 64.</dd>
763 <p>More calling conventions can be added/defined on an as-needed basis, to
764 support Pascal conventions or any other well-known target-independent
769 <!-- ======================================================================= -->
771 <a name="visibility">Visibility Styles</a>
776 <p>All Global Variables and Functions have one of the following visibility
780 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
781 <dd>On targets that use the ELF object file format, default visibility means
782 that the declaration is visible to other modules and, in shared libraries,
783 means that the declared entity may be overridden. On Darwin, default
784 visibility means that the declaration is visible to other modules. Default
785 visibility corresponds to "external linkage" in the language.</dd>
787 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
788 <dd>Two declarations of an object with hidden visibility refer to the same
789 object if they are in the same shared object. Usually, hidden visibility
790 indicates that the symbol will not be placed into the dynamic symbol
791 table, so no other module (executable or shared library) can reference it
794 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
795 <dd>On ELF, protected visibility indicates that the symbol will be placed in
796 the dynamic symbol table, but that references within the defining module
797 will bind to the local symbol. That is, the symbol cannot be overridden by
803 <!-- ======================================================================= -->
805 <a name="namedtypes">Named Types</a>
810 <p>LLVM IR allows you to specify name aliases for certain types. This can make
811 it easier to read the IR and make the IR more condensed (particularly when
812 recursive types are involved). An example of a name specification is:</p>
814 <pre class="doc_code">
815 %mytype = type { %mytype*, i32 }
818 <p>You may give a name to any <a href="#typesystem">type</a> except
819 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
820 is expected with the syntax "%mytype".</p>
822 <p>Note that type names are aliases for the structural type that they indicate,
823 and that you can therefore specify multiple names for the same type. This
824 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
825 uses structural typing, the name is not part of the type. When printing out
826 LLVM IR, the printer will pick <em>one name</em> to render all types of a
827 particular shape. This means that if you have code where two different
828 source types end up having the same LLVM type, that the dumper will sometimes
829 print the "wrong" or unexpected type. This is an important design point and
830 isn't going to change.</p>
834 <!-- ======================================================================= -->
836 <a name="globalvars">Global Variables</a>
841 <p>Global variables define regions of memory allocated at compilation time
842 instead of run-time. Global variables may optionally be initialized, may
843 have an explicit section to be placed in, and may have an optional explicit
844 alignment specified.</p>
846 <p>A variable may be defined as <tt>thread_local</tt>, which
847 means that it will not be shared by threads (each thread will have a
848 separated copy of the variable). Not all targets support thread-local
849 variables. Optionally, a TLS model may be specified:</p>
852 <dt><b><tt>localdynamic</tt></b>:</dt>
853 <dd>For variables that are only used within the current shared library.</dd>
855 <dt><b><tt>initialexec</tt></b>:</dt>
856 <dd>For variables in modules that will not be loaded dynamically.</dd>
858 <dt><b><tt>localexec</tt></b>:</dt>
859 <dd>For variables defined in the executable and only used within it.</dd>
862 <p>The models correspond to the ELF TLS models; see
863 <a href="http://people.redhat.com/drepper/tls.pdf">ELF
864 Handling For Thread-Local Storage</a> for more information on under which
865 circumstances the different models may be used. The target may choose a
866 different TLS model if the specified model is not supported, or if a better
867 choice of model can be made.</p>
869 <p>A variable may be defined as a global
870 "constant," which indicates that the contents of the variable
871 will <b>never</b> be modified (enabling better optimization, allowing the
872 global data to be placed in the read-only section of an executable, etc).
873 Note that variables that need runtime initialization cannot be marked
874 "constant" as there is a store to the variable.</p>
876 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
877 constant, even if the final definition of the global is not. This capability
878 can be used to enable slightly better optimization of the program, but
879 requires the language definition to guarantee that optimizations based on the
880 'constantness' are valid for the translation units that do not include the
883 <p>As SSA values, global variables define pointer values that are in scope
884 (i.e. they dominate) all basic blocks in the program. Global variables
885 always define a pointer to their "content" type because they describe a
886 region of memory, and all memory objects in LLVM are accessed through
889 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
890 that the address is not significant, only the content. Constants marked
891 like this can be merged with other constants if they have the same
892 initializer. Note that a constant with significant address <em>can</em>
893 be merged with a <tt>unnamed_addr</tt> constant, the result being a
894 constant whose address is significant.</p>
896 <p>A global variable may be declared to reside in a target-specific numbered
897 address space. For targets that support them, address spaces may affect how
898 optimizations are performed and/or what target instructions are used to
899 access the variable. The default address space is zero. The address space
900 qualifier must precede any other attributes.</p>
902 <p>LLVM allows an explicit section to be specified for globals. If the target
903 supports it, it will emit globals to the section specified.</p>
905 <p>An explicit alignment may be specified for a global, which must be a power
906 of 2. If not present, or if the alignment is set to zero, the alignment of
907 the global is set by the target to whatever it feels convenient. If an
908 explicit alignment is specified, the global is forced to have exactly that
909 alignment. Targets and optimizers are not allowed to over-align the global
910 if the global has an assigned section. In this case, the extra alignment
911 could be observable: for example, code could assume that the globals are
912 densely packed in their section and try to iterate over them as an array,
913 alignment padding would break this iteration.</p>
915 <p>For example, the following defines a global in a numbered address space with
916 an initializer, section, and alignment:</p>
918 <pre class="doc_code">
919 @G = addrspace(5) constant float 1.0, section "foo", align 4
922 <p>The following example defines a thread-local global with
923 the <tt>initialexec</tt> TLS model:</p>
925 <pre class="doc_code">
926 @G = thread_local(initialexec) global i32 0, align 4
932 <!-- ======================================================================= -->
934 <a name="functionstructure">Functions</a>
939 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
940 optional <a href="#linkage">linkage type</a>, an optional
941 <a href="#visibility">visibility style</a>, an optional
942 <a href="#callingconv">calling convention</a>,
943 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
944 <a href="#paramattrs">parameter attribute</a> for the return type, a function
945 name, a (possibly empty) argument list (each with optional
946 <a href="#paramattrs">parameter attributes</a>), optional
947 <a href="#fnattrs">function attributes</a>, an optional section, an optional
948 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
949 curly brace, a list of basic blocks, and a closing curly brace.</p>
951 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
952 optional <a href="#linkage">linkage type</a>, an optional
953 <a href="#visibility">visibility style</a>, an optional
954 <a href="#callingconv">calling convention</a>,
955 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
956 <a href="#paramattrs">parameter attribute</a> for the return type, a function
957 name, a possibly empty list of arguments, an optional alignment, and an
958 optional <a href="#gc">garbage collector name</a>.</p>
960 <p>A function definition contains a list of basic blocks, forming the CFG
961 (Control Flow Graph) for the function. Each basic block may optionally start
962 with a label (giving the basic block a symbol table entry), contains a list
963 of instructions, and ends with a <a href="#terminators">terminator</a>
964 instruction (such as a branch or function return).</p>
966 <p>The first basic block in a function is special in two ways: it is immediately
967 executed on entrance to the function, and it is not allowed to have
968 predecessor basic blocks (i.e. there can not be any branches to the entry
969 block of a function). Because the block can have no predecessors, it also
970 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
972 <p>LLVM allows an explicit section to be specified for functions. If the target
973 supports it, it will emit functions to the section specified.</p>
975 <p>An explicit alignment may be specified for a function. If not present, or if
976 the alignment is set to zero, the alignment of the function is set by the
977 target to whatever it feels convenient. If an explicit alignment is
978 specified, the function is forced to have at least that much alignment. All
979 alignments must be a power of 2.</p>
981 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
982 be significant and two identical functions can be merged.</p>
985 <pre class="doc_code">
986 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
987 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
988 <ResultType> @<FunctionName> ([argument list])
989 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
990 [<a href="#gc">gc</a>] { ... }
995 <!-- ======================================================================= -->
997 <a name="aliasstructure">Aliases</a>
1002 <p>Aliases act as "second name" for the aliasee value (which can be either
1003 function, global variable, another alias or bitcast of global value). Aliases
1004 may have an optional <a href="#linkage">linkage type</a>, and an
1005 optional <a href="#visibility">visibility style</a>.</p>
1008 <pre class="doc_code">
1009 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
1014 <!-- ======================================================================= -->
1016 <a name="namedmetadatastructure">Named Metadata</a>
1021 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
1022 nodes</a> (but not metadata strings) are the only valid operands for
1023 a named metadata.</p>
1026 <pre class="doc_code">
1027 ; Some unnamed metadata nodes, which are referenced by the named metadata.
1028 !0 = metadata !{metadata !"zero"}
1029 !1 = metadata !{metadata !"one"}
1030 !2 = metadata !{metadata !"two"}
1032 !name = !{!0, !1, !2}
1037 <!-- ======================================================================= -->
1039 <a name="paramattrs">Parameter Attributes</a>
1044 <p>The return type and each parameter of a function type may have a set of
1045 <i>parameter attributes</i> associated with them. Parameter attributes are
1046 used to communicate additional information about the result or parameters of
1047 a function. Parameter attributes are considered to be part of the function,
1048 not of the function type, so functions with different parameter attributes
1049 can have the same function type.</p>
1051 <p>Parameter attributes are simple keywords that follow the type specified. If
1052 multiple parameter attributes are needed, they are space separated. For
1055 <pre class="doc_code">
1056 declare i32 @printf(i8* noalias nocapture, ...)
1057 declare i32 @atoi(i8 zeroext)
1058 declare signext i8 @returns_signed_char()
1061 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1062 <tt>readonly</tt>) come immediately after the argument list.</p>
1064 <p>Currently, only the following parameter attributes are defined:</p>
1067 <dt><tt><b>zeroext</b></tt></dt>
1068 <dd>This indicates to the code generator that the parameter or return value
1069 should be zero-extended to the extent required by the target's ABI (which
1070 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1071 parameter) or the callee (for a return value).</dd>
1073 <dt><tt><b>signext</b></tt></dt>
1074 <dd>This indicates to the code generator that the parameter or return value
1075 should be sign-extended to the extent required by the target's ABI (which
1076 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1079 <dt><tt><b>inreg</b></tt></dt>
1080 <dd>This indicates that this parameter or return value should be treated in a
1081 special target-dependent fashion during while emitting code for a function
1082 call or return (usually, by putting it in a register as opposed to memory,
1083 though some targets use it to distinguish between two different kinds of
1084 registers). Use of this attribute is target-specific.</dd>
1086 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1087 <dd><p>This indicates that the pointer parameter should really be passed by
1088 value to the function. The attribute implies that a hidden copy of the
1090 is made between the caller and the callee, so the callee is unable to
1091 modify the value in the caller. This attribute is only valid on LLVM
1092 pointer arguments. It is generally used to pass structs and arrays by
1093 value, but is also valid on pointers to scalars. The copy is considered
1094 to belong to the caller not the callee (for example,
1095 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1096 <tt>byval</tt> parameters). This is not a valid attribute for return
1099 <p>The byval attribute also supports specifying an alignment with
1100 the align attribute. It indicates the alignment of the stack slot to
1101 form and the known alignment of the pointer specified to the call site. If
1102 the alignment is not specified, then the code generator makes a
1103 target-specific assumption.</p></dd>
1105 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1106 <dd>This indicates that the pointer parameter specifies the address of a
1107 structure that is the return value of the function in the source program.
1108 This pointer must be guaranteed by the caller to be valid: loads and
1109 stores to the structure may be assumed by the callee to not to trap. This
1110 may only be applied to the first parameter. This is not a valid attribute
1111 for return values. </dd>
1113 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1114 <dd>This indicates that pointer values
1115 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1116 value do not alias pointer values which are not <i>based</i> on it,
1117 ignoring certain "irrelevant" dependencies.
1118 For a call to the parent function, dependencies between memory
1119 references from before or after the call and from those during the call
1120 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1121 return value used in that call.
1122 The caller shares the responsibility with the callee for ensuring that
1123 these requirements are met.
1124 For further details, please see the discussion of the NoAlias response in
1125 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1127 Note that this definition of <tt>noalias</tt> is intentionally
1128 similar to the definition of <tt>restrict</tt> in C99 for function
1129 arguments, though it is slightly weaker.
1131 For function return values, C99's <tt>restrict</tt> is not meaningful,
1132 while LLVM's <tt>noalias</tt> is.
1135 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1136 <dd>This indicates that the callee does not make any copies of the pointer
1137 that outlive the callee itself. This is not a valid attribute for return
1140 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1141 <dd>This indicates that the pointer parameter can be excised using the
1142 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1143 attribute for return values.</dd>
1148 <!-- ======================================================================= -->
1150 <a name="gc">Garbage Collector Names</a>
1155 <p>Each function may specify a garbage collector name, which is simply a
1158 <pre class="doc_code">
1159 define void @f() gc "name" { ... }
1162 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1163 collector which will cause the compiler to alter its output in order to
1164 support the named garbage collection algorithm.</p>
1168 <!-- ======================================================================= -->
1170 <a name="fnattrs">Function Attributes</a>
1175 <p>Function attributes are set to communicate additional information about a
1176 function. Function attributes are considered to be part of the function, not
1177 of the function type, so functions with different parameter attributes can
1178 have the same function type.</p>
1180 <p>Function attributes are simple keywords that follow the type specified. If
1181 multiple attributes are needed, they are space separated. For example:</p>
1183 <pre class="doc_code">
1184 define void @f() noinline { ... }
1185 define void @f() alwaysinline { ... }
1186 define void @f() alwaysinline optsize { ... }
1187 define void @f() optsize { ... }
1191 <dt><tt><b>address_safety</b></tt></dt>
1192 <dd>This attribute indicates that the address safety analysis
1193 is enabled for this function. </dd>
1195 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1196 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1197 the backend should forcibly align the stack pointer. Specify the
1198 desired alignment, which must be a power of two, in parentheses.
1200 <dt><tt><b>alwaysinline</b></tt></dt>
1201 <dd>This attribute indicates that the inliner should attempt to inline this
1202 function into callers whenever possible, ignoring any active inlining size
1203 threshold for this caller.</dd>
1205 <dt><tt><b>nonlazybind</b></tt></dt>
1206 <dd>This attribute suppresses lazy symbol binding for the function. This
1207 may make calls to the function faster, at the cost of extra program
1208 startup time if the function is not called during program startup.</dd>
1210 <dt><tt><b>ia_nsdialect</b></tt></dt>
1211 <dd>This attribute indicates the associated inline assembly call is using a
1212 non-standard assembly dialect. The standard dialect is ATT, which is
1213 assumed when this attribute is not present. When present, the dialect
1214 is assumed to be Intel. Currently, ATT and Intel are the only supported
1217 <dt><tt><b>inlinehint</b></tt></dt>
1218 <dd>This attribute indicates that the source code contained a hint that inlining
1219 this function is desirable (such as the "inline" keyword in C/C++). It
1220 is just a hint; it imposes no requirements on the inliner.</dd>
1222 <dt><tt><b>naked</b></tt></dt>
1223 <dd>This attribute disables prologue / epilogue emission for the function.
1224 This can have very system-specific consequences.</dd>
1226 <dt><tt><b>noimplicitfloat</b></tt></dt>
1227 <dd>This attributes disables implicit floating point instructions.</dd>
1229 <dt><tt><b>noinline</b></tt></dt>
1230 <dd>This attribute indicates that the inliner should never inline this
1231 function in any situation. This attribute may not be used together with
1232 the <tt>alwaysinline</tt> attribute.</dd>
1234 <dt><tt><b>noredzone</b></tt></dt>
1235 <dd>This attribute indicates that the code generator should not use a red
1236 zone, even if the target-specific ABI normally permits it.</dd>
1238 <dt><tt><b>noreturn</b></tt></dt>
1239 <dd>This function attribute indicates that the function never returns
1240 normally. This produces undefined behavior at runtime if the function
1241 ever does dynamically return.</dd>
1243 <dt><tt><b>nounwind</b></tt></dt>
1244 <dd>This function attribute indicates that the function never returns with an
1245 unwind or exceptional control flow. If the function does unwind, its
1246 runtime behavior is undefined.</dd>
1248 <dt><tt><b>optsize</b></tt></dt>
1249 <dd>This attribute suggests that optimization passes and code generator passes
1250 make choices that keep the code size of this function low, and otherwise
1251 do optimizations specifically to reduce code size.</dd>
1253 <dt><tt><b>readnone</b></tt></dt>
1254 <dd>This attribute indicates that the function computes its result (or decides
1255 to unwind an exception) based strictly on its arguments, without
1256 dereferencing any pointer arguments or otherwise accessing any mutable
1257 state (e.g. memory, control registers, etc) visible to caller functions.
1258 It does not write through any pointer arguments
1259 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1260 changes any state visible to callers. This means that it cannot unwind
1261 exceptions by calling the <tt>C++</tt> exception throwing methods.</dd>
1263 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1264 <dd>This attribute indicates that the function does not write through any
1265 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1266 arguments) or otherwise modify any state (e.g. memory, control registers,
1267 etc) visible to caller functions. It may dereference pointer arguments
1268 and read state that may be set in the caller. A readonly function always
1269 returns the same value (or unwinds an exception identically) when called
1270 with the same set of arguments and global state. It cannot unwind an
1271 exception by calling the <tt>C++</tt> exception throwing methods.</dd>
1273 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1274 <dd>This attribute indicates that this function can return twice. The
1275 C <code>setjmp</code> is an example of such a function. The compiler
1276 disables some optimizations (like tail calls) in the caller of these
1279 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1280 <dd>This attribute indicates that the function should emit a stack smashing
1281 protector. It is in the form of a "canary"—a random value placed on
1282 the stack before the local variables that's checked upon return from the
1283 function to see if it has been overwritten. A heuristic is used to
1284 determine if a function needs stack protectors or not.<br>
1286 If a function that has an <tt>ssp</tt> attribute is inlined into a
1287 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1288 function will have an <tt>ssp</tt> attribute.</dd>
1290 <dt><tt><b>sspreq</b></tt></dt>
1291 <dd>This attribute indicates that the function should <em>always</em> emit a
1292 stack smashing protector. This overrides
1293 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1295 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1296 function that doesn't have an <tt>sspreq</tt> attribute or which has
1297 an <tt>ssp</tt> attribute, then the resulting function will have
1298 an <tt>sspreq</tt> attribute.</dd>
1300 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1301 <dd>This attribute indicates that the ABI being targeted requires that
1302 an unwind table entry be produce for this function even if we can
1303 show that no exceptions passes by it. This is normally the case for
1304 the ELF x86-64 abi, but it can be disabled for some compilation
1310 <!-- ======================================================================= -->
1312 <a name="moduleasm">Module-Level Inline Assembly</a>
1317 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1318 the GCC "file scope inline asm" blocks. These blocks are internally
1319 concatenated by LLVM and treated as a single unit, but may be separated in
1320 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1322 <pre class="doc_code">
1323 module asm "inline asm code goes here"
1324 module asm "more can go here"
1327 <p>The strings can contain any character by escaping non-printable characters.
1328 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1331 <p>The inline asm code is simply printed to the machine code .s file when
1332 assembly code is generated.</p>
1336 <!-- ======================================================================= -->
1338 <a name="datalayout">Data Layout</a>
1343 <p>A module may specify a target specific data layout string that specifies how
1344 data is to be laid out in memory. The syntax for the data layout is
1347 <pre class="doc_code">
1348 target datalayout = "<i>layout specification</i>"
1351 <p>The <i>layout specification</i> consists of a list of specifications
1352 separated by the minus sign character ('-'). Each specification starts with
1353 a letter and may include other information after the letter to define some
1354 aspect of the data layout. The specifications accepted are as follows:</p>
1358 <dd>Specifies that the target lays out data in big-endian form. That is, the
1359 bits with the most significance have the lowest address location.</dd>
1362 <dd>Specifies that the target lays out data in little-endian form. That is,
1363 the bits with the least significance have the lowest address
1366 <dt><tt>S<i>size</i></tt></dt>
1367 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1368 of stack variables is limited to the natural stack alignment to avoid
1369 dynamic stack realignment. The stack alignment must be a multiple of
1370 8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1371 which does not prevent any alignment promotions.</dd>
1373 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1374 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1375 <i>preferred</i> alignments. All sizes are in bits. Specifying
1376 the <i>pref</i> alignment is optional. If omitted, the
1377 preceding <tt>:</tt> should be omitted too.</dd>
1379 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1380 <dd>This specifies the alignment for an integer type of a given bit
1381 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1383 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1384 <dd>This specifies the alignment for a vector type of a given bit
1387 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1388 <dd>This specifies the alignment for a floating point type of a given bit
1389 <i>size</i>. Only values of <i>size</i> that are supported by the target
1390 will work. 32 (float) and 64 (double) are supported on all targets;
1391 80 or 128 (different flavors of long double) are also supported on some
1394 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1395 <dd>This specifies the alignment for an aggregate type of a given bit
1398 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1399 <dd>This specifies the alignment for a stack object of a given bit
1402 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1403 <dd>This specifies a set of native integer widths for the target CPU
1404 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1405 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1406 this set are considered to support most general arithmetic
1407 operations efficiently.</dd>
1410 <p>When constructing the data layout for a given target, LLVM starts with a
1411 default set of specifications which are then (possibly) overridden by the
1412 specifications in the <tt>datalayout</tt> keyword. The default specifications
1413 are given in this list:</p>
1416 <li><tt>E</tt> - big endian</li>
1417 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1418 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1419 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1420 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1421 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1422 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1423 alignment of 64-bits</li>
1424 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1425 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1426 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1427 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1428 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1429 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1432 <p>When LLVM is determining the alignment for a given type, it uses the
1433 following rules:</p>
1436 <li>If the type sought is an exact match for one of the specifications, that
1437 specification is used.</li>
1439 <li>If no match is found, and the type sought is an integer type, then the
1440 smallest integer type that is larger than the bitwidth of the sought type
1441 is used. If none of the specifications are larger than the bitwidth then
1442 the largest integer type is used. For example, given the default
1443 specifications above, the i7 type will use the alignment of i8 (next
1444 largest) while both i65 and i256 will use the alignment of i64 (largest
1447 <li>If no match is found, and the type sought is a vector type, then the
1448 largest vector type that is smaller than the sought vector type will be
1449 used as a fall back. This happens because <128 x double> can be
1450 implemented in terms of 64 <2 x double>, for example.</li>
1453 <p>The function of the data layout string may not be what you expect. Notably,
1454 this is not a specification from the frontend of what alignment the code
1455 generator should use.</p>
1457 <p>Instead, if specified, the target data layout is required to match what the
1458 ultimate <em>code generator</em> expects. This string is used by the
1459 mid-level optimizers to
1460 improve code, and this only works if it matches what the ultimate code
1461 generator uses. If you would like to generate IR that does not embed this
1462 target-specific detail into the IR, then you don't have to specify the
1463 string. This will disable some optimizations that require precise layout
1464 information, but this also prevents those optimizations from introducing
1465 target specificity into the IR.</p>
1471 <!-- ======================================================================= -->
1473 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1478 <p>Any memory access must be done through a pointer value associated
1479 with an address range of the memory access, otherwise the behavior
1480 is undefined. Pointer values are associated with address ranges
1481 according to the following rules:</p>
1484 <li>A pointer value is associated with the addresses associated with
1485 any value it is <i>based</i> on.
1486 <li>An address of a global variable is associated with the address
1487 range of the variable's storage.</li>
1488 <li>The result value of an allocation instruction is associated with
1489 the address range of the allocated storage.</li>
1490 <li>A null pointer in the default address-space is associated with
1492 <li>An integer constant other than zero or a pointer value returned
1493 from a function not defined within LLVM may be associated with address
1494 ranges allocated through mechanisms other than those provided by
1495 LLVM. Such ranges shall not overlap with any ranges of addresses
1496 allocated by mechanisms provided by LLVM.</li>
1499 <p>A pointer value is <i>based</i> on another pointer value according
1500 to the following rules:</p>
1503 <li>A pointer value formed from a
1504 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1505 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1506 <li>The result value of a
1507 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1508 of the <tt>bitcast</tt>.</li>
1509 <li>A pointer value formed by an
1510 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1511 pointer values that contribute (directly or indirectly) to the
1512 computation of the pointer's value.</li>
1513 <li>The "<i>based</i> on" relationship is transitive.</li>
1516 <p>Note that this definition of <i>"based"</i> is intentionally
1517 similar to the definition of <i>"based"</i> in C99, though it is
1518 slightly weaker.</p>
1520 <p>LLVM IR does not associate types with memory. The result type of a
1521 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1522 alignment of the memory from which to load, as well as the
1523 interpretation of the value. The first operand type of a
1524 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1525 and alignment of the store.</p>
1527 <p>Consequently, type-based alias analysis, aka TBAA, aka
1528 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1529 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1530 additional information which specialized optimization passes may use
1531 to implement type-based alias analysis.</p>
1535 <!-- ======================================================================= -->
1537 <a name="volatile">Volatile Memory Accesses</a>
1542 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1543 href="#i_store"><tt>store</tt></a>s, and <a
1544 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1545 The optimizers must not change the number of volatile operations or change their
1546 order of execution relative to other volatile operations. The optimizers
1547 <i>may</i> change the order of volatile operations relative to non-volatile
1548 operations. This is not Java's "volatile" and has no cross-thread
1549 synchronization behavior.</p>
1553 <!-- ======================================================================= -->
1555 <a name="memmodel">Memory Model for Concurrent Operations</a>
1560 <p>The LLVM IR does not define any way to start parallel threads of execution
1561 or to register signal handlers. Nonetheless, there are platform-specific
1562 ways to create them, and we define LLVM IR's behavior in their presence. This
1563 model is inspired by the C++0x memory model.</p>
1565 <p>For a more informal introduction to this model, see the
1566 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1568 <p>We define a <i>happens-before</i> partial order as the least partial order
1571 <li>Is a superset of single-thread program order, and</li>
1572 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1573 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1574 by platform-specific techniques, like pthread locks, thread
1575 creation, thread joining, etc., and by atomic instructions.
1576 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1580 <p>Note that program order does not introduce <i>happens-before</i> edges
1581 between a thread and signals executing inside that thread.</p>
1583 <p>Every (defined) read operation (load instructions, memcpy, atomic
1584 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1585 (defined) write operations (store instructions, atomic
1586 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1587 initialized globals are considered to have a write of the initializer which is
1588 atomic and happens before any other read or write of the memory in question.
1589 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1590 any write to the same byte, except:</p>
1593 <li>If <var>write<sub>1</sub></var> happens before
1594 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1595 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1596 does not see <var>write<sub>1</sub></var>.
1597 <li>If <var>R<sub>byte</sub></var> happens before
1598 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1599 see <var>write<sub>3</sub></var>.
1602 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1604 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1605 is supposed to give guarantees which can support
1606 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1607 addresses which do not behave like normal memory. It does not generally
1608 provide cross-thread synchronization.)
1609 <li>Otherwise, if there is no write to the same byte that happens before
1610 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1611 <tt>undef</tt> for that byte.
1612 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1613 <var>R<sub>byte</sub></var> returns the value written by that
1615 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1616 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1617 values written. See the <a href="#ordering">Atomic Memory Ordering
1618 Constraints</a> section for additional constraints on how the choice
1620 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1623 <p><var>R</var> returns the value composed of the series of bytes it read.
1624 This implies that some bytes within the value may be <tt>undef</tt>
1625 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1626 defines the semantics of the operation; it doesn't mean that targets will
1627 emit more than one instruction to read the series of bytes.</p>
1629 <p>Note that in cases where none of the atomic intrinsics are used, this model
1630 places only one restriction on IR transformations on top of what is required
1631 for single-threaded execution: introducing a store to a byte which might not
1632 otherwise be stored is not allowed in general. (Specifically, in the case
1633 where another thread might write to and read from an address, introducing a
1634 store can change a load that may see exactly one write into a load that may
1635 see multiple writes.)</p>
1637 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1638 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1639 none of the backends currently in the tree fall into this category; however,
1640 there might be targets which care. If there are, we want a paragraph
1643 Targets may specify that stores narrower than a certain width are not
1644 available; on such a target, for the purposes of this model, treat any
1645 non-atomic write with an alignment or width less than the minimum width
1646 as if it writes to the relevant surrounding bytes.
1651 <!-- ======================================================================= -->
1653 <a name="ordering">Atomic Memory Ordering Constraints</a>
1658 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1659 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1660 <a href="#i_fence"><code>fence</code></a>,
1661 <a href="#i_load"><code>atomic load</code></a>, and
1662 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1663 that determines which other atomic instructions on the same address they
1664 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1665 but are somewhat more colloquial. If these descriptions aren't precise enough,
1666 check those specs (see spec references in the
1667 <a href="Atomics.html#introduction">atomics guide</a>).
1668 <a href="#i_fence"><code>fence</code></a> instructions
1669 treat these orderings somewhat differently since they don't take an address.
1670 See that instruction's documentation for details.</p>
1672 <p>For a simpler introduction to the ordering constraints, see the
1673 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1676 <dt><code>unordered</code></dt>
1677 <dd>The set of values that can be read is governed by the happens-before
1678 partial order. A value cannot be read unless some operation wrote it.
1679 This is intended to provide a guarantee strong enough to model Java's
1680 non-volatile shared variables. This ordering cannot be specified for
1681 read-modify-write operations; it is not strong enough to make them atomic
1682 in any interesting way.</dd>
1683 <dt><code>monotonic</code></dt>
1684 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1685 total order for modifications by <code>monotonic</code> operations on each
1686 address. All modification orders must be compatible with the happens-before
1687 order. There is no guarantee that the modification orders can be combined to
1688 a global total order for the whole program (and this often will not be
1689 possible). The read in an atomic read-modify-write operation
1690 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1691 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1692 reads the value in the modification order immediately before the value it
1693 writes. If one atomic read happens before another atomic read of the same
1694 address, the later read must see the same value or a later value in the
1695 address's modification order. This disallows reordering of
1696 <code>monotonic</code> (or stronger) operations on the same address. If an
1697 address is written <code>monotonic</code>ally by one thread, and other threads
1698 <code>monotonic</code>ally read that address repeatedly, the other threads must
1699 eventually see the write. This corresponds to the C++0x/C1x
1700 <code>memory_order_relaxed</code>.</dd>
1701 <dt><code>acquire</code></dt>
1702 <dd>In addition to the guarantees of <code>monotonic</code>,
1703 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1704 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1705 <dt><code>release</code></dt>
1706 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1707 writes a value which is subsequently read by an <code>acquire</code> operation,
1708 it <i>synchronizes-with</i> that operation. (This isn't a complete
1709 description; see the C++0x definition of a release sequence.) This corresponds
1710 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1711 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1712 <code>acquire</code> and <code>release</code> operation on its address.
1713 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1714 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1715 <dd>In addition to the guarantees of <code>acq_rel</code>
1716 (<code>acquire</code> for an operation which only reads, <code>release</code>
1717 for an operation which only writes), there is a global total order on all
1718 sequentially-consistent operations on all addresses, which is consistent with
1719 the <i>happens-before</i> partial order and with the modification orders of
1720 all the affected addresses. Each sequentially-consistent read sees the last
1721 preceding write to the same address in this global order. This corresponds
1722 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1725 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1726 it only <i>synchronizes with</i> or participates in modification and seq_cst
1727 total orderings with other operations running in the same thread (for example,
1728 in signal handlers).</p>
1734 <!-- *********************************************************************** -->
1735 <h2><a name="typesystem">Type System</a></h2>
1736 <!-- *********************************************************************** -->
1740 <p>The LLVM type system is one of the most important features of the
1741 intermediate representation. Being typed enables a number of optimizations
1742 to be performed on the intermediate representation directly, without having
1743 to do extra analyses on the side before the transformation. A strong type
1744 system makes it easier to read the generated code and enables novel analyses
1745 and transformations that are not feasible to perform on normal three address
1746 code representations.</p>
1748 <!-- ======================================================================= -->
1750 <a name="t_classifications">Type Classifications</a>
1755 <p>The types fall into a few useful classifications:</p>
1757 <table border="1" cellspacing="0" cellpadding="4">
1759 <tr><th>Classification</th><th>Types</th></tr>
1761 <td><a href="#t_integer">integer</a></td>
1762 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1765 <td><a href="#t_floating">floating point</a></td>
1766 <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1769 <td><a name="t_firstclass">first class</a></td>
1770 <td><a href="#t_integer">integer</a>,
1771 <a href="#t_floating">floating point</a>,
1772 <a href="#t_pointer">pointer</a>,
1773 <a href="#t_vector">vector</a>,
1774 <a href="#t_struct">structure</a>,
1775 <a href="#t_array">array</a>,
1776 <a href="#t_label">label</a>,
1777 <a href="#t_metadata">metadata</a>.
1781 <td><a href="#t_primitive">primitive</a></td>
1782 <td><a href="#t_label">label</a>,
1783 <a href="#t_void">void</a>,
1784 <a href="#t_integer">integer</a>,
1785 <a href="#t_floating">floating point</a>,
1786 <a href="#t_x86mmx">x86mmx</a>,
1787 <a href="#t_metadata">metadata</a>.</td>
1790 <td><a href="#t_derived">derived</a></td>
1791 <td><a href="#t_array">array</a>,
1792 <a href="#t_function">function</a>,
1793 <a href="#t_pointer">pointer</a>,
1794 <a href="#t_struct">structure</a>,
1795 <a href="#t_vector">vector</a>,
1796 <a href="#t_opaque">opaque</a>.
1802 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1803 important. Values of these types are the only ones which can be produced by
1808 <!-- ======================================================================= -->
1810 <a name="t_primitive">Primitive Types</a>
1815 <p>The primitive types are the fundamental building blocks of the LLVM
1818 <!-- _______________________________________________________________________ -->
1820 <a name="t_integer">Integer Type</a>
1826 <p>The integer type is a very simple type that simply specifies an arbitrary
1827 bit width for the integer type desired. Any bit width from 1 bit to
1828 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1835 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1839 <table class="layout">
1841 <td class="left"><tt>i1</tt></td>
1842 <td class="left">a single-bit integer.</td>
1845 <td class="left"><tt>i32</tt></td>
1846 <td class="left">a 32-bit integer.</td>
1849 <td class="left"><tt>i1942652</tt></td>
1850 <td class="left">a really big integer of over 1 million bits.</td>
1856 <!-- _______________________________________________________________________ -->
1858 <a name="t_floating">Floating Point Types</a>
1865 <tr><th>Type</th><th>Description</th></tr>
1866 <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr>
1867 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1868 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1869 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1870 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1871 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1877 <!-- _______________________________________________________________________ -->
1879 <a name="t_x86mmx">X86mmx Type</a>
1885 <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>
1894 <!-- _______________________________________________________________________ -->
1896 <a name="t_void">Void Type</a>
1902 <p>The void type does not represent any value and has no size.</p>
1911 <!-- _______________________________________________________________________ -->
1913 <a name="t_label">Label Type</a>
1919 <p>The label type represents code labels.</p>
1928 <!-- _______________________________________________________________________ -->
1930 <a name="t_metadata">Metadata Type</a>
1936 <p>The metadata type represents embedded metadata. No derived types may be
1937 created from metadata except for <a href="#t_function">function</a>
1949 <!-- ======================================================================= -->
1951 <a name="t_derived">Derived Types</a>
1956 <p>The real power in LLVM comes from the derived types in the system. This is
1957 what allows a programmer to represent arrays, functions, pointers, and other
1958 useful types. Each of these types contain one or more element types which
1959 may be a primitive type, or another derived type. For example, it is
1960 possible to have a two dimensional array, using an array as the element type
1961 of another array.</p>
1963 <!-- _______________________________________________________________________ -->
1965 <a name="t_aggregate">Aggregate Types</a>
1970 <p>Aggregate Types are a subset of derived types that can contain multiple
1971 member types. <a href="#t_array">Arrays</a> and
1972 <a href="#t_struct">structs</a> are aggregate types.
1973 <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p>
1977 <!-- _______________________________________________________________________ -->
1979 <a name="t_array">Array Type</a>
1985 <p>The array type is a very simple derived type that arranges elements
1986 sequentially in memory. The array type requires a size (number of elements)
1987 and an underlying data type.</p>
1991 [<# elements> x <elementtype>]
1994 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1995 be any type with a size.</p>
1998 <table class="layout">
2000 <td class="left"><tt>[40 x i32]</tt></td>
2001 <td class="left">Array of 40 32-bit integer values.</td>
2004 <td class="left"><tt>[41 x i32]</tt></td>
2005 <td class="left">Array of 41 32-bit integer values.</td>
2008 <td class="left"><tt>[4 x i8]</tt></td>
2009 <td class="left">Array of 4 8-bit integer values.</td>
2012 <p>Here are some examples of multidimensional arrays:</p>
2013 <table class="layout">
2015 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
2016 <td class="left">3x4 array of 32-bit integer values.</td>
2019 <td class="left"><tt>[12 x [10 x float]]</tt></td>
2020 <td class="left">12x10 array of single precision floating point values.</td>
2023 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
2024 <td class="left">2x3x4 array of 16-bit integer values.</td>
2028 <p>There is no restriction on indexing beyond the end of the array implied by
2029 a static type (though there are restrictions on indexing beyond the bounds
2030 of an allocated object in some cases). This means that single-dimension
2031 'variable sized array' addressing can be implemented in LLVM with a zero
2032 length array type. An implementation of 'pascal style arrays' in LLVM could
2033 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
2037 <!-- _______________________________________________________________________ -->
2039 <a name="t_function">Function Type</a>
2045 <p>The function type can be thought of as a function signature. It consists of
2046 a return type and a list of formal parameter types. The return type of a
2047 function type is a first class type or a void type.</p>
2051 <returntype> (<parameter list>)
2054 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
2055 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
2056 which indicates that the function takes a variable number of arguments.
2057 Variable argument functions can access their arguments with
2058 the <a href="#int_varargs">variable argument handling intrinsic</a>
2059 functions. '<tt><returntype></tt>' is any type except
2060 <a href="#t_label">label</a>.</p>
2063 <table class="layout">
2065 <td class="left"><tt>i32 (i32)</tt></td>
2066 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2068 </tr><tr class="layout">
2069 <td class="left"><tt>float (i16, i32 *) *
2071 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2072 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2073 returning <tt>float</tt>.
2075 </tr><tr class="layout">
2076 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2077 <td class="left">A vararg function that takes at least one
2078 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2079 which returns an integer. This is the signature for <tt>printf</tt> in
2082 </tr><tr class="layout">
2083 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2084 <td class="left">A function taking an <tt>i32</tt>, returning a
2085 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2092 <!-- _______________________________________________________________________ -->
2094 <a name="t_struct">Structure Type</a>
2100 <p>The structure type is used to represent a collection of data members together
2101 in memory. The elements of a structure may be any type that has a size.</p>
2103 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2104 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2105 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2106 Structures in registers are accessed using the
2107 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2108 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2110 <p>Structures may optionally be "packed" structures, which indicate that the
2111 alignment of the struct is one byte, and that there is no padding between
2112 the elements. In non-packed structs, padding between field types is inserted
2113 as defined by the TargetData string in the module, which is required to match
2114 what the underlying code generator expects.</p>
2116 <p>Structures can either be "literal" or "identified". A literal structure is
2117 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2118 types are always defined at the top level with a name. Literal types are
2119 uniqued by their contents and can never be recursive or opaque since there is
2120 no way to write one. Identified types can be recursive, can be opaqued, and are
2126 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2127 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2131 <table class="layout">
2133 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2134 <td class="left">A triple of three <tt>i32</tt> values</td>
2137 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2138 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2139 second element is a <a href="#t_pointer">pointer</a> to a
2140 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2141 an <tt>i32</tt>.</td>
2144 <td class="left"><tt><{ i8, i32 }></tt></td>
2145 <td class="left">A packed struct known to be 5 bytes in size.</td>
2151 <!-- _______________________________________________________________________ -->
2153 <a name="t_opaque">Opaque Structure Types</a>
2159 <p>Opaque structure types are used to represent named structure types that do
2160 not have a body specified. This corresponds (for example) to the C notion of
2161 a forward declared structure.</p>
2170 <table class="layout">
2172 <td class="left"><tt>opaque</tt></td>
2173 <td class="left">An opaque type.</td>
2181 <!-- _______________________________________________________________________ -->
2183 <a name="t_pointer">Pointer Type</a>
2189 <p>The pointer type is used to specify memory locations.
2190 Pointers are commonly used to reference objects in memory.</p>
2192 <p>Pointer types may have an optional address space attribute defining the
2193 numbered address space where the pointed-to object resides. The default
2194 address space is number zero. The semantics of non-zero address
2195 spaces are target-specific.</p>
2197 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2198 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2206 <table class="layout">
2208 <td class="left"><tt>[4 x i32]*</tt></td>
2209 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2210 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2213 <td class="left"><tt>i32 (i32*) *</tt></td>
2214 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2215 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2219 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2220 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2221 that resides in address space #5.</td>
2227 <!-- _______________________________________________________________________ -->
2229 <a name="t_vector">Vector Type</a>
2235 <p>A vector type is a simple derived type that represents a vector of elements.
2236 Vector types are used when multiple primitive data are operated in parallel
2237 using a single instruction (SIMD). A vector type requires a size (number of
2238 elements) and an underlying primitive data type. Vector types are considered
2239 <a href="#t_firstclass">first class</a>.</p>
2243 < <# elements> x <elementtype> >
2246 <p>The number of elements is a constant integer value larger than 0; elementtype
2247 may be any integer or floating point type, or a pointer to these types.
2248 Vectors of size zero are not allowed. </p>
2251 <table class="layout">
2253 <td class="left"><tt><4 x i32></tt></td>
2254 <td class="left">Vector of 4 32-bit integer values.</td>
2257 <td class="left"><tt><8 x float></tt></td>
2258 <td class="left">Vector of 8 32-bit floating-point values.</td>
2261 <td class="left"><tt><2 x i64></tt></td>
2262 <td class="left">Vector of 2 64-bit integer values.</td>
2265 <td class="left"><tt><4 x i64*></tt></td>
2266 <td class="left">Vector of 4 pointers to 64-bit integer values.</td>
2276 <!-- *********************************************************************** -->
2277 <h2><a name="constants">Constants</a></h2>
2278 <!-- *********************************************************************** -->
2282 <p>LLVM has several different basic types of constants. This section describes
2283 them all and their syntax.</p>
2285 <!-- ======================================================================= -->
2287 <a name="simpleconstants">Simple Constants</a>
2293 <dt><b>Boolean constants</b></dt>
2294 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2295 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2297 <dt><b>Integer constants</b></dt>
2298 <dd>Standard integers (such as '4') are constants of
2299 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2300 with integer types.</dd>
2302 <dt><b>Floating point constants</b></dt>
2303 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2304 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2305 notation (see below). The assembler requires the exact decimal value of a
2306 floating-point constant. For example, the assembler accepts 1.25 but
2307 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2308 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2310 <dt><b>Null pointer constants</b></dt>
2311 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2312 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2315 <p>The one non-intuitive notation for constants is the hexadecimal form of
2316 floating point constants. For example, the form '<tt>double
2317 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2318 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2319 constants are required (and the only time that they are generated by the
2320 disassembler) is when a floating point constant must be emitted but it cannot
2321 be represented as a decimal floating point number in a reasonable number of
2322 digits. For example, NaN's, infinities, and other special values are
2323 represented in their IEEE hexadecimal format so that assembly and disassembly
2324 do not cause any bits to change in the constants.</p>
2326 <p>When using the hexadecimal form, constants of types half, float, and double are
2327 represented using the 16-digit form shown above (which matches the IEEE754
2328 representation for double); half and float values must, however, be exactly
2329 representable as IEE754 half and single precision, respectively.
2330 Hexadecimal format is always used
2331 for long double, and there are three forms of long double. The 80-bit format
2332 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2333 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2334 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2335 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2336 currently supported target uses this format. Long doubles will only work if
2337 they match the long double format on your target. The IEEE 16-bit format
2338 (half precision) is represented by <tt>0xH</tt> followed by 4 hexadecimal
2339 digits. All hexadecimal formats are big-endian (sign bit at the left).</p>
2341 <p>There are no constants of type x86mmx.</p>
2344 <!-- ======================================================================= -->
2346 <a name="aggregateconstants"></a> <!-- old anchor -->
2347 <a name="complexconstants">Complex Constants</a>
2352 <p>Complex constants are a (potentially recursive) combination of simple
2353 constants and smaller complex constants.</p>
2356 <dt><b>Structure constants</b></dt>
2357 <dd>Structure constants are represented with notation similar to structure
2358 type definitions (a comma separated list of elements, surrounded by braces
2359 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2360 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2361 Structure constants must have <a href="#t_struct">structure type</a>, and
2362 the number and types of elements must match those specified by the
2365 <dt><b>Array constants</b></dt>
2366 <dd>Array constants are represented with notation similar to array type
2367 definitions (a comma separated list of elements, surrounded by square
2368 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2369 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2370 the number and types of elements must match those specified by the
2373 <dt><b>Vector constants</b></dt>
2374 <dd>Vector constants are represented with notation similar to vector type
2375 definitions (a comma separated list of elements, surrounded by
2376 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2377 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2378 have <a href="#t_vector">vector type</a>, and the number and types of
2379 elements must match those specified by the type.</dd>
2381 <dt><b>Zero initialization</b></dt>
2382 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2383 value to zero of <em>any</em> type, including scalar and
2384 <a href="#t_aggregate">aggregate</a> types.
2385 This is often used to avoid having to print large zero initializers
2386 (e.g. for large arrays) and is always exactly equivalent to using explicit
2387 zero initializers.</dd>
2389 <dt><b>Metadata node</b></dt>
2390 <dd>A metadata node is a structure-like constant with
2391 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2392 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2393 be interpreted as part of the instruction stream, metadata is a place to
2394 attach additional information such as debug info.</dd>
2399 <!-- ======================================================================= -->
2401 <a name="globalconstants">Global Variable and Function Addresses</a>
2406 <p>The addresses of <a href="#globalvars">global variables</a>
2407 and <a href="#functionstructure">functions</a> are always implicitly valid
2408 (link-time) constants. These constants are explicitly referenced when
2409 the <a href="#identifiers">identifier for the global</a> is used and always
2410 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2411 legal LLVM file:</p>
2413 <pre class="doc_code">
2416 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2421 <!-- ======================================================================= -->
2423 <a name="undefvalues">Undefined Values</a>
2428 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2429 indicates that the user of the value may receive an unspecified bit-pattern.
2430 Undefined values may be of any type (other than '<tt>label</tt>'
2431 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2433 <p>Undefined values are useful because they indicate to the compiler that the
2434 program is well defined no matter what value is used. This gives the
2435 compiler more freedom to optimize. Here are some examples of (potentially
2436 surprising) transformations that are valid (in pseudo IR):</p>
2439 <pre class="doc_code">
2449 <p>This is safe because all of the output bits are affected by the undef bits.
2450 Any output bit can have a zero or one depending on the input bits.</p>
2452 <pre class="doc_code">
2463 <p>These logical operations have bits that are not always affected by the input.
2464 For example, if <tt>%X</tt> has a zero bit, then the output of the
2465 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2466 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2467 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2468 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2469 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2470 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2471 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2473 <pre class="doc_code">
2474 %A = select undef, %X, %Y
2475 %B = select undef, 42, %Y
2476 %C = select %X, %Y, undef
2487 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2488 branch) conditions can go <em>either way</em>, but they have to come from one
2489 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2490 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2491 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2492 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2493 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2496 <pre class="doc_code">
2497 %A = xor undef, undef
2515 <p>This example points out that two '<tt>undef</tt>' operands are not
2516 necessarily the same. This can be surprising to people (and also matches C
2517 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2518 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2519 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2520 its value over its "live range". This is true because the variable doesn't
2521 actually <em>have a live range</em>. Instead, the value is logically read
2522 from arbitrary registers that happen to be around when needed, so the value
2523 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2524 need to have the same semantics or the core LLVM "replace all uses with"
2525 concept would not hold.</p>
2527 <pre class="doc_code">
2535 <p>These examples show the crucial difference between an <em>undefined
2536 value</em> and <em>undefined behavior</em>. An undefined value (like
2537 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2538 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2539 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2540 defined on SNaN's. However, in the second example, we can make a more
2541 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2542 arbitrary value, we are allowed to assume that it could be zero. Since a
2543 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2544 the operation does not execute at all. This allows us to delete the divide and
2545 all code after it. Because the undefined operation "can't happen", the
2546 optimizer can assume that it occurs in dead code.</p>
2548 <pre class="doc_code">
2549 a: store undef -> %X
2550 b: store %X -> undef
2556 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2557 undefined value can be assumed to not have any effect; we can assume that the
2558 value is overwritten with bits that happen to match what was already there.
2559 However, a store <em>to</em> an undefined location could clobber arbitrary
2560 memory, therefore, it has undefined behavior.</p>
2564 <!-- ======================================================================= -->
2566 <a name="poisonvalues">Poison Values</a>
2571 <p>Poison values are similar to <a href="#undefvalues">undef values</a>, however
2572 they also represent the fact that an instruction or constant expression which
2573 cannot evoke side effects has nevertheless detected a condition which results
2574 in undefined behavior.</p>
2576 <p>There is currently no way of representing a poison value in the IR; they
2577 only exist when produced by operations such as
2578 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2580 <p>Poison value behavior is defined in terms of value <i>dependence</i>:</p>
2583 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2584 their operands.</li>
2586 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2587 to their dynamic predecessor basic block.</li>
2589 <li>Function arguments depend on the corresponding actual argument values in
2590 the dynamic callers of their functions.</li>
2592 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2593 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2594 control back to them.</li>
2596 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2597 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>,
2598 or exception-throwing call instructions that dynamically transfer control
2601 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2602 referenced memory addresses, following the order in the IR
2603 (including loads and stores implied by intrinsics such as
2604 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2606 <!-- TODO: In the case of multiple threads, this only applies if the store
2607 "happens-before" the load or store. -->
2609 <!-- TODO: floating-point exception state -->
2611 <li>An instruction with externally visible side effects depends on the most
2612 recent preceding instruction with externally visible side effects, following
2613 the order in the IR. (This includes
2614 <a href="#volatile">volatile operations</a>.)</li>
2616 <li>An instruction <i>control-depends</i> on a
2617 <a href="#terminators">terminator instruction</a>
2618 if the terminator instruction has multiple successors and the instruction
2619 is always executed when control transfers to one of the successors, and
2620 may not be executed when control is transferred to another.</li>
2622 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2623 instruction if the set of instructions it otherwise depends on would be
2624 different if the terminator had transferred control to a different
2627 <li>Dependence is transitive.</li>
2631 <p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>,
2632 with the additional affect that any instruction which has a <i>dependence</i>
2633 on a poison value has undefined behavior.</p>
2635 <p>Here are some examples:</p>
2637 <pre class="doc_code">
2639 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2640 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2641 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2642 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2644 store i32 %poison, i32* @g ; Poison value stored to memory.
2645 %poison2 = load i32* @g ; Poison value loaded back from memory.
2647 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2649 %narrowaddr = bitcast i32* @g to i16*
2650 %wideaddr = bitcast i32* @g to i64*
2651 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2652 %poison4 = load i64* %wideaddr ; Returns a poison value.
2654 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2655 br i1 %cmp, label %true, label %end ; Branch to either destination.
2658 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2659 ; it has undefined behavior.
2663 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2664 ; Both edges into this PHI are
2665 ; control-dependent on %cmp, so this
2666 ; always results in a poison value.
2668 store volatile i32 0, i32* @g ; This would depend on the store in %true
2669 ; if %cmp is true, or the store in %entry
2670 ; otherwise, so this is undefined behavior.
2672 br i1 %cmp, label %second_true, label %second_end
2673 ; The same branch again, but this time the
2674 ; true block doesn't have side effects.
2681 store volatile i32 0, i32* @g ; This time, the instruction always depends
2682 ; on the store in %end. Also, it is
2683 ; control-equivalent to %end, so this is
2684 ; well-defined (ignoring earlier undefined
2685 ; behavior in this example).
2690 <!-- ======================================================================= -->
2692 <a name="blockaddress">Addresses of Basic Blocks</a>
2697 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2699 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2700 basic block in the specified function, and always has an i8* type. Taking
2701 the address of the entry block is illegal.</p>
2703 <p>This value only has defined behavior when used as an operand to the
2704 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2705 comparisons against null. Pointer equality tests between labels addresses
2706 results in undefined behavior — though, again, comparison against null
2707 is ok, and no label is equal to the null pointer. This may be passed around
2708 as an opaque pointer sized value as long as the bits are not inspected. This
2709 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2710 long as the original value is reconstituted before the <tt>indirectbr</tt>
2713 <p>Finally, some targets may provide defined semantics when using the value as
2714 the operand to an inline assembly, but that is target specific.</p>
2719 <!-- ======================================================================= -->
2721 <a name="constantexprs">Constant Expressions</a>
2726 <p>Constant expressions are used to allow expressions involving other constants
2727 to be used as constants. Constant expressions may be of
2728 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2729 operation that does not have side effects (e.g. load and call are not
2730 supported). The following is the syntax for constant expressions:</p>
2733 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2734 <dd>Truncate a constant to another type. The bit size of CST must be larger
2735 than the bit size of TYPE. Both types must be integers.</dd>
2737 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2738 <dd>Zero extend a constant to another type. The bit size of CST must be
2739 smaller than the bit size of TYPE. Both types must be integers.</dd>
2741 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2742 <dd>Sign extend a constant to another type. The bit size of CST must be
2743 smaller than the bit size of TYPE. Both types must be integers.</dd>
2745 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2746 <dd>Truncate a floating point constant to another floating point type. The
2747 size of CST must be larger than the size of TYPE. Both types must be
2748 floating point.</dd>
2750 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2751 <dd>Floating point extend a constant to another type. The size of CST must be
2752 smaller or equal to the size of TYPE. Both types must be floating
2755 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2756 <dd>Convert a floating point constant to the corresponding unsigned integer
2757 constant. TYPE must be a scalar or vector integer type. CST must be of
2758 scalar or vector floating point type. Both CST and TYPE must be scalars,
2759 or vectors of the same number of elements. If the value won't fit in the
2760 integer type, the results are undefined.</dd>
2762 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2763 <dd>Convert a floating point constant to the corresponding signed integer
2764 constant. TYPE must be a scalar or vector integer type. CST must be of
2765 scalar or vector floating point type. Both CST and TYPE must be scalars,
2766 or vectors of the same number of elements. If the value won't fit in the
2767 integer type, the results are undefined.</dd>
2769 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2770 <dd>Convert an unsigned integer constant to the corresponding floating point
2771 constant. TYPE must be a scalar or vector floating point type. CST must be
2772 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2773 vectors of the same number of elements. If the value won't fit in the
2774 floating point type, the results are undefined.</dd>
2776 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2777 <dd>Convert a signed integer constant to the corresponding floating point
2778 constant. TYPE must be a scalar or vector floating point type. CST must be
2779 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2780 vectors of the same number of elements. If the value won't fit in the
2781 floating point type, the results are undefined.</dd>
2783 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2784 <dd>Convert a pointer typed constant to the corresponding integer constant
2785 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2786 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2787 make it fit in <tt>TYPE</tt>.</dd>
2789 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2790 <dd>Convert an integer constant to a pointer constant. TYPE must be a pointer
2791 type. CST must be of integer type. The CST value is zero extended,
2792 truncated, or unchanged to make it fit in a pointer size. This one is
2793 <i>really</i> dangerous!</dd>
2795 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2796 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2797 are the same as those for the <a href="#i_bitcast">bitcast
2798 instruction</a>.</dd>
2800 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2801 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2802 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2803 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2804 instruction, the index list may have zero or more indexes, which are
2805 required to make sense for the type of "CSTPTR".</dd>
2807 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2808 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2810 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2811 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2813 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2814 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2816 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2817 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2820 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2821 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2824 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2825 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2828 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2829 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2830 constants. The index list is interpreted in a similar manner as indices in
2831 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2832 index value must be specified.</dd>
2834 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2835 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2836 constants. The index list is interpreted in a similar manner as indices in
2837 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2838 index value must be specified.</dd>
2840 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2841 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2842 be any of the <a href="#binaryops">binary</a>
2843 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2844 on operands are the same as those for the corresponding instruction
2845 (e.g. no bitwise operations on floating point values are allowed).</dd>
2852 <!-- *********************************************************************** -->
2853 <h2><a name="othervalues">Other Values</a></h2>
2854 <!-- *********************************************************************** -->
2856 <!-- ======================================================================= -->
2858 <a name="inlineasm">Inline Assembler Expressions</a>
2863 <p>LLVM supports inline assembler expressions (as opposed
2864 to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of
2865 a special value. This value represents the inline assembler as a string
2866 (containing the instructions to emit), a list of operand constraints (stored
2867 as a string), a flag that indicates whether or not the inline asm
2868 expression has side effects, and a flag indicating whether the function
2869 containing the asm needs to align its stack conservatively. An example
2870 inline assembler expression is:</p>
2872 <pre class="doc_code">
2873 i32 (i32) asm "bswap $0", "=r,r"
2876 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2877 a <a href="#i_call"><tt>call</tt></a> or an
2878 <a href="#i_invoke"><tt>invoke</tt></a> instruction.
2879 Thus, typically we have:</p>
2881 <pre class="doc_code">
2882 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2885 <p>Inline asms with side effects not visible in the constraint list must be
2886 marked as having side effects. This is done through the use of the
2887 '<tt>sideeffect</tt>' keyword, like so:</p>
2889 <pre class="doc_code">
2890 call void asm sideeffect "eieio", ""()
2893 <p>In some cases inline asms will contain code that will not work unless the
2894 stack is aligned in some way, such as calls or SSE instructions on x86,
2895 yet will not contain code that does that alignment within the asm.
2896 The compiler should make conservative assumptions about what the asm might
2897 contain and should generate its usual stack alignment code in the prologue
2898 if the '<tt>alignstack</tt>' keyword is present:</p>
2900 <pre class="doc_code">
2901 call void asm alignstack "eieio", ""()
2904 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2908 <p>TODO: The format of the asm and constraints string still need to be
2909 documented here. Constraints on what can be done (e.g. duplication, moving,
2910 etc need to be documented). This is probably best done by reference to
2911 another document that covers inline asm from a holistic perspective.</p>
2914 <!-- _______________________________________________________________________ -->
2916 <a name="inlineasm_md">Inline Asm Metadata</a>
2921 <p>The call instructions that wrap inline asm nodes may have a
2922 "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant
2923 integers. If present, the code generator will use the integer as the
2924 location cookie value when report errors through the <tt>LLVMContext</tt>
2925 error reporting mechanisms. This allows a front-end to correlate backend
2926 errors that occur with inline asm back to the source code that produced it.
2929 <pre class="doc_code">
2930 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2932 !42 = !{ i32 1234567 }
2935 <p>It is up to the front-end to make sense of the magic numbers it places in the
2936 IR. If the MDNode contains multiple constants, the code generator will use
2937 the one that corresponds to the line of the asm that the error occurs on.</p>
2943 <!-- ======================================================================= -->
2945 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2950 <p>LLVM IR allows metadata to be attached to instructions in the program that
2951 can convey extra information about the code to the optimizers and code
2952 generator. One example application of metadata is source-level debug
2953 information. There are two metadata primitives: strings and nodes. All
2954 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2955 preceding exclamation point ('<tt>!</tt>').</p>
2957 <p>A metadata string is a string surrounded by double quotes. It can contain
2958 any character by escaping non-printable characters with "<tt>\xx</tt>" where
2959 "<tt>xx</tt>" is the two digit hex code. For example:
2960 "<tt>!"test\00"</tt>".</p>
2962 <p>Metadata nodes are represented with notation similar to structure constants
2963 (a comma separated list of elements, surrounded by braces and preceded by an
2964 exclamation point). Metadata nodes can have any values as their operand. For
2967 <div class="doc_code">
2969 !{ metadata !"test\00", i32 10}
2973 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2974 metadata nodes, which can be looked up in the module symbol table. For
2977 <div class="doc_code">
2979 !foo = metadata !{!4, !3}
2983 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2984 function is using two metadata arguments:</p>
2986 <div class="doc_code">
2988 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2992 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2993 attached to the <tt>add</tt> instruction using the <tt>!dbg</tt>
2996 <div class="doc_code">
2998 %indvar.next = add i64 %indvar, 1, !dbg !21
3002 <p>More information about specific metadata nodes recognized by the optimizers
3003 and code generator is found below.</p>
3005 <!-- _______________________________________________________________________ -->
3007 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
3012 <p>In LLVM IR, memory does not have types, so LLVM's own type system is not
3013 suitable for doing TBAA. Instead, metadata is added to the IR to describe
3014 a type system of a higher level language. This can be used to implement
3015 typical C/C++ TBAA, but it can also be used to implement custom alias
3016 analysis behavior for other languages.</p>
3018 <p>The current metadata format is very simple. TBAA metadata nodes have up to
3019 three fields, e.g.:</p>
3021 <div class="doc_code">
3023 !0 = metadata !{ metadata !"an example type tree" }
3024 !1 = metadata !{ metadata !"int", metadata !0 }
3025 !2 = metadata !{ metadata !"float", metadata !0 }
3026 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
3030 <p>The first field is an identity field. It can be any value, usually
3031 a metadata string, which uniquely identifies the type. The most important
3032 name in the tree is the name of the root node. Two trees with
3033 different root node names are entirely disjoint, even if they
3034 have leaves with common names.</p>
3036 <p>The second field identifies the type's parent node in the tree, or
3037 is null or omitted for a root node. A type is considered to alias
3038 all of its descendants and all of its ancestors in the tree. Also,
3039 a type is considered to alias all types in other trees, so that
3040 bitcode produced from multiple front-ends is handled conservatively.</p>
3042 <p>If the third field is present, it's an integer which if equal to 1
3043 indicates that the type is "constant" (meaning
3044 <tt>pointsToConstantMemory</tt> should return true; see
3045 <a href="AliasAnalysis.html#OtherItfs">other useful
3046 <tt>AliasAnalysis</tt> methods</a>).</p>
3050 <!-- _______________________________________________________________________ -->
3052 <a name="fpmath">'<tt>fpmath</tt>' Metadata</a>
3057 <p><tt>fpmath</tt> metadata may be attached to any instruction of floating point
3058 type. It can be used to express the maximum acceptable error in the result of
3059 that instruction, in ULPs, thus potentially allowing the compiler to use a
3060 more efficient but less accurate method of computing it. ULP is defined as
3065 <p>If <tt>x</tt> is a real number that lies between two finite consecutive
3066 floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one
3067 of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the
3068 distance between the two non-equal finite floating-point numbers nearest
3069 <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p>
3073 <p>The metadata node shall consist of a single positive floating point number
3074 representing the maximum relative error, for example:</p>
3076 <div class="doc_code">
3078 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3084 <!-- _______________________________________________________________________ -->
3086 <a name="range">'<tt>range</tt>' Metadata</a>
3090 <p><tt>range</tt> metadata may be attached only to loads of integer types. It
3091 expresses the possible ranges the loaded value is in. The ranges are
3092 represented with a flattened list of integers. The loaded value is known to
3093 be in the union of the ranges defined by each consecutive pair. Each pair
3094 has the following properties:</p>
3096 <li>The type must match the type loaded by the instruction.</li>
3097 <li>The pair <tt>a,b</tt> represents the range <tt>[a,b)</tt>.</li>
3098 <li>Both <tt>a</tt> and <tt>b</tt> are constants.</li>
3099 <li>The range is allowed to wrap.</li>
3100 <li>The range should not represent the full or empty set. That is,
3101 <tt>a!=b</tt>. </li>
3103 <p> In addition, the pairs must be in signed order of the lower bound and
3104 they must be non-contiguous.</p>
3107 <div class="doc_code">
3109 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
3110 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3111 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
3112 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
3114 !0 = metadata !{ i8 0, i8 2 }
3115 !1 = metadata !{ i8 255, i8 2 }
3116 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
3117 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
3125 <!-- *********************************************************************** -->
3127 <a name="module_flags">Module Flags Metadata</a>
3129 <!-- *********************************************************************** -->
3133 <p>Information about the module as a whole is difficult to convey to LLVM's
3134 subsystems. The LLVM IR isn't sufficient to transmit this
3135 information. The <tt>llvm.module.flags</tt> named metadata exists in order to
3136 facilitate this. These flags are in the form of key / value pairs —
3137 much like a dictionary — making it easy for any subsystem who cares
3138 about a flag to look it up.</p>
3140 <p>The <tt>llvm.module.flags</tt> metadata contains a list of metadata
3141 triplets. Each triplet has the following form:</p>
3144 <li>The first element is a <i>behavior</i> flag, which specifies the behavior
3145 when two (or more) modules are merged together, and it encounters two (or
3146 more) metadata with the same ID. The supported behaviors are described
3149 <li>The second element is a metadata string that is a unique ID for the
3150 metadata. How each ID is interpreted is documented below.</li>
3152 <li>The third element is the value of the flag.</li>
3155 <p>When two (or more) modules are merged together, the resulting
3156 <tt>llvm.module.flags</tt> metadata is the union of the
3157 modules' <tt>llvm.module.flags</tt> metadata. The only exception being a flag
3158 with the <i>Override</i> behavior, which may override another flag's value
3161 <p>The following behaviors are supported:</p>
3163 <table border="1" cellspacing="0" cellpadding="4">
3173 <dt><b>Error</b></dt>
3174 <dd>Emits an error if two values disagree. It is an error to have an ID
3175 with both an Error and a Warning behavior.</dd>
3183 <dt><b>Warning</b></dt>
3184 <dd>Emits a warning if two values disagree.</dd>
3192 <dt><b>Require</b></dt>
3193 <dd>Emits an error when the specified value is not present or doesn't
3194 have the specified value. It is an error for two (or more)
3195 <tt>llvm.module.flags</tt> with the same ID to have the Require
3196 behavior but different values. There may be multiple Require flags
3205 <dt><b>Override</b></dt>
3206 <dd>Uses the specified value if the two values disagree. It is an
3207 error for two (or more) <tt>llvm.module.flags</tt> with the same
3208 ID to have the Override behavior but different values.</dd>
3215 <p>An example of module flags:</p>
3217 <pre class="doc_code">
3218 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3219 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3220 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3221 !3 = metadata !{ i32 3, metadata !"qux",
3223 metadata !"foo", i32 1
3226 !llvm.module.flags = !{ !0, !1, !2, !3 }
3230 <li><p>Metadata <tt>!0</tt> has the ID <tt>!"foo"</tt> and the value '1'. The
3231 behavior if two or more <tt>!"foo"</tt> flags are seen is to emit an
3232 error if their values are not equal.</p></li>
3234 <li><p>Metadata <tt>!1</tt> has the ID <tt>!"bar"</tt> and the value '37'. The
3235 behavior if two or more <tt>!"bar"</tt> flags are seen is to use the
3236 value '37' if their values are not equal.</p></li>
3238 <li><p>Metadata <tt>!2</tt> has the ID <tt>!"qux"</tt> and the value '42'. The
3239 behavior if two or more <tt>!"qux"</tt> flags are seen is to emit a
3240 warning if their values are not equal.</p></li>
3242 <li><p>Metadata <tt>!3</tt> has the ID <tt>!"qux"</tt> and the value:</p>
3244 <pre class="doc_code">
3245 metadata !{ metadata !"foo", i32 1 }
3248 <p>The behavior is to emit an error if the <tt>llvm.module.flags</tt> does
3249 not contain a flag with the ID <tt>!"foo"</tt> that has the value
3250 '1'. If two or more <tt>!"qux"</tt> flags exist, then they must have
3251 the same value or an error will be issued.</p></li>
3255 <!-- ======================================================================= -->
3257 <a name="objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a>
3262 <p>On the Mach-O platform, Objective-C stores metadata about garbage collection
3263 in a special section called "image info". The metadata consists of a version
3264 number and a bitmask specifying what types of garbage collection are
3265 supported (if any) by the file. If two or more modules are linked together
3266 their garbage collection metadata needs to be merged rather than appended
3269 <p>The Objective-C garbage collection module flags metadata consists of the
3270 following key-value pairs:</p>
3272 <table border="1" cellspacing="0" cellpadding="4">
3280 <td><tt>Objective-C Version</tt></td>
3281 <td align="left"><b>[Required]</b> — The Objective-C ABI
3282 version. Valid values are 1 and 2.</td>
3285 <td><tt>Objective-C Image Info Version</tt></td>
3286 <td align="left"><b>[Required]</b> — The version of the image info
3287 section. Currently always 0.</td>
3290 <td><tt>Objective-C Image Info Section</tt></td>
3291 <td align="left"><b>[Required]</b> — The section to place the
3292 metadata. Valid values are <tt>"__OBJC, __image_info, regular"</tt> for
3293 Objective-C ABI version 1, and <tt>"__DATA,__objc_imageinfo, regular,
3294 no_dead_strip"</tt> for Objective-C ABI version 2.</td>
3297 <td><tt>Objective-C Garbage Collection</tt></td>
3298 <td align="left"><b>[Required]</b> — Specifies whether garbage
3299 collection is supported or not. Valid values are 0, for no garbage
3300 collection, and 2, for garbage collection supported.</td>
3303 <td><tt>Objective-C GC Only</tt></td>
3304 <td align="left"><b>[Optional]</b> — Specifies that only garbage
3305 collection is supported. If present, its value must be 6. This flag
3306 requires that the <tt>Objective-C Garbage Collection</tt> flag have the
3312 <p>Some important flag interactions:</p>
3315 <li>If a module with <tt>Objective-C Garbage Collection</tt> set to 0 is
3316 merged with a module with <tt>Objective-C Garbage Collection</tt> set to
3317 2, then the resulting module has the <tt>Objective-C Garbage
3318 Collection</tt> flag set to 0.</li>
3320 <li>A module with <tt>Objective-C Garbage Collection</tt> set to 0 cannot be
3321 merged with a module with <tt>Objective-C GC Only</tt> set to 6.</li>
3328 <!-- *********************************************************************** -->
3330 <a name="intrinsic_globals">Intrinsic Global Variables</a>
3332 <!-- *********************************************************************** -->
3334 <p>LLVM has a number of "magic" global variables that contain data that affect
3335 code generation or other IR semantics. These are documented here. All globals
3336 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
3337 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
3340 <!-- ======================================================================= -->
3342 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
3347 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
3348 href="#linkage_appending">appending linkage</a>. This array contains a list of
3349 pointers to global variables and functions which may optionally have a pointer
3350 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
3352 <div class="doc_code">
3357 @llvm.used = appending global [2 x i8*] [
3359 i8* bitcast (i32* @Y to i8*)
3360 ], section "llvm.metadata"
3364 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
3365 compiler, assembler, and linker are required to treat the symbol as if there
3366 is a reference to the global that it cannot see. For example, if a variable
3367 has internal linkage and no references other than that from
3368 the <tt>@llvm.used</tt> list, it cannot be deleted. This is commonly used to
3369 represent references from inline asms and other things the compiler cannot
3370 "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p>
3372 <p>On some targets, the code generator must emit a directive to the assembler or
3373 object file to prevent the assembler and linker from molesting the
3378 <!-- ======================================================================= -->
3380 <a name="intg_compiler_used">
3381 The '<tt>llvm.compiler.used</tt>' Global Variable
3387 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
3388 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
3389 touching the symbol. On targets that support it, this allows an intelligent
3390 linker to optimize references to the symbol without being impeded as it would
3391 be by <tt>@llvm.used</tt>.</p>
3393 <p>This is a rare construct that should only be used in rare circumstances, and
3394 should not be exposed to source languages.</p>
3398 <!-- ======================================================================= -->
3400 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
3405 <div class="doc_code">
3407 %0 = type { i32, void ()* }
3408 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3412 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor
3413 functions and associated priorities. The functions referenced by this array
3414 will be called in ascending order of priority (i.e. lowest first) when the
3415 module is loaded. The order of functions with the same priority is not
3420 <!-- ======================================================================= -->
3422 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3427 <div class="doc_code">
3429 %0 = type { i32, void ()* }
3430 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3434 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions
3435 and associated priorities. The functions referenced by this array will be
3436 called in descending order of priority (i.e. highest first) when the module
3437 is loaded. The order of functions with the same priority is not defined.</p>
3443 <!-- *********************************************************************** -->
3444 <h2><a name="instref">Instruction Reference</a></h2>
3445 <!-- *********************************************************************** -->
3449 <p>The LLVM instruction set consists of several different classifications of
3450 instructions: <a href="#terminators">terminator
3451 instructions</a>, <a href="#binaryops">binary instructions</a>,
3452 <a href="#bitwiseops">bitwise binary instructions</a>,
3453 <a href="#memoryops">memory instructions</a>, and
3454 <a href="#otherops">other instructions</a>.</p>
3456 <!-- ======================================================================= -->
3458 <a name="terminators">Terminator Instructions</a>
3463 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3464 in a program ends with a "Terminator" instruction, which indicates which
3465 block should be executed after the current block is finished. These
3466 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3467 control flow, not values (the one exception being the
3468 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3470 <p>The terminator instructions are:
3471 '<a href="#i_ret"><tt>ret</tt></a>',
3472 '<a href="#i_br"><tt>br</tt></a>',
3473 '<a href="#i_switch"><tt>switch</tt></a>',
3474 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3475 '<a href="#i_invoke"><tt>invoke</tt></a>',
3476 '<a href="#i_resume"><tt>resume</tt></a>', and
3477 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3479 <!-- _______________________________________________________________________ -->
3481 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3488 ret <type> <value> <i>; Return a value from a non-void function</i>
3489 ret void <i>; Return from void function</i>
3493 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3494 a value) from a function back to the caller.</p>
3496 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3497 value and then causes control flow, and one that just causes control flow to
3501 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3502 return value. The type of the return value must be a
3503 '<a href="#t_firstclass">first class</a>' type.</p>
3505 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3506 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3507 value or a return value with a type that does not match its type, or if it
3508 has a void return type and contains a '<tt>ret</tt>' instruction with a
3512 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3513 the calling function's context. If the caller is a
3514 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3515 instruction after the call. If the caller was an
3516 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3517 the beginning of the "normal" destination block. If the instruction returns
3518 a value, that value shall set the call or invoke instruction's return
3523 ret i32 5 <i>; Return an integer value of 5</i>
3524 ret void <i>; Return from a void function</i>
3525 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3529 <!-- _______________________________________________________________________ -->
3531 <a name="i_br">'<tt>br</tt>' Instruction</a>
3538 br i1 <cond>, label <iftrue>, label <iffalse>
3539 br label <dest> <i>; Unconditional branch</i>
3543 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3544 different basic block in the current function. There are two forms of this
3545 instruction, corresponding to a conditional branch and an unconditional
3549 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3550 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3551 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3555 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3556 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3557 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3558 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3563 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3564 br i1 %cond, label %IfEqual, label %IfUnequal
3566 <a href="#i_ret">ret</a> i32 1
3568 <a href="#i_ret">ret</a> i32 0
3573 <!-- _______________________________________________________________________ -->
3575 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3582 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3586 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3587 several different places. It is a generalization of the '<tt>br</tt>'
3588 instruction, allowing a branch to occur to one of many possible
3592 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3593 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3594 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3595 The table is not allowed to contain duplicate constant entries.</p>
3598 <p>The <tt>switch</tt> instruction specifies a table of values and
3599 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3600 is searched for the given value. If the value is found, control flow is
3601 transferred to the corresponding destination; otherwise, control flow is
3602 transferred to the default destination.</p>
3604 <h5>Implementation:</h5>
3605 <p>Depending on properties of the target machine and the particular
3606 <tt>switch</tt> instruction, this instruction may be code generated in
3607 different ways. For example, it could be generated as a series of chained
3608 conditional branches or with a lookup table.</p>
3612 <i>; Emulate a conditional br instruction</i>
3613 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3614 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3616 <i>; Emulate an unconditional br instruction</i>
3617 switch i32 0, label %dest [ ]
3619 <i>; Implement a jump table:</i>
3620 switch i32 %val, label %otherwise [ i32 0, label %onzero
3622 i32 2, label %ontwo ]
3628 <!-- _______________________________________________________________________ -->
3630 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3637 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3642 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3643 within the current function, whose address is specified by
3644 "<tt>address</tt>". Address must be derived from a <a
3645 href="#blockaddress">blockaddress</a> constant.</p>
3649 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3650 rest of the arguments indicate the full set of possible destinations that the
3651 address may point to. Blocks are allowed to occur multiple times in the
3652 destination list, though this isn't particularly useful.</p>
3654 <p>This destination list is required so that dataflow analysis has an accurate
3655 understanding of the CFG.</p>
3659 <p>Control transfers to the block specified in the address argument. All
3660 possible destination blocks must be listed in the label list, otherwise this
3661 instruction has undefined behavior. This implies that jumps to labels
3662 defined in other functions have undefined behavior as well.</p>
3664 <h5>Implementation:</h5>
3666 <p>This is typically implemented with a jump through a register.</p>
3670 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3676 <!-- _______________________________________________________________________ -->
3678 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3685 <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>]
3686 to label <normal label> unwind label <exception label>
3690 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3691 function, with the possibility of control flow transfer to either the
3692 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3693 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3694 control flow will return to the "normal" label. If the callee (or any
3695 indirect callees) returns via the "<a href="#i_resume"><tt>resume</tt></a>"
3696 instruction or other exception handling mechanism, control is interrupted and
3697 continued at the dynamically nearest "exception" label.</p>
3699 <p>The '<tt>exception</tt>' label is a
3700 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3701 exception. As such, '<tt>exception</tt>' label is required to have the
3702 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3703 the information about the behavior of the program after unwinding
3704 happens, as its first non-PHI instruction. The restrictions on the
3705 "<tt>landingpad</tt>" instruction's tightly couples it to the
3706 "<tt>invoke</tt>" instruction, so that the important information contained
3707 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3711 <p>This instruction requires several arguments:</p>
3714 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3715 convention</a> the call should use. If none is specified, the call
3716 defaults to using C calling conventions.</li>
3718 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3719 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3720 '<tt>inreg</tt>' attributes are valid here.</li>
3722 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3723 function value being invoked. In most cases, this is a direct function
3724 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3725 off an arbitrary pointer to function value.</li>
3727 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3728 function to be invoked. </li>
3730 <li>'<tt>function args</tt>': argument list whose types match the function
3731 signature argument types and parameter attributes. All arguments must be
3732 of <a href="#t_firstclass">first class</a> type. If the function
3733 signature indicates the function accepts a variable number of arguments,
3734 the extra arguments can be specified.</li>
3736 <li>'<tt>normal label</tt>': the label reached when the called function
3737 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3739 <li>'<tt>exception label</tt>': the label reached when a callee returns via
3740 the <a href="#i_resume"><tt>resume</tt></a> instruction or other exception
3741 handling mechanism.</li>
3743 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3744 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3745 '<tt>readnone</tt>' attributes are valid here.</li>
3749 <p>This instruction is designed to operate as a standard
3750 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3751 primary difference is that it establishes an association with a label, which
3752 is used by the runtime library to unwind the stack.</p>
3754 <p>This instruction is used in languages with destructors to ensure that proper
3755 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3756 exception. Additionally, this is important for implementation of
3757 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3759 <p>For the purposes of the SSA form, the definition of the value returned by the
3760 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3761 block to the "normal" label. If the callee unwinds then no return value is
3766 %retval = invoke i32 @Test(i32 15) to label %Continue
3767 unwind label %TestCleanup <i>; {i32}:retval set</i>
3768 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3769 unwind label %TestCleanup <i>; {i32}:retval set</i>
3774 <!-- _______________________________________________________________________ -->
3777 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3784 resume <type> <value>
3788 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3792 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3793 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3797 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3798 (in-flight) exception whose unwinding was interrupted with
3799 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3803 resume { i8*, i32 } %exn
3808 <!-- _______________________________________________________________________ -->
3811 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3822 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3823 instruction is used to inform the optimizer that a particular portion of the
3824 code is not reachable. This can be used to indicate that the code after a
3825 no-return function cannot be reached, and other facts.</p>
3828 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3834 <!-- ======================================================================= -->
3836 <a name="binaryops">Binary Operations</a>
3841 <p>Binary operators are used to do most of the computation in a program. They
3842 require two operands of the same type, execute an operation on them, and
3843 produce a single value. The operands might represent multiple data, as is
3844 the case with the <a href="#t_vector">vector</a> data type. The result value
3845 has the same type as its operands.</p>
3847 <p>There are several different binary operators:</p>
3849 <!-- _______________________________________________________________________ -->
3851 <a name="i_add">'<tt>add</tt>' Instruction</a>
3858 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3859 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3860 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3861 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3865 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3868 <p>The two arguments to the '<tt>add</tt>' instruction must
3869 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3870 integer values. Both arguments must have identical types.</p>
3873 <p>The value produced is the integer sum of the two operands.</p>
3875 <p>If the sum has unsigned overflow, the result returned is the mathematical
3876 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3878 <p>Because LLVM integers use a two's complement representation, this instruction
3879 is appropriate for both signed and unsigned integers.</p>
3881 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3882 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3883 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3884 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3885 respectively, occurs.</p>
3889 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3894 <!-- _______________________________________________________________________ -->
3896 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3903 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3907 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3910 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3911 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3912 floating point values. Both arguments must have identical types.</p>
3915 <p>The value produced is the floating point sum of the two operands.</p>
3919 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3924 <!-- _______________________________________________________________________ -->
3926 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3933 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3934 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3935 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3936 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3940 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3943 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3944 '<tt>neg</tt>' instruction present in most other intermediate
3945 representations.</p>
3948 <p>The two arguments to the '<tt>sub</tt>' instruction must
3949 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3950 integer values. Both arguments must have identical types.</p>
3953 <p>The value produced is the integer difference of the two operands.</p>
3955 <p>If the difference has unsigned overflow, the result returned is the
3956 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3959 <p>Because LLVM integers use a two's complement representation, this instruction
3960 is appropriate for both signed and unsigned integers.</p>
3962 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3963 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3964 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3965 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3966 respectively, occurs.</p>
3970 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3971 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3976 <!-- _______________________________________________________________________ -->
3978 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3985 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3989 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3992 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3993 '<tt>fneg</tt>' instruction present in most other intermediate
3994 representations.</p>
3997 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3998 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3999 floating point values. Both arguments must have identical types.</p>
4002 <p>The value produced is the floating point difference of the two operands.</p>
4006 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
4007 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
4012 <!-- _______________________________________________________________________ -->
4014 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
4021 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4022 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4023 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4024 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4028 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
4031 <p>The two arguments to the '<tt>mul</tt>' instruction must
4032 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4033 integer values. Both arguments must have identical types.</p>
4036 <p>The value produced is the integer product of the two operands.</p>
4038 <p>If the result of the multiplication has unsigned overflow, the result
4039 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
4040 width of the result.</p>
4042 <p>Because LLVM integers use a two's complement representation, and the result
4043 is the same width as the operands, this instruction returns the correct
4044 result for both signed and unsigned integers. If a full product
4045 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
4046 be sign-extended or zero-extended as appropriate to the width of the full
4049 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
4050 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
4051 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
4052 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4053 respectively, occurs.</p>
4057 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
4062 <!-- _______________________________________________________________________ -->
4064 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
4071 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4075 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
4078 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
4079 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4080 floating point values. Both arguments must have identical types.</p>
4083 <p>The value produced is the floating point product of the two operands.</p>
4087 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
4092 <!-- _______________________________________________________________________ -->
4094 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
4101 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4102 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4106 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
4109 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
4110 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4111 values. Both arguments must have identical types.</p>
4114 <p>The value produced is the unsigned integer quotient of the two operands.</p>
4116 <p>Note that unsigned integer division and signed integer division are distinct
4117 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
4119 <p>Division by zero leads to undefined behavior.</p>
4121 <p>If the <tt>exact</tt> keyword is present, the result value of the
4122 <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a
4123 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
4128 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4133 <!-- _______________________________________________________________________ -->
4135 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
4142 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4143 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4147 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
4150 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
4151 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4152 values. Both arguments must have identical types.</p>
4155 <p>The value produced is the signed integer quotient of the two operands rounded
4158 <p>Note that signed integer division and unsigned integer division are distinct
4159 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
4161 <p>Division by zero leads to undefined behavior. Overflow also leads to
4162 undefined behavior; this is a rare case, but can occur, for example, by doing
4163 a 32-bit division of -2147483648 by -1.</p>
4165 <p>If the <tt>exact</tt> keyword is present, the result value of the
4166 <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would
4171 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4176 <!-- _______________________________________________________________________ -->
4178 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
4185 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4189 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
4192 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
4193 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4194 floating point values. Both arguments must have identical types.</p>
4197 <p>The value produced is the floating point quotient of the two operands.</p>
4201 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
4206 <!-- _______________________________________________________________________ -->
4208 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
4215 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4219 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
4220 division of its two arguments.</p>
4223 <p>The two arguments to the '<tt>urem</tt>' instruction must be
4224 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4225 values. Both arguments must have identical types.</p>
4228 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
4229 This instruction always performs an unsigned division to get the
4232 <p>Note that unsigned integer remainder and signed integer remainder are
4233 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
4235 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
4239 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4244 <!-- _______________________________________________________________________ -->
4246 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
4253 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4257 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
4258 division of its two operands. This instruction can also take
4259 <a href="#t_vector">vector</a> versions of the values in which case the
4260 elements must be integers.</p>
4263 <p>The two arguments to the '<tt>srem</tt>' instruction must be
4264 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4265 values. Both arguments must have identical types.</p>
4268 <p>This instruction returns the <i>remainder</i> of a division (where the result
4269 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
4270 <i>modulo</i> operator (where the result is either zero or has the same sign
4271 as the divisor, <tt>op2</tt>) of a value.
4272 For more information about the difference,
4273 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
4274 Math Forum</a>. For a table of how this is implemented in various languages,
4275 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
4276 Wikipedia: modulo operation</a>.</p>
4278 <p>Note that signed integer remainder and unsigned integer remainder are
4279 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
4281 <p>Taking the remainder of a division by zero leads to undefined behavior.
4282 Overflow also leads to undefined behavior; this is a rare case, but can
4283 occur, for example, by taking the remainder of a 32-bit division of
4284 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
4285 lets srem be implemented using instructions that return both the result of
4286 the division and the remainder.)</p>
4290 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4295 <!-- _______________________________________________________________________ -->
4297 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
4304 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4308 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
4309 its two operands.</p>
4312 <p>The two arguments to the '<tt>frem</tt>' instruction must be
4313 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4314 floating point values. Both arguments must have identical types.</p>
4317 <p>This instruction returns the <i>remainder</i> of a division. The remainder
4318 has the same sign as the dividend.</p>
4322 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
4329 <!-- ======================================================================= -->
4331 <a name="bitwiseops">Bitwise Binary Operations</a>
4336 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
4337 program. They are generally very efficient instructions and can commonly be
4338 strength reduced from other instructions. They require two operands of the
4339 same type, execute an operation on them, and produce a single value. The
4340 resulting value is the same type as its operands.</p>
4342 <!-- _______________________________________________________________________ -->
4344 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
4351 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4352 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4353 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4354 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4358 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
4359 a specified number of bits.</p>
4362 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
4363 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4364 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4367 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
4368 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
4369 is (statically or dynamically) negative or equal to or larger than the number
4370 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4371 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4372 shift amount in <tt>op2</tt>.</p>
4374 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
4375 <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits. If
4376 the <tt>nsw</tt> keyword is present, then the shift produces a
4377 <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree
4378 with the resultant sign bit. As such, NUW/NSW have the same semantics as
4379 they would if the shift were expressed as a mul instruction with the same
4380 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
4384 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
4385 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
4386 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
4387 <result> = shl i32 1, 32 <i>; undefined</i>
4388 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4393 <!-- _______________________________________________________________________ -->
4395 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4402 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4403 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4407 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4408 operand shifted to the right a specified number of bits with zero fill.</p>
4411 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4412 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4413 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4416 <p>This instruction always performs a logical shift right operation. The most
4417 significant bits of the result will be filled with zero bits after the shift.
4418 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4419 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4420 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4421 shift amount in <tt>op2</tt>.</p>
4423 <p>If the <tt>exact</tt> keyword is present, the result value of the
4424 <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4425 shifted out are non-zero.</p>
4430 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4431 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4432 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4433 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4434 <result> = lshr i32 1, 32 <i>; undefined</i>
4435 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4440 <!-- _______________________________________________________________________ -->
4442 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4449 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4450 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4454 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4455 operand shifted to the right a specified number of bits with sign
4459 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4460 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4461 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4464 <p>This instruction always performs an arithmetic shift right operation, The
4465 most significant bits of the result will be filled with the sign bit
4466 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4467 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4468 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4469 the corresponding shift amount in <tt>op2</tt>.</p>
4471 <p>If the <tt>exact</tt> keyword is present, the result value of the
4472 <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4473 shifted out are non-zero.</p>
4477 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4478 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4479 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4480 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4481 <result> = ashr i32 1, 32 <i>; undefined</i>
4482 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4487 <!-- _______________________________________________________________________ -->
4489 <a name="i_and">'<tt>and</tt>' Instruction</a>
4496 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4500 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4504 <p>The two arguments to the '<tt>and</tt>' instruction must be
4505 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4506 values. Both arguments must have identical types.</p>
4509 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4511 <table border="1" cellspacing="0" cellpadding="4">
4543 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4544 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4545 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4548 <!-- _______________________________________________________________________ -->
4550 <a name="i_or">'<tt>or</tt>' Instruction</a>
4557 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4561 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4565 <p>The two arguments to the '<tt>or</tt>' instruction must be
4566 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4567 values. Both arguments must have identical types.</p>
4570 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4572 <table border="1" cellspacing="0" cellpadding="4">
4604 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4605 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4606 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4611 <!-- _______________________________________________________________________ -->
4613 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4620 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4624 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4625 its two operands. The <tt>xor</tt> is used to implement the "one's
4626 complement" operation, which is the "~" operator in C.</p>
4629 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4630 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4631 values. Both arguments must have identical types.</p>
4634 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4636 <table border="1" cellspacing="0" cellpadding="4">
4668 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4669 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4670 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4671 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4678 <!-- ======================================================================= -->
4680 <a name="vectorops">Vector Operations</a>
4685 <p>LLVM supports several instructions to represent vector operations in a
4686 target-independent manner. These instructions cover the element-access and
4687 vector-specific operations needed to process vectors effectively. While LLVM
4688 does directly support these vector operations, many sophisticated algorithms
4689 will want to use target-specific intrinsics to take full advantage of a
4690 specific target.</p>
4692 <!-- _______________________________________________________________________ -->
4694 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4701 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4705 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4706 from a vector at a specified index.</p>
4710 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4711 of <a href="#t_vector">vector</a> type. The second operand is an index
4712 indicating the position from which to extract the element. The index may be
4716 <p>The result is a scalar of the same type as the element type of
4717 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4718 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4719 results are undefined.</p>
4723 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4728 <!-- _______________________________________________________________________ -->
4730 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4737 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4741 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4742 vector at a specified index.</p>
4745 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4746 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4747 whose type must equal the element type of the first operand. The third
4748 operand is an index indicating the position at which to insert the value.
4749 The index may be a variable.</p>
4752 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4753 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4754 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4755 results are undefined.</p>
4759 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4764 <!-- _______________________________________________________________________ -->
4766 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4773 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4777 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4778 from two input vectors, returning a vector with the same element type as the
4779 input and length that is the same as the shuffle mask.</p>
4782 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4783 with the same type. The third argument is a shuffle mask whose
4784 element type is always 'i32'. The result of the instruction is a vector
4785 whose length is the same as the shuffle mask and whose element type is the
4786 same as the element type of the first two operands.</p>
4788 <p>The shuffle mask operand is required to be a constant vector with either
4789 constant integer or undef values.</p>
4792 <p>The elements of the two input vectors are numbered from left to right across
4793 both of the vectors. The shuffle mask operand specifies, for each element of
4794 the result vector, which element of the two input vectors the result element
4795 gets. The element selector may be undef (meaning "don't care") and the
4796 second operand may be undef if performing a shuffle from only one vector.</p>
4800 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4801 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4802 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4803 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4804 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4805 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4806 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4807 <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>
4814 <!-- ======================================================================= -->
4816 <a name="aggregateops">Aggregate Operations</a>
4821 <p>LLVM supports several instructions for working with
4822 <a href="#t_aggregate">aggregate</a> values.</p>
4824 <!-- _______________________________________________________________________ -->
4826 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4833 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4837 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4838 from an <a href="#t_aggregate">aggregate</a> value.</p>
4841 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4842 of <a href="#t_struct">struct</a> or
4843 <a href="#t_array">array</a> type. The operands are constant indices to
4844 specify which value to extract in a similar manner as indices in a
4845 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4846 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4848 <li>Since the value being indexed is not a pointer, the first index is
4849 omitted and assumed to be zero.</li>
4850 <li>At least one index must be specified.</li>
4851 <li>Not only struct indices but also array indices must be in
4856 <p>The result is the value at the position in the aggregate specified by the
4861 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4866 <!-- _______________________________________________________________________ -->
4868 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4875 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4879 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4880 in an <a href="#t_aggregate">aggregate</a> value.</p>
4883 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4884 of <a href="#t_struct">struct</a> or
4885 <a href="#t_array">array</a> type. The second operand is a first-class
4886 value to insert. The following operands are constant indices indicating
4887 the position at which to insert the value in a similar manner as indices in a
4888 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4889 value to insert must have the same type as the value identified by the
4893 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4894 that of <tt>val</tt> except that the value at the position specified by the
4895 indices is that of <tt>elt</tt>.</p>
4899 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4900 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4901 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4908 <!-- ======================================================================= -->
4910 <a name="memoryops">Memory Access and Addressing Operations</a>
4915 <p>A key design point of an SSA-based representation is how it represents
4916 memory. In LLVM, no memory locations are in SSA form, which makes things
4917 very simple. This section describes how to read, write, and allocate
4920 <!-- _______________________________________________________________________ -->
4922 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4929 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4933 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4934 currently executing function, to be automatically released when this function
4935 returns to its caller. The object is always allocated in the generic address
4936 space (address space zero).</p>
4939 <p>The '<tt>alloca</tt>' instruction
4940 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4941 runtime stack, returning a pointer of the appropriate type to the program.
4942 If "NumElements" is specified, it is the number of elements allocated,
4943 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4944 specified, the value result of the allocation is guaranteed to be aligned to
4945 at least that boundary. If not specified, or if zero, the target can choose
4946 to align the allocation on any convenient boundary compatible with the
4949 <p>'<tt>type</tt>' may be any sized type.</p>
4952 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4953 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4954 memory is automatically released when the function returns. The
4955 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4956 variables that must have an address available. When the function returns
4957 (either with the <tt><a href="#i_ret">ret</a></tt>
4958 or <tt><a href="#i_resume">resume</a></tt> instructions), the memory is
4959 reclaimed. Allocating zero bytes is legal, but the result is undefined.
4960 The order in which memory is allocated (ie., which way the stack grows) is
4967 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4968 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4969 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4970 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4975 <!-- _______________________________________________________________________ -->
4977 <a name="i_load">'<tt>load</tt>' Instruction</a>
4984 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
4985 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
4986 !<index> = !{ i32 1 }
4990 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4993 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4994 from which to load. The pointer must point to
4995 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4996 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4997 number or order of execution of this <tt>load</tt> with other <a
4998 href="#volatile">volatile operations</a>.</p>
5000 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
5001 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5002 argument. The <code>release</code> and <code>acq_rel</code> orderings are
5003 not valid on <code>load</code> instructions. Atomic loads produce <a
5004 href="#memorymodel">defined</a> results when they may see multiple atomic
5005 stores. The type of the pointee must be an integer type whose bit width
5006 is a power of two greater than or equal to eight and less than or equal
5007 to a target-specific size limit. <code>align</code> must be explicitly
5008 specified on atomic loads, and the load has undefined behavior if the
5009 alignment is not set to a value which is at least the size in bytes of
5010 the pointee. <code>!nontemporal</code> does not have any defined semantics
5011 for atomic loads.</p>
5013 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
5014 operation (that is, the alignment of the memory address). A value of 0 or an
5015 omitted <tt>align</tt> argument means that the operation has the preferential
5016 alignment for the target. It is the responsibility of the code emitter to
5017 ensure that the alignment information is correct. Overestimating the
5018 alignment results in undefined behavior. Underestimating the alignment may
5019 produce less efficient code. An alignment of 1 is always safe.</p>
5021 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
5022 metatadata name <index> corresponding to a metadata node with
5023 one <tt>i32</tt> entry of value 1. The existence of
5024 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
5025 and code generator that this load is not expected to be reused in the cache.
5026 The code generator may select special instructions to save cache bandwidth,
5027 such as the <tt>MOVNT</tt> instruction on x86.</p>
5029 <p>The optional <tt>!invariant.load</tt> metadata must reference a single
5030 metatadata name <index> corresponding to a metadata node with no
5031 entries. The existence of the <tt>!invariant.load</tt> metatadata on the
5032 instruction tells the optimizer and code generator that this load address
5033 points to memory which does not change value during program execution.
5034 The optimizer may then move this load around, for example, by hoisting it
5035 out of loops using loop invariant code motion.</p>
5038 <p>The location of memory pointed to is loaded. If the value being loaded is of
5039 scalar type then the number of bytes read does not exceed the minimum number
5040 of bytes needed to hold all bits of the type. For example, loading an
5041 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
5042 <tt>i20</tt> with a size that is not an integral number of bytes, the result
5043 is undefined if the value was not originally written using a store of the
5048 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5049 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
5050 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
5055 <!-- _______________________________________________________________________ -->
5057 <a name="i_store">'<tt>store</tt>' Instruction</a>
5064 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
5065 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
5069 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
5072 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
5073 and an address at which to store it. The type of the
5074 '<tt><pointer></tt>' operand must be a pointer to
5075 the <a href="#t_firstclass">first class</a> type of the
5076 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
5077 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
5078 order of execution of this <tt>store</tt> with other <a
5079 href="#volatile">volatile operations</a>.</p>
5081 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
5082 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5083 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
5084 valid on <code>store</code> instructions. Atomic loads produce <a
5085 href="#memorymodel">defined</a> results when they may see multiple atomic
5086 stores. The type of the pointee must be an integer type whose bit width
5087 is a power of two greater than or equal to eight and less than or equal
5088 to a target-specific size limit. <code>align</code> must be explicitly
5089 specified on atomic stores, and the store has undefined behavior if the
5090 alignment is not set to a value which is at least the size in bytes of
5091 the pointee. <code>!nontemporal</code> does not have any defined semantics
5092 for atomic stores.</p>
5094 <p>The optional constant "align" argument specifies the alignment of the
5095 operation (that is, the alignment of the memory address). A value of 0 or an
5096 omitted "align" argument means that the operation has the preferential
5097 alignment for the target. It is the responsibility of the code emitter to
5098 ensure that the alignment information is correct. Overestimating the
5099 alignment results in an undefined behavior. Underestimating the alignment may
5100 produce less efficient code. An alignment of 1 is always safe.</p>
5102 <p>The optional !nontemporal metadata must reference a single metatadata
5103 name <index> corresponding to a metadata node with one i32 entry of
5104 value 1. The existence of the !nontemporal metatadata on the
5105 instruction tells the optimizer and code generator that this load is
5106 not expected to be reused in the cache. The code generator may
5107 select special instructions to save cache bandwidth, such as the
5108 MOVNT instruction on x86.</p>
5112 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
5113 location specified by the '<tt><pointer></tt>' operand. If
5114 '<tt><value></tt>' is of scalar type then the number of bytes written
5115 does not exceed the minimum number of bytes needed to hold all bits of the
5116 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
5117 writing a value of a type like <tt>i20</tt> with a size that is not an
5118 integral number of bytes, it is unspecified what happens to the extra bits
5119 that do not belong to the type, but they will typically be overwritten.</p>
5123 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5124 store i32 3, i32* %ptr <i>; yields {void}</i>
5125 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
5130 <!-- _______________________________________________________________________ -->
5132 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
5139 fence [singlethread] <ordering> <i>; yields {void}</i>
5143 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
5144 between operations.</p>
5146 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
5147 href="#ordering">ordering</a> argument which defines what
5148 <i>synchronizes-with</i> edges they add. They can only be given
5149 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
5150 <code>seq_cst</code> orderings.</p>
5153 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
5154 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
5155 <code>acquire</code> ordering semantics if and only if there exist atomic
5156 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
5157 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
5158 <var>X</var> modifies <var>M</var> (either directly or through some side effect
5159 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
5160 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
5161 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
5162 than an explicit <code>fence</code>, one (but not both) of the atomic operations
5163 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
5164 <code>acquire</code> (resp.) ordering constraint and still
5165 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
5166 <i>happens-before</i> edge.</p>
5168 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
5169 having both <code>acquire</code> and <code>release</code> semantics specified
5170 above, participates in the global program order of other <code>seq_cst</code>
5171 operations and/or fences.</p>
5173 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
5174 specifies that the fence only synchronizes with other fences in the same
5175 thread. (This is useful for interacting with signal handlers.)</p>
5179 fence acquire <i>; yields {void}</i>
5180 fence singlethread seq_cst <i>; yields {void}</i>
5185 <!-- _______________________________________________________________________ -->
5187 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
5194 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
5198 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
5199 It loads a value in memory and compares it to a given value. If they are
5200 equal, it stores a new value into the memory.</p>
5203 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
5204 address to operate on, a value to compare to the value currently be at that
5205 address, and a new value to place at that address if the compared values are
5206 equal. The type of '<var><cmp></var>' must be an integer type whose
5207 bit width is a power of two greater than or equal to eight and less than
5208 or equal to a target-specific size limit. '<var><cmp></var>' and
5209 '<var><new></var>' must have the same type, and the type of
5210 '<var><pointer></var>' must be a pointer to that type. If the
5211 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
5212 optimizer is not allowed to modify the number or order of execution
5213 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
5216 <!-- FIXME: Extend allowed types. -->
5218 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
5219 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
5221 <p>The optional "<code>singlethread</code>" argument declares that the
5222 <code>cmpxchg</code> is only atomic with respect to code (usually signal
5223 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
5224 cmpxchg is atomic with respect to all other code in the system.</p>
5226 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
5227 the size in memory of the operand.
5230 <p>The contents of memory at the location specified by the
5231 '<tt><pointer></tt>' operand is read and compared to
5232 '<tt><cmp></tt>'; if the read value is the equal,
5233 '<tt><new></tt>' is written. The original value at the location
5236 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
5237 purpose of identifying <a href="#release_sequence">release sequences</a>. A
5238 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
5239 parameter determined by dropping any <code>release</code> part of the
5240 <code>cmpxchg</code>'s ordering.</p>
5243 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
5244 optimization work on ARM.)
5246 FIXME: Is a weaker ordering constraint on failure helpful in practice?
5252 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
5253 <a href="#i_br">br</a> label %loop
5256 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
5257 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
5258 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
5259 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
5260 <a href="#i_br">br</a> i1 %success, label %done, label %loop
5268 <!-- _______________________________________________________________________ -->
5270 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
5277 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
5281 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
5284 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
5285 operation to apply, an address whose value to modify, an argument to the
5286 operation. The operation must be one of the following keywords:</p>
5301 <p>The type of '<var><value></var>' must be an integer type whose
5302 bit width is a power of two greater than or equal to eight and less than
5303 or equal to a target-specific size limit. The type of the
5304 '<code><pointer></code>' operand must be a pointer to that type.
5305 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
5306 optimizer is not allowed to modify the number or order of execution of this
5307 <code>atomicrmw</code> with other <a href="#volatile">volatile
5310 <!-- FIXME: Extend allowed types. -->
5313 <p>The contents of memory at the location specified by the
5314 '<tt><pointer></tt>' operand are atomically read, modified, and written
5315 back. The original value at the location is returned. The modification is
5316 specified by the <var>operation</var> argument:</p>
5319 <li>xchg: <code>*ptr = val</code></li>
5320 <li>add: <code>*ptr = *ptr + val</code></li>
5321 <li>sub: <code>*ptr = *ptr - val</code></li>
5322 <li>and: <code>*ptr = *ptr & val</code></li>
5323 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
5324 <li>or: <code>*ptr = *ptr | val</code></li>
5325 <li>xor: <code>*ptr = *ptr ^ val</code></li>
5326 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
5327 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
5328 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
5329 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
5334 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
5339 <!-- _______________________________________________________________________ -->
5341 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
5348 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5349 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5350 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5354 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
5355 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
5356 It performs address calculation only and does not access memory.</p>
5359 <p>The first argument is always a pointer or a vector of pointers,
5360 and forms the basis of the
5361 calculation. The remaining arguments are indices that indicate which of the
5362 elements of the aggregate object are indexed. The interpretation of each
5363 index is dependent on the type being indexed into. The first index always
5364 indexes the pointer value given as the first argument, the second index
5365 indexes a value of the type pointed to (not necessarily the value directly
5366 pointed to, since the first index can be non-zero), etc. The first type
5367 indexed into must be a pointer value, subsequent types can be arrays,
5368 vectors, and structs. Note that subsequent types being indexed into
5369 can never be pointers, since that would require loading the pointer before
5370 continuing calculation.</p>
5372 <p>The type of each index argument depends on the type it is indexing into.
5373 When indexing into a (optionally packed) structure, only <tt>i32</tt>
5374 integer <b>constants</b> are allowed. When indexing into an array, pointer
5375 or vector, integers of any width are allowed, and they are not required to be
5376 constant. These integers are treated as signed values where relevant.</p>
5378 <p>For example, let's consider a C code fragment and how it gets compiled to
5381 <pre class="doc_code">
5393 int *foo(struct ST *s) {
5394 return &s[1].Z.B[5][13];
5398 <p>The LLVM code generated by Clang is:</p>
5400 <pre class="doc_code">
5401 %struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 }
5402 %struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT }
5404 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5406 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5412 <p>In the example above, the first index is indexing into the
5413 '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a
5414 '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a
5415 structure. The second index indexes into the third element of the structure,
5416 yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>'
5417 type, another structure. The third index indexes into the second element of
5418 the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The
5419 two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>'
5420 type. The '<tt>getelementptr</tt>' instruction returns a pointer to this
5421 element, thus computing a value of '<tt>i32*</tt>' type.</p>
5423 <p>Note that it is perfectly legal to index partially through a structure,
5424 returning a pointer to an inner element. Because of this, the LLVM code for
5425 the given testcase is equivalent to:</p>
5427 <pre class="doc_code">
5428 define i32* @foo(%struct.ST* %s) {
5429 %t1 = getelementptr %struct.ST* %s, i32 1 <i>; yields %struct.ST*:%t1</i>
5430 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 <i>; yields %struct.RT*:%t2</i>
5431 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5432 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5433 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5438 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5439 <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the
5440 base pointer is not an <i>in bounds</i> address of an allocated object,
5441 or if any of the addresses that would be formed by successive addition of
5442 the offsets implied by the indices to the base address with infinitely
5443 precise signed arithmetic are not an <i>in bounds</i> address of that
5444 allocated object. The <i>in bounds</i> addresses for an allocated object
5445 are all the addresses that point into the object, plus the address one
5447 In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword
5448 applies to each of the computations element-wise. </p>
5450 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5451 the base address with silently-wrapping two's complement arithmetic. If the
5452 offsets have a different width from the pointer, they are sign-extended or
5453 truncated to the width of the pointer. The result value of the
5454 <tt>getelementptr</tt> may be outside the object pointed to by the base
5455 pointer. The result value may not necessarily be used to access memory
5456 though, even if it happens to point into allocated storage. See the
5457 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5460 <p>The getelementptr instruction is often confusing. For some more insight into
5461 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5465 <i>; yields [12 x i8]*:aptr</i>
5466 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5467 <i>; yields i8*:vptr</i>
5468 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5469 <i>; yields i8*:eptr</i>
5470 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5471 <i>; yields i32*:iptr</i>
5472 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5475 <p>In cases where the pointer argument is a vector of pointers, only a
5476 single index may be used, and the number of vector elements has to be
5477 the same. For example: </p>
5478 <pre class="doc_code">
5479 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5486 <!-- ======================================================================= -->
5488 <a name="convertops">Conversion Operations</a>
5493 <p>The instructions in this category are the conversion instructions (casting)
5494 which all take a single operand and a type. They perform various bit
5495 conversions on the operand.</p>
5497 <!-- _______________________________________________________________________ -->
5499 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5506 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5510 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5511 type <tt>ty2</tt>.</p>
5514 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5515 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5516 of the same number of integers.
5517 The bit size of the <tt>value</tt> must be larger than
5518 the bit size of the destination type, <tt>ty2</tt>.
5519 Equal sized types are not allowed.</p>
5522 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5523 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5524 source size must be larger than the destination size, <tt>trunc</tt> cannot
5525 be a <i>no-op cast</i>. It will always truncate bits.</p>
5529 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5530 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5531 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5532 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5537 <!-- _______________________________________________________________________ -->
5539 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5546 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5550 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5555 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5556 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5557 of the same number of integers.
5558 The bit size of the <tt>value</tt> must be smaller than
5559 the bit size of the destination type,
5563 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5564 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5566 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5570 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5571 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5572 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5577 <!-- _______________________________________________________________________ -->
5579 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5586 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5590 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5593 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5594 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5595 of the same number of integers.
5596 The bit size of the <tt>value</tt> must be smaller than
5597 the bit size of the destination type,
5601 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5602 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5603 of the type <tt>ty2</tt>.</p>
5605 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5609 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5610 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5611 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5616 <!-- _______________________________________________________________________ -->
5618 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5625 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5629 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5633 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5634 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5635 to cast it to. The size of <tt>value</tt> must be larger than the size of
5636 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5637 <i>no-op cast</i>.</p>
5640 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5641 <a href="#t_floating">floating point</a> type to a smaller
5642 <a href="#t_floating">floating point</a> type. If the value cannot fit
5643 within the destination type, <tt>ty2</tt>, then the results are
5648 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5649 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5654 <!-- _______________________________________________________________________ -->
5656 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5663 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5667 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5668 floating point value.</p>
5671 <p>The '<tt>fpext</tt>' instruction takes a
5672 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5673 a <a href="#t_floating">floating point</a> type to cast it to. The source
5674 type must be smaller than the destination type.</p>
5677 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5678 <a href="#t_floating">floating point</a> type to a larger
5679 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5680 used to make a <i>no-op cast</i> because it always changes bits. Use
5681 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5685 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5686 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5691 <!-- _______________________________________________________________________ -->
5693 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5700 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5704 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5705 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5708 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5709 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5710 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5711 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5712 vector integer type with the same number of elements as <tt>ty</tt></p>
5715 <p>The '<tt>fptoui</tt>' instruction converts its
5716 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5717 towards zero) unsigned integer value. If the value cannot fit
5718 in <tt>ty2</tt>, the results are undefined.</p>
5722 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5723 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5724 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5729 <!-- _______________________________________________________________________ -->
5731 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5738 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5742 <p>The '<tt>fptosi</tt>' instruction converts
5743 <a href="#t_floating">floating point</a> <tt>value</tt> to
5744 type <tt>ty2</tt>.</p>
5747 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5748 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5749 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5750 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5751 vector integer type with the same number of elements as <tt>ty</tt></p>
5754 <p>The '<tt>fptosi</tt>' instruction converts its
5755 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5756 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5757 the results are undefined.</p>
5761 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5762 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5763 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5768 <!-- _______________________________________________________________________ -->
5770 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5777 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5781 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5782 integer and converts that value to the <tt>ty2</tt> type.</p>
5785 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5786 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5787 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5788 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5789 floating point type with the same number of elements as <tt>ty</tt></p>
5792 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5793 integer quantity and converts it to the corresponding floating point
5794 value. If the value cannot fit in the floating point value, the results are
5799 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5800 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5805 <!-- _______________________________________________________________________ -->
5807 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5814 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5818 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5819 and converts that value to the <tt>ty2</tt> type.</p>
5822 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5823 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5824 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5825 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5826 floating point type with the same number of elements as <tt>ty</tt></p>
5829 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5830 quantity and converts it to the corresponding floating point value. If the
5831 value cannot fit in the floating point value, the results are undefined.</p>
5835 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5836 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5841 <!-- _______________________________________________________________________ -->
5843 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5850 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5854 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of
5855 pointers <tt>value</tt> to
5856 the integer (or vector of integers) type <tt>ty2</tt>.</p>
5859 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5860 must be a a value of type <a href="#t_pointer">pointer</a> or a vector of
5861 pointers, and a type to cast it to
5862 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector
5863 of integers type.</p>
5866 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5867 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5868 truncating or zero extending that value to the size of the integer type. If
5869 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5870 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5871 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5876 %X = ptrtoint i32* %P to i8 <i>; yields truncation on 32-bit architecture</i>
5877 %Y = ptrtoint i32* %P to i64 <i>; yields zero extension on 32-bit architecture</i>
5878 %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>
5883 <!-- _______________________________________________________________________ -->
5885 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5892 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5896 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5897 pointer type, <tt>ty2</tt>.</p>
5900 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5901 value to cast, and a type to cast it to, which must be a
5902 <a href="#t_pointer">pointer</a> type.</p>
5905 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5906 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5907 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5908 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5909 than the size of a pointer then a zero extension is done. If they are the
5910 same size, nothing is done (<i>no-op cast</i>).</p>
5914 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5915 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5916 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5917 %Z = inttoptr <4 x i32> %G to <4 x i8*><i>; yields truncation of vector G to four pointers</i>
5922 <!-- _______________________________________________________________________ -->
5924 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5931 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5935 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5936 <tt>ty2</tt> without changing any bits.</p>
5939 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5940 non-aggregate first class value, and a type to cast it to, which must also be
5941 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5942 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5943 identical. If the source type is a pointer, the destination type must also be
5944 a pointer. This instruction supports bitwise conversion of vectors to
5945 integers and to vectors of other types (as long as they have the same
5949 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5950 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5951 this conversion. The conversion is done as if the <tt>value</tt> had been
5952 stored to memory and read back as type <tt>ty2</tt>.
5953 Pointer (or vector of pointers) types may only be converted to other pointer
5954 (or vector of pointers) types with this instruction. To convert
5955 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5956 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5960 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
5961 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
5962 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
5963 %Z = bitcast <2 x i32*> %V to <2 x i64*> <i>; yields <2 x i64*></i>
5970 <!-- ======================================================================= -->
5972 <a name="otherops">Other Operations</a>
5977 <p>The instructions in this category are the "miscellaneous" instructions, which
5978 defy better classification.</p>
5980 <!-- _______________________________________________________________________ -->
5982 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5989 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5993 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5994 boolean values based on comparison of its two integer, integer vector,
5995 pointer, or pointer vector operands.</p>
5998 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
5999 the condition code indicating the kind of comparison to perform. It is not a
6000 value, just a keyword. The possible condition code are:</p>
6003 <li><tt>eq</tt>: equal</li>
6004 <li><tt>ne</tt>: not equal </li>
6005 <li><tt>ugt</tt>: unsigned greater than</li>
6006 <li><tt>uge</tt>: unsigned greater or equal</li>
6007 <li><tt>ult</tt>: unsigned less than</li>
6008 <li><tt>ule</tt>: unsigned less or equal</li>
6009 <li><tt>sgt</tt>: signed greater than</li>
6010 <li><tt>sge</tt>: signed greater or equal</li>
6011 <li><tt>slt</tt>: signed less than</li>
6012 <li><tt>sle</tt>: signed less or equal</li>
6015 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
6016 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
6017 typed. They must also be identical types.</p>
6020 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
6021 condition code given as <tt>cond</tt>. The comparison performed always yields
6022 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
6023 result, as follows:</p>
6026 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
6027 <tt>false</tt> otherwise. No sign interpretation is necessary or
6030 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
6031 <tt>false</tt> otherwise. No sign interpretation is necessary or
6034 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
6035 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6037 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
6038 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6039 to <tt>op2</tt>.</li>
6041 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
6042 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6044 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
6045 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6047 <li><tt>sgt</tt>: interprets the operands as signed values and yields
6048 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6050 <li><tt>sge</tt>: interprets the operands as signed values and yields
6051 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6052 to <tt>op2</tt>.</li>
6054 <li><tt>slt</tt>: interprets the operands as signed values and yields
6055 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6057 <li><tt>sle</tt>: interprets the operands as signed values and yields
6058 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6061 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
6062 values are compared as if they were integers.</p>
6064 <p>If the operands are integer vectors, then they are compared element by
6065 element. The result is an <tt>i1</tt> vector with the same number of elements
6066 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
6070 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
6071 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
6072 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
6073 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
6074 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
6075 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
6078 <p>Note that the code generator does not yet support vector types with
6079 the <tt>icmp</tt> instruction.</p>
6083 <!-- _______________________________________________________________________ -->
6085 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
6092 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
6096 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
6097 values based on comparison of its operands.</p>
6099 <p>If the operands are floating point scalars, then the result type is a boolean
6100 (<a href="#t_integer"><tt>i1</tt></a>).</p>
6102 <p>If the operands are floating point vectors, then the result type is a vector
6103 of boolean with the same number of elements as the operands being
6107 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
6108 the condition code indicating the kind of comparison to perform. It is not a
6109 value, just a keyword. The possible condition code are:</p>
6112 <li><tt>false</tt>: no comparison, always returns false</li>
6113 <li><tt>oeq</tt>: ordered and equal</li>
6114 <li><tt>ogt</tt>: ordered and greater than </li>
6115 <li><tt>oge</tt>: ordered and greater than or equal</li>
6116 <li><tt>olt</tt>: ordered and less than </li>
6117 <li><tt>ole</tt>: ordered and less than or equal</li>
6118 <li><tt>one</tt>: ordered and not equal</li>
6119 <li><tt>ord</tt>: ordered (no nans)</li>
6120 <li><tt>ueq</tt>: unordered or equal</li>
6121 <li><tt>ugt</tt>: unordered or greater than </li>
6122 <li><tt>uge</tt>: unordered or greater than or equal</li>
6123 <li><tt>ult</tt>: unordered or less than </li>
6124 <li><tt>ule</tt>: unordered or less than or equal</li>
6125 <li><tt>une</tt>: unordered or not equal</li>
6126 <li><tt>uno</tt>: unordered (either nans)</li>
6127 <li><tt>true</tt>: no comparison, always returns true</li>
6130 <p><i>Ordered</i> means that neither operand is a QNAN while
6131 <i>unordered</i> means that either operand may be a QNAN.</p>
6133 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
6134 a <a href="#t_floating">floating point</a> type or
6135 a <a href="#t_vector">vector</a> of floating point type. They must have
6136 identical types.</p>
6139 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
6140 according to the condition code given as <tt>cond</tt>. If the operands are
6141 vectors, then the vectors are compared element by element. Each comparison
6142 performed always yields an <a href="#t_integer">i1</a> result, as
6146 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
6148 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6149 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6151 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6152 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6154 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6155 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6157 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6158 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6160 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6161 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6163 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6164 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6166 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
6168 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
6169 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6171 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
6172 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6174 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
6175 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6177 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
6178 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6180 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
6181 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6183 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
6184 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6186 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
6188 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
6193 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
6194 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
6195 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
6196 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
6199 <p>Note that the code generator does not yet support vector types with
6200 the <tt>fcmp</tt> instruction.</p>
6204 <!-- _______________________________________________________________________ -->
6206 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
6213 <result> = phi <ty> [ <val0>, <label0>], ...
6217 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
6218 SSA graph representing the function.</p>
6221 <p>The type of the incoming values is specified with the first type field. After
6222 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
6223 one pair for each predecessor basic block of the current block. Only values
6224 of <a href="#t_firstclass">first class</a> type may be used as the value
6225 arguments to the PHI node. Only labels may be used as the label
6228 <p>There must be no non-phi instructions between the start of a basic block and
6229 the PHI instructions: i.e. PHI instructions must be first in a basic
6232 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
6233 occur on the edge from the corresponding predecessor block to the current
6234 block (but after any definition of an '<tt>invoke</tt>' instruction's return
6235 value on the same edge).</p>
6238 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
6239 specified by the pair corresponding to the predecessor basic block that
6240 executed just prior to the current block.</p>
6244 Loop: ; Infinite loop that counts from 0 on up...
6245 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6246 %nextindvar = add i32 %indvar, 1
6252 <!-- _______________________________________________________________________ -->
6254 <a name="i_select">'<tt>select</tt>' Instruction</a>
6261 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
6263 <i>selty</i> is either i1 or {<N x i1>}
6267 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
6268 condition, without branching.</p>
6272 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
6273 values indicating the condition, and two values of the
6274 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
6275 vectors and the condition is a scalar, then entire vectors are selected, not
6276 individual elements.</p>
6279 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
6280 first value argument; otherwise, it returns the second value argument.</p>
6282 <p>If the condition is a vector of i1, then the value arguments must be vectors
6283 of the same size, and the selection is done element by element.</p>
6287 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
6292 <!-- _______________________________________________________________________ -->
6294 <a name="i_call">'<tt>call</tt>' Instruction</a>
6301 <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>]
6305 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
6308 <p>This instruction requires several arguments:</p>
6311 <li>The optional "tail" marker indicates that the callee function does not
6312 access any allocas or varargs in the caller. Note that calls may be
6313 marked "tail" even if they do not occur before
6314 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
6315 present, the function call is eligible for tail call optimization,
6316 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
6317 optimized into a jump</a>. The code generator may optimize calls marked
6318 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
6319 sibling call optimization</a> when the caller and callee have
6320 matching signatures, or 2) forced tail call optimization when the
6321 following extra requirements are met:
6323 <li>Caller and callee both have the calling
6324 convention <tt>fastcc</tt>.</li>
6325 <li>The call is in tail position (ret immediately follows call and ret
6326 uses value of call or is void).</li>
6327 <li>Option <tt>-tailcallopt</tt> is enabled,
6328 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
6329 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
6330 constraints are met.</a></li>
6334 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
6335 convention</a> the call should use. If none is specified, the call
6336 defaults to using C calling conventions. The calling convention of the
6337 call must match the calling convention of the target function, or else the
6338 behavior is undefined.</li>
6340 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
6341 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
6342 '<tt>inreg</tt>' attributes are valid here.</li>
6344 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
6345 type of the return value. Functions that return no value are marked
6346 <tt><a href="#t_void">void</a></tt>.</li>
6348 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
6349 being invoked. The argument types must match the types implied by this
6350 signature. This type can be omitted if the function is not varargs and if
6351 the function type does not return a pointer to a function.</li>
6353 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
6354 be invoked. In most cases, this is a direct function invocation, but
6355 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
6356 to function value.</li>
6358 <li>'<tt>function args</tt>': argument list whose types match the function
6359 signature argument types and parameter attributes. All arguments must be
6360 of <a href="#t_firstclass">first class</a> type. If the function
6361 signature indicates the function accepts a variable number of arguments,
6362 the extra arguments can be specified.</li>
6364 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
6365 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
6366 '<tt>readnone</tt>' attributes are valid here.</li>
6370 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
6371 a specified function, with its incoming arguments bound to the specified
6372 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
6373 function, control flow continues with the instruction after the function
6374 call, and the return value of the function is bound to the result
6379 %retval = call i32 @test(i32 %argc)
6380 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
6381 %X = tail call i32 @foo() <i>; yields i32</i>
6382 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
6383 call void %foo(i8 97 signext)
6385 %struct.A = type { i32, i8 }
6386 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
6387 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
6388 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
6389 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
6390 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
6393 <p>llvm treats calls to some functions with names and arguments that match the
6394 standard C99 library as being the C99 library functions, and may perform
6395 optimizations or generate code for them under that assumption. This is
6396 something we'd like to change in the future to provide better support for
6397 freestanding environments and non-C-based languages.</p>
6401 <!-- _______________________________________________________________________ -->
6403 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
6410 <resultval> = va_arg <va_list*> <arglist>, <argty>
6414 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6415 the "variable argument" area of a function call. It is used to implement the
6416 <tt>va_arg</tt> macro in C.</p>
6419 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6420 argument. It returns a value of the specified argument type and increments
6421 the <tt>va_list</tt> to point to the next argument. The actual type
6422 of <tt>va_list</tt> is target specific.</p>
6425 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6426 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6427 to the next argument. For more information, see the variable argument
6428 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6430 <p>It is legal for this instruction to be called in a function which does not
6431 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6434 <p><tt>va_arg</tt> is an LLVM instruction instead of
6435 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6439 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6441 <p>Note that the code generator does not yet fully support va_arg on many
6442 targets. Also, it does not currently support va_arg with aggregate types on
6447 <!-- _______________________________________________________________________ -->
6449 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6456 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6457 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6459 <clause> := catch <type> <value>
6460 <clause> := filter <array constant type> <array constant>
6464 <p>The '<tt>landingpad</tt>' instruction is used by
6465 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6466 system</a> to specify that a basic block is a landing pad — one where
6467 the exception lands, and corresponds to the code found in the
6468 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6469 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6470 re-entry to the function. The <tt>resultval</tt> has the
6471 type <tt>resultty</tt>.</p>
6474 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6475 function associated with the unwinding mechanism. The optional
6476 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6478 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6479 or <tt>filter</tt> — and contains the global variable representing the
6480 "type" that may be caught or filtered respectively. Unlike the
6481 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6482 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6483 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6484 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6487 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6488 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6489 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6490 calling conventions, how the personality function results are represented in
6491 LLVM IR is target specific.</p>
6493 <p>The clauses are applied in order from top to bottom. If two
6494 <tt>landingpad</tt> instructions are merged together through inlining, the
6495 clauses from the calling function are appended to the list of clauses.
6496 When the call stack is being unwound due to an exception being thrown, the
6497 exception is compared against each <tt>clause</tt> in turn. If it doesn't
6498 match any of the clauses, and the <tt>cleanup</tt> flag is not set, then
6499 unwinding continues further up the call stack.</p>
6501 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6504 <li>A landing pad block is a basic block which is the unwind destination of an
6505 '<tt>invoke</tt>' instruction.</li>
6506 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6507 first non-PHI instruction.</li>
6508 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6510 <li>A basic block that is not a landing pad block may not include a
6511 '<tt>landingpad</tt>' instruction.</li>
6512 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6513 personality function.</li>
6518 ;; A landing pad which can catch an integer.
6519 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6521 ;; A landing pad that is a cleanup.
6522 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6524 ;; A landing pad which can catch an integer and can only throw a double.
6525 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6527 filter [1 x i8**] [@_ZTId]
6536 <!-- *********************************************************************** -->
6537 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6538 <!-- *********************************************************************** -->
6542 <p>LLVM supports the notion of an "intrinsic function". These functions have
6543 well known names and semantics and are required to follow certain
6544 restrictions. Overall, these intrinsics represent an extension mechanism for
6545 the LLVM language that does not require changing all of the transformations
6546 in LLVM when adding to the language (or the bitcode reader/writer, the
6547 parser, etc...).</p>
6549 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6550 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6551 begin with this prefix. Intrinsic functions must always be external
6552 functions: you cannot define the body of intrinsic functions. Intrinsic
6553 functions may only be used in call or invoke instructions: it is illegal to
6554 take the address of an intrinsic function. Additionally, because intrinsic
6555 functions are part of the LLVM language, it is required if any are added that
6556 they be documented here.</p>
6558 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6559 family of functions that perform the same operation but on different data
6560 types. Because LLVM can represent over 8 million different integer types,
6561 overloading is used commonly to allow an intrinsic function to operate on any
6562 integer type. One or more of the argument types or the result type can be
6563 overloaded to accept any integer type. Argument types may also be defined as
6564 exactly matching a previous argument's type or the result type. This allows
6565 an intrinsic function which accepts multiple arguments, but needs all of them
6566 to be of the same type, to only be overloaded with respect to a single
6567 argument or the result.</p>
6569 <p>Overloaded intrinsics will have the names of its overloaded argument types
6570 encoded into its function name, each preceded by a period. Only those types
6571 which are overloaded result in a name suffix. Arguments whose type is matched
6572 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6573 can take an integer of any width and returns an integer of exactly the same
6574 integer width. This leads to a family of functions such as
6575 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6576 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6577 suffix is required. Because the argument's type is matched against the return
6578 type, it does not require its own name suffix.</p>
6580 <p>To learn how to add an intrinsic function, please see the
6581 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6583 <!-- ======================================================================= -->
6585 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6590 <p>Variable argument support is defined in LLVM with
6591 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6592 intrinsic functions. These functions are related to the similarly named
6593 macros defined in the <tt><stdarg.h></tt> header file.</p>
6595 <p>All of these functions operate on arguments that use a target-specific value
6596 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6597 not define what this type is, so all transformations should be prepared to
6598 handle these functions regardless of the type used.</p>
6600 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6601 instruction and the variable argument handling intrinsic functions are
6604 <pre class="doc_code">
6605 define i32 @test(i32 %X, ...) {
6606 ; Initialize variable argument processing
6608 %ap2 = bitcast i8** %ap to i8*
6609 call void @llvm.va_start(i8* %ap2)
6611 ; Read a single integer argument
6612 %tmp = va_arg i8** %ap, i32
6614 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6616 %aq2 = bitcast i8** %aq to i8*
6617 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6618 call void @llvm.va_end(i8* %aq2)
6620 ; Stop processing of arguments.
6621 call void @llvm.va_end(i8* %ap2)
6625 declare void @llvm.va_start(i8*)
6626 declare void @llvm.va_copy(i8*, i8*)
6627 declare void @llvm.va_end(i8*)
6630 <!-- _______________________________________________________________________ -->
6632 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6640 declare void %llvm.va_start(i8* <arglist>)
6644 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6645 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6648 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6651 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6652 macro available in C. In a target-dependent way, it initializes
6653 the <tt>va_list</tt> element to which the argument points, so that the next
6654 call to <tt>va_arg</tt> will produce the first variable argument passed to
6655 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6656 need to know the last argument of the function as the compiler can figure
6661 <!-- _______________________________________________________________________ -->
6663 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6670 declare void @llvm.va_end(i8* <arglist>)
6674 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6675 which has been initialized previously
6676 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6677 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6680 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6683 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6684 macro available in C. In a target-dependent way, it destroys
6685 the <tt>va_list</tt> element to which the argument points. Calls
6686 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6687 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6688 with calls to <tt>llvm.va_end</tt>.</p>
6692 <!-- _______________________________________________________________________ -->
6694 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6701 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6705 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6706 from the source argument list to the destination argument list.</p>
6709 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6710 The second argument is a pointer to a <tt>va_list</tt> element to copy
6714 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6715 macro available in C. In a target-dependent way, it copies the
6716 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6717 element. This intrinsic is necessary because
6718 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6719 arbitrarily complex and require, for example, memory allocation.</p>
6725 <!-- ======================================================================= -->
6727 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6732 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6733 Collection</a> (GC) requires the implementation and generation of these
6734 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6735 roots on the stack</a>, as well as garbage collector implementations that
6736 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6737 barriers. Front-ends for type-safe garbage collected languages should generate
6738 these intrinsics to make use of the LLVM garbage collectors. For more details,
6739 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6742 <p>The garbage collection intrinsics only operate on objects in the generic
6743 address space (address space zero).</p>
6745 <!-- _______________________________________________________________________ -->
6747 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6754 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6758 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6759 the code generator, and allows some metadata to be associated with it.</p>
6762 <p>The first argument specifies the address of a stack object that contains the
6763 root pointer. The second pointer (which must be either a constant or a
6764 global value address) contains the meta-data to be associated with the
6768 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6769 location. At compile-time, the code generator generates information to allow
6770 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6771 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6776 <!-- _______________________________________________________________________ -->
6778 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6785 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6789 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6790 locations, allowing garbage collector implementations that require read
6794 <p>The second argument is the address to read from, which should be an address
6795 allocated from the garbage collector. The first object is a pointer to the
6796 start of the referenced object, if needed by the language runtime (otherwise
6800 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6801 instruction, but may be replaced with substantially more complex code by the
6802 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6803 may only be used in a function which <a href="#gc">specifies a GC
6808 <!-- _______________________________________________________________________ -->
6810 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6817 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6821 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6822 locations, allowing garbage collector implementations that require write
6823 barriers (such as generational or reference counting collectors).</p>
6826 <p>The first argument is the reference to store, the second is the start of the
6827 object to store it to, and the third is the address of the field of Obj to
6828 store to. If the runtime does not require a pointer to the object, Obj may
6832 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6833 instruction, but may be replaced with substantially more complex code by the
6834 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6835 may only be used in a function which <a href="#gc">specifies a GC
6842 <!-- ======================================================================= -->
6844 <a name="int_codegen">Code Generator Intrinsics</a>
6849 <p>These intrinsics are provided by LLVM to expose special features that may
6850 only be implemented with code generator support.</p>
6852 <!-- _______________________________________________________________________ -->
6854 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6861 declare i8 *@llvm.returnaddress(i32 <level>)
6865 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6866 target-specific value indicating the return address of the current function
6867 or one of its callers.</p>
6870 <p>The argument to this intrinsic indicates which function to return the address
6871 for. Zero indicates the calling function, one indicates its caller, etc.
6872 The argument is <b>required</b> to be a constant integer value.</p>
6875 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6876 indicating the return address of the specified call frame, or zero if it
6877 cannot be identified. The value returned by this intrinsic is likely to be
6878 incorrect or 0 for arguments other than zero, so it should only be used for
6879 debugging purposes.</p>
6881 <p>Note that calling this intrinsic does not prevent function inlining or other
6882 aggressive transformations, so the value returned may not be that of the
6883 obvious source-language caller.</p>
6887 <!-- _______________________________________________________________________ -->
6889 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6896 declare i8* @llvm.frameaddress(i32 <level>)
6900 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6901 target-specific frame pointer value for the specified stack frame.</p>
6904 <p>The argument to this intrinsic indicates which function to return the frame
6905 pointer for. Zero indicates the calling function, one indicates its caller,
6906 etc. The argument is <b>required</b> to be a constant integer value.</p>
6909 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6910 indicating the frame address of the specified call frame, or zero if it
6911 cannot be identified. The value returned by this intrinsic is likely to be
6912 incorrect or 0 for arguments other than zero, so it should only be used for
6913 debugging purposes.</p>
6915 <p>Note that calling this intrinsic does not prevent function inlining or other
6916 aggressive transformations, so the value returned may not be that of the
6917 obvious source-language caller.</p>
6921 <!-- _______________________________________________________________________ -->
6923 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6930 declare i8* @llvm.stacksave()
6934 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6935 of the function stack, for use
6936 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6937 useful for implementing language features like scoped automatic variable
6938 sized arrays in C99.</p>
6941 <p>This intrinsic returns a opaque pointer value that can be passed
6942 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6943 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6944 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6945 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6946 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6947 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6951 <!-- _______________________________________________________________________ -->
6953 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6960 declare void @llvm.stackrestore(i8* %ptr)
6964 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6965 the function stack to the state it was in when the
6966 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6967 executed. This is useful for implementing language features like scoped
6968 automatic variable sized arrays in C99.</p>
6971 <p>See the description
6972 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6976 <!-- _______________________________________________________________________ -->
6978 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6985 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6989 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6990 insert a prefetch instruction if supported; otherwise, it is a noop.
6991 Prefetches have no effect on the behavior of the program but can change its
6992 performance characteristics.</p>
6995 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6996 specifier determining if the fetch should be for a read (0) or write (1),
6997 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
6998 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6999 specifies whether the prefetch is performed on the data (1) or instruction (0)
7000 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
7001 must be constant integers.</p>
7004 <p>This intrinsic does not modify the behavior of the program. In particular,
7005 prefetches cannot trap and do not produce a value. On targets that support
7006 this intrinsic, the prefetch can provide hints to the processor cache for
7007 better performance.</p>
7011 <!-- _______________________________________________________________________ -->
7013 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
7020 declare void @llvm.pcmarker(i32 <id>)
7024 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
7025 Counter (PC) in a region of code to simulators and other tools. The method
7026 is target specific, but it is expected that the marker will use exported
7027 symbols to transmit the PC of the marker. The marker makes no guarantees
7028 that it will remain with any specific instruction after optimizations. It is
7029 possible that the presence of a marker will inhibit optimizations. The
7030 intended use is to be inserted after optimizations to allow correlations of
7031 simulation runs.</p>
7034 <p><tt>id</tt> is a numerical id identifying the marker.</p>
7037 <p>This intrinsic does not modify the behavior of the program. Backends that do
7038 not support this intrinsic may ignore it.</p>
7042 <!-- _______________________________________________________________________ -->
7044 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
7051 declare i64 @llvm.readcyclecounter()
7055 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
7056 counter register (or similar low latency, high accuracy clocks) on those
7057 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7058 should map to RPCC. As the backing counters overflow quickly (on the order
7059 of 9 seconds on alpha), this should only be used for small timings.</p>
7062 <p>When directly supported, reading the cycle counter should not modify any
7063 memory. Implementations are allowed to either return a application specific
7064 value or a system wide value. On backends without support, this is lowered
7065 to a constant 0.</p>
7071 <!-- ======================================================================= -->
7073 <a name="int_libc">Standard C Library Intrinsics</a>
7078 <p>LLVM provides intrinsics for a few important standard C library functions.
7079 These intrinsics allow source-language front-ends to pass information about
7080 the alignment of the pointer arguments to the code generator, providing
7081 opportunity for more efficient code generation.</p>
7083 <!-- _______________________________________________________________________ -->
7085 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
7091 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
7092 integer bit width and for different address spaces. Not all targets support
7093 all bit widths however.</p>
7096 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7097 i32 <len>, i32 <align>, i1 <isvolatile>)
7098 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7099 i64 <len>, i32 <align>, i1 <isvolatile>)
7103 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7104 source location to the destination location.</p>
7106 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
7107 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7108 and the pointers can be in specified address spaces.</p>
7112 <p>The first argument is a pointer to the destination, the second is a pointer
7113 to the source. The third argument is an integer argument specifying the
7114 number of bytes to copy, the fourth argument is the alignment of the
7115 source and destination locations, and the fifth is a boolean indicating a
7116 volatile access.</p>
7118 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7119 then the caller guarantees that both the source and destination pointers are
7120 aligned to that boundary.</p>
7122 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7123 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
7124 The detailed access behavior is not very cleanly specified and it is unwise
7125 to depend on it.</p>
7129 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7130 source location to the destination location, which are not allowed to
7131 overlap. It copies "len" bytes of memory over. If the argument is known to
7132 be aligned to some boundary, this can be specified as the fourth argument,
7133 otherwise it should be set to 0 or 1.</p>
7137 <!-- _______________________________________________________________________ -->
7139 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
7145 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
7146 width and for different address space. Not all targets support all bit
7150 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7151 i32 <len>, i32 <align>, i1 <isvolatile>)
7152 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7153 i64 <len>, i32 <align>, i1 <isvolatile>)
7157 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
7158 source location to the destination location. It is similar to the
7159 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
7162 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
7163 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7164 and the pointers can be in specified address spaces.</p>
7168 <p>The first argument is a pointer to the destination, the second is a pointer
7169 to the source. The third argument is an integer argument specifying the
7170 number of bytes to copy, the fourth argument is the alignment of the
7171 source and destination locations, and the fifth is a boolean indicating a
7172 volatile access.</p>
7174 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7175 then the caller guarantees that the source and destination pointers are
7176 aligned to that boundary.</p>
7178 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7179 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
7180 The detailed access behavior is not very cleanly specified and it is unwise
7181 to depend on it.</p>
7185 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
7186 source location to the destination location, which may overlap. It copies
7187 "len" bytes of memory over. If the argument is known to be aligned to some
7188 boundary, this can be specified as the fourth argument, otherwise it should
7189 be set to 0 or 1.</p>
7193 <!-- _______________________________________________________________________ -->
7195 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
7201 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
7202 width and for different address spaces. However, not all targets support all
7206 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7207 i32 <len>, i32 <align>, i1 <isvolatile>)
7208 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7209 i64 <len>, i32 <align>, i1 <isvolatile>)
7213 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
7214 particular byte value.</p>
7216 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
7217 intrinsic does not return a value and takes extra alignment/volatile
7218 arguments. Also, the destination can be in an arbitrary address space.</p>
7221 <p>The first argument is a pointer to the destination to fill, the second is the
7222 byte value with which to fill it, the third argument is an integer argument
7223 specifying the number of bytes to fill, and the fourth argument is the known
7224 alignment of the destination location.</p>
7226 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7227 then the caller guarantees that the destination pointer is aligned to that
7230 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7231 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
7232 The detailed access behavior is not very cleanly specified and it is unwise
7233 to depend on it.</p>
7236 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
7237 at the destination location. If the argument is known to be aligned to some
7238 boundary, this can be specified as the fourth argument, otherwise it should
7239 be set to 0 or 1.</p>
7243 <!-- _______________________________________________________________________ -->
7245 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
7251 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
7252 floating point or vector of floating point type. Not all targets support all
7256 declare float @llvm.sqrt.f32(float %Val)
7257 declare double @llvm.sqrt.f64(double %Val)
7258 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7259 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7260 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7264 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
7265 returning the same value as the libm '<tt>sqrt</tt>' functions would.
7266 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
7267 behavior for negative numbers other than -0.0 (which allows for better
7268 optimization, because there is no need to worry about errno being
7269 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
7272 <p>The argument and return value are floating point numbers of the same
7276 <p>This function returns the sqrt of the specified operand if it is a
7277 nonnegative floating point number.</p>
7281 <!-- _______________________________________________________________________ -->
7283 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
7289 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
7290 floating point or vector of floating point type. Not all targets support all
7294 declare float @llvm.powi.f32(float %Val, i32 %power)
7295 declare double @llvm.powi.f64(double %Val, i32 %power)
7296 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7297 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7298 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7302 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
7303 specified (positive or negative) power. The order of evaluation of
7304 multiplications is not defined. When a vector of floating point type is
7305 used, the second argument remains a scalar integer value.</p>
7308 <p>The second argument is an integer power, and the first is a value to raise to
7312 <p>This function returns the first value raised to the second power with an
7313 unspecified sequence of rounding operations.</p>
7317 <!-- _______________________________________________________________________ -->
7319 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
7325 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
7326 floating point or vector of floating point type. Not all targets support all
7330 declare float @llvm.sin.f32(float %Val)
7331 declare double @llvm.sin.f64(double %Val)
7332 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7333 declare fp128 @llvm.sin.f128(fp128 %Val)
7334 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7338 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
7341 <p>The argument and return value are floating point numbers of the same
7345 <p>This function returns the sine of the specified operand, returning the same
7346 values as the libm <tt>sin</tt> functions would, and handles error conditions
7347 in the same way.</p>
7351 <!-- _______________________________________________________________________ -->
7353 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
7359 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
7360 floating point or vector of floating point type. Not all targets support all
7364 declare float @llvm.cos.f32(float %Val)
7365 declare double @llvm.cos.f64(double %Val)
7366 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7367 declare fp128 @llvm.cos.f128(fp128 %Val)
7368 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7372 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
7375 <p>The argument and return value are floating point numbers of the same
7379 <p>This function returns the cosine of the specified operand, returning the same
7380 values as the libm <tt>cos</tt> functions would, and handles error conditions
7381 in the same way.</p>
7385 <!-- _______________________________________________________________________ -->
7387 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
7393 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
7394 floating point or vector of floating point type. Not all targets support all
7398 declare float @llvm.pow.f32(float %Val, float %Power)
7399 declare double @llvm.pow.f64(double %Val, double %Power)
7400 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7401 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7402 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7406 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
7407 specified (positive or negative) power.</p>
7410 <p>The second argument is a floating point power, and the first is a value to
7411 raise to that power.</p>
7414 <p>This function returns the first value raised to the second power, returning
7415 the same values as the libm <tt>pow</tt> functions would, and handles error
7416 conditions in the same way.</p>
7420 <!-- _______________________________________________________________________ -->
7422 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7428 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7429 floating point or vector of floating point type. Not all targets support all
7433 declare float @llvm.exp.f32(float %Val)
7434 declare double @llvm.exp.f64(double %Val)
7435 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7436 declare fp128 @llvm.exp.f128(fp128 %Val)
7437 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7441 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7444 <p>The argument and return value are floating point numbers of the same
7448 <p>This function returns the same values as the libm <tt>exp</tt> functions
7449 would, and handles error conditions in the same way.</p>
7453 <!-- _______________________________________________________________________ -->
7455 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7461 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7462 floating point or vector of floating point type. Not all targets support all
7466 declare float @llvm.log.f32(float %Val)
7467 declare double @llvm.log.f64(double %Val)
7468 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7469 declare fp128 @llvm.log.f128(fp128 %Val)
7470 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7474 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7477 <p>The argument and return value are floating point numbers of the same
7481 <p>This function returns the same values as the libm <tt>log</tt> functions
7482 would, and handles error conditions in the same way.</p>
7486 <!-- _______________________________________________________________________ -->
7488 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7494 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7495 floating point or vector of floating point type. Not all targets support all
7499 declare float @llvm.fma.f32(float %a, float %b, float %c)
7500 declare double @llvm.fma.f64(double %a, double %b, double %c)
7501 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7502 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7503 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7507 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7511 <p>The argument and return value are floating point numbers of the same
7515 <p>This function returns the same values as the libm <tt>fma</tt> functions
7520 <!-- _______________________________________________________________________ -->
7522 <a name="int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a>
7528 <p>This is an overloaded intrinsic. You can use <tt>llvm.fabs</tt> on any
7529 floating point or vector of floating point type. Not all targets support all
7533 declare float @llvm.fabs.f32(float %Val)
7534 declare double @llvm.fabs.f64(double %Val)
7535 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7536 declare fp128 @llvm.fabs.f128(fp128 %Val)
7537 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7541 <p>The '<tt>llvm.fabs.*</tt>' intrinsics return the absolute value of
7545 <p>The argument and return value are floating point numbers of the same
7549 <p>This function returns the same values as the libm <tt>fabs</tt> functions
7550 would, and handles error conditions in the same way.</p>
7554 <!-- _______________________________________________________________________ -->
7556 <a name="int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a>
7562 <p>This is an overloaded intrinsic. You can use <tt>llvm.floor</tt> on any
7563 floating point or vector of floating point type. Not all targets support all
7567 declare float @llvm.floor.f32(float %Val)
7568 declare double @llvm.floor.f64(double %Val)
7569 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7570 declare fp128 @llvm.floor.f128(fp128 %Val)
7571 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7575 <p>The '<tt>llvm.floor.*</tt>' intrinsics return the floor of
7579 <p>The argument and return value are floating point numbers of the same
7583 <p>This function returns the same values as the libm <tt>floor</tt> functions
7584 would, and handles error conditions in the same way.</p>
7590 <!-- ======================================================================= -->
7592 <a name="int_manip">Bit Manipulation Intrinsics</a>
7597 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7598 These allow efficient code generation for some algorithms.</p>
7600 <!-- _______________________________________________________________________ -->
7602 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7608 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7609 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7612 declare i16 @llvm.bswap.i16(i16 <id>)
7613 declare i32 @llvm.bswap.i32(i32 <id>)
7614 declare i64 @llvm.bswap.i64(i64 <id>)
7618 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7619 values with an even number of bytes (positive multiple of 16 bits). These
7620 are useful for performing operations on data that is not in the target's
7621 native byte order.</p>
7624 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7625 and low byte of the input i16 swapped. Similarly,
7626 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7627 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7628 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7629 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7630 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7631 more, respectively).</p>
7635 <!-- _______________________________________________________________________ -->
7637 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7643 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7644 width, or on any vector with integer elements. Not all targets support all
7645 bit widths or vector types, however.</p>
7648 declare i8 @llvm.ctpop.i8(i8 <src>)
7649 declare i16 @llvm.ctpop.i16(i16 <src>)
7650 declare i32 @llvm.ctpop.i32(i32 <src>)
7651 declare i64 @llvm.ctpop.i64(i64 <src>)
7652 declare i256 @llvm.ctpop.i256(i256 <src>)
7653 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
7657 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7661 <p>The only argument is the value to be counted. The argument may be of any
7662 integer type, or a vector with integer elements.
7663 The return type must match the argument type.</p>
7666 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7667 element of a vector.</p>
7671 <!-- _______________________________________________________________________ -->
7673 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7679 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7680 integer bit width, or any vector whose elements are integers. Not all
7681 targets support all bit widths or vector types, however.</p>
7684 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
7685 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
7686 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
7687 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
7688 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
7689 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7693 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7694 leading zeros in a variable.</p>
7697 <p>The first argument is the value to be counted. This argument may be of any
7698 integer type, or a vectory with integer element type. The return type
7699 must match the first argument type.</p>
7701 <p>The second argument must be a constant and is a flag to indicate whether the
7702 intrinsic should ensure that a zero as the first argument produces a defined
7703 result. Historically some architectures did not provide a defined result for
7704 zero values as efficiently, and many algorithms are now predicated on
7705 avoiding zero-value inputs.</p>
7708 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7709 zeros in a variable, or within each element of the vector.
7710 If <tt>src == 0</tt> then the result is the size in bits of the type of
7711 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7712 For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7716 <!-- _______________________________________________________________________ -->
7718 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7724 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7725 integer bit width, or any vector of integer elements. Not all targets
7726 support all bit widths or vector types, however.</p>
7729 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
7730 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
7731 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
7732 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
7733 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
7734 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
7738 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7742 <p>The first argument is the value to be counted. This argument may be of any
7743 integer type, or a vectory with integer element type. The return type
7744 must match the first argument type.</p>
7746 <p>The second argument must be a constant and is a flag to indicate whether the
7747 intrinsic should ensure that a zero as the first argument produces a defined
7748 result. Historically some architectures did not provide a defined result for
7749 zero values as efficiently, and many algorithms are now predicated on
7750 avoiding zero-value inputs.</p>
7753 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7754 zeros in a variable, or within each element of a vector.
7755 If <tt>src == 0</tt> then the result is the size in bits of the type of
7756 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7757 For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7763 <!-- ======================================================================= -->
7765 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7770 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7772 <!-- _______________________________________________________________________ -->
7774 <a name="int_sadd_overflow">
7775 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7782 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7783 on any integer bit width.</p>
7786 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7787 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7788 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7792 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7793 a signed addition of the two arguments, and indicate whether an overflow
7794 occurred during the signed summation.</p>
7797 <p>The arguments (%a and %b) and the first element of the result structure may
7798 be of integer types of any bit width, but they must have the same bit
7799 width. The second element of the result structure must be of
7800 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7801 undergo signed addition.</p>
7804 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7805 a signed addition of the two variables. They return a structure — the
7806 first element of which is the signed summation, and the second element of
7807 which is a bit specifying if the signed summation resulted in an
7812 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7813 %sum = extractvalue {i32, i1} %res, 0
7814 %obit = extractvalue {i32, i1} %res, 1
7815 br i1 %obit, label %overflow, label %normal
7820 <!-- _______________________________________________________________________ -->
7822 <a name="int_uadd_overflow">
7823 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7830 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7831 on any integer bit width.</p>
7834 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7835 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7836 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7840 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7841 an unsigned addition of the two arguments, and indicate whether a carry
7842 occurred during the unsigned summation.</p>
7845 <p>The arguments (%a and %b) and the first element of the result structure may
7846 be of integer types of any bit width, but they must have the same bit
7847 width. The second element of the result structure must be of
7848 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7849 undergo unsigned addition.</p>
7852 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7853 an unsigned addition of the two arguments. They return a structure —
7854 the first element of which is the sum, and the second element of which is a
7855 bit specifying if the unsigned summation resulted in a carry.</p>
7859 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7860 %sum = extractvalue {i32, i1} %res, 0
7861 %obit = extractvalue {i32, i1} %res, 1
7862 br i1 %obit, label %carry, label %normal
7867 <!-- _______________________________________________________________________ -->
7869 <a name="int_ssub_overflow">
7870 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7877 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7878 on any integer bit width.</p>
7881 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7882 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7883 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7887 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7888 a signed subtraction of the two arguments, and indicate whether an overflow
7889 occurred during the signed subtraction.</p>
7892 <p>The arguments (%a and %b) and the first element of the result structure may
7893 be of integer types of any bit width, but they must have the same bit
7894 width. The second element of the result structure must be of
7895 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7896 undergo signed subtraction.</p>
7899 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7900 a signed subtraction of the two arguments. They return a structure —
7901 the first element of which is the subtraction, and the second element of
7902 which is a bit specifying if the signed subtraction resulted in an
7907 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7908 %sum = extractvalue {i32, i1} %res, 0
7909 %obit = extractvalue {i32, i1} %res, 1
7910 br i1 %obit, label %overflow, label %normal
7915 <!-- _______________________________________________________________________ -->
7917 <a name="int_usub_overflow">
7918 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7925 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7926 on any integer bit width.</p>
7929 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7930 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7931 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7935 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7936 an unsigned subtraction of the two arguments, and indicate whether an
7937 overflow occurred during the unsigned subtraction.</p>
7940 <p>The arguments (%a and %b) and the first element of the result structure may
7941 be of integer types of any bit width, but they must have the same bit
7942 width. The second element of the result structure must be of
7943 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7944 undergo unsigned subtraction.</p>
7947 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7948 an unsigned subtraction of the two arguments. They return a structure —
7949 the first element of which is the subtraction, and the second element of
7950 which is a bit specifying if the unsigned subtraction resulted in an
7955 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7956 %sum = extractvalue {i32, i1} %res, 0
7957 %obit = extractvalue {i32, i1} %res, 1
7958 br i1 %obit, label %overflow, label %normal
7963 <!-- _______________________________________________________________________ -->
7965 <a name="int_smul_overflow">
7966 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7973 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7974 on any integer bit width.</p>
7977 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7978 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7979 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7984 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7985 a signed multiplication of the two arguments, and indicate whether an
7986 overflow occurred during the signed multiplication.</p>
7989 <p>The arguments (%a and %b) and the first element of the result structure may
7990 be of integer types of any bit width, but they must have the same bit
7991 width. The second element of the result structure must be of
7992 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7993 undergo signed multiplication.</p>
7996 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7997 a signed multiplication of the two arguments. They return a structure —
7998 the first element of which is the multiplication, and the second element of
7999 which is a bit specifying if the signed multiplication resulted in an
8004 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8005 %sum = extractvalue {i32, i1} %res, 0
8006 %obit = extractvalue {i32, i1} %res, 1
8007 br i1 %obit, label %overflow, label %normal
8012 <!-- _______________________________________________________________________ -->
8014 <a name="int_umul_overflow">
8015 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
8022 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
8023 on any integer bit width.</p>
8026 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8027 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8028 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8032 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
8033 a unsigned multiplication of the two arguments, and indicate whether an
8034 overflow occurred during the unsigned multiplication.</p>
8037 <p>The arguments (%a and %b) and the first element of the result structure may
8038 be of integer types of any bit width, but they must have the same bit
8039 width. The second element of the result structure must be of
8040 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8041 undergo unsigned multiplication.</p>
8044 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
8045 an unsigned multiplication of the two arguments. They return a structure
8046 — the first element of which is the multiplication, and the second
8047 element of which is a bit specifying if the unsigned multiplication resulted
8052 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8053 %sum = extractvalue {i32, i1} %res, 0
8054 %obit = extractvalue {i32, i1} %res, 1
8055 br i1 %obit, label %overflow, label %normal
8062 <!-- ======================================================================= -->
8064 <a name="spec_arithmetic">Specialised Arithmetic Intrinsics</a>
8067 <!-- _______________________________________________________________________ -->
8070 <a name="fmuladd">'<tt>llvm.fmuladd.*</tt>' Intrinsic</a>
8077 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8078 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8082 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsic functions represent multiply-add
8083 expressions that can be fused if the code generator determines that the fused
8084 expression would be legal and efficient.</p>
8087 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsics each take three arguments: two
8088 multiplicands, a and b, and an addend c.</p>
8091 <p>The expression:</p>
8093 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8095 <p>is equivalent to the expression a * b + c, except that rounding will not be
8096 performed between the multiplication and addition steps if the code generator
8097 fuses the operations. Fusion is not guaranteed, even if the target platform
8098 supports it. If a fused multiply-add is required the corresponding llvm.fma.*
8099 intrinsic function should be used instead.</p>
8103 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8108 <!-- ======================================================================= -->
8110 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
8115 <p>For most target platforms, half precision floating point is a storage-only
8116 format. This means that it is
8117 a dense encoding (in memory) but does not support computation in the
8120 <p>This means that code must first load the half-precision floating point
8121 value as an i16, then convert it to float with <a
8122 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
8123 Computation can then be performed on the float value (including extending to
8124 double etc). To store the value back to memory, it is first converted to
8125 float if needed, then converted to i16 with
8126 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
8127 storing as an i16 value.</p>
8129 <!-- _______________________________________________________________________ -->
8131 <a name="int_convert_to_fp16">
8132 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
8140 declare i16 @llvm.convert.to.fp16(f32 %a)
8144 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8145 a conversion from single precision floating point format to half precision
8146 floating point format.</p>
8149 <p>The intrinsic function contains single argument - the value to be
8153 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8154 a conversion from single precision floating point format to half precision
8155 floating point format. The return value is an <tt>i16</tt> which
8156 contains the converted number.</p>
8160 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8161 store i16 %res, i16* @x, align 2
8166 <!-- _______________________________________________________________________ -->
8168 <a name="int_convert_from_fp16">
8169 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
8177 declare f32 @llvm.convert.from.fp16(i16 %a)
8181 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
8182 a conversion from half precision floating point format to single precision
8183 floating point format.</p>
8186 <p>The intrinsic function contains single argument - the value to be
8190 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
8191 conversion from half single precision floating point format to single
8192 precision floating point format. The input half-float value is represented by
8193 an <tt>i16</tt> value.</p>
8197 %a = load i16* @x, align 2
8198 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8205 <!-- ======================================================================= -->
8207 <a name="int_debugger">Debugger Intrinsics</a>
8212 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
8213 prefix), are described in
8214 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
8215 Level Debugging</a> document.</p>
8219 <!-- ======================================================================= -->
8221 <a name="int_eh">Exception Handling Intrinsics</a>
8226 <p>The LLVM exception handling intrinsics (which all start with
8227 <tt>llvm.eh.</tt> prefix), are described in
8228 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
8229 Handling</a> document.</p>
8233 <!-- ======================================================================= -->
8235 <a name="int_trampoline">Trampoline Intrinsics</a>
8240 <p>These intrinsics make it possible to excise one parameter, marked with
8241 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
8242 The result is a callable
8243 function pointer lacking the nest parameter - the caller does not need to
8244 provide a value for it. Instead, the value to use is stored in advance in a
8245 "trampoline", a block of memory usually allocated on the stack, which also
8246 contains code to splice the nest value into the argument list. This is used
8247 to implement the GCC nested function address extension.</p>
8249 <p>For example, if the function is
8250 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
8251 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
8254 <pre class="doc_code">
8255 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8256 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8257 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8258 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8259 %fp = bitcast i8* %p to i32 (i32, i32)*
8262 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
8263 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
8265 <!-- _______________________________________________________________________ -->
8268 '<tt>llvm.init.trampoline</tt>' Intrinsic
8276 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8280 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
8281 turning it into a trampoline.</p>
8284 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
8285 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
8286 sufficiently aligned block of memory; this memory is written to by the
8287 intrinsic. Note that the size and the alignment are target-specific - LLVM
8288 currently provides no portable way of determining them, so a front-end that
8289 generates this intrinsic needs to have some target-specific knowledge.
8290 The <tt>func</tt> argument must hold a function bitcast to
8291 an <tt>i8*</tt>.</p>
8294 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
8295 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
8296 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
8297 which can be <a href="#int_trampoline">bitcast (to a new function) and
8298 called</a>. The new function's signature is the same as that of
8299 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
8300 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
8301 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
8302 with the same argument list, but with <tt>nval</tt> used for the missing
8303 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
8304 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
8305 to the returned function pointer is undefined.</p>
8308 <!-- _______________________________________________________________________ -->
8311 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
8319 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8323 <p>This performs any required machine-specific adjustment to the address of a
8324 trampoline (passed as <tt>tramp</tt>).</p>
8327 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
8328 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
8332 <p>On some architectures the address of the code to be executed needs to be
8333 different to the address where the trampoline is actually stored. This
8334 intrinsic returns the executable address corresponding to <tt>tramp</tt>
8335 after performing the required machine specific adjustments.
8336 The pointer returned can then be <a href="#int_trampoline"> bitcast and
8344 <!-- ======================================================================= -->
8346 <a name="int_memorymarkers">Memory Use Markers</a>
8351 <p>This class of intrinsics exists to information about the lifetime of memory
8352 objects and ranges where variables are immutable.</p>
8354 <!-- _______________________________________________________________________ -->
8356 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8363 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8367 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8368 object's lifetime.</p>
8371 <p>The first argument is a constant integer representing the size of the
8372 object, or -1 if it is variable sized. The second argument is a pointer to
8376 <p>This intrinsic indicates that before this point in the code, the value of the
8377 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8378 never be used and has an undefined value. A load from the pointer that
8379 precedes this intrinsic can be replaced with
8380 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8384 <!-- _______________________________________________________________________ -->
8386 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8393 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8397 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8398 object's lifetime.</p>
8401 <p>The first argument is a constant integer representing the size of the
8402 object, or -1 if it is variable sized. The second argument is a pointer to
8406 <p>This intrinsic indicates that after this point in the code, the value of the
8407 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8408 never be used and has an undefined value. Any stores into the memory object
8409 following this intrinsic may be removed as dead.
8413 <!-- _______________________________________________________________________ -->
8415 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8422 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8426 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8427 a memory object will not change.</p>
8430 <p>The first argument is a constant integer representing the size of the
8431 object, or -1 if it is variable sized. The second argument is a pointer to
8435 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8436 the return value, the referenced memory location is constant and
8441 <!-- _______________________________________________________________________ -->
8443 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8450 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8454 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8455 a memory object are mutable.</p>
8458 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8459 The second argument is a constant integer representing the size of the
8460 object, or -1 if it is variable sized and the third argument is a pointer
8464 <p>This intrinsic indicates that the memory is mutable again.</p>
8470 <!-- ======================================================================= -->
8472 <a name="int_general">General Intrinsics</a>
8477 <p>This class of intrinsics is designed to be generic and has no specific
8480 <!-- _______________________________________________________________________ -->
8482 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8489 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8493 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8496 <p>The first argument is a pointer to a value, the second is a pointer to a
8497 global string, the third is a pointer to a global string which is the source
8498 file name, and the last argument is the line number.</p>
8501 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8502 This can be useful for special purpose optimizations that want to look for
8503 these annotations. These have no other defined use; they are ignored by code
8504 generation and optimization.</p>
8508 <!-- _______________________________________________________________________ -->
8510 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8516 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8517 any integer bit width.</p>
8520 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8521 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8522 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8523 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8524 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8528 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8531 <p>The first argument is an integer value (result of some expression), the
8532 second is a pointer to a global string, the third is a pointer to a global
8533 string which is the source file name, and the last argument is the line
8534 number. It returns the value of the first argument.</p>
8537 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8538 arbitrary strings. This can be useful for special purpose optimizations that
8539 want to look for these annotations. These have no other defined use; they
8540 are ignored by code generation and optimization.</p>
8544 <!-- _______________________________________________________________________ -->
8546 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8553 declare void @llvm.trap() noreturn nounwind
8557 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8563 <p>This intrinsic is lowered to the target dependent trap instruction. If the
8564 target does not have a trap instruction, this intrinsic will be lowered to
8565 a call of the <tt>abort()</tt> function.</p>
8569 <!-- _______________________________________________________________________ -->
8571 <a name="int_debugtrap">'<tt>llvm.debugtrap</tt>' Intrinsic</a>
8578 declare void @llvm.debugtrap() nounwind
8582 <p>The '<tt>llvm.debugtrap</tt>' intrinsic.</p>
8588 <p>This intrinsic is lowered to code which is intended to cause an execution
8589 trap with the intention of requesting the attention of a debugger.</p>
8593 <!-- _______________________________________________________________________ -->
8595 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8602 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8606 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8607 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8608 ensure that it is placed on the stack before local variables.</p>
8611 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8612 arguments. The first argument is the value loaded from the stack
8613 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8614 that has enough space to hold the value of the guard.</p>
8617 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8618 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8619 stack. This is to ensure that if a local variable on the stack is
8620 overwritten, it will destroy the value of the guard. When the function exits,
8621 the guard on the stack is checked against the original guard. If they are
8622 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8627 <!-- _______________________________________________________________________ -->
8629 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8636 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
8637 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
8641 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8642 the optimizers to determine at compile time whether a) an operation (like
8643 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8644 runtime check for overflow isn't necessary. An object in this context means
8645 an allocation of a specific class, structure, array, or other object.</p>
8648 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8649 argument is a pointer to or into the <tt>object</tt>. The second argument
8650 is a boolean and determines whether <tt>llvm.objectsize</tt> returns 0 (if
8651 true) or -1 (if false) when the object size is unknown.
8652 The second argument only accepts constants.</p>
8655 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to a constant representing
8656 the size of the object concerned. If the size cannot be determined at compile
8657 time, <tt>llvm.objectsize</tt> returns <tt>i32/i64 -1 or 0</tt>
8658 (depending on the <tt>min</tt> argument).</p>
8661 <!-- _______________________________________________________________________ -->
8663 <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a>
8670 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
8671 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
8675 <p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the
8676 most probable) value of <tt>val</tt>, which can be used by optimizers.</p>
8679 <p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first
8680 argument is a value. The second argument is an expected value, this needs to
8681 be a constant value, variables are not allowed.</p>
8684 <p>This intrinsic is lowered to the <tt>val</tt>.</p>
8687 <!-- _______________________________________________________________________ -->
8689 <a name="int_donothing">'<tt>llvm.donothing</tt>' Intrinsic</a>
8696 declare void @llvm.donothing() nounwind readnone
8700 <p>The <tt>llvm.donothing</tt> intrinsic doesn't perform any operation. It's the
8701 only intrinsic that can be called with an invoke instruction.</p>
8707 <p>This intrinsic does nothing, and it's removed by optimizers and ignored by
8714 <!-- *********************************************************************** -->
8717 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8718 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8719 <a href="http://validator.w3.org/check/referer"><img
8720 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8722 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8723 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8724 Last modified: $Date$