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>
58 <li><a href="#fastmath">Fast-Math Flags</a></li>
61 <li><a href="#typesystem">Type System</a>
63 <li><a href="#t_classifications">Type Classifications</a></li>
64 <li><a href="#t_primitive">Primitive Types</a>
66 <li><a href="#t_integer">Integer Type</a></li>
67 <li><a href="#t_floating">Floating Point Types</a></li>
68 <li><a href="#t_x86mmx">X86mmx Type</a></li>
69 <li><a href="#t_void">Void Type</a></li>
70 <li><a href="#t_label">Label Type</a></li>
71 <li><a href="#t_metadata">Metadata Type</a></li>
74 <li><a href="#t_derived">Derived Types</a>
76 <li><a href="#t_aggregate">Aggregate Types</a>
78 <li><a href="#t_array">Array Type</a></li>
79 <li><a href="#t_struct">Structure Type</a></li>
80 <li><a href="#t_opaque">Opaque Structure Types</a></li>
81 <li><a href="#t_vector">Vector Type</a></li>
84 <li><a href="#t_function">Function Type</a></li>
85 <li><a href="#t_pointer">Pointer Type</a></li>
90 <li><a href="#constants">Constants</a>
92 <li><a href="#simpleconstants">Simple Constants</a></li>
93 <li><a href="#complexconstants">Complex Constants</a></li>
94 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
95 <li><a href="#undefvalues">Undefined Values</a></li>
96 <li><a href="#poisonvalues">Poison Values</a></li>
97 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
98 <li><a href="#constantexprs">Constant Expressions</a></li>
101 <li><a href="#othervalues">Other Values</a>
103 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
104 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a>
106 <li><a href="#tbaa">'<tt>tbaa</tt>' Metadata</a></li>
107 <li><a href="#tbaa.struct">'<tt>tbaa.struct</tt>' Metadata</a></li>
108 <li><a href="#fpmath">'<tt>fpmath</tt>' Metadata</a></li>
109 <li><a href="#range">'<tt>range</tt>' Metadata</a></li>
114 <li><a href="#module_flags">Module Flags Metadata</a>
116 <li><a href="#objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a></li>
119 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
121 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
122 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
123 Global Variable</a></li>
124 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
125 Global Variable</a></li>
126 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
127 Global Variable</a></li>
130 <li><a href="#instref">Instruction Reference</a>
132 <li><a href="#terminators">Terminator Instructions</a>
134 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
135 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
136 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
137 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
138 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
139 <li><a href="#i_resume">'<tt>resume</tt>' Instruction</a></li>
140 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
143 <li><a href="#binaryops">Binary Operations</a>
145 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
146 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
147 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
148 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
149 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
150 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
151 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
152 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
153 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
154 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
155 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
156 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
159 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
161 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
162 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
163 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
164 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
165 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
166 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
169 <li><a href="#vectorops">Vector Operations</a>
171 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
172 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
173 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
176 <li><a href="#aggregateops">Aggregate Operations</a>
178 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
179 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
182 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
184 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
185 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
186 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
187 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
188 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
189 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
190 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
193 <li><a href="#convertops">Conversion Operations</a>
195 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
196 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
197 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
198 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
199 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
200 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
201 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
202 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
203 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
204 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
205 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
206 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
209 <li><a href="#otherops">Other Operations</a>
211 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
212 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
213 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
214 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
215 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
216 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
217 <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
222 <li><a href="#intrinsics">Intrinsic Functions</a>
224 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
226 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
227 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
228 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
231 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
233 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
234 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
235 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
238 <li><a href="#int_codegen">Code Generator Intrinsics</a>
240 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
241 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
242 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
243 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
244 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
245 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
246 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
249 <li><a href="#int_libc">Standard C Library Intrinsics</a>
251 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
252 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
253 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
254 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
255 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
256 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
257 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
258 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
259 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
260 <li><a href="#int_exp2">'<tt>llvm.exp2.*</tt>' Intrinsic</a></li>
261 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
262 <li><a href="#int_log10">'<tt>llvm.log10.*</tt>' Intrinsic</a></li>
263 <li><a href="#int_log2">'<tt>llvm.log2.*</tt>' Intrinsic</a></li>
264 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
265 <li><a href="#int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a></li>
266 <li><a href="#int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a></li>
267 <li><a href="#int_ceil">'<tt>llvm.ceil.*</tt>' Intrinsic</a></li>
268 <li><a href="#int_trunc">'<tt>llvm.trunc.*</tt>' Intrinsic</a></li>
269 <li><a href="#int_rint">'<tt>llvm.rint.*</tt>' Intrinsic</a></li>
270 <li><a href="#int_nearbyint">'<tt>llvm.nearbyint.*</tt>' Intrinsic</a></li>
273 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
275 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
276 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
277 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
278 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
281 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
283 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
284 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
285 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
286 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
287 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
288 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
291 <li><a href="#spec_arithmetic">Specialised Arithmetic Intrinsics</a>
293 <li><a href="#fmuladd">'<tt>llvm.fmuladd</tt> Intrinsic</a></li>
296 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
298 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
299 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
302 <li><a href="#int_debugger">Debugger intrinsics</a></li>
303 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
304 <li><a href="#int_trampoline">Trampoline Intrinsics</a>
306 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
307 <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
310 <li><a href="#int_memorymarkers">Memory Use Markers</a>
312 <li><a href="#int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a></li>
313 <li><a href="#int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a></li>
314 <li><a href="#int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a></li>
315 <li><a href="#int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a></li>
318 <li><a href="#int_general">General intrinsics</a>
320 <li><a href="#int_var_annotation">
321 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
322 <li><a href="#int_annotation">
323 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
324 <li><a href="#int_trap">
325 '<tt>llvm.trap</tt>' Intrinsic</a></li>
326 <li><a href="#int_debugtrap">
327 '<tt>llvm.debugtrap</tt>' Intrinsic</a></li>
328 <li><a href="#int_stackprotector">
329 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
330 <li><a href="#int_objectsize">
331 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
332 <li><a href="#int_expect">
333 '<tt>llvm.expect</tt>' Intrinsic</a></li>
334 <li><a href="#int_donothing">
335 '<tt>llvm.donothing</tt>' Intrinsic</a></li>
342 <div class="doc_author">
343 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
344 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
347 <!-- *********************************************************************** -->
348 <h2><a name="abstract">Abstract</a></h2>
349 <!-- *********************************************************************** -->
353 <p>This document is a reference manual for the LLVM assembly language. LLVM is
354 a Static Single Assignment (SSA) based representation that provides type
355 safety, low-level operations, flexibility, and the capability of representing
356 'all' high-level languages cleanly. It is the common code representation
357 used throughout all phases of the LLVM compilation strategy.</p>
361 <!-- *********************************************************************** -->
362 <h2><a name="introduction">Introduction</a></h2>
363 <!-- *********************************************************************** -->
367 <p>The LLVM code representation is designed to be used in three different forms:
368 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
369 for fast loading by a Just-In-Time compiler), and as a human readable
370 assembly language representation. This allows LLVM to provide a powerful
371 intermediate representation for efficient compiler transformations and
372 analysis, while providing a natural means to debug and visualize the
373 transformations. The three different forms of LLVM are all equivalent. This
374 document describes the human readable representation and notation.</p>
376 <p>The LLVM representation aims to be light-weight and low-level while being
377 expressive, typed, and extensible at the same time. It aims to be a
378 "universal IR" of sorts, by being at a low enough level that high-level ideas
379 may be cleanly mapped to it (similar to how microprocessors are "universal
380 IR's", allowing many source languages to be mapped to them). By providing
381 type information, LLVM can be used as the target of optimizations: for
382 example, through pointer analysis, it can be proven that a C automatic
383 variable is never accessed outside of the current function, allowing it to
384 be promoted to a simple SSA value instead of a memory location.</p>
386 <!-- _______________________________________________________________________ -->
388 <a name="wellformed">Well-Formedness</a>
393 <p>It is important to note that this document describes 'well formed' LLVM
394 assembly language. There is a difference between what the parser accepts and
395 what is considered 'well formed'. For example, the following instruction is
396 syntactically okay, but not well formed:</p>
398 <pre class="doc_code">
399 %x = <a href="#i_add">add</a> i32 1, %x
402 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
403 LLVM infrastructure provides a verification pass that may be used to verify
404 that an LLVM module is well formed. This pass is automatically run by the
405 parser after parsing input assembly and by the optimizer before it outputs
406 bitcode. The violations pointed out by the verifier pass indicate bugs in
407 transformation passes or input to the parser.</p>
413 <!-- Describe the typesetting conventions here. -->
415 <!-- *********************************************************************** -->
416 <h2><a name="identifiers">Identifiers</a></h2>
417 <!-- *********************************************************************** -->
421 <p>LLVM identifiers come in two basic types: global and local. Global
422 identifiers (functions, global variables) begin with the <tt>'@'</tt>
423 character. Local identifiers (register names, types) begin with
424 the <tt>'%'</tt> character. Additionally, there are three different formats
425 for identifiers, for different purposes:</p>
428 <li>Named values are represented as a string of characters with their prefix.
429 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
430 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
431 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
432 other characters in their names can be surrounded with quotes. Special
433 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
434 ASCII code for the character in hexadecimal. In this way, any character
435 can be used in a name value, even quotes themselves.</li>
437 <li>Unnamed values are represented as an unsigned numeric value with their
438 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
440 <li>Constants, which are described in a <a href="#constants">section about
441 constants</a>, below.</li>
444 <p>LLVM requires that values start with a prefix for two reasons: Compilers
445 don't need to worry about name clashes with reserved words, and the set of
446 reserved words may be expanded in the future without penalty. Additionally,
447 unnamed identifiers allow a compiler to quickly come up with a temporary
448 variable without having to avoid symbol table conflicts.</p>
450 <p>Reserved words in LLVM are very similar to reserved words in other
451 languages. There are keywords for different opcodes
452 ('<tt><a href="#i_add">add</a></tt>',
453 '<tt><a href="#i_bitcast">bitcast</a></tt>',
454 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
455 ('<tt><a href="#t_void">void</a></tt>',
456 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
457 reserved words cannot conflict with variable names, because none of them
458 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
460 <p>Here is an example of LLVM code to multiply the integer variable
461 '<tt>%X</tt>' by 8:</p>
465 <pre class="doc_code">
466 %result = <a href="#i_mul">mul</a> i32 %X, 8
469 <p>After strength reduction:</p>
471 <pre class="doc_code">
472 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
475 <p>And the hard way:</p>
477 <pre class="doc_code">
478 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
479 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
480 %result = <a href="#i_add">add</a> i32 %1, %1
483 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
484 lexical features of LLVM:</p>
487 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
490 <li>Unnamed temporaries are created when the result of a computation is not
491 assigned to a named value.</li>
493 <li>Unnamed temporaries are numbered sequentially</li>
496 <p>It also shows a convention that we follow in this document. When
497 demonstrating instructions, we will follow an instruction with a comment that
498 defines the type and name of value produced. Comments are shown in italic
503 <!-- *********************************************************************** -->
504 <h2><a name="highlevel">High Level Structure</a></h2>
505 <!-- *********************************************************************** -->
507 <!-- ======================================================================= -->
509 <a name="modulestructure">Module Structure</a>
514 <p>LLVM programs are composed of <tt>Module</tt>s, each of which is a
515 translation unit of the input programs. Each module consists of functions,
516 global variables, and symbol table entries. Modules may be combined together
517 with the LLVM linker, which merges function (and global variable)
518 definitions, resolves forward declarations, and merges symbol table
519 entries. Here is an example of the "hello world" module:</p>
521 <pre class="doc_code">
522 <i>; Declare the string constant as a global constant.</i>
523 <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"
525 <i>; External declaration of the puts function</i>
526 <a href="#functionstructure">declare</a> i32 @puts(i8* <a href="#nocapture">nocapture</a>) <a href="#fnattrs">nounwind</a>
528 <i>; Definition of main function</i>
529 define i32 @main() { <i>; i32()* </i>
530 <i>; Convert [13 x i8]* to i8 *...</i>
531 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.str, i64 0, i64 0
533 <i>; Call puts function to write out the string to stdout.</i>
534 <a href="#i_call">call</a> i32 @puts(i8* %cast210)
535 <a href="#i_ret">ret</a> i32 0
538 <i>; Named metadata</i>
539 !1 = metadata !{i32 42}
543 <p>This example is made up of a <a href="#globalvars">global variable</a> named
544 "<tt>.str</tt>", an external declaration of the "<tt>puts</tt>" function,
545 a <a href="#functionstructure">function definition</a> for
546 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
549 <p>In general, a module is made up of a list of global values (where both
550 functions and global variables are global values). Global values are
551 represented by a pointer to a memory location (in this case, a pointer to an
552 array of char, and a pointer to a function), and have one of the
553 following <a href="#linkage">linkage types</a>.</p>
557 <!-- ======================================================================= -->
559 <a name="linkage">Linkage Types</a>
564 <p>All Global Variables and Functions have one of the following types of
568 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
569 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
570 by objects in the current module. In particular, linking code into a
571 module with an private global value may cause the private to be renamed as
572 necessary to avoid collisions. Because the symbol is private to the
573 module, all references can be updated. This doesn't show up in any symbol
574 table in the object file.</dd>
576 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
577 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
578 assembler and evaluated by the linker. Unlike normal strong symbols, they
579 are removed by the linker from the final linked image (executable or
580 dynamic library).</dd>
582 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
583 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
584 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
585 linker. The symbols are removed by the linker from the final linked image
586 (executable or dynamic library).</dd>
588 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
589 <dd>Similar to private, but the value shows as a local symbol
590 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
591 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
593 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
594 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
595 into the object file corresponding to the LLVM module. They exist to
596 allow inlining and other optimizations to take place given knowledge of
597 the definition of the global, which is known to be somewhere outside the
598 module. Globals with <tt>available_externally</tt> linkage are allowed to
599 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
600 This linkage type is only allowed on definitions, not declarations.</dd>
602 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
603 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
604 the same name when linkage occurs. This can be used to implement
605 some forms of inline functions, templates, or other code which must be
606 generated in each translation unit that uses it, but where the body may
607 be overridden with a more definitive definition later. Unreferenced
608 <tt>linkonce</tt> globals are allowed to be discarded. Note that
609 <tt>linkonce</tt> linkage does not actually allow the optimizer to
610 inline the body of this function into callers because it doesn't know if
611 this definition of the function is the definitive definition within the
612 program or whether it will be overridden by a stronger definition.
613 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
616 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
617 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
618 <tt>linkonce</tt> linkage, except that unreferenced globals with
619 <tt>weak</tt> linkage may not be discarded. This is used for globals that
620 are declared "weak" in C source code.</dd>
622 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
623 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
624 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
626 Symbols with "<tt>common</tt>" linkage are merged in the same way as
627 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
628 <tt>common</tt> symbols may not have an explicit section,
629 must have a zero initializer, and may not be marked '<a
630 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
631 have common linkage.</dd>
634 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
635 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
636 pointer to array type. When two global variables with appending linkage
637 are linked together, the two global arrays are appended together. This is
638 the LLVM, typesafe, equivalent of having the system linker append together
639 "sections" with identical names when .o files are linked.</dd>
641 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
642 <dd>The semantics of this linkage follow the ELF object file model: the symbol
643 is weak until linked, if not linked, the symbol becomes null instead of
644 being an undefined reference.</dd>
646 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
647 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
648 <dd>Some languages allow differing globals to be merged, such as two functions
649 with different semantics. Other languages, such as <tt>C++</tt>, ensure
650 that only equivalent globals are ever merged (the "one definition rule"
651 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
652 and <tt>weak_odr</tt> linkage types to indicate that the global will only
653 be merged with equivalent globals. These linkage types are otherwise the
654 same as their non-<tt>odr</tt> versions.</dd>
656 <dt><tt><b><a name="linkage_linkonce_odr_auto_hide">linkonce_odr_auto_hide</a></b></tt></dt>
657 <dd>Similar to "<tt>linkonce_odr</tt>", but nothing in the translation unit
658 takes the address of this definition. For instance, functions that had an
659 inline definition, but the compiler decided not to inline it.
660 <tt>linkonce_odr_auto_hide</tt> may have only <tt>default</tt> visibility.
661 The symbols are removed by the linker from the final linked image
662 (executable or dynamic library).</dd>
664 <dt><tt><b><a name="linkage_external">external</a></b></tt></dt>
665 <dd>If none of the above identifiers are used, the global is externally
666 visible, meaning that it participates in linkage and can be used to
667 resolve external symbol references.</dd>
670 <p>The next two types of linkage are targeted for Microsoft Windows platform
671 only. They are designed to support importing (exporting) symbols from (to)
672 DLLs (Dynamic Link Libraries).</p>
675 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
676 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
677 or variable via a global pointer to a pointer that is set up by the DLL
678 exporting the symbol. On Microsoft Windows targets, the pointer name is
679 formed by combining <code>__imp_</code> and the function or variable
682 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
683 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
684 pointer to a pointer in a DLL, so that it can be referenced with the
685 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
686 name is formed by combining <code>__imp_</code> and the function or
690 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
691 another module defined a "<tt>.LC0</tt>" variable and was linked with this
692 one, one of the two would be renamed, preventing a collision. Since
693 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
694 declarations), they are accessible outside of the current module.</p>
696 <p>It is illegal for a function <i>declaration</i> to have any linkage type
697 other than <tt>external</tt>, <tt>dllimport</tt>
698 or <tt>extern_weak</tt>.</p>
700 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
701 or <tt>weak_odr</tt> linkages.</p>
705 <!-- ======================================================================= -->
707 <a name="callingconv">Calling Conventions</a>
712 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
713 and <a href="#i_invoke">invokes</a> can all have an optional calling
714 convention specified for the call. The calling convention of any pair of
715 dynamic caller/callee must match, or the behavior of the program is
716 undefined. The following calling conventions are supported by LLVM, and more
717 may be added in the future:</p>
720 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
721 <dd>This calling convention (the default if no other calling convention is
722 specified) matches the target C calling conventions. This calling
723 convention supports varargs function calls and tolerates some mismatch in
724 the declared prototype and implemented declaration of the function (as
727 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
728 <dd>This calling convention attempts to make calls as fast as possible
729 (e.g. by passing things in registers). This calling convention allows the
730 target to use whatever tricks it wants to produce fast code for the
731 target, without having to conform to an externally specified ABI
732 (Application Binary Interface).
733 <a href="CodeGenerator.html#id80">Tail calls can only be optimized
734 when this, the GHC or the HiPE convention is used.</a> This calling
735 convention does not support varargs and requires the prototype of all
736 callees to exactly match the prototype of the function definition.</dd>
738 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
739 <dd>This calling convention attempts to make code in the caller as efficient
740 as possible under the assumption that the call is not commonly executed.
741 As such, these calls often preserve all registers so that the call does
742 not break any live ranges in the caller side. This calling convention
743 does not support varargs and requires the prototype of all callees to
744 exactly match the prototype of the function definition.</dd>
746 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
747 <dd>This calling convention has been implemented specifically for use by the
748 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
749 It passes everything in registers, going to extremes to achieve this by
750 disabling callee save registers. This calling convention should not be
751 used lightly but only for specific situations such as an alternative to
752 the <em>register pinning</em> performance technique often used when
753 implementing functional programming languages. At the moment only X86
754 supports this convention and it has the following limitations:
756 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
757 floating point types are supported.</li>
758 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
759 6 floating point parameters.</li>
761 This calling convention supports
762 <a href="CodeGenerator.html#id80">tail call optimization</a> but
763 requires both the caller and callee are using it.
766 <dt><b>"<tt>cc <em>11</em></tt>" - The HiPE calling convention</b>:</dt>
767 <dd>This calling convention has been implemented specifically for use by the
768 <a href="http://www.it.uu.se/research/group/hipe/">High-Performance Erlang
769 (HiPE)</a> compiler, <em>the</em> native code compiler of the
770 <a href="http://www.erlang.org/download.shtml">Ericsson's Open Source
771 Erlang/OTP system</a>. It uses more registers for argument passing than
772 the ordinary C calling convention and defines no callee-saved registers.
773 The calling convention properly supports
774 <a href="CodeGenerator.html#id80">tail call optimization</a> but requires
775 that both the caller and the callee use it. It uses a <em>register
776 pinning</em> mechanism, similar to GHC's convention, for keeping
777 frequently accessed runtime components pinned to specific hardware
778 registers. At the moment only X86 supports this convention (both 32 and 64
781 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
782 <dd>Any calling convention may be specified by number, allowing
783 target-specific calling conventions to be used. Target specific calling
784 conventions start at 64.</dd>
787 <p>More calling conventions can be added/defined on an as-needed basis, to
788 support Pascal conventions or any other well-known target-independent
793 <!-- ======================================================================= -->
795 <a name="visibility">Visibility Styles</a>
800 <p>All Global Variables and Functions have one of the following visibility
804 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
805 <dd>On targets that use the ELF object file format, default visibility means
806 that the declaration is visible to other modules and, in shared libraries,
807 means that the declared entity may be overridden. On Darwin, default
808 visibility means that the declaration is visible to other modules. Default
809 visibility corresponds to "external linkage" in the language.</dd>
811 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
812 <dd>Two declarations of an object with hidden visibility refer to the same
813 object if they are in the same shared object. Usually, hidden visibility
814 indicates that the symbol will not be placed into the dynamic symbol
815 table, so no other module (executable or shared library) can reference it
818 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
819 <dd>On ELF, protected visibility indicates that the symbol will be placed in
820 the dynamic symbol table, but that references within the defining module
821 will bind to the local symbol. That is, the symbol cannot be overridden by
827 <!-- ======================================================================= -->
829 <a name="namedtypes">Named Types</a>
834 <p>LLVM IR allows you to specify name aliases for certain types. This can make
835 it easier to read the IR and make the IR more condensed (particularly when
836 recursive types are involved). An example of a name specification is:</p>
838 <pre class="doc_code">
839 %mytype = type { %mytype*, i32 }
842 <p>You may give a name to any <a href="#typesystem">type</a> except
843 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
844 is expected with the syntax "%mytype".</p>
846 <p>Note that type names are aliases for the structural type that they indicate,
847 and that you can therefore specify multiple names for the same type. This
848 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
849 uses structural typing, the name is not part of the type. When printing out
850 LLVM IR, the printer will pick <em>one name</em> to render all types of a
851 particular shape. This means that if you have code where two different
852 source types end up having the same LLVM type, that the dumper will sometimes
853 print the "wrong" or unexpected type. This is an important design point and
854 isn't going to change.</p>
858 <!-- ======================================================================= -->
860 <a name="globalvars">Global Variables</a>
865 <p>Global variables define regions of memory allocated at compilation time
866 instead of run-time. Global variables may optionally be initialized, may
867 have an explicit section to be placed in, and may have an optional explicit
868 alignment specified.</p>
870 <p>A variable may be defined as <tt>thread_local</tt>, which
871 means that it will not be shared by threads (each thread will have a
872 separated copy of the variable). Not all targets support thread-local
873 variables. Optionally, a TLS model may be specified:</p>
876 <dt><b><tt>localdynamic</tt></b>:</dt>
877 <dd>For variables that are only used within the current shared library.</dd>
879 <dt><b><tt>initialexec</tt></b>:</dt>
880 <dd>For variables in modules that will not be loaded dynamically.</dd>
882 <dt><b><tt>localexec</tt></b>:</dt>
883 <dd>For variables defined in the executable and only used within it.</dd>
886 <p>The models correspond to the ELF TLS models; see
887 <a href="http://people.redhat.com/drepper/tls.pdf">ELF
888 Handling For Thread-Local Storage</a> for more information on under which
889 circumstances the different models may be used. The target may choose a
890 different TLS model if the specified model is not supported, or if a better
891 choice of model can be made.</p>
893 <p>A variable may be defined as a global
894 "constant," which indicates that the contents of the variable
895 will <b>never</b> be modified (enabling better optimization, allowing the
896 global data to be placed in the read-only section of an executable, etc).
897 Note that variables that need runtime initialization cannot be marked
898 "constant" as there is a store to the variable.</p>
900 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
901 constant, even if the final definition of the global is not. This capability
902 can be used to enable slightly better optimization of the program, but
903 requires the language definition to guarantee that optimizations based on the
904 'constantness' are valid for the translation units that do not include the
907 <p>As SSA values, global variables define pointer values that are in scope
908 (i.e. they dominate) all basic blocks in the program. Global variables
909 always define a pointer to their "content" type because they describe a
910 region of memory, and all memory objects in LLVM are accessed through
913 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
914 that the address is not significant, only the content. Constants marked
915 like this can be merged with other constants if they have the same
916 initializer. Note that a constant with significant address <em>can</em>
917 be merged with a <tt>unnamed_addr</tt> constant, the result being a
918 constant whose address is significant.</p>
920 <p>A global variable may be declared to reside in a target-specific numbered
921 address space. For targets that support them, address spaces may affect how
922 optimizations are performed and/or what target instructions are used to
923 access the variable. The default address space is zero. The address space
924 qualifier must precede any other attributes.</p>
926 <p>LLVM allows an explicit section to be specified for globals. If the target
927 supports it, it will emit globals to the section specified.</p>
929 <p>An explicit alignment may be specified for a global, which must be a power
930 of 2. If not present, or if the alignment is set to zero, the alignment of
931 the global is set by the target to whatever it feels convenient. If an
932 explicit alignment is specified, the global is forced to have exactly that
933 alignment. Targets and optimizers are not allowed to over-align the global
934 if the global has an assigned section. In this case, the extra alignment
935 could be observable: for example, code could assume that the globals are
936 densely packed in their section and try to iterate over them as an array,
937 alignment padding would break this iteration.</p>
939 <p>For example, the following defines a global in a numbered address space with
940 an initializer, section, and alignment:</p>
942 <pre class="doc_code">
943 @G = addrspace(5) constant float 1.0, section "foo", align 4
946 <p>The following example defines a thread-local global with
947 the <tt>initialexec</tt> TLS model:</p>
949 <pre class="doc_code">
950 @G = thread_local(initialexec) global i32 0, align 4
956 <!-- ======================================================================= -->
958 <a name="functionstructure">Functions</a>
963 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
964 optional <a href="#linkage">linkage type</a>, an optional
965 <a href="#visibility">visibility style</a>, an optional
966 <a href="#callingconv">calling convention</a>,
967 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
968 <a href="#paramattrs">parameter attribute</a> for the return type, a function
969 name, a (possibly empty) argument list (each with optional
970 <a href="#paramattrs">parameter attributes</a>), optional
971 <a href="#fnattrs">function attributes</a>, an optional section, an optional
972 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
973 curly brace, a list of basic blocks, and a closing curly brace.</p>
975 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
976 optional <a href="#linkage">linkage type</a>, an optional
977 <a href="#visibility">visibility style</a>, an optional
978 <a href="#callingconv">calling convention</a>,
979 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
980 <a href="#paramattrs">parameter attribute</a> for the return type, a function
981 name, a possibly empty list of arguments, an optional alignment, and an
982 optional <a href="#gc">garbage collector name</a>.</p>
984 <p>A function definition contains a list of basic blocks, forming the CFG
985 (Control Flow Graph) for the function. Each basic block may optionally start
986 with a label (giving the basic block a symbol table entry), contains a list
987 of instructions, and ends with a <a href="#terminators">terminator</a>
988 instruction (such as a branch or function return).</p>
990 <p>The first basic block in a function is special in two ways: it is immediately
991 executed on entrance to the function, and it is not allowed to have
992 predecessor basic blocks (i.e. there can not be any branches to the entry
993 block of a function). Because the block can have no predecessors, it also
994 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
996 <p>LLVM allows an explicit section to be specified for functions. If the target
997 supports it, it will emit functions to the section specified.</p>
999 <p>An explicit alignment may be specified for a function. If not present, or if
1000 the alignment is set to zero, the alignment of the function is set by the
1001 target to whatever it feels convenient. If an explicit alignment is
1002 specified, the function is forced to have at least that much alignment. All
1003 alignments must be a power of 2.</p>
1005 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
1006 be significant and two identical functions can be merged.</p>
1009 <pre class="doc_code">
1010 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
1011 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
1012 <ResultType> @<FunctionName> ([argument list])
1013 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
1014 [<a href="#gc">gc</a>] { ... }
1019 <!-- ======================================================================= -->
1021 <a name="aliasstructure">Aliases</a>
1026 <p>Aliases act as "second name" for the aliasee value (which can be either
1027 function, global variable, another alias or bitcast of global value). Aliases
1028 may have an optional <a href="#linkage">linkage type</a>, and an
1029 optional <a href="#visibility">visibility style</a>.</p>
1032 <pre class="doc_code">
1033 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
1038 <!-- ======================================================================= -->
1040 <a name="namedmetadatastructure">Named Metadata</a>
1045 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
1046 nodes</a> (but not metadata strings) are the only valid operands for
1047 a named metadata.</p>
1050 <pre class="doc_code">
1051 ; Some unnamed metadata nodes, which are referenced by the named metadata.
1052 !0 = metadata !{metadata !"zero"}
1053 !1 = metadata !{metadata !"one"}
1054 !2 = metadata !{metadata !"two"}
1056 !name = !{!0, !1, !2}
1061 <!-- ======================================================================= -->
1063 <a name="paramattrs">Parameter Attributes</a>
1068 <p>The return type and each parameter of a function type may have a set of
1069 <i>parameter attributes</i> associated with them. Parameter attributes are
1070 used to communicate additional information about the result or parameters of
1071 a function. Parameter attributes are considered to be part of the function,
1072 not of the function type, so functions with different parameter attributes
1073 can have the same function type.</p>
1075 <p>Parameter attributes are simple keywords that follow the type specified. If
1076 multiple parameter attributes are needed, they are space separated. For
1079 <pre class="doc_code">
1080 declare i32 @printf(i8* noalias nocapture, ...)
1081 declare i32 @atoi(i8 zeroext)
1082 declare signext i8 @returns_signed_char()
1085 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1086 <tt>readonly</tt>) come immediately after the argument list.</p>
1088 <p>Currently, only the following parameter attributes are defined:</p>
1091 <dt><tt><b>zeroext</b></tt></dt>
1092 <dd>This indicates to the code generator that the parameter or return value
1093 should be zero-extended to the extent required by the target's ABI (which
1094 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1095 parameter) or the callee (for a return value).</dd>
1097 <dt><tt><b>signext</b></tt></dt>
1098 <dd>This indicates to the code generator that the parameter or return value
1099 should be sign-extended to the extent required by the target's ABI (which
1100 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1103 <dt><tt><b>inreg</b></tt></dt>
1104 <dd>This indicates that this parameter or return value should be treated in a
1105 special target-dependent fashion during while emitting code for a function
1106 call or return (usually, by putting it in a register as opposed to memory,
1107 though some targets use it to distinguish between two different kinds of
1108 registers). Use of this attribute is target-specific.</dd>
1110 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1111 <dd><p>This indicates that the pointer parameter should really be passed by
1112 value to the function. The attribute implies that a hidden copy of the
1114 is made between the caller and the callee, so the callee is unable to
1115 modify the value in the caller. This attribute is only valid on LLVM
1116 pointer arguments. It is generally used to pass structs and arrays by
1117 value, but is also valid on pointers to scalars. The copy is considered
1118 to belong to the caller not the callee (for example,
1119 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1120 <tt>byval</tt> parameters). This is not a valid attribute for return
1123 <p>The byval attribute also supports specifying an alignment with
1124 the align attribute. It indicates the alignment of the stack slot to
1125 form and the known alignment of the pointer specified to the call site. If
1126 the alignment is not specified, then the code generator makes a
1127 target-specific assumption.</p></dd>
1129 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1130 <dd>This indicates that the pointer parameter specifies the address of a
1131 structure that is the return value of the function in the source program.
1132 This pointer must be guaranteed by the caller to be valid: loads and
1133 stores to the structure may be assumed by the callee to not to trap and
1134 to be properly aligned. This may only be applied to the first parameter.
1135 This is not a valid attribute for return values. </dd>
1137 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1138 <dd>This indicates that pointer values
1139 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1140 value do not alias pointer values which are not <i>based</i> on it,
1141 ignoring certain "irrelevant" dependencies.
1142 For a call to the parent function, dependencies between memory
1143 references from before or after the call and from those during the call
1144 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1145 return value used in that call.
1146 The caller shares the responsibility with the callee for ensuring that
1147 these requirements are met.
1148 For further details, please see the discussion of the NoAlias response in
1149 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1151 Note that this definition of <tt>noalias</tt> is intentionally
1152 similar to the definition of <tt>restrict</tt> in C99 for function
1153 arguments, though it is slightly weaker.
1155 For function return values, C99's <tt>restrict</tt> is not meaningful,
1156 while LLVM's <tt>noalias</tt> is.
1159 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1160 <dd>This indicates that the callee does not make any copies of the pointer
1161 that outlive the callee itself. This is not a valid attribute for return
1164 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1165 <dd>This indicates that the pointer parameter can be excised using the
1166 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1167 attribute for return values.</dd>
1172 <!-- ======================================================================= -->
1174 <a name="gc">Garbage Collector Names</a>
1179 <p>Each function may specify a garbage collector name, which is simply a
1182 <pre class="doc_code">
1183 define void @f() gc "name" { ... }
1186 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1187 collector which will cause the compiler to alter its output in order to
1188 support the named garbage collection algorithm.</p>
1192 <!-- ======================================================================= -->
1194 <a name="fnattrs">Function Attributes</a>
1199 <p>Function attributes are set to communicate additional information about a
1200 function. Function attributes are considered to be part of the function, not
1201 of the function type, so functions with different function attributes can
1202 have the same function type.</p>
1204 <p>Function attributes are simple keywords that follow the type specified. If
1205 multiple attributes are needed, they are space separated. For example:</p>
1207 <pre class="doc_code">
1208 define void @f() noinline { ... }
1209 define void @f() alwaysinline { ... }
1210 define void @f() alwaysinline optsize { ... }
1211 define void @f() optsize { ... }
1215 <dt><tt><b>address_safety</b></tt></dt>
1216 <dd>This attribute indicates that the address safety analysis
1217 is enabled for this function. </dd>
1219 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1220 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1221 the backend should forcibly align the stack pointer. Specify the
1222 desired alignment, which must be a power of two, in parentheses.
1224 <dt><tt><b>alwaysinline</b></tt></dt>
1225 <dd>This attribute indicates that the inliner should attempt to inline this
1226 function into callers whenever possible, ignoring any active inlining size
1227 threshold for this caller.</dd>
1229 <dt><tt><b>nonlazybind</b></tt></dt>
1230 <dd>This attribute suppresses lazy symbol binding for the function. This
1231 may make calls to the function faster, at the cost of extra program
1232 startup time if the function is not called during program startup.</dd>
1234 <dt><tt><b>inlinehint</b></tt></dt>
1235 <dd>This attribute indicates that the source code contained a hint that inlining
1236 this function is desirable (such as the "inline" keyword in C/C++). It
1237 is just a hint; it imposes no requirements on the inliner.</dd>
1239 <dt><tt><b>naked</b></tt></dt>
1240 <dd>This attribute disables prologue / epilogue emission for the function.
1241 This can have very system-specific consequences.</dd>
1243 <dt><tt><b>noimplicitfloat</b></tt></dt>
1244 <dd>This attributes disables implicit floating point instructions.</dd>
1246 <dt><tt><b>noinline</b></tt></dt>
1247 <dd>This attribute indicates that the inliner should never inline this
1248 function in any situation. This attribute may not be used together with
1249 the <tt>alwaysinline</tt> attribute.</dd>
1251 <dt><tt><b>noredzone</b></tt></dt>
1252 <dd>This attribute indicates that the code generator should not use a red
1253 zone, even if the target-specific ABI normally permits it.</dd>
1255 <dt><tt><b>noreturn</b></tt></dt>
1256 <dd>This function attribute indicates that the function never returns
1257 normally. This produces undefined behavior at runtime if the function
1258 ever does dynamically return.</dd>
1260 <dt><tt><b>nounwind</b></tt></dt>
1261 <dd>This function attribute indicates that the function never returns with an
1262 unwind or exceptional control flow. If the function does unwind, its
1263 runtime behavior is undefined.</dd>
1265 <dt><tt><b>optsize</b></tt></dt>
1266 <dd>This attribute suggests that optimization passes and code generator passes
1267 make choices that keep the code size of this function low, and otherwise
1268 do optimizations specifically to reduce code size.</dd>
1270 <dt><tt><b>readnone</b></tt></dt>
1271 <dd>This attribute indicates that the function computes its result (or decides
1272 to unwind an exception) based strictly on its arguments, without
1273 dereferencing any pointer arguments or otherwise accessing any mutable
1274 state (e.g. memory, control registers, etc) visible to caller functions.
1275 It does not write through any pointer arguments
1276 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1277 changes any state visible to callers. This means that it cannot unwind
1278 exceptions by calling the <tt>C++</tt> exception throwing methods.</dd>
1280 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1281 <dd>This attribute indicates that the function does not write through any
1282 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1283 arguments) or otherwise modify any state (e.g. memory, control registers,
1284 etc) visible to caller functions. It may dereference pointer arguments
1285 and read state that may be set in the caller. A readonly function always
1286 returns the same value (or unwinds an exception identically) when called
1287 with the same set of arguments and global state. It cannot unwind an
1288 exception by calling the <tt>C++</tt> exception throwing methods.</dd>
1290 <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1291 <dd>This attribute indicates that this function can return twice. The
1292 C <code>setjmp</code> is an example of such a function. The compiler
1293 disables some optimizations (like tail calls) in the caller of these
1296 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1297 <dd>This attribute indicates that the function should emit a stack smashing
1298 protector. It is in the form of a "canary"—a random value placed on
1299 the stack before the local variables that's checked upon return from the
1300 function to see if it has been overwritten. A heuristic is used to
1301 determine if a function needs stack protectors or not.<br>
1303 If a function that has an <tt>ssp</tt> attribute is inlined into a
1304 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1305 function will have an <tt>ssp</tt> attribute.</dd>
1307 <dt><tt><b>sspreq</b></tt></dt>
1308 <dd>This attribute indicates that the function should <em>always</em> emit a
1309 stack smashing protector. This overrides
1310 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1312 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1313 function that doesn't have an <tt>sspreq</tt> attribute or which has
1314 an <tt>ssp</tt> attribute, then the resulting function will have
1315 an <tt>sspreq</tt> attribute.</dd>
1317 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1318 <dd>This attribute indicates that the ABI being targeted requires that
1319 an unwind table entry be produce for this function even if we can
1320 show that no exceptions passes by it. This is normally the case for
1321 the ELF x86-64 abi, but it can be disabled for some compilation
1327 <!-- ======================================================================= -->
1329 <a name="moduleasm">Module-Level Inline Assembly</a>
1334 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1335 the GCC "file scope inline asm" blocks. These blocks are internally
1336 concatenated by LLVM and treated as a single unit, but may be separated in
1337 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1339 <pre class="doc_code">
1340 module asm "inline asm code goes here"
1341 module asm "more can go here"
1344 <p>The strings can contain any character by escaping non-printable characters.
1345 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1348 <p>The inline asm code is simply printed to the machine code .s file when
1349 assembly code is generated.</p>
1353 <!-- ======================================================================= -->
1355 <a name="datalayout">Data Layout</a>
1360 <p>A module may specify a target specific data layout string that specifies how
1361 data is to be laid out in memory. The syntax for the data layout is
1364 <pre class="doc_code">
1365 target datalayout = "<i>layout specification</i>"
1368 <p>The <i>layout specification</i> consists of a list of specifications
1369 separated by the minus sign character ('-'). Each specification starts with
1370 a letter and may include other information after the letter to define some
1371 aspect of the data layout. The specifications accepted are as follows:</p>
1375 <dd>Specifies that the target lays out data in big-endian form. That is, the
1376 bits with the most significance have the lowest address location.</dd>
1379 <dd>Specifies that the target lays out data in little-endian form. That is,
1380 the bits with the least significance have the lowest address
1383 <dt><tt>S<i>size</i></tt></dt>
1384 <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1385 of stack variables is limited to the natural stack alignment to avoid
1386 dynamic stack realignment. The stack alignment must be a multiple of
1387 8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1388 which does not prevent any alignment promotions.</dd>
1390 <dt><tt>p[n]:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1391 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1392 <i>preferred</i> alignments for address space <i>n</i>. All sizes are in
1393 bits. Specifying the <i>pref</i> alignment is optional. If omitted, the
1394 preceding <tt>:</tt> should be omitted too. The address space,
1395 <i>n</i> is optional, and if not specified, denotes the default address
1396 space 0. The value of <i>n</i> must be in the range [1,2^23).</dd>
1398 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1399 <dd>This specifies the alignment for an integer type of a given bit
1400 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1402 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1403 <dd>This specifies the alignment for a vector type of a given bit
1406 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1407 <dd>This specifies the alignment for a floating point type of a given bit
1408 <i>size</i>. Only values of <i>size</i> that are supported by the target
1409 will work. 32 (float) and 64 (double) are supported on all targets;
1410 80 or 128 (different flavors of long double) are also supported on some
1413 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1414 <dd>This specifies the alignment for an aggregate type of a given bit
1417 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1418 <dd>This specifies the alignment for a stack object of a given bit
1421 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1422 <dd>This specifies a set of native integer widths for the target CPU
1423 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1424 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1425 this set are considered to support most general arithmetic
1426 operations efficiently.</dd>
1429 <p>When constructing the data layout for a given target, LLVM starts with a
1430 default set of specifications which are then (possibly) overridden by the
1431 specifications in the <tt>datalayout</tt> keyword. The default specifications
1432 are given in this list:</p>
1435 <li><tt>E</tt> - big endian</li>
1436 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1437 <li><tt>p1:32:32:32</tt> - 32-bit pointers with 32-bit alignment for
1438 address space 1</li>
1439 <li><tt>p2:16:32:32</tt> - 16-bit pointers with 32-bit alignment for
1440 address space 2</li>
1441 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1442 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1443 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1444 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1445 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1446 alignment of 64-bits</li>
1447 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1448 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1449 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1450 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1451 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1452 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1455 <p>When LLVM is determining the alignment for a given type, it uses the
1456 following rules:</p>
1459 <li>If the type sought is an exact match for one of the specifications, that
1460 specification is used.</li>
1462 <li>If no match is found, and the type sought is an integer type, then the
1463 smallest integer type that is larger than the bitwidth of the sought type
1464 is used. If none of the specifications are larger than the bitwidth then
1465 the largest integer type is used. For example, given the default
1466 specifications above, the i7 type will use the alignment of i8 (next
1467 largest) while both i65 and i256 will use the alignment of i64 (largest
1470 <li>If no match is found, and the type sought is a vector type, then the
1471 largest vector type that is smaller than the sought vector type will be
1472 used as a fall back. This happens because <128 x double> can be
1473 implemented in terms of 64 <2 x double>, for example.</li>
1476 <p>The function of the data layout string may not be what you expect. Notably,
1477 this is not a specification from the frontend of what alignment the code
1478 generator should use.</p>
1480 <p>Instead, if specified, the target data layout is required to match what the
1481 ultimate <em>code generator</em> expects. This string is used by the
1482 mid-level optimizers to
1483 improve code, and this only works if it matches what the ultimate code
1484 generator uses. If you would like to generate IR that does not embed this
1485 target-specific detail into the IR, then you don't have to specify the
1486 string. This will disable some optimizations that require precise layout
1487 information, but this also prevents those optimizations from introducing
1488 target specificity into the IR.</p>
1494 <!-- ======================================================================= -->
1496 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1501 <p>Any memory access must be done through a pointer value associated
1502 with an address range of the memory access, otherwise the behavior
1503 is undefined. Pointer values are associated with address ranges
1504 according to the following rules:</p>
1507 <li>A pointer value is associated with the addresses associated with
1508 any value it is <i>based</i> on.
1509 <li>An address of a global variable is associated with the address
1510 range of the variable's storage.</li>
1511 <li>The result value of an allocation instruction is associated with
1512 the address range of the allocated storage.</li>
1513 <li>A null pointer in the default address-space is associated with
1515 <li>An integer constant other than zero or a pointer value returned
1516 from a function not defined within LLVM may be associated with address
1517 ranges allocated through mechanisms other than those provided by
1518 LLVM. Such ranges shall not overlap with any ranges of addresses
1519 allocated by mechanisms provided by LLVM.</li>
1522 <p>A pointer value is <i>based</i> on another pointer value according
1523 to the following rules:</p>
1526 <li>A pointer value formed from a
1527 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1528 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1529 <li>The result value of a
1530 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1531 of the <tt>bitcast</tt>.</li>
1532 <li>A pointer value formed by an
1533 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1534 pointer values that contribute (directly or indirectly) to the
1535 computation of the pointer's value.</li>
1536 <li>The "<i>based</i> on" relationship is transitive.</li>
1539 <p>Note that this definition of <i>"based"</i> is intentionally
1540 similar to the definition of <i>"based"</i> in C99, though it is
1541 slightly weaker.</p>
1543 <p>LLVM IR does not associate types with memory. The result type of a
1544 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1545 alignment of the memory from which to load, as well as the
1546 interpretation of the value. The first operand type of a
1547 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1548 and alignment of the store.</p>
1550 <p>Consequently, type-based alias analysis, aka TBAA, aka
1551 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1552 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1553 additional information which specialized optimization passes may use
1554 to implement type-based alias analysis.</p>
1558 <!-- ======================================================================= -->
1560 <a name="volatile">Volatile Memory Accesses</a>
1565 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1566 href="#i_store"><tt>store</tt></a>s, and <a
1567 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1568 The optimizers must not change the number of volatile operations or change their
1569 order of execution relative to other volatile operations. The optimizers
1570 <i>may</i> change the order of volatile operations relative to non-volatile
1571 operations. This is not Java's "volatile" and has no cross-thread
1572 synchronization behavior.</p>
1576 <!-- ======================================================================= -->
1578 <a name="memmodel">Memory Model for Concurrent Operations</a>
1583 <p>The LLVM IR does not define any way to start parallel threads of execution
1584 or to register signal handlers. Nonetheless, there are platform-specific
1585 ways to create them, and we define LLVM IR's behavior in their presence. This
1586 model is inspired by the C++0x memory model.</p>
1588 <p>For a more informal introduction to this model, see the
1589 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1591 <p>We define a <i>happens-before</i> partial order as the least partial order
1594 <li>Is a superset of single-thread program order, and</li>
1595 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1596 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1597 by platform-specific techniques, like pthread locks, thread
1598 creation, thread joining, etc., and by atomic instructions.
1599 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1603 <p>Note that program order does not introduce <i>happens-before</i> edges
1604 between a thread and signals executing inside that thread.</p>
1606 <p>Every (defined) read operation (load instructions, memcpy, atomic
1607 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1608 (defined) write operations (store instructions, atomic
1609 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1610 initialized globals are considered to have a write of the initializer which is
1611 atomic and happens before any other read or write of the memory in question.
1612 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1613 any write to the same byte, except:</p>
1616 <li>If <var>write<sub>1</sub></var> happens before
1617 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1618 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1619 does not see <var>write<sub>1</sub></var>.
1620 <li>If <var>R<sub>byte</sub></var> happens before
1621 <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1622 see <var>write<sub>3</sub></var>.
1625 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1627 <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1628 is supposed to give guarantees which can support
1629 <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1630 addresses which do not behave like normal memory. It does not generally
1631 provide cross-thread synchronization.)
1632 <li>Otherwise, if there is no write to the same byte that happens before
1633 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1634 <tt>undef</tt> for that byte.
1635 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1636 <var>R<sub>byte</sub></var> returns the value written by that
1638 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1639 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1640 values written. See the <a href="#ordering">Atomic Memory Ordering
1641 Constraints</a> section for additional constraints on how the choice
1643 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1646 <p><var>R</var> returns the value composed of the series of bytes it read.
1647 This implies that some bytes within the value may be <tt>undef</tt>
1648 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1649 defines the semantics of the operation; it doesn't mean that targets will
1650 emit more than one instruction to read the series of bytes.</p>
1652 <p>Note that in cases where none of the atomic intrinsics are used, this model
1653 places only one restriction on IR transformations on top of what is required
1654 for single-threaded execution: introducing a store to a byte which might not
1655 otherwise be stored is not allowed in general. (Specifically, in the case
1656 where another thread might write to and read from an address, introducing a
1657 store can change a load that may see exactly one write into a load that may
1658 see multiple writes.)</p>
1660 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1661 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1662 none of the backends currently in the tree fall into this category; however,
1663 there might be targets which care. If there are, we want a paragraph
1666 Targets may specify that stores narrower than a certain width are not
1667 available; on such a target, for the purposes of this model, treat any
1668 non-atomic write with an alignment or width less than the minimum width
1669 as if it writes to the relevant surrounding bytes.
1674 <!-- ======================================================================= -->
1676 <a name="ordering">Atomic Memory Ordering Constraints</a>
1681 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1682 <a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1683 <a href="#i_fence"><code>fence</code></a>,
1684 <a href="#i_load"><code>atomic load</code></a>, and
1685 <a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1686 that determines which other atomic instructions on the same address they
1687 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1688 but are somewhat more colloquial. If these descriptions aren't precise enough,
1689 check those specs (see spec references in the
1690 <a href="Atomics.html#introduction">atomics guide</a>).
1691 <a href="#i_fence"><code>fence</code></a> instructions
1692 treat these orderings somewhat differently since they don't take an address.
1693 See that instruction's documentation for details.</p>
1695 <p>For a simpler introduction to the ordering constraints, see the
1696 <a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1699 <dt><code>unordered</code></dt>
1700 <dd>The set of values that can be read is governed by the happens-before
1701 partial order. A value cannot be read unless some operation wrote it.
1702 This is intended to provide a guarantee strong enough to model Java's
1703 non-volatile shared variables. This ordering cannot be specified for
1704 read-modify-write operations; it is not strong enough to make them atomic
1705 in any interesting way.</dd>
1706 <dt><code>monotonic</code></dt>
1707 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1708 total order for modifications by <code>monotonic</code> operations on each
1709 address. All modification orders must be compatible with the happens-before
1710 order. There is no guarantee that the modification orders can be combined to
1711 a global total order for the whole program (and this often will not be
1712 possible). The read in an atomic read-modify-write operation
1713 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1714 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1715 reads the value in the modification order immediately before the value it
1716 writes. If one atomic read happens before another atomic read of the same
1717 address, the later read must see the same value or a later value in the
1718 address's modification order. This disallows reordering of
1719 <code>monotonic</code> (or stronger) operations on the same address. If an
1720 address is written <code>monotonic</code>ally by one thread, and other threads
1721 <code>monotonic</code>ally read that address repeatedly, the other threads must
1722 eventually see the write. This corresponds to the C++0x/C1x
1723 <code>memory_order_relaxed</code>.</dd>
1724 <dt><code>acquire</code></dt>
1725 <dd>In addition to the guarantees of <code>monotonic</code>,
1726 a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1727 operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1728 <dt><code>release</code></dt>
1729 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1730 writes a value which is subsequently read by an <code>acquire</code> operation,
1731 it <i>synchronizes-with</i> that operation. (This isn't a complete
1732 description; see the C++0x definition of a release sequence.) This corresponds
1733 to the C++0x/C1x <code>memory_order_release</code>.</dd>
1734 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1735 <code>acquire</code> and <code>release</code> operation on its address.
1736 This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1737 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1738 <dd>In addition to the guarantees of <code>acq_rel</code>
1739 (<code>acquire</code> for an operation which only reads, <code>release</code>
1740 for an operation which only writes), there is a global total order on all
1741 sequentially-consistent operations on all addresses, which is consistent with
1742 the <i>happens-before</i> partial order and with the modification orders of
1743 all the affected addresses. Each sequentially-consistent read sees the last
1744 preceding write to the same address in this global order. This corresponds
1745 to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1748 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1749 it only <i>synchronizes with</i> or participates in modification and seq_cst
1750 total orderings with other operations running in the same thread (for example,
1751 in signal handlers).</p>
1755 <!-- ======================================================================= -->
1757 <a name="fastmath">Fast-Math Flags</a>
1762 <p> LLVM IR floating-point binary ops (<a href="#i_fadd"><code>fadd</code></a>,
1763 <a href="#i_fsub"><code>fsub</code></a>, <a
1764 href="#i_fmul"><code>fmul</code></a>, <a href="#i_fdiv"><code>fdiv</code></a>,
1765 <a href="#i_frem"><code>frem</code></a>) have the following flags
1766 that can set to enable otherwise unsafe floating point operations</p>
1768 <dt><code>nnan</dt></code>
1770 No NaNs - Allow optimizations to assume the arguments and result are not
1771 NaN. Such optimizations are required to retain defined behavior over NaNs, but
1772 the value of the result is undefined.
1775 <dt><code>ninf</code></dt>
1777 No Infs - Allow optimizations to assume the arguments and result are not
1778 +/-Inf. Such optimizations are required to retain defined behavior over +/-Inf,
1779 but the value of the result is undefined.
1782 <dt><code>nsz</code></dt>
1784 No Signed Zeros - Allow optimizations to treat the sign of a zero argument or
1785 result as insignificant.
1788 <dt><code>arcp</code></dt>
1790 Allow Reciprocal - Allow optimizations to use the reciprocal of an argument
1791 rather than perform division.
1794 <dt><code>fast</code></TD>
1796 Fast - Allow algebraically equivalent transformations that may dramatically
1797 change results in floating point (e.g. reassociate). This flag implies all the
1805 <!-- *********************************************************************** -->
1806 <h2><a name="typesystem">Type System</a></h2>
1807 <!-- *********************************************************************** -->
1811 <p>The LLVM type system is one of the most important features of the
1812 intermediate representation. Being typed enables a number of optimizations
1813 to be performed on the intermediate representation directly, without having
1814 to do extra analyses on the side before the transformation. A strong type
1815 system makes it easier to read the generated code and enables novel analyses
1816 and transformations that are not feasible to perform on normal three address
1817 code representations.</p>
1819 <!-- ======================================================================= -->
1821 <a name="t_classifications">Type Classifications</a>
1826 <p>The types fall into a few useful classifications:</p>
1828 <table border="1" cellspacing="0" cellpadding="4">
1830 <tr><th>Classification</th><th>Types</th></tr>
1832 <td><a href="#t_integer">integer</a></td>
1833 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1836 <td><a href="#t_floating">floating point</a></td>
1837 <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1840 <td><a name="t_firstclass">first class</a></td>
1841 <td><a href="#t_integer">integer</a>,
1842 <a href="#t_floating">floating point</a>,
1843 <a href="#t_pointer">pointer</a>,
1844 <a href="#t_vector">vector</a>,
1845 <a href="#t_struct">structure</a>,
1846 <a href="#t_array">array</a>,
1847 <a href="#t_label">label</a>,
1848 <a href="#t_metadata">metadata</a>.
1852 <td><a href="#t_primitive">primitive</a></td>
1853 <td><a href="#t_label">label</a>,
1854 <a href="#t_void">void</a>,
1855 <a href="#t_integer">integer</a>,
1856 <a href="#t_floating">floating point</a>,
1857 <a href="#t_x86mmx">x86mmx</a>,
1858 <a href="#t_metadata">metadata</a>.</td>
1861 <td><a href="#t_derived">derived</a></td>
1862 <td><a href="#t_array">array</a>,
1863 <a href="#t_function">function</a>,
1864 <a href="#t_pointer">pointer</a>,
1865 <a href="#t_struct">structure</a>,
1866 <a href="#t_vector">vector</a>,
1867 <a href="#t_opaque">opaque</a>.
1873 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1874 important. Values of these types are the only ones which can be produced by
1879 <!-- ======================================================================= -->
1881 <a name="t_primitive">Primitive Types</a>
1886 <p>The primitive types are the fundamental building blocks of the LLVM
1889 <!-- _______________________________________________________________________ -->
1891 <a name="t_integer">Integer Type</a>
1897 <p>The integer type is a very simple type that simply specifies an arbitrary
1898 bit width for the integer type desired. Any bit width from 1 bit to
1899 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1906 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1910 <table class="layout">
1912 <td class="left"><tt>i1</tt></td>
1913 <td class="left">a single-bit integer.</td>
1916 <td class="left"><tt>i32</tt></td>
1917 <td class="left">a 32-bit integer.</td>
1920 <td class="left"><tt>i1942652</tt></td>
1921 <td class="left">a really big integer of over 1 million bits.</td>
1927 <!-- _______________________________________________________________________ -->
1929 <a name="t_floating">Floating Point Types</a>
1936 <tr><th>Type</th><th>Description</th></tr>
1937 <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr>
1938 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1939 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1940 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1941 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1942 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1948 <!-- _______________________________________________________________________ -->
1950 <a name="t_x86mmx">X86mmx Type</a>
1956 <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>
1965 <!-- _______________________________________________________________________ -->
1967 <a name="t_void">Void Type</a>
1973 <p>The void type does not represent any value and has no size.</p>
1982 <!-- _______________________________________________________________________ -->
1984 <a name="t_label">Label Type</a>
1990 <p>The label type represents code labels.</p>
1999 <!-- _______________________________________________________________________ -->
2001 <a name="t_metadata">Metadata Type</a>
2007 <p>The metadata type represents embedded metadata. No derived types may be
2008 created from metadata except for <a href="#t_function">function</a>
2020 <!-- ======================================================================= -->
2022 <a name="t_derived">Derived Types</a>
2027 <p>The real power in LLVM comes from the derived types in the system. This is
2028 what allows a programmer to represent arrays, functions, pointers, and other
2029 useful types. Each of these types contain one or more element types which
2030 may be a primitive type, or another derived type. For example, it is
2031 possible to have a two dimensional array, using an array as the element type
2032 of another array.</p>
2034 <!-- _______________________________________________________________________ -->
2036 <a name="t_aggregate">Aggregate Types</a>
2041 <p>Aggregate Types are a subset of derived types that can contain multiple
2042 member types. <a href="#t_array">Arrays</a> and
2043 <a href="#t_struct">structs</a> are aggregate types.
2044 <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p>
2048 <!-- _______________________________________________________________________ -->
2050 <a name="t_array">Array Type</a>
2056 <p>The array type is a very simple derived type that arranges elements
2057 sequentially in memory. The array type requires a size (number of elements)
2058 and an underlying data type.</p>
2062 [<# elements> x <elementtype>]
2065 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
2066 be any type with a size.</p>
2069 <table class="layout">
2071 <td class="left"><tt>[40 x i32]</tt></td>
2072 <td class="left">Array of 40 32-bit integer values.</td>
2075 <td class="left"><tt>[41 x i32]</tt></td>
2076 <td class="left">Array of 41 32-bit integer values.</td>
2079 <td class="left"><tt>[4 x i8]</tt></td>
2080 <td class="left">Array of 4 8-bit integer values.</td>
2083 <p>Here are some examples of multidimensional arrays:</p>
2084 <table class="layout">
2086 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
2087 <td class="left">3x4 array of 32-bit integer values.</td>
2090 <td class="left"><tt>[12 x [10 x float]]</tt></td>
2091 <td class="left">12x10 array of single precision floating point values.</td>
2094 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
2095 <td class="left">2x3x4 array of 16-bit integer values.</td>
2099 <p>There is no restriction on indexing beyond the end of the array implied by
2100 a static type (though there are restrictions on indexing beyond the bounds
2101 of an allocated object in some cases). This means that single-dimension
2102 'variable sized array' addressing can be implemented in LLVM with a zero
2103 length array type. An implementation of 'pascal style arrays' in LLVM could
2104 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
2108 <!-- _______________________________________________________________________ -->
2110 <a name="t_function">Function Type</a>
2116 <p>The function type can be thought of as a function signature. It consists of
2117 a return type and a list of formal parameter types. The return type of a
2118 function type is a first class type or a void type.</p>
2122 <returntype> (<parameter list>)
2125 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
2126 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
2127 which indicates that the function takes a variable number of arguments.
2128 Variable argument functions can access their arguments with
2129 the <a href="#int_varargs">variable argument handling intrinsic</a>
2130 functions. '<tt><returntype></tt>' is any type except
2131 <a href="#t_label">label</a>.</p>
2134 <table class="layout">
2136 <td class="left"><tt>i32 (i32)</tt></td>
2137 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2139 </tr><tr class="layout">
2140 <td class="left"><tt>float (i16, i32 *) *
2142 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2143 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2144 returning <tt>float</tt>.
2146 </tr><tr class="layout">
2147 <td class="left"><tt>i32 (i8*, ...)</tt></td>
2148 <td class="left">A vararg function that takes at least one
2149 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2150 which returns an integer. This is the signature for <tt>printf</tt> in
2153 </tr><tr class="layout">
2154 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2155 <td class="left">A function taking an <tt>i32</tt>, returning a
2156 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2163 <!-- _______________________________________________________________________ -->
2165 <a name="t_struct">Structure Type</a>
2171 <p>The structure type is used to represent a collection of data members together
2172 in memory. The elements of a structure may be any type that has a size.</p>
2174 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2175 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2176 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2177 Structures in registers are accessed using the
2178 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2179 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2181 <p>Structures may optionally be "packed" structures, which indicate that the
2182 alignment of the struct is one byte, and that there is no padding between
2183 the elements. In non-packed structs, padding between field types is inserted
2184 as defined by the DataLayout string in the module, which is required to match
2185 what the underlying code generator expects.</p>
2187 <p>Structures can either be "literal" or "identified". A literal structure is
2188 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2189 types are always defined at the top level with a name. Literal types are
2190 uniqued by their contents and can never be recursive or opaque since there is
2191 no way to write one. Identified types can be recursive, can be opaqued, and are
2197 %T1 = type { <type list> } <i>; Identified normal struct type</i>
2198 %T2 = type <{ <type list> }> <i>; Identified packed struct type</i>
2202 <table class="layout">
2204 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2205 <td class="left">A triple of three <tt>i32</tt> values</td>
2208 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2209 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2210 second element is a <a href="#t_pointer">pointer</a> to a
2211 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2212 an <tt>i32</tt>.</td>
2215 <td class="left"><tt><{ i8, i32 }></tt></td>
2216 <td class="left">A packed struct known to be 5 bytes in size.</td>
2222 <!-- _______________________________________________________________________ -->
2224 <a name="t_opaque">Opaque Structure Types</a>
2230 <p>Opaque structure types are used to represent named structure types that do
2231 not have a body specified. This corresponds (for example) to the C notion of
2232 a forward declared structure.</p>
2241 <table class="layout">
2243 <td class="left"><tt>opaque</tt></td>
2244 <td class="left">An opaque type.</td>
2252 <!-- _______________________________________________________________________ -->
2254 <a name="t_pointer">Pointer Type</a>
2260 <p>The pointer type is used to specify memory locations.
2261 Pointers are commonly used to reference objects in memory.</p>
2263 <p>Pointer types may have an optional address space attribute defining the
2264 numbered address space where the pointed-to object resides. The default
2265 address space is number zero. The semantics of non-zero address
2266 spaces are target-specific.</p>
2268 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2269 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2277 <table class="layout">
2279 <td class="left"><tt>[4 x i32]*</tt></td>
2280 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2281 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2284 <td class="left"><tt>i32 (i32*) *</tt></td>
2285 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2286 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2290 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2291 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2292 that resides in address space #5.</td>
2298 <!-- _______________________________________________________________________ -->
2300 <a name="t_vector">Vector Type</a>
2306 <p>A vector type is a simple derived type that represents a vector of elements.
2307 Vector types are used when multiple primitive data are operated in parallel
2308 using a single instruction (SIMD). A vector type requires a size (number of
2309 elements) and an underlying primitive data type. Vector types are considered
2310 <a href="#t_firstclass">first class</a>.</p>
2314 < <# elements> x <elementtype> >
2317 <p>The number of elements is a constant integer value larger than 0; elementtype
2318 may be any integer or floating point type, or a pointer to these types.
2319 Vectors of size zero are not allowed. </p>
2322 <table class="layout">
2324 <td class="left"><tt><4 x i32></tt></td>
2325 <td class="left">Vector of 4 32-bit integer values.</td>
2328 <td class="left"><tt><8 x float></tt></td>
2329 <td class="left">Vector of 8 32-bit floating-point values.</td>
2332 <td class="left"><tt><2 x i64></tt></td>
2333 <td class="left">Vector of 2 64-bit integer values.</td>
2336 <td class="left"><tt><4 x i64*></tt></td>
2337 <td class="left">Vector of 4 pointers to 64-bit integer values.</td>
2347 <!-- *********************************************************************** -->
2348 <h2><a name="constants">Constants</a></h2>
2349 <!-- *********************************************************************** -->
2353 <p>LLVM has several different basic types of constants. This section describes
2354 them all and their syntax.</p>
2356 <!-- ======================================================================= -->
2358 <a name="simpleconstants">Simple Constants</a>
2364 <dt><b>Boolean constants</b></dt>
2365 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2366 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2368 <dt><b>Integer constants</b></dt>
2369 <dd>Standard integers (such as '4') are constants of
2370 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2371 with integer types.</dd>
2373 <dt><b>Floating point constants</b></dt>
2374 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2375 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2376 notation (see below). The assembler requires the exact decimal value of a
2377 floating-point constant. For example, the assembler accepts 1.25 but
2378 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2379 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2381 <dt><b>Null pointer constants</b></dt>
2382 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2383 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2386 <p>The one non-intuitive notation for constants is the hexadecimal form of
2387 floating point constants. For example, the form '<tt>double
2388 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2389 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2390 constants are required (and the only time that they are generated by the
2391 disassembler) is when a floating point constant must be emitted but it cannot
2392 be represented as a decimal floating point number in a reasonable number of
2393 digits. For example, NaN's, infinities, and other special values are
2394 represented in their IEEE hexadecimal format so that assembly and disassembly
2395 do not cause any bits to change in the constants.</p>
2397 <p>When using the hexadecimal form, constants of types half, float, and double are
2398 represented using the 16-digit form shown above (which matches the IEEE754
2399 representation for double); half and float values must, however, be exactly
2400 representable as IEE754 half and single precision, respectively.
2401 Hexadecimal format is always used
2402 for long double, and there are three forms of long double. The 80-bit format
2403 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2404 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2405 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2406 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2407 currently supported target uses this format. Long doubles will only work if
2408 they match the long double format on your target. The IEEE 16-bit format
2409 (half precision) is represented by <tt>0xH</tt> followed by 4 hexadecimal
2410 digits. All hexadecimal formats are big-endian (sign bit at the left).</p>
2412 <p>There are no constants of type x86mmx.</p>
2415 <!-- ======================================================================= -->
2417 <a name="aggregateconstants"></a> <!-- old anchor -->
2418 <a name="complexconstants">Complex Constants</a>
2423 <p>Complex constants are a (potentially recursive) combination of simple
2424 constants and smaller complex constants.</p>
2427 <dt><b>Structure constants</b></dt>
2428 <dd>Structure constants are represented with notation similar to structure
2429 type definitions (a comma separated list of elements, surrounded by braces
2430 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2431 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2432 Structure constants must have <a href="#t_struct">structure type</a>, and
2433 the number and types of elements must match those specified by the
2436 <dt><b>Array constants</b></dt>
2437 <dd>Array constants are represented with notation similar to array type
2438 definitions (a comma separated list of elements, surrounded by square
2439 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2440 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2441 the number and types of elements must match those specified by the
2444 <dt><b>Vector constants</b></dt>
2445 <dd>Vector constants are represented with notation similar to vector type
2446 definitions (a comma separated list of elements, surrounded by
2447 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2448 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2449 have <a href="#t_vector">vector type</a>, and the number and types of
2450 elements must match those specified by the type.</dd>
2452 <dt><b>Zero initialization</b></dt>
2453 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2454 value to zero of <em>any</em> type, including scalar and
2455 <a href="#t_aggregate">aggregate</a> types.
2456 This is often used to avoid having to print large zero initializers
2457 (e.g. for large arrays) and is always exactly equivalent to using explicit
2458 zero initializers.</dd>
2460 <dt><b>Metadata node</b></dt>
2461 <dd>A metadata node is a structure-like constant with
2462 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2463 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2464 be interpreted as part of the instruction stream, metadata is a place to
2465 attach additional information such as debug info.</dd>
2470 <!-- ======================================================================= -->
2472 <a name="globalconstants">Global Variable and Function Addresses</a>
2477 <p>The addresses of <a href="#globalvars">global variables</a>
2478 and <a href="#functionstructure">functions</a> are always implicitly valid
2479 (link-time) constants. These constants are explicitly referenced when
2480 the <a href="#identifiers">identifier for the global</a> is used and always
2481 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2482 legal LLVM file:</p>
2484 <pre class="doc_code">
2487 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2492 <!-- ======================================================================= -->
2494 <a name="undefvalues">Undefined Values</a>
2499 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2500 indicates that the user of the value may receive an unspecified bit-pattern.
2501 Undefined values may be of any type (other than '<tt>label</tt>'
2502 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2504 <p>Undefined values are useful because they indicate to the compiler that the
2505 program is well defined no matter what value is used. This gives the
2506 compiler more freedom to optimize. Here are some examples of (potentially
2507 surprising) transformations that are valid (in pseudo IR):</p>
2510 <pre class="doc_code">
2520 <p>This is safe because all of the output bits are affected by the undef bits.
2521 Any output bit can have a zero or one depending on the input bits.</p>
2523 <pre class="doc_code">
2534 <p>These logical operations have bits that are not always affected by the input.
2535 For example, if <tt>%X</tt> has a zero bit, then the output of the
2536 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2537 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2538 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2539 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2540 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2541 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2542 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2544 <pre class="doc_code">
2545 %A = select undef, %X, %Y
2546 %B = select undef, 42, %Y
2547 %C = select %X, %Y, undef
2558 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2559 branch) conditions can go <em>either way</em>, but they have to come from one
2560 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2561 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2562 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2563 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2564 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2567 <pre class="doc_code">
2568 %A = xor undef, undef
2586 <p>This example points out that two '<tt>undef</tt>' operands are not
2587 necessarily the same. This can be surprising to people (and also matches C
2588 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2589 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2590 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2591 its value over its "live range". This is true because the variable doesn't
2592 actually <em>have a live range</em>. Instead, the value is logically read
2593 from arbitrary registers that happen to be around when needed, so the value
2594 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2595 need to have the same semantics or the core LLVM "replace all uses with"
2596 concept would not hold.</p>
2598 <pre class="doc_code">
2606 <p>These examples show the crucial difference between an <em>undefined
2607 value</em> and <em>undefined behavior</em>. An undefined value (like
2608 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2609 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2610 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2611 defined on SNaN's. However, in the second example, we can make a more
2612 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2613 arbitrary value, we are allowed to assume that it could be zero. Since a
2614 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2615 the operation does not execute at all. This allows us to delete the divide and
2616 all code after it. Because the undefined operation "can't happen", the
2617 optimizer can assume that it occurs in dead code.</p>
2619 <pre class="doc_code">
2620 a: store undef -> %X
2621 b: store %X -> undef
2627 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2628 undefined value can be assumed to not have any effect; we can assume that the
2629 value is overwritten with bits that happen to match what was already there.
2630 However, a store <em>to</em> an undefined location could clobber arbitrary
2631 memory, therefore, it has undefined behavior.</p>
2635 <!-- ======================================================================= -->
2637 <a name="poisonvalues">Poison Values</a>
2642 <p>Poison values are similar to <a href="#undefvalues">undef values</a>, however
2643 they also represent the fact that an instruction or constant expression which
2644 cannot evoke side effects has nevertheless detected a condition which results
2645 in undefined behavior.</p>
2647 <p>There is currently no way of representing a poison value in the IR; they
2648 only exist when produced by operations such as
2649 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2651 <p>Poison value behavior is defined in terms of value <i>dependence</i>:</p>
2654 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2655 their operands.</li>
2657 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2658 to their dynamic predecessor basic block.</li>
2660 <li>Function arguments depend on the corresponding actual argument values in
2661 the dynamic callers of their functions.</li>
2663 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2664 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2665 control back to them.</li>
2667 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2668 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>,
2669 or exception-throwing call instructions that dynamically transfer control
2672 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2673 referenced memory addresses, following the order in the IR
2674 (including loads and stores implied by intrinsics such as
2675 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2677 <!-- TODO: In the case of multiple threads, this only applies if the store
2678 "happens-before" the load or store. -->
2680 <!-- TODO: floating-point exception state -->
2682 <li>An instruction with externally visible side effects depends on the most
2683 recent preceding instruction with externally visible side effects, following
2684 the order in the IR. (This includes
2685 <a href="#volatile">volatile operations</a>.)</li>
2687 <li>An instruction <i>control-depends</i> on a
2688 <a href="#terminators">terminator instruction</a>
2689 if the terminator instruction has multiple successors and the instruction
2690 is always executed when control transfers to one of the successors, and
2691 may not be executed when control is transferred to another.</li>
2693 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2694 instruction if the set of instructions it otherwise depends on would be
2695 different if the terminator had transferred control to a different
2698 <li>Dependence is transitive.</li>
2702 <p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>,
2703 with the additional affect that any instruction which has a <i>dependence</i>
2704 on a poison value has undefined behavior.</p>
2706 <p>Here are some examples:</p>
2708 <pre class="doc_code">
2710 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2711 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2712 %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2713 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2715 store i32 %poison, i32* @g ; Poison value stored to memory.
2716 %poison2 = load i32* @g ; Poison value loaded back from memory.
2718 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2720 %narrowaddr = bitcast i32* @g to i16*
2721 %wideaddr = bitcast i32* @g to i64*
2722 %poison3 = load i16* %narrowaddr ; Returns a poison value.
2723 %poison4 = load i64* %wideaddr ; Returns a poison value.
2725 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2726 br i1 %cmp, label %true, label %end ; Branch to either destination.
2729 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2730 ; it has undefined behavior.
2734 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2735 ; Both edges into this PHI are
2736 ; control-dependent on %cmp, so this
2737 ; always results in a poison value.
2739 store volatile i32 0, i32* @g ; This would depend on the store in %true
2740 ; if %cmp is true, or the store in %entry
2741 ; otherwise, so this is undefined behavior.
2743 br i1 %cmp, label %second_true, label %second_end
2744 ; The same branch again, but this time the
2745 ; true block doesn't have side effects.
2752 store volatile i32 0, i32* @g ; This time, the instruction always depends
2753 ; on the store in %end. Also, it is
2754 ; control-equivalent to %end, so this is
2755 ; well-defined (ignoring earlier undefined
2756 ; behavior in this example).
2761 <!-- ======================================================================= -->
2763 <a name="blockaddress">Addresses of Basic Blocks</a>
2768 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2770 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2771 basic block in the specified function, and always has an i8* type. Taking
2772 the address of the entry block is illegal.</p>
2774 <p>This value only has defined behavior when used as an operand to the
2775 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2776 comparisons against null. Pointer equality tests between labels addresses
2777 results in undefined behavior — though, again, comparison against null
2778 is ok, and no label is equal to the null pointer. This may be passed around
2779 as an opaque pointer sized value as long as the bits are not inspected. This
2780 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2781 long as the original value is reconstituted before the <tt>indirectbr</tt>
2784 <p>Finally, some targets may provide defined semantics when using the value as
2785 the operand to an inline assembly, but that is target specific.</p>
2790 <!-- ======================================================================= -->
2792 <a name="constantexprs">Constant Expressions</a>
2797 <p>Constant expressions are used to allow expressions involving other constants
2798 to be used as constants. Constant expressions may be of
2799 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2800 operation that does not have side effects (e.g. load and call are not
2801 supported). The following is the syntax for constant expressions:</p>
2804 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2805 <dd>Truncate a constant to another type. The bit size of CST must be larger
2806 than the bit size of TYPE. Both types must be integers.</dd>
2808 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2809 <dd>Zero extend a constant to another type. The bit size of CST must be
2810 smaller than the bit size of TYPE. Both types must be integers.</dd>
2812 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2813 <dd>Sign extend a constant to another type. The bit size of CST must be
2814 smaller than the bit size of TYPE. Both types must be integers.</dd>
2816 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2817 <dd>Truncate a floating point constant to another floating point type. The
2818 size of CST must be larger than the size of TYPE. Both types must be
2819 floating point.</dd>
2821 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2822 <dd>Floating point extend a constant to another type. The size of CST must be
2823 smaller or equal to the size of TYPE. Both types must be floating
2826 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2827 <dd>Convert a floating point constant to the corresponding unsigned integer
2828 constant. TYPE must be a scalar or vector integer type. CST must be of
2829 scalar or vector floating point type. Both CST and TYPE must be scalars,
2830 or vectors of the same number of elements. If the value won't fit in the
2831 integer type, the results are undefined.</dd>
2833 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2834 <dd>Convert a floating point constant to the corresponding signed integer
2835 constant. TYPE must be a scalar or vector integer type. CST must be of
2836 scalar or vector floating point type. Both CST and TYPE must be scalars,
2837 or vectors of the same number of elements. If the value won't fit in the
2838 integer type, the results are undefined.</dd>
2840 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2841 <dd>Convert an unsigned integer constant to the corresponding floating point
2842 constant. TYPE must be a scalar or vector floating point type. CST must be
2843 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2844 vectors of the same number of elements. If the value won't fit in the
2845 floating point type, the results are undefined.</dd>
2847 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2848 <dd>Convert a signed integer constant to the corresponding floating point
2849 constant. TYPE must be a scalar or vector floating point type. CST must be
2850 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2851 vectors of the same number of elements. If the value won't fit in the
2852 floating point type, the results are undefined.</dd>
2854 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2855 <dd>Convert a pointer typed constant to the corresponding integer constant
2856 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2857 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2858 make it fit in <tt>TYPE</tt>.</dd>
2860 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2861 <dd>Convert an integer constant to a pointer constant. TYPE must be a pointer
2862 type. CST must be of integer type. The CST value is zero extended,
2863 truncated, or unchanged to make it fit in a pointer size. This one is
2864 <i>really</i> dangerous!</dd>
2866 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2867 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2868 are the same as those for the <a href="#i_bitcast">bitcast
2869 instruction</a>.</dd>
2871 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2872 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2873 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2874 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2875 instruction, the index list may have zero or more indexes, which are
2876 required to make sense for the type of "CSTPTR".</dd>
2878 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2879 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2881 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2882 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2884 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2885 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2887 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2888 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2891 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2892 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2895 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2896 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2899 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2900 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2901 constants. The index list is interpreted in a similar manner as indices in
2902 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2903 index value must be specified.</dd>
2905 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2906 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2907 constants. The index list is interpreted in a similar manner as indices in
2908 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2909 index value must be specified.</dd>
2911 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2912 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2913 be any of the <a href="#binaryops">binary</a>
2914 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2915 on operands are the same as those for the corresponding instruction
2916 (e.g. no bitwise operations on floating point values are allowed).</dd>
2923 <!-- *********************************************************************** -->
2924 <h2><a name="othervalues">Other Values</a></h2>
2925 <!-- *********************************************************************** -->
2927 <!-- ======================================================================= -->
2929 <a name="inlineasm">Inline Assembler Expressions</a>
2934 <p>LLVM supports inline assembler expressions (as opposed
2935 to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of
2936 a special value. This value represents the inline assembler as a string
2937 (containing the instructions to emit), a list of operand constraints (stored
2938 as a string), a flag that indicates whether or not the inline asm
2939 expression has side effects, and a flag indicating whether the function
2940 containing the asm needs to align its stack conservatively. An example
2941 inline assembler expression is:</p>
2943 <pre class="doc_code">
2944 i32 (i32) asm "bswap $0", "=r,r"
2947 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2948 a <a href="#i_call"><tt>call</tt></a> or an
2949 <a href="#i_invoke"><tt>invoke</tt></a> instruction.
2950 Thus, typically we have:</p>
2952 <pre class="doc_code">
2953 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2956 <p>Inline asms with side effects not visible in the constraint list must be
2957 marked as having side effects. This is done through the use of the
2958 '<tt>sideeffect</tt>' keyword, like so:</p>
2960 <pre class="doc_code">
2961 call void asm sideeffect "eieio", ""()
2964 <p>In some cases inline asms will contain code that will not work unless the
2965 stack is aligned in some way, such as calls or SSE instructions on x86,
2966 yet will not contain code that does that alignment within the asm.
2967 The compiler should make conservative assumptions about what the asm might
2968 contain and should generate its usual stack alignment code in the prologue
2969 if the '<tt>alignstack</tt>' keyword is present:</p>
2971 <pre class="doc_code">
2972 call void asm alignstack "eieio", ""()
2975 <p>Inline asms also support using non-standard assembly dialects. The assumed
2976 dialect is ATT. When the '<tt>inteldialect</tt>' keyword is present, the
2977 inline asm is using the Intel dialect. Currently, ATT and Intel are the
2978 only supported dialects. An example is:</p>
2980 <pre class="doc_code">
2981 call void asm inteldialect "eieio", ""()
2984 <p>If multiple keywords appear the '<tt>sideeffect</tt>' keyword must come
2985 first, the '<tt>alignstack</tt>' keyword second and the
2986 '<tt>inteldialect</tt>' keyword last.</p>
2989 <p>TODO: The format of the asm and constraints string still need to be
2990 documented here. Constraints on what can be done (e.g. duplication, moving,
2991 etc need to be documented). This is probably best done by reference to
2992 another document that covers inline asm from a holistic perspective.</p>
2995 <!-- _______________________________________________________________________ -->
2997 <a name="inlineasm_md">Inline Asm Metadata</a>
3002 <p>The call instructions that wrap inline asm nodes may have a
3003 "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant
3004 integers. If present, the code generator will use the integer as the
3005 location cookie value when report errors through the <tt>LLVMContext</tt>
3006 error reporting mechanisms. This allows a front-end to correlate backend
3007 errors that occur with inline asm back to the source code that produced it.
3010 <pre class="doc_code">
3011 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
3013 !42 = !{ i32 1234567 }
3016 <p>It is up to the front-end to make sense of the magic numbers it places in the
3017 IR. If the MDNode contains multiple constants, the code generator will use
3018 the one that corresponds to the line of the asm that the error occurs on.</p>
3024 <!-- ======================================================================= -->
3026 <a name="metadata">Metadata Nodes and Metadata Strings</a>
3031 <p>LLVM IR allows metadata to be attached to instructions in the program that
3032 can convey extra information about the code to the optimizers and code
3033 generator. One example application of metadata is source-level debug
3034 information. There are two metadata primitives: strings and nodes. All
3035 metadata has the <tt>metadata</tt> type and is identified in syntax by a
3036 preceding exclamation point ('<tt>!</tt>').</p>
3038 <p>A metadata string is a string surrounded by double quotes. It can contain
3039 any character by escaping non-printable characters with "<tt>\xx</tt>" where
3040 "<tt>xx</tt>" is the two digit hex code. For example:
3041 "<tt>!"test\00"</tt>".</p>
3043 <p>Metadata nodes are represented with notation similar to structure constants
3044 (a comma separated list of elements, surrounded by braces and preceded by an
3045 exclamation point). Metadata nodes can have any values as their operand. For
3048 <div class="doc_code">
3050 !{ metadata !"test\00", i32 10}
3054 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
3055 metadata nodes, which can be looked up in the module symbol table. For
3058 <div class="doc_code">
3060 !foo = metadata !{!4, !3}
3064 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
3065 function is using two metadata arguments:</p>
3067 <div class="doc_code">
3069 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
3073 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
3074 attached to the <tt>add</tt> instruction using the <tt>!dbg</tt>
3077 <div class="doc_code">
3079 %indvar.next = add i64 %indvar, 1, !dbg !21
3083 <p>More information about specific metadata nodes recognized by the optimizers
3084 and code generator is found below.</p>
3086 <!-- _______________________________________________________________________ -->
3088 <a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
3093 <p>In LLVM IR, memory does not have types, so LLVM's own type system is not
3094 suitable for doing TBAA. Instead, metadata is added to the IR to describe
3095 a type system of a higher level language. This can be used to implement
3096 typical C/C++ TBAA, but it can also be used to implement custom alias
3097 analysis behavior for other languages.</p>
3099 <p>The current metadata format is very simple. TBAA metadata nodes have up to
3100 three fields, e.g.:</p>
3102 <div class="doc_code">
3104 !0 = metadata !{ metadata !"an example type tree" }
3105 !1 = metadata !{ metadata !"int", metadata !0 }
3106 !2 = metadata !{ metadata !"float", metadata !0 }
3107 !3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
3111 <p>The first field is an identity field. It can be any value, usually
3112 a metadata string, which uniquely identifies the type. The most important
3113 name in the tree is the name of the root node. Two trees with
3114 different root node names are entirely disjoint, even if they
3115 have leaves with common names.</p>
3117 <p>The second field identifies the type's parent node in the tree, or
3118 is null or omitted for a root node. A type is considered to alias
3119 all of its descendants and all of its ancestors in the tree. Also,
3120 a type is considered to alias all types in other trees, so that
3121 bitcode produced from multiple front-ends is handled conservatively.</p>
3123 <p>If the third field is present, it's an integer which if equal to 1
3124 indicates that the type is "constant" (meaning
3125 <tt>pointsToConstantMemory</tt> should return true; see
3126 <a href="AliasAnalysis.html#OtherItfs">other useful
3127 <tt>AliasAnalysis</tt> methods</a>).</p>
3131 <!-- _______________________________________________________________________ -->
3133 <a name="tbaa.struct">'<tt>tbaa.struct</tt>' Metadata</a>
3138 <p>The <a href="#int_memcpy"><tt>llvm.memcpy</tt></a> is often used to implement
3139 aggregate assignment operations in C and similar languages, however it is
3140 defined to copy a contiguous region of memory, which is more than strictly
3141 necessary for aggregate types which contain holes due to padding. Also, it
3142 doesn't contain any TBAA information about the fields of the aggregate.</p>
3144 <p><tt>!tbaa.struct</tt> metadata can describe which memory subregions in a memcpy
3145 are padding and what the TBAA tags of the struct are.</p>
3147 <p>The current metadata format is very simple. <tt>!tbaa.struct</tt> metadata nodes
3148 are a list of operands which are in conceptual groups of three. For each
3149 group of three, the first operand gives the byte offset of a field in bytes,
3150 the second gives its size in bytes, and the third gives its
3153 <div class="doc_code">
3155 !4 = metadata !{ i64 0, i64 4, metadata !1, i64 8, i64 4, metadata !2 }
3159 <p>This describes a struct with two fields. The first is at offset 0 bytes
3160 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
3161 and has size 4 bytes and has tbaa tag !2.</p>
3163 <p>Note that the fields need not be contiguous. In this example, there is a
3164 4 byte gap between the two fields. This gap represents padding which
3165 does not carry useful data and need not be preserved.</p>
3169 <!-- _______________________________________________________________________ -->
3171 <a name="fpmath">'<tt>fpmath</tt>' Metadata</a>
3176 <p><tt>fpmath</tt> metadata may be attached to any instruction of floating point
3177 type. It can be used to express the maximum acceptable error in the result of
3178 that instruction, in ULPs, thus potentially allowing the compiler to use a
3179 more efficient but less accurate method of computing it. ULP is defined as
3184 <p>If <tt>x</tt> is a real number that lies between two finite consecutive
3185 floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one
3186 of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the
3187 distance between the two non-equal finite floating-point numbers nearest
3188 <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p>
3192 <p>The metadata node shall consist of a single positive floating point number
3193 representing the maximum relative error, for example:</p>
3195 <div class="doc_code">
3197 !0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3203 <!-- _______________________________________________________________________ -->
3205 <a name="range">'<tt>range</tt>' Metadata</a>
3209 <p><tt>range</tt> metadata may be attached only to loads of integer types. It
3210 expresses the possible ranges the loaded value is in. The ranges are
3211 represented with a flattened list of integers. The loaded value is known to
3212 be in the union of the ranges defined by each consecutive pair. Each pair
3213 has the following properties:</p>
3215 <li>The type must match the type loaded by the instruction.</li>
3216 <li>The pair <tt>a,b</tt> represents the range <tt>[a,b)</tt>.</li>
3217 <li>Both <tt>a</tt> and <tt>b</tt> are constants.</li>
3218 <li>The range is allowed to wrap.</li>
3219 <li>The range should not represent the full or empty set. That is,
3220 <tt>a!=b</tt>. </li>
3222 <p> In addition, the pairs must be in signed order of the lower bound and
3223 they must be non-contiguous.</p>
3226 <div class="doc_code">
3228 %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
3229 %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3230 %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
3231 %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
3233 !0 = metadata !{ i8 0, i8 2 }
3234 !1 = metadata !{ i8 255, i8 2 }
3235 !2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
3236 !3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
3244 <!-- *********************************************************************** -->
3246 <a name="module_flags">Module Flags Metadata</a>
3248 <!-- *********************************************************************** -->
3252 <p>Information about the module as a whole is difficult to convey to LLVM's
3253 subsystems. The LLVM IR isn't sufficient to transmit this
3254 information. The <tt>llvm.module.flags</tt> named metadata exists in order to
3255 facilitate this. These flags are in the form of key / value pairs —
3256 much like a dictionary — making it easy for any subsystem who cares
3257 about a flag to look it up.</p>
3259 <p>The <tt>llvm.module.flags</tt> metadata contains a list of metadata
3260 triplets. Each triplet has the following form:</p>
3263 <li>The first element is a <i>behavior</i> flag, which specifies the behavior
3264 when two (or more) modules are merged together, and it encounters two (or
3265 more) metadata with the same ID. The supported behaviors are described
3268 <li>The second element is a metadata string that is a unique ID for the
3269 metadata. How each ID is interpreted is documented below.</li>
3271 <li>The third element is the value of the flag.</li>
3274 <p>When two (or more) modules are merged together, the resulting
3275 <tt>llvm.module.flags</tt> metadata is the union of the
3276 modules' <tt>llvm.module.flags</tt> metadata. The only exception being a flag
3277 with the <i>Override</i> behavior, which may override another flag's value
3280 <p>The following behaviors are supported:</p>
3282 <table border="1" cellspacing="0" cellpadding="4">
3292 <dt><b>Error</b></dt>
3293 <dd>Emits an error if two values disagree. It is an error to have an ID
3294 with both an Error and a Warning behavior.</dd>
3302 <dt><b>Warning</b></dt>
3303 <dd>Emits a warning if two values disagree.</dd>
3311 <dt><b>Require</b></dt>
3312 <dd>Emits an error when the specified value is not present or doesn't
3313 have the specified value. It is an error for two (or more)
3314 <tt>llvm.module.flags</tt> with the same ID to have the Require
3315 behavior but different values. There may be multiple Require flags
3324 <dt><b>Override</b></dt>
3325 <dd>Uses the specified value if the two values disagree. It is an
3326 error for two (or more) <tt>llvm.module.flags</tt> with the same
3327 ID to have the Override behavior but different values.</dd>
3334 <p>An example of module flags:</p>
3336 <pre class="doc_code">
3337 !0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3338 !1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3339 !2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3340 !3 = metadata !{ i32 3, metadata !"qux",
3342 metadata !"foo", i32 1
3345 !llvm.module.flags = !{ !0, !1, !2, !3 }
3349 <li><p>Metadata <tt>!0</tt> has the ID <tt>!"foo"</tt> and the value '1'. The
3350 behavior if two or more <tt>!"foo"</tt> flags are seen is to emit an
3351 error if their values are not equal.</p></li>
3353 <li><p>Metadata <tt>!1</tt> has the ID <tt>!"bar"</tt> and the value '37'. The
3354 behavior if two or more <tt>!"bar"</tt> flags are seen is to use the
3355 value '37' if their values are not equal.</p></li>
3357 <li><p>Metadata <tt>!2</tt> has the ID <tt>!"qux"</tt> and the value '42'. The
3358 behavior if two or more <tt>!"qux"</tt> flags are seen is to emit a
3359 warning if their values are not equal.</p></li>
3361 <li><p>Metadata <tt>!3</tt> has the ID <tt>!"qux"</tt> and the value:</p>
3363 <pre class="doc_code">
3364 metadata !{ metadata !"foo", i32 1 }
3367 <p>The behavior is to emit an error if the <tt>llvm.module.flags</tt> does
3368 not contain a flag with the ID <tt>!"foo"</tt> that has the value
3369 '1'. If two or more <tt>!"qux"</tt> flags exist, then they must have
3370 the same value or an error will be issued.</p></li>
3374 <!-- ======================================================================= -->
3376 <a name="objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a>
3381 <p>On the Mach-O platform, Objective-C stores metadata about garbage collection
3382 in a special section called "image info". The metadata consists of a version
3383 number and a bitmask specifying what types of garbage collection are
3384 supported (if any) by the file. If two or more modules are linked together
3385 their garbage collection metadata needs to be merged rather than appended
3388 <p>The Objective-C garbage collection module flags metadata consists of the
3389 following key-value pairs:</p>
3391 <table border="1" cellspacing="0" cellpadding="4">
3399 <td><tt>Objective-C Version</tt></td>
3400 <td align="left"><b>[Required]</b> — The Objective-C ABI
3401 version. Valid values are 1 and 2.</td>
3404 <td><tt>Objective-C Image Info Version</tt></td>
3405 <td align="left"><b>[Required]</b> — The version of the image info
3406 section. Currently always 0.</td>
3409 <td><tt>Objective-C Image Info Section</tt></td>
3410 <td align="left"><b>[Required]</b> — The section to place the
3411 metadata. Valid values are <tt>"__OBJC, __image_info, regular"</tt> for
3412 Objective-C ABI version 1, and <tt>"__DATA,__objc_imageinfo, regular,
3413 no_dead_strip"</tt> for Objective-C ABI version 2.</td>
3416 <td><tt>Objective-C Garbage Collection</tt></td>
3417 <td align="left"><b>[Required]</b> — Specifies whether garbage
3418 collection is supported or not. Valid values are 0, for no garbage
3419 collection, and 2, for garbage collection supported.</td>
3422 <td><tt>Objective-C GC Only</tt></td>
3423 <td align="left"><b>[Optional]</b> — Specifies that only garbage
3424 collection is supported. If present, its value must be 6. This flag
3425 requires that the <tt>Objective-C Garbage Collection</tt> flag have the
3431 <p>Some important flag interactions:</p>
3434 <li>If a module with <tt>Objective-C Garbage Collection</tt> set to 0 is
3435 merged with a module with <tt>Objective-C Garbage Collection</tt> set to
3436 2, then the resulting module has the <tt>Objective-C Garbage
3437 Collection</tt> flag set to 0.</li>
3439 <li>A module with <tt>Objective-C Garbage Collection</tt> set to 0 cannot be
3440 merged with a module with <tt>Objective-C GC Only</tt> set to 6.</li>
3447 <!-- *********************************************************************** -->
3449 <a name="intrinsic_globals">Intrinsic Global Variables</a>
3451 <!-- *********************************************************************** -->
3453 <p>LLVM has a number of "magic" global variables that contain data that affect
3454 code generation or other IR semantics. These are documented here. All globals
3455 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
3456 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
3459 <!-- ======================================================================= -->
3461 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
3466 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
3467 href="#linkage_appending">appending linkage</a>. This array contains a list of
3468 pointers to global variables and functions which may optionally have a pointer
3469 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
3471 <div class="doc_code">
3476 @llvm.used = appending global [2 x i8*] [
3478 i8* bitcast (i32* @Y to i8*)
3479 ], section "llvm.metadata"
3483 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
3484 compiler, assembler, and linker are required to treat the symbol as if there
3485 is a reference to the global that it cannot see. For example, if a variable
3486 has internal linkage and no references other than that from
3487 the <tt>@llvm.used</tt> list, it cannot be deleted. This is commonly used to
3488 represent references from inline asms and other things the compiler cannot
3489 "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p>
3491 <p>On some targets, the code generator must emit a directive to the assembler or
3492 object file to prevent the assembler and linker from molesting the
3497 <!-- ======================================================================= -->
3499 <a name="intg_compiler_used">
3500 The '<tt>llvm.compiler.used</tt>' Global Variable
3506 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
3507 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
3508 touching the symbol. On targets that support it, this allows an intelligent
3509 linker to optimize references to the symbol without being impeded as it would
3510 be by <tt>@llvm.used</tt>.</p>
3512 <p>This is a rare construct that should only be used in rare circumstances, and
3513 should not be exposed to source languages.</p>
3517 <!-- ======================================================================= -->
3519 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
3524 <div class="doc_code">
3526 %0 = type { i32, void ()* }
3527 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3531 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor
3532 functions and associated priorities. The functions referenced by this array
3533 will be called in ascending order of priority (i.e. lowest first) when the
3534 module is loaded. The order of functions with the same priority is not
3539 <!-- ======================================================================= -->
3541 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3546 <div class="doc_code">
3548 %0 = type { i32, void ()* }
3549 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3553 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions
3554 and associated priorities. The functions referenced by this array will be
3555 called in descending order of priority (i.e. highest first) when the module
3556 is loaded. The order of functions with the same priority is not defined.</p>
3562 <!-- *********************************************************************** -->
3563 <h2><a name="instref">Instruction Reference</a></h2>
3564 <!-- *********************************************************************** -->
3568 <p>The LLVM instruction set consists of several different classifications of
3569 instructions: <a href="#terminators">terminator
3570 instructions</a>, <a href="#binaryops">binary instructions</a>,
3571 <a href="#bitwiseops">bitwise binary instructions</a>,
3572 <a href="#memoryops">memory instructions</a>, and
3573 <a href="#otherops">other instructions</a>.</p>
3575 <!-- ======================================================================= -->
3577 <a name="terminators">Terminator Instructions</a>
3582 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3583 in a program ends with a "Terminator" instruction, which indicates which
3584 block should be executed after the current block is finished. These
3585 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3586 control flow, not values (the one exception being the
3587 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3589 <p>The terminator instructions are:
3590 '<a href="#i_ret"><tt>ret</tt></a>',
3591 '<a href="#i_br"><tt>br</tt></a>',
3592 '<a href="#i_switch"><tt>switch</tt></a>',
3593 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3594 '<a href="#i_invoke"><tt>invoke</tt></a>',
3595 '<a href="#i_resume"><tt>resume</tt></a>', and
3596 '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3598 <!-- _______________________________________________________________________ -->
3600 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3607 ret <type> <value> <i>; Return a value from a non-void function</i>
3608 ret void <i>; Return from void function</i>
3612 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3613 a value) from a function back to the caller.</p>
3615 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3616 value and then causes control flow, and one that just causes control flow to
3620 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3621 return value. The type of the return value must be a
3622 '<a href="#t_firstclass">first class</a>' type.</p>
3624 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3625 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3626 value or a return value with a type that does not match its type, or if it
3627 has a void return type and contains a '<tt>ret</tt>' instruction with a
3631 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3632 the calling function's context. If the caller is a
3633 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3634 instruction after the call. If the caller was an
3635 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3636 the beginning of the "normal" destination block. If the instruction returns
3637 a value, that value shall set the call or invoke instruction's return
3642 ret i32 5 <i>; Return an integer value of 5</i>
3643 ret void <i>; Return from a void function</i>
3644 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3648 <!-- _______________________________________________________________________ -->
3650 <a name="i_br">'<tt>br</tt>' Instruction</a>
3657 br i1 <cond>, label <iftrue>, label <iffalse>
3658 br label <dest> <i>; Unconditional branch</i>
3662 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3663 different basic block in the current function. There are two forms of this
3664 instruction, corresponding to a conditional branch and an unconditional
3668 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3669 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3670 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3674 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3675 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3676 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3677 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3682 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3683 br i1 %cond, label %IfEqual, label %IfUnequal
3685 <a href="#i_ret">ret</a> i32 1
3687 <a href="#i_ret">ret</a> i32 0
3692 <!-- _______________________________________________________________________ -->
3694 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3701 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3705 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3706 several different places. It is a generalization of the '<tt>br</tt>'
3707 instruction, allowing a branch to occur to one of many possible
3711 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3712 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3713 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3714 The table is not allowed to contain duplicate constant entries.</p>
3717 <p>The <tt>switch</tt> instruction specifies a table of values and
3718 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3719 is searched for the given value. If the value is found, control flow is
3720 transferred to the corresponding destination; otherwise, control flow is
3721 transferred to the default destination.</p>
3723 <h5>Implementation:</h5>
3724 <p>Depending on properties of the target machine and the particular
3725 <tt>switch</tt> instruction, this instruction may be code generated in
3726 different ways. For example, it could be generated as a series of chained
3727 conditional branches or with a lookup table.</p>
3731 <i>; Emulate a conditional br instruction</i>
3732 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3733 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3735 <i>; Emulate an unconditional br instruction</i>
3736 switch i32 0, label %dest [ ]
3738 <i>; Implement a jump table:</i>
3739 switch i32 %val, label %otherwise [ i32 0, label %onzero
3741 i32 2, label %ontwo ]
3747 <!-- _______________________________________________________________________ -->
3749 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3756 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3761 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3762 within the current function, whose address is specified by
3763 "<tt>address</tt>". Address must be derived from a <a
3764 href="#blockaddress">blockaddress</a> constant.</p>
3768 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3769 rest of the arguments indicate the full set of possible destinations that the
3770 address may point to. Blocks are allowed to occur multiple times in the
3771 destination list, though this isn't particularly useful.</p>
3773 <p>This destination list is required so that dataflow analysis has an accurate
3774 understanding of the CFG.</p>
3778 <p>Control transfers to the block specified in the address argument. All
3779 possible destination blocks must be listed in the label list, otherwise this
3780 instruction has undefined behavior. This implies that jumps to labels
3781 defined in other functions have undefined behavior as well.</p>
3783 <h5>Implementation:</h5>
3785 <p>This is typically implemented with a jump through a register.</p>
3789 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3795 <!-- _______________________________________________________________________ -->
3797 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3804 <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>]
3805 to label <normal label> unwind label <exception label>
3809 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3810 function, with the possibility of control flow transfer to either the
3811 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3812 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3813 control flow will return to the "normal" label. If the callee (or any
3814 indirect callees) returns via the "<a href="#i_resume"><tt>resume</tt></a>"
3815 instruction or other exception handling mechanism, control is interrupted and
3816 continued at the dynamically nearest "exception" label.</p>
3818 <p>The '<tt>exception</tt>' label is a
3819 <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3820 exception. As such, '<tt>exception</tt>' label is required to have the
3821 "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3822 the information about the behavior of the program after unwinding
3823 happens, as its first non-PHI instruction. The restrictions on the
3824 "<tt>landingpad</tt>" instruction's tightly couples it to the
3825 "<tt>invoke</tt>" instruction, so that the important information contained
3826 within the "<tt>landingpad</tt>" instruction can't be lost through normal
3830 <p>This instruction requires several arguments:</p>
3833 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3834 convention</a> the call should use. If none is specified, the call
3835 defaults to using C calling conventions.</li>
3837 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3838 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3839 '<tt>inreg</tt>' attributes are valid here.</li>
3841 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3842 function value being invoked. In most cases, this is a direct function
3843 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3844 off an arbitrary pointer to function value.</li>
3846 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3847 function to be invoked. </li>
3849 <li>'<tt>function args</tt>': argument list whose types match the function
3850 signature argument types and parameter attributes. All arguments must be
3851 of <a href="#t_firstclass">first class</a> type. If the function
3852 signature indicates the function accepts a variable number of arguments,
3853 the extra arguments can be specified.</li>
3855 <li>'<tt>normal label</tt>': the label reached when the called function
3856 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3858 <li>'<tt>exception label</tt>': the label reached when a callee returns via
3859 the <a href="#i_resume"><tt>resume</tt></a> instruction or other exception
3860 handling mechanism.</li>
3862 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3863 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3864 '<tt>readnone</tt>' attributes are valid here.</li>
3868 <p>This instruction is designed to operate as a standard
3869 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3870 primary difference is that it establishes an association with a label, which
3871 is used by the runtime library to unwind the stack.</p>
3873 <p>This instruction is used in languages with destructors to ensure that proper
3874 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3875 exception. Additionally, this is important for implementation of
3876 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3878 <p>For the purposes of the SSA form, the definition of the value returned by the
3879 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3880 block to the "normal" label. If the callee unwinds then no return value is
3885 %retval = invoke i32 @Test(i32 15) to label %Continue
3886 unwind label %TestCleanup <i>; {i32}:retval set</i>
3887 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3888 unwind label %TestCleanup <i>; {i32}:retval set</i>
3893 <!-- _______________________________________________________________________ -->
3896 <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3903 resume <type> <value>
3907 <p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3911 <p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3912 same type as the result of any '<tt>landingpad</tt>' instruction in the same
3916 <p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3917 (in-flight) exception whose unwinding was interrupted with
3918 a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3922 resume { i8*, i32 } %exn
3927 <!-- _______________________________________________________________________ -->
3930 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3941 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3942 instruction is used to inform the optimizer that a particular portion of the
3943 code is not reachable. This can be used to indicate that the code after a
3944 no-return function cannot be reached, and other facts.</p>
3947 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3953 <!-- ======================================================================= -->
3955 <a name="binaryops">Binary Operations</a>
3960 <p>Binary operators are used to do most of the computation in a program. They
3961 require two operands of the same type, execute an operation on them, and
3962 produce a single value. The operands might represent multiple data, as is
3963 the case with the <a href="#t_vector">vector</a> data type. The result value
3964 has the same type as its operands.</p>
3966 <p>There are several different binary operators:</p>
3968 <!-- _______________________________________________________________________ -->
3970 <a name="i_add">'<tt>add</tt>' Instruction</a>
3977 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3978 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3979 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3980 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3984 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3987 <p>The two arguments to the '<tt>add</tt>' instruction must
3988 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3989 integer values. Both arguments must have identical types.</p>
3992 <p>The value produced is the integer sum of the two operands.</p>
3994 <p>If the sum has unsigned overflow, the result returned is the mathematical
3995 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3997 <p>Because LLVM integers use a two's complement representation, this instruction
3998 is appropriate for both signed and unsigned integers.</p>
4000 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
4001 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
4002 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
4003 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4004 respectively, occurs.</p>
4008 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
4013 <!-- _______________________________________________________________________ -->
4015 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
4022 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4026 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
4029 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
4030 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4031 floating point values. Both arguments must have identical types.</p>
4034 <p>The value produced is the floating point sum of the two operands. This
4035 instruction can also take any number of <a href="#fastmath">fast-math
4036 flags</a>, which are optimization hints to enable otherwise unsafe floating
4037 point optimizations:</p>
4041 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
4046 <!-- _______________________________________________________________________ -->
4048 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
4055 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4056 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4057 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4058 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4062 <p>The '<tt>sub</tt>' instruction returns the difference of its two
4065 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
4066 '<tt>neg</tt>' instruction present in most other intermediate
4067 representations.</p>
4070 <p>The two arguments to the '<tt>sub</tt>' instruction must
4071 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4072 integer values. Both arguments must have identical types.</p>
4075 <p>The value produced is the integer difference of the two operands.</p>
4077 <p>If the difference has unsigned overflow, the result returned is the
4078 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
4081 <p>Because LLVM integers use a two's complement representation, this instruction
4082 is appropriate for both signed and unsigned integers.</p>
4084 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
4085 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
4086 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
4087 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4088 respectively, occurs.</p>
4092 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
4093 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
4098 <!-- _______________________________________________________________________ -->
4100 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
4107 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4111 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
4114 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
4115 '<tt>fneg</tt>' instruction present in most other intermediate
4116 representations.</p>
4119 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
4120 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4121 floating point values. Both arguments must have identical types.</p>
4124 <p>The value produced is the floating point difference of the two operands.
4125 This instruction can also take any number of <a href="#fastmath">fast-math
4126 flags</a>, which are optimization hints to enable otherwise unsafe floating
4127 point optimizations:</p>
4131 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
4132 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
4137 <!-- _______________________________________________________________________ -->
4139 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
4146 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4147 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4148 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4149 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4153 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
4156 <p>The two arguments to the '<tt>mul</tt>' instruction must
4157 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4158 integer values. Both arguments must have identical types.</p>
4161 <p>The value produced is the integer product of the two operands.</p>
4163 <p>If the result of the multiplication has unsigned overflow, the result
4164 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
4165 width of the result.</p>
4167 <p>Because LLVM integers use a two's complement representation, and the result
4168 is the same width as the operands, this instruction returns the correct
4169 result for both signed and unsigned integers. If a full product
4170 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
4171 be sign-extended or zero-extended as appropriate to the width of the full
4174 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
4175 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
4176 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
4177 is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4178 respectively, occurs.</p>
4182 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
4187 <!-- _______________________________________________________________________ -->
4189 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
4196 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4200 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
4203 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
4204 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4205 floating point values. Both arguments must have identical types.</p>
4208 <p>The value produced is the floating point product of the two operands. This
4209 instruction can also take any number of <a href="#fastmath">fast-math
4210 flags</a>, which are optimization hints to enable otherwise unsafe floating
4211 point optimizations:</p>
4215 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
4220 <!-- _______________________________________________________________________ -->
4222 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
4229 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4230 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4234 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
4237 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
4238 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4239 values. Both arguments must have identical types.</p>
4242 <p>The value produced is the unsigned integer quotient of the two operands.</p>
4244 <p>Note that unsigned integer division and signed integer division are distinct
4245 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
4247 <p>Division by zero leads to undefined behavior.</p>
4249 <p>If the <tt>exact</tt> keyword is present, the result value of the
4250 <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a
4251 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
4256 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4261 <!-- _______________________________________________________________________ -->
4263 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
4270 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4271 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4275 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
4278 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
4279 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4280 values. Both arguments must have identical types.</p>
4283 <p>The value produced is the signed integer quotient of the two operands rounded
4286 <p>Note that signed integer division and unsigned integer division are distinct
4287 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
4289 <p>Division by zero leads to undefined behavior. Overflow also leads to
4290 undefined behavior; this is a rare case, but can occur, for example, by doing
4291 a 32-bit division of -2147483648 by -1.</p>
4293 <p>If the <tt>exact</tt> keyword is present, the result value of the
4294 <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would
4299 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
4304 <!-- _______________________________________________________________________ -->
4306 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
4313 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4317 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
4320 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
4321 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4322 floating point values. Both arguments must have identical types.</p>
4325 <p>The value produced is the floating point quotient of the two operands. This
4326 instruction can also take any number of <a href="#fastmath">fast-math
4327 flags</a>, which are optimization hints to enable otherwise unsafe floating
4328 point optimizations:</p>
4333 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
4338 <!-- _______________________________________________________________________ -->
4340 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
4347 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4351 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
4352 division of its two arguments.</p>
4355 <p>The two arguments to the '<tt>urem</tt>' instruction must be
4356 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4357 values. Both arguments must have identical types.</p>
4360 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
4361 This instruction always performs an unsigned division to get the
4364 <p>Note that unsigned integer remainder and signed integer remainder are
4365 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
4367 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
4371 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4376 <!-- _______________________________________________________________________ -->
4378 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
4385 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4389 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
4390 division of its two operands. This instruction can also take
4391 <a href="#t_vector">vector</a> versions of the values in which case the
4392 elements must be integers.</p>
4395 <p>The two arguments to the '<tt>srem</tt>' instruction must be
4396 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4397 values. Both arguments must have identical types.</p>
4400 <p>This instruction returns the <i>remainder</i> of a division (where the result
4401 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
4402 <i>modulo</i> operator (where the result is either zero or has the same sign
4403 as the divisor, <tt>op2</tt>) of a value.
4404 For more information about the difference,
4405 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
4406 Math Forum</a>. For a table of how this is implemented in various languages,
4407 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
4408 Wikipedia: modulo operation</a>.</p>
4410 <p>Note that signed integer remainder and unsigned integer remainder are
4411 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
4413 <p>Taking the remainder of a division by zero leads to undefined behavior.
4414 Overflow also leads to undefined behavior; this is a rare case, but can
4415 occur, for example, by taking the remainder of a 32-bit division of
4416 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
4417 lets srem be implemented using instructions that return both the result of
4418 the division and the remainder.)</p>
4422 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
4427 <!-- _______________________________________________________________________ -->
4429 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
4436 <result> = frem [fast-math flags]* <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4440 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
4441 its two operands.</p>
4444 <p>The two arguments to the '<tt>frem</tt>' instruction must be
4445 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4446 floating point values. Both arguments must have identical types.</p>
4449 <p>This instruction returns the <i>remainder</i> of a division. The remainder
4450 has the same sign as the dividend. This instruction can also take any number
4451 of <a href="#fastmath">fast-math flags</a>, which are optimization hints to
4452 enable otherwise unsafe floating point optimizations:</p>
4456 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
4463 <!-- ======================================================================= -->
4465 <a name="bitwiseops">Bitwise Binary Operations</a>
4470 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
4471 program. They are generally very efficient instructions and can commonly be
4472 strength reduced from other instructions. They require two operands of the
4473 same type, execute an operation on them, and produce a single value. The
4474 resulting value is the same type as its operands.</p>
4476 <!-- _______________________________________________________________________ -->
4478 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
4485 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4486 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4487 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4488 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4492 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
4493 a specified number of bits.</p>
4496 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
4497 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4498 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4501 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
4502 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
4503 is (statically or dynamically) negative or equal to or larger than the number
4504 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4505 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4506 shift amount in <tt>op2</tt>.</p>
4508 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
4509 <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits. If
4510 the <tt>nsw</tt> keyword is present, then the shift produces a
4511 <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree
4512 with the resultant sign bit. As such, NUW/NSW have the same semantics as
4513 they would if the shift were expressed as a mul instruction with the same
4514 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
4518 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
4519 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
4520 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
4521 <result> = shl i32 1, 32 <i>; undefined</i>
4522 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
4527 <!-- _______________________________________________________________________ -->
4529 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4536 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4537 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4541 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4542 operand shifted to the right a specified number of bits with zero fill.</p>
4545 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4546 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4547 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4550 <p>This instruction always performs a logical shift right operation. The most
4551 significant bits of the result will be filled with zero bits after the shift.
4552 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4553 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4554 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4555 shift amount in <tt>op2</tt>.</p>
4557 <p>If the <tt>exact</tt> keyword is present, the result value of the
4558 <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4559 shifted out are non-zero.</p>
4564 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
4565 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
4566 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
4567 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
4568 <result> = lshr i32 1, 32 <i>; undefined</i>
4569 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
4574 <!-- _______________________________________________________________________ -->
4576 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4583 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4584 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4588 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4589 operand shifted to the right a specified number of bits with sign
4593 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4594 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4595 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4598 <p>This instruction always performs an arithmetic shift right operation, The
4599 most significant bits of the result will be filled with the sign bit
4600 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4601 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4602 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4603 the corresponding shift amount in <tt>op2</tt>.</p>
4605 <p>If the <tt>exact</tt> keyword is present, the result value of the
4606 <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4607 shifted out are non-zero.</p>
4611 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4612 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4613 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4614 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4615 <result> = ashr i32 1, 32 <i>; undefined</i>
4616 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4621 <!-- _______________________________________________________________________ -->
4623 <a name="i_and">'<tt>and</tt>' Instruction</a>
4630 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4634 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4638 <p>The two arguments to the '<tt>and</tt>' instruction must be
4639 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4640 values. Both arguments must have identical types.</p>
4643 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4645 <table border="1" cellspacing="0" cellpadding="4">
4677 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4678 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4679 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4682 <!-- _______________________________________________________________________ -->
4684 <a name="i_or">'<tt>or</tt>' Instruction</a>
4691 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4695 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4699 <p>The two arguments to the '<tt>or</tt>' instruction must be
4700 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4701 values. Both arguments must have identical types.</p>
4704 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4706 <table border="1" cellspacing="0" cellpadding="4">
4738 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4739 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4740 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4745 <!-- _______________________________________________________________________ -->
4747 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4754 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4758 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4759 its two operands. The <tt>xor</tt> is used to implement the "one's
4760 complement" operation, which is the "~" operator in C.</p>
4763 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4764 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4765 values. Both arguments must have identical types.</p>
4768 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4770 <table border="1" cellspacing="0" cellpadding="4">
4802 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4803 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4804 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4805 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4812 <!-- ======================================================================= -->
4814 <a name="vectorops">Vector Operations</a>
4819 <p>LLVM supports several instructions to represent vector operations in a
4820 target-independent manner. These instructions cover the element-access and
4821 vector-specific operations needed to process vectors effectively. While LLVM
4822 does directly support these vector operations, many sophisticated algorithms
4823 will want to use target-specific intrinsics to take full advantage of a
4824 specific target.</p>
4826 <!-- _______________________________________________________________________ -->
4828 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4835 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4839 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4840 from a vector at a specified index.</p>
4844 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4845 of <a href="#t_vector">vector</a> type. The second operand is an index
4846 indicating the position from which to extract the element. The index may be
4850 <p>The result is a scalar of the same type as the element type of
4851 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4852 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4853 results are undefined.</p>
4857 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4862 <!-- _______________________________________________________________________ -->
4864 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4871 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4875 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4876 vector at a specified index.</p>
4879 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4880 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4881 whose type must equal the element type of the first operand. The third
4882 operand is an index indicating the position at which to insert the value.
4883 The index may be a variable.</p>
4886 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4887 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4888 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4889 results are undefined.</p>
4893 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4898 <!-- _______________________________________________________________________ -->
4900 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4907 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4911 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4912 from two input vectors, returning a vector with the same element type as the
4913 input and length that is the same as the shuffle mask.</p>
4916 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4917 with the same type. The third argument is a shuffle mask whose
4918 element type is always 'i32'. The result of the instruction is a vector
4919 whose length is the same as the shuffle mask and whose element type is the
4920 same as the element type of the first two operands.</p>
4922 <p>The shuffle mask operand is required to be a constant vector with either
4923 constant integer or undef values.</p>
4926 <p>The elements of the two input vectors are numbered from left to right across
4927 both of the vectors. The shuffle mask operand specifies, for each element of
4928 the result vector, which element of the two input vectors the result element
4929 gets. The element selector may be undef (meaning "don't care") and the
4930 second operand may be undef if performing a shuffle from only one vector.</p>
4934 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4935 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4936 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4937 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4938 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4939 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4940 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4941 <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>
4948 <!-- ======================================================================= -->
4950 <a name="aggregateops">Aggregate Operations</a>
4955 <p>LLVM supports several instructions for working with
4956 <a href="#t_aggregate">aggregate</a> values.</p>
4958 <!-- _______________________________________________________________________ -->
4960 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4967 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4971 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4972 from an <a href="#t_aggregate">aggregate</a> value.</p>
4975 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4976 of <a href="#t_struct">struct</a> or
4977 <a href="#t_array">array</a> type. The operands are constant indices to
4978 specify which value to extract in a similar manner as indices in a
4979 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4980 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4982 <li>Since the value being indexed is not a pointer, the first index is
4983 omitted and assumed to be zero.</li>
4984 <li>At least one index must be specified.</li>
4985 <li>Not only struct indices but also array indices must be in
4990 <p>The result is the value at the position in the aggregate specified by the
4995 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
5000 <!-- _______________________________________________________________________ -->
5002 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
5009 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
5013 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
5014 in an <a href="#t_aggregate">aggregate</a> value.</p>
5017 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
5018 of <a href="#t_struct">struct</a> or
5019 <a href="#t_array">array</a> type. The second operand is a first-class
5020 value to insert. The following operands are constant indices indicating
5021 the position at which to insert the value in a similar manner as indices in a
5022 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
5023 value to insert must have the same type as the value identified by the
5027 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
5028 that of <tt>val</tt> except that the value at the position specified by the
5029 indices is that of <tt>elt</tt>.</p>
5033 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
5034 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
5035 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
5042 <!-- ======================================================================= -->
5044 <a name="memoryops">Memory Access and Addressing Operations</a>
5049 <p>A key design point of an SSA-based representation is how it represents
5050 memory. In LLVM, no memory locations are in SSA form, which makes things
5051 very simple. This section describes how to read, write, and allocate
5054 <!-- _______________________________________________________________________ -->
5056 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
5063 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
5067 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
5068 currently executing function, to be automatically released when this function
5069 returns to its caller. The object is always allocated in the generic address
5070 space (address space zero).</p>
5073 <p>The '<tt>alloca</tt>' instruction
5074 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
5075 runtime stack, returning a pointer of the appropriate type to the program.
5076 If "NumElements" is specified, it is the number of elements allocated,
5077 otherwise "NumElements" is defaulted to be one. If a constant alignment is
5078 specified, the value result of the allocation is guaranteed to be aligned to
5079 at least that boundary. If not specified, or if zero, the target can choose
5080 to align the allocation on any convenient boundary compatible with the
5083 <p>'<tt>type</tt>' may be any sized type.</p>
5086 <p>Memory is allocated; a pointer is returned. The operation is undefined if
5087 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
5088 memory is automatically released when the function returns. The
5089 '<tt>alloca</tt>' instruction is commonly used to represent automatic
5090 variables that must have an address available. When the function returns
5091 (either with the <tt><a href="#i_ret">ret</a></tt>
5092 or <tt><a href="#i_resume">resume</a></tt> instructions), the memory is
5093 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5094 The order in which memory is allocated (ie., which way the stack grows) is
5101 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
5102 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
5103 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
5104 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
5109 <!-- _______________________________________________________________________ -->
5111 <a name="i_load">'<tt>load</tt>' Instruction</a>
5118 <result> = load [volatile] <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>]
5119 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5120 !<index> = !{ i32 1 }
5124 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
5127 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
5128 from which to load. The pointer must point to
5129 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
5130 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
5131 number or order of execution of this <tt>load</tt> with other <a
5132 href="#volatile">volatile operations</a>.</p>
5134 <p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
5135 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5136 argument. The <code>release</code> and <code>acq_rel</code> orderings are
5137 not valid on <code>load</code> instructions. Atomic loads produce <a
5138 href="#memorymodel">defined</a> results when they may see multiple atomic
5139 stores. The type of the pointee must be an integer type whose bit width
5140 is a power of two greater than or equal to eight and less than or equal
5141 to a target-specific size limit. <code>align</code> must be explicitly
5142 specified on atomic loads, and the load has undefined behavior if the
5143 alignment is not set to a value which is at least the size in bytes of
5144 the pointee. <code>!nontemporal</code> does not have any defined semantics
5145 for atomic loads.</p>
5147 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
5148 operation (that is, the alignment of the memory address). A value of 0 or an
5149 omitted <tt>align</tt> argument means that the operation has the abi
5150 alignment for the target. It is the responsibility of the code emitter to
5151 ensure that the alignment information is correct. Overestimating the
5152 alignment results in undefined behavior. Underestimating the alignment may
5153 produce less efficient code. An alignment of 1 is always safe.</p>
5155 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
5156 metatadata name <index> corresponding to a metadata node with
5157 one <tt>i32</tt> entry of value 1. The existence of
5158 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
5159 and code generator that this load is not expected to be reused in the cache.
5160 The code generator may select special instructions to save cache bandwidth,
5161 such as the <tt>MOVNT</tt> instruction on x86.</p>
5163 <p>The optional <tt>!invariant.load</tt> metadata must reference a single
5164 metatadata name <index> corresponding to a metadata node with no
5165 entries. The existence of the <tt>!invariant.load</tt> metatadata on the
5166 instruction tells the optimizer and code generator that this load address
5167 points to memory which does not change value during program execution.
5168 The optimizer may then move this load around, for example, by hoisting it
5169 out of loops using loop invariant code motion.</p>
5172 <p>The location of memory pointed to is loaded. If the value being loaded is of
5173 scalar type then the number of bytes read does not exceed the minimum number
5174 of bytes needed to hold all bits of the type. For example, loading an
5175 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
5176 <tt>i20</tt> with a size that is not an integral number of bytes, the result
5177 is undefined if the value was not originally written using a store of the
5182 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5183 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
5184 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
5189 <!-- _______________________________________________________________________ -->
5191 <a name="i_store">'<tt>store</tt>' Instruction</a>
5198 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
5199 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> <i>; yields {void}</i>
5203 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
5206 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
5207 and an address at which to store it. The type of the
5208 '<tt><pointer></tt>' operand must be a pointer to
5209 the <a href="#t_firstclass">first class</a> type of the
5210 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
5211 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
5212 order of execution of this <tt>store</tt> with other <a
5213 href="#volatile">volatile operations</a>.</p>
5215 <p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
5216 <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5217 argument. The <code>acquire</code> and <code>acq_rel</code> orderings aren't
5218 valid on <code>store</code> instructions. Atomic loads produce <a
5219 href="#memorymodel">defined</a> results when they may see multiple atomic
5220 stores. The type of the pointee must be an integer type whose bit width
5221 is a power of two greater than or equal to eight and less than or equal
5222 to a target-specific size limit. <code>align</code> must be explicitly
5223 specified on atomic stores, and the store has undefined behavior if the
5224 alignment is not set to a value which is at least the size in bytes of
5225 the pointee. <code>!nontemporal</code> does not have any defined semantics
5226 for atomic stores.</p>
5228 <p>The optional constant "align" argument specifies the alignment of the
5229 operation (that is, the alignment of the memory address). A value of 0 or an
5230 omitted "align" argument means that the operation has the abi
5231 alignment for the target. It is the responsibility of the code emitter to
5232 ensure that the alignment information is correct. Overestimating the
5233 alignment results in an undefined behavior. Underestimating the alignment may
5234 produce less efficient code. An alignment of 1 is always safe.</p>
5236 <p>The optional !nontemporal metadata must reference a single metatadata
5237 name <index> corresponding to a metadata node with one i32 entry of
5238 value 1. The existence of the !nontemporal metatadata on the
5239 instruction tells the optimizer and code generator that this load is
5240 not expected to be reused in the cache. The code generator may
5241 select special instructions to save cache bandwidth, such as the
5242 MOVNT instruction on x86.</p>
5246 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
5247 location specified by the '<tt><pointer></tt>' operand. If
5248 '<tt><value></tt>' is of scalar type then the number of bytes written
5249 does not exceed the minimum number of bytes needed to hold all bits of the
5250 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
5251 writing a value of a type like <tt>i20</tt> with a size that is not an
5252 integral number of bytes, it is unspecified what happens to the extra bits
5253 that do not belong to the type, but they will typically be overwritten.</p>
5257 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
5258 store i32 3, i32* %ptr <i>; yields {void}</i>
5259 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
5264 <!-- _______________________________________________________________________ -->
5266 <a name="i_fence">'<tt>fence</tt>' Instruction</a>
5273 fence [singlethread] <ordering> <i>; yields {void}</i>
5277 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
5278 between operations.</p>
5280 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
5281 href="#ordering">ordering</a> argument which defines what
5282 <i>synchronizes-with</i> edges they add. They can only be given
5283 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
5284 <code>seq_cst</code> orderings.</p>
5287 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
5288 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
5289 <code>acquire</code> ordering semantics if and only if there exist atomic
5290 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
5291 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
5292 <var>X</var> modifies <var>M</var> (either directly or through some side effect
5293 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
5294 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
5295 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
5296 than an explicit <code>fence</code>, one (but not both) of the atomic operations
5297 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
5298 <code>acquire</code> (resp.) ordering constraint and still
5299 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
5300 <i>happens-before</i> edge.</p>
5302 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
5303 having both <code>acquire</code> and <code>release</code> semantics specified
5304 above, participates in the global program order of other <code>seq_cst</code>
5305 operations and/or fences.</p>
5307 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
5308 specifies that the fence only synchronizes with other fences in the same
5309 thread. (This is useful for interacting with signal handlers.)</p>
5313 fence acquire <i>; yields {void}</i>
5314 fence singlethread seq_cst <i>; yields {void}</i>
5319 <!-- _______________________________________________________________________ -->
5321 <a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
5328 cmpxchg [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
5332 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
5333 It loads a value in memory and compares it to a given value. If they are
5334 equal, it stores a new value into the memory.</p>
5337 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
5338 address to operate on, a value to compare to the value currently be at that
5339 address, and a new value to place at that address if the compared values are
5340 equal. The type of '<var><cmp></var>' must be an integer type whose
5341 bit width is a power of two greater than or equal to eight and less than
5342 or equal to a target-specific size limit. '<var><cmp></var>' and
5343 '<var><new></var>' must have the same type, and the type of
5344 '<var><pointer></var>' must be a pointer to that type. If the
5345 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
5346 optimizer is not allowed to modify the number or order of execution
5347 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
5350 <!-- FIXME: Extend allowed types. -->
5352 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
5353 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
5355 <p>The optional "<code>singlethread</code>" argument declares that the
5356 <code>cmpxchg</code> is only atomic with respect to code (usually signal
5357 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
5358 cmpxchg is atomic with respect to all other code in the system.</p>
5360 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
5361 the size in memory of the operand.
5364 <p>The contents of memory at the location specified by the
5365 '<tt><pointer></tt>' operand is read and compared to
5366 '<tt><cmp></tt>'; if the read value is the equal,
5367 '<tt><new></tt>' is written. The original value at the location
5370 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
5371 purpose of identifying <a href="#release_sequence">release sequences</a>. A
5372 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
5373 parameter determined by dropping any <code>release</code> part of the
5374 <code>cmpxchg</code>'s ordering.</p>
5377 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
5378 optimization work on ARM.)
5380 FIXME: Is a weaker ordering constraint on failure helpful in practice?
5386 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
5387 <a href="#i_br">br</a> label %loop
5390 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
5391 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
5392 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
5393 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
5394 <a href="#i_br">br</a> i1 %success, label %done, label %loop
5402 <!-- _______________________________________________________________________ -->
5404 <a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
5411 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
5415 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
5418 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
5419 operation to apply, an address whose value to modify, an argument to the
5420 operation. The operation must be one of the following keywords:</p>
5435 <p>The type of '<var><value></var>' must be an integer type whose
5436 bit width is a power of two greater than or equal to eight and less than
5437 or equal to a target-specific size limit. The type of the
5438 '<code><pointer></code>' operand must be a pointer to that type.
5439 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
5440 optimizer is not allowed to modify the number or order of execution of this
5441 <code>atomicrmw</code> with other <a href="#volatile">volatile
5444 <!-- FIXME: Extend allowed types. -->
5447 <p>The contents of memory at the location specified by the
5448 '<tt><pointer></tt>' operand are atomically read, modified, and written
5449 back. The original value at the location is returned. The modification is
5450 specified by the <var>operation</var> argument:</p>
5453 <li>xchg: <code>*ptr = val</code></li>
5454 <li>add: <code>*ptr = *ptr + val</code></li>
5455 <li>sub: <code>*ptr = *ptr - val</code></li>
5456 <li>and: <code>*ptr = *ptr & val</code></li>
5457 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
5458 <li>or: <code>*ptr = *ptr | val</code></li>
5459 <li>xor: <code>*ptr = *ptr ^ val</code></li>
5460 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
5461 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
5462 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
5463 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
5468 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
5473 <!-- _______________________________________________________________________ -->
5475 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
5482 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
5483 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
5484 <result> = getelementptr <ptr vector> ptrval, <vector index type> idx
5488 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
5489 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
5490 It performs address calculation only and does not access memory.</p>
5493 <p>The first argument is always a pointer or a vector of pointers,
5494 and forms the basis of the
5495 calculation. The remaining arguments are indices that indicate which of the
5496 elements of the aggregate object are indexed. The interpretation of each
5497 index is dependent on the type being indexed into. The first index always
5498 indexes the pointer value given as the first argument, the second index
5499 indexes a value of the type pointed to (not necessarily the value directly
5500 pointed to, since the first index can be non-zero), etc. The first type
5501 indexed into must be a pointer value, subsequent types can be arrays,
5502 vectors, and structs. Note that subsequent types being indexed into
5503 can never be pointers, since that would require loading the pointer before
5504 continuing calculation.</p>
5506 <p>The type of each index argument depends on the type it is indexing into.
5507 When indexing into a (optionally packed) structure, only <tt>i32</tt>
5508 integer <b>constants</b> are allowed (when using a vector of indices they
5509 must all be the <b>same</b> <tt>i32</tt> integer constant). When indexing
5510 into an array, pointer or vector, integers of any width are allowed, and
5511 they are not required to be constant. These integers are treated as signed
5512 values where relevant.</p>
5514 <p>For example, let's consider a C code fragment and how it gets compiled to
5517 <pre class="doc_code">
5529 int *foo(struct ST *s) {
5530 return &s[1].Z.B[5][13];
5534 <p>The LLVM code generated by Clang is:</p>
5536 <pre class="doc_code">
5537 %struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 }
5538 %struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT }
5540 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5542 %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5548 <p>In the example above, the first index is indexing into the
5549 '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a
5550 '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a
5551 structure. The second index indexes into the third element of the structure,
5552 yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>'
5553 type, another structure. The third index indexes into the second element of
5554 the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The
5555 two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>'
5556 type. The '<tt>getelementptr</tt>' instruction returns a pointer to this
5557 element, thus computing a value of '<tt>i32*</tt>' type.</p>
5559 <p>Note that it is perfectly legal to index partially through a structure,
5560 returning a pointer to an inner element. Because of this, the LLVM code for
5561 the given testcase is equivalent to:</p>
5563 <pre class="doc_code">
5564 define i32* @foo(%struct.ST* %s) {
5565 %t1 = getelementptr %struct.ST* %s, i32 1 <i>; yields %struct.ST*:%t1</i>
5566 %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2 <i>; yields %struct.RT*:%t2</i>
5567 %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
5568 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
5569 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
5574 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
5575 <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the
5576 base pointer is not an <i>in bounds</i> address of an allocated object,
5577 or if any of the addresses that would be formed by successive addition of
5578 the offsets implied by the indices to the base address with infinitely
5579 precise signed arithmetic are not an <i>in bounds</i> address of that
5580 allocated object. The <i>in bounds</i> addresses for an allocated object
5581 are all the addresses that point into the object, plus the address one
5583 In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword
5584 applies to each of the computations element-wise. </p>
5586 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5587 the base address with silently-wrapping two's complement arithmetic. If the
5588 offsets have a different width from the pointer, they are sign-extended or
5589 truncated to the width of the pointer. The result value of the
5590 <tt>getelementptr</tt> may be outside the object pointed to by the base
5591 pointer. The result value may not necessarily be used to access memory
5592 though, even if it happens to point into allocated storage. See the
5593 <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5596 <p>The getelementptr instruction is often confusing. For some more insight into
5597 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5601 <i>; yields [12 x i8]*:aptr</i>
5602 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5603 <i>; yields i8*:vptr</i>
5604 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
5605 <i>; yields i8*:eptr</i>
5606 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5607 <i>; yields i32*:iptr</i>
5608 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5611 <p>In cases where the pointer argument is a vector of pointers, each index must
5612 be a vector with the same number of elements. For example: </p>
5613 <pre class="doc_code">
5614 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5621 <!-- ======================================================================= -->
5623 <a name="convertops">Conversion Operations</a>
5628 <p>The instructions in this category are the conversion instructions (casting)
5629 which all take a single operand and a type. They perform various bit
5630 conversions on the operand.</p>
5632 <!-- _______________________________________________________________________ -->
5634 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5641 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5645 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5646 type <tt>ty2</tt>.</p>
5649 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5650 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5651 of the same number of integers.
5652 The bit size of the <tt>value</tt> must be larger than
5653 the bit size of the destination type, <tt>ty2</tt>.
5654 Equal sized types are not allowed.</p>
5657 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5658 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5659 source size must be larger than the destination size, <tt>trunc</tt> cannot
5660 be a <i>no-op cast</i>. It will always truncate bits.</p>
5664 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5665 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5666 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5667 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5672 <!-- _______________________________________________________________________ -->
5674 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5681 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5685 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5690 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5691 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5692 of the same number of integers.
5693 The bit size of the <tt>value</tt> must be smaller than
5694 the bit size of the destination type,
5698 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5699 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5701 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5705 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5706 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5707 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5712 <!-- _______________________________________________________________________ -->
5714 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5721 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5725 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5728 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5729 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5730 of the same number of integers.
5731 The bit size of the <tt>value</tt> must be smaller than
5732 the bit size of the destination type,
5736 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5737 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5738 of the type <tt>ty2</tt>.</p>
5740 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5744 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5745 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5746 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5751 <!-- _______________________________________________________________________ -->
5753 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5760 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5764 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5768 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5769 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5770 to cast it to. The size of <tt>value</tt> must be larger than the size of
5771 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5772 <i>no-op cast</i>.</p>
5775 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5776 <a href="#t_floating">floating point</a> type to a smaller
5777 <a href="#t_floating">floating point</a> type. If the value cannot fit
5778 within the destination type, <tt>ty2</tt>, then the results are
5783 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5784 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5789 <!-- _______________________________________________________________________ -->
5791 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5798 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5802 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5803 floating point value.</p>
5806 <p>The '<tt>fpext</tt>' instruction takes a
5807 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5808 a <a href="#t_floating">floating point</a> type to cast it to. The source
5809 type must be smaller than the destination type.</p>
5812 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5813 <a href="#t_floating">floating point</a> type to a larger
5814 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5815 used to make a <i>no-op cast</i> because it always changes bits. Use
5816 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5820 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5821 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5826 <!-- _______________________________________________________________________ -->
5828 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5835 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5839 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5840 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5843 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5844 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5845 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5846 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5847 vector integer type with the same number of elements as <tt>ty</tt></p>
5850 <p>The '<tt>fptoui</tt>' instruction converts its
5851 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5852 towards zero) unsigned integer value. If the value cannot fit
5853 in <tt>ty2</tt>, the results are undefined.</p>
5857 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5858 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5859 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5864 <!-- _______________________________________________________________________ -->
5866 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5873 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5877 <p>The '<tt>fptosi</tt>' instruction converts
5878 <a href="#t_floating">floating point</a> <tt>value</tt> to
5879 type <tt>ty2</tt>.</p>
5882 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5883 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5884 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5885 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5886 vector integer type with the same number of elements as <tt>ty</tt></p>
5889 <p>The '<tt>fptosi</tt>' instruction converts its
5890 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5891 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5892 the results are undefined.</p>
5896 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5897 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5898 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5903 <!-- _______________________________________________________________________ -->
5905 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5912 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5916 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5917 integer and converts that value to the <tt>ty2</tt> type.</p>
5920 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5921 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5922 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5923 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5924 floating point type with the same number of elements as <tt>ty</tt></p>
5927 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5928 integer quantity and converts it to the corresponding floating point
5929 value. If the value cannot fit in the floating point value, the results are
5934 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5935 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5940 <!-- _______________________________________________________________________ -->
5942 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5949 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5953 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5954 and converts that value to the <tt>ty2</tt> type.</p>
5957 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5958 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5959 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5960 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5961 floating point type with the same number of elements as <tt>ty</tt></p>
5964 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5965 quantity and converts it to the corresponding floating point value. If the
5966 value cannot fit in the floating point value, the results are undefined.</p>
5970 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5971 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5976 <!-- _______________________________________________________________________ -->
5978 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5985 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5989 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of
5990 pointers <tt>value</tt> to
5991 the integer (or vector of integers) type <tt>ty2</tt>.</p>
5994 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5995 must be a a value of type <a href="#t_pointer">pointer</a> or a vector of
5996 pointers, and a type to cast it to
5997 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector
5998 of integers type.</p>
6001 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
6002 <tt>ty2</tt> by interpreting the pointer value as an integer and either
6003 truncating or zero extending that value to the size of the integer type. If
6004 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
6005 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
6006 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
6011 %X = ptrtoint i32* %P to i8 <i>; yields truncation on 32-bit architecture</i>
6012 %Y = ptrtoint i32* %P to i64 <i>; yields zero extension on 32-bit architecture</i>
6013 %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>
6018 <!-- _______________________________________________________________________ -->
6020 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
6027 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
6031 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
6032 pointer type, <tt>ty2</tt>.</p>
6035 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
6036 value to cast, and a type to cast it to, which must be a
6037 <a href="#t_pointer">pointer</a> type.</p>
6040 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
6041 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
6042 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
6043 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
6044 than the size of a pointer then a zero extension is done. If they are the
6045 same size, nothing is done (<i>no-op cast</i>).</p>
6049 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
6050 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
6051 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
6052 %Z = inttoptr <4 x i32> %G to <4 x i8*><i>; yields truncation of vector G to four pointers</i>
6057 <!-- _______________________________________________________________________ -->
6059 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
6066 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
6070 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
6071 <tt>ty2</tt> without changing any bits.</p>
6074 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
6075 non-aggregate first class value, and a type to cast it to, which must also be
6076 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
6077 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
6078 identical. If the source type is a pointer, the destination type must also be
6079 a pointer. This instruction supports bitwise conversion of vectors to
6080 integers and to vectors of other types (as long as they have the same
6084 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
6085 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
6086 this conversion. The conversion is done as if the <tt>value</tt> had been
6087 stored to memory and read back as type <tt>ty2</tt>.
6088 Pointer (or vector of pointers) types may only be converted to other pointer
6089 (or vector of pointers) types with this instruction. To convert
6090 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
6091 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
6095 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
6096 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
6097 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
6098 %Z = bitcast <2 x i32*> %V to <2 x i64*> <i>; yields <2 x i64*></i>
6105 <!-- ======================================================================= -->
6107 <a name="otherops">Other Operations</a>
6112 <p>The instructions in this category are the "miscellaneous" instructions, which
6113 defy better classification.</p>
6115 <!-- _______________________________________________________________________ -->
6117 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
6124 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
6128 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
6129 boolean values based on comparison of its two integer, integer vector,
6130 pointer, or pointer vector operands.</p>
6133 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
6134 the condition code indicating the kind of comparison to perform. It is not a
6135 value, just a keyword. The possible condition code are:</p>
6138 <li><tt>eq</tt>: equal</li>
6139 <li><tt>ne</tt>: not equal </li>
6140 <li><tt>ugt</tt>: unsigned greater than</li>
6141 <li><tt>uge</tt>: unsigned greater or equal</li>
6142 <li><tt>ult</tt>: unsigned less than</li>
6143 <li><tt>ule</tt>: unsigned less or equal</li>
6144 <li><tt>sgt</tt>: signed greater than</li>
6145 <li><tt>sge</tt>: signed greater or equal</li>
6146 <li><tt>slt</tt>: signed less than</li>
6147 <li><tt>sle</tt>: signed less or equal</li>
6150 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
6151 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
6152 typed. They must also be identical types.</p>
6155 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
6156 condition code given as <tt>cond</tt>. The comparison performed always yields
6157 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
6158 result, as follows:</p>
6161 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
6162 <tt>false</tt> otherwise. No sign interpretation is necessary or
6165 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
6166 <tt>false</tt> otherwise. No sign interpretation is necessary or
6169 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
6170 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6172 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
6173 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6174 to <tt>op2</tt>.</li>
6176 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
6177 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6179 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
6180 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6182 <li><tt>sgt</tt>: interprets the operands as signed values and yields
6183 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6185 <li><tt>sge</tt>: interprets the operands as signed values and yields
6186 <tt>true</tt> if <tt>op1</tt> is greater than or equal
6187 to <tt>op2</tt>.</li>
6189 <li><tt>slt</tt>: interprets the operands as signed values and yields
6190 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6192 <li><tt>sle</tt>: interprets the operands as signed values and yields
6193 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6196 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
6197 values are compared as if they were integers.</p>
6199 <p>If the operands are integer vectors, then they are compared element by
6200 element. The result is an <tt>i1</tt> vector with the same number of elements
6201 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
6205 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
6206 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
6207 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
6208 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
6209 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
6210 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
6213 <p>Note that the code generator does not yet support vector types with
6214 the <tt>icmp</tt> instruction.</p>
6218 <!-- _______________________________________________________________________ -->
6220 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
6227 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
6231 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
6232 values based on comparison of its operands.</p>
6234 <p>If the operands are floating point scalars, then the result type is a boolean
6235 (<a href="#t_integer"><tt>i1</tt></a>).</p>
6237 <p>If the operands are floating point vectors, then the result type is a vector
6238 of boolean with the same number of elements as the operands being
6242 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
6243 the condition code indicating the kind of comparison to perform. It is not a
6244 value, just a keyword. The possible condition code are:</p>
6247 <li><tt>false</tt>: no comparison, always returns false</li>
6248 <li><tt>oeq</tt>: ordered and equal</li>
6249 <li><tt>ogt</tt>: ordered and greater than </li>
6250 <li><tt>oge</tt>: ordered and greater than or equal</li>
6251 <li><tt>olt</tt>: ordered and less than </li>
6252 <li><tt>ole</tt>: ordered and less than or equal</li>
6253 <li><tt>one</tt>: ordered and not equal</li>
6254 <li><tt>ord</tt>: ordered (no nans)</li>
6255 <li><tt>ueq</tt>: unordered or equal</li>
6256 <li><tt>ugt</tt>: unordered or greater than </li>
6257 <li><tt>uge</tt>: unordered or greater than or equal</li>
6258 <li><tt>ult</tt>: unordered or less than </li>
6259 <li><tt>ule</tt>: unordered or less than or equal</li>
6260 <li><tt>une</tt>: unordered or not equal</li>
6261 <li><tt>uno</tt>: unordered (either nans)</li>
6262 <li><tt>true</tt>: no comparison, always returns true</li>
6265 <p><i>Ordered</i> means that neither operand is a QNAN while
6266 <i>unordered</i> means that either operand may be a QNAN.</p>
6268 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
6269 a <a href="#t_floating">floating point</a> type or
6270 a <a href="#t_vector">vector</a> of floating point type. They must have
6271 identical types.</p>
6274 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
6275 according to the condition code given as <tt>cond</tt>. If the operands are
6276 vectors, then the vectors are compared element by element. Each comparison
6277 performed always yields an <a href="#t_integer">i1</a> result, as
6281 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
6283 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6284 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6286 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6287 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6289 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6290 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6292 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6293 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6295 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6296 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6298 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6299 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6301 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
6303 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
6304 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6306 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
6307 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6309 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
6310 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6312 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
6313 <tt>op1</tt> is less than <tt>op2</tt>.</li>
6315 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
6316 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6318 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
6319 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6321 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
6323 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
6328 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
6329 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
6330 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
6331 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
6334 <p>Note that the code generator does not yet support vector types with
6335 the <tt>fcmp</tt> instruction.</p>
6339 <!-- _______________________________________________________________________ -->
6341 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
6348 <result> = phi <ty> [ <val0>, <label0>], ...
6352 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
6353 SSA graph representing the function.</p>
6356 <p>The type of the incoming values is specified with the first type field. After
6357 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
6358 one pair for each predecessor basic block of the current block. Only values
6359 of <a href="#t_firstclass">first class</a> type may be used as the value
6360 arguments to the PHI node. Only labels may be used as the label
6363 <p>There must be no non-phi instructions between the start of a basic block and
6364 the PHI instructions: i.e. PHI instructions must be first in a basic
6367 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
6368 occur on the edge from the corresponding predecessor block to the current
6369 block (but after any definition of an '<tt>invoke</tt>' instruction's return
6370 value on the same edge).</p>
6373 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
6374 specified by the pair corresponding to the predecessor basic block that
6375 executed just prior to the current block.</p>
6379 Loop: ; Infinite loop that counts from 0 on up...
6380 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6381 %nextindvar = add i32 %indvar, 1
6387 <!-- _______________________________________________________________________ -->
6389 <a name="i_select">'<tt>select</tt>' Instruction</a>
6396 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
6398 <i>selty</i> is either i1 or {<N x i1>}
6402 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
6403 condition, without branching.</p>
6407 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
6408 values indicating the condition, and two values of the
6409 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
6410 vectors and the condition is a scalar, then entire vectors are selected, not
6411 individual elements.</p>
6414 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
6415 first value argument; otherwise, it returns the second value argument.</p>
6417 <p>If the condition is a vector of i1, then the value arguments must be vectors
6418 of the same size, and the selection is done element by element.</p>
6422 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
6427 <!-- _______________________________________________________________________ -->
6429 <a name="i_call">'<tt>call</tt>' Instruction</a>
6436 <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>]
6440 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
6443 <p>This instruction requires several arguments:</p>
6446 <li>The optional "tail" marker indicates that the callee function does not
6447 access any allocas or varargs in the caller. Note that calls may be
6448 marked "tail" even if they do not occur before
6449 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
6450 present, the function call is eligible for tail call optimization,
6451 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
6452 optimized into a jump</a>. The code generator may optimize calls marked
6453 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
6454 sibling call optimization</a> when the caller and callee have
6455 matching signatures, or 2) forced tail call optimization when the
6456 following extra requirements are met:
6458 <li>Caller and callee both have the calling
6459 convention <tt>fastcc</tt>.</li>
6460 <li>The call is in tail position (ret immediately follows call and ret
6461 uses value of call or is void).</li>
6462 <li>Option <tt>-tailcallopt</tt> is enabled,
6463 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
6464 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
6465 constraints are met.</a></li>
6469 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
6470 convention</a> the call should use. If none is specified, the call
6471 defaults to using C calling conventions. The calling convention of the
6472 call must match the calling convention of the target function, or else the
6473 behavior is undefined.</li>
6475 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
6476 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
6477 '<tt>inreg</tt>' attributes are valid here.</li>
6479 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
6480 type of the return value. Functions that return no value are marked
6481 <tt><a href="#t_void">void</a></tt>.</li>
6483 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
6484 being invoked. The argument types must match the types implied by this
6485 signature. This type can be omitted if the function is not varargs and if
6486 the function type does not return a pointer to a function.</li>
6488 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
6489 be invoked. In most cases, this is a direct function invocation, but
6490 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
6491 to function value.</li>
6493 <li>'<tt>function args</tt>': argument list whose types match the function
6494 signature argument types and parameter attributes. All arguments must be
6495 of <a href="#t_firstclass">first class</a> type. If the function
6496 signature indicates the function accepts a variable number of arguments,
6497 the extra arguments can be specified.</li>
6499 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
6500 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
6501 '<tt>readnone</tt>' attributes are valid here.</li>
6505 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
6506 a specified function, with its incoming arguments bound to the specified
6507 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
6508 function, control flow continues with the instruction after the function
6509 call, and the return value of the function is bound to the result
6514 %retval = call i32 @test(i32 %argc)
6515 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
6516 %X = tail call i32 @foo() <i>; yields i32</i>
6517 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
6518 call void %foo(i8 97 signext)
6520 %struct.A = type { i32, i8 }
6521 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
6522 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
6523 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
6524 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
6525 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
6528 <p>llvm treats calls to some functions with names and arguments that match the
6529 standard C99 library as being the C99 library functions, and may perform
6530 optimizations or generate code for them under that assumption. This is
6531 something we'd like to change in the future to provide better support for
6532 freestanding environments and non-C-based languages.</p>
6536 <!-- _______________________________________________________________________ -->
6538 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
6545 <resultval> = va_arg <va_list*> <arglist>, <argty>
6549 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6550 the "variable argument" area of a function call. It is used to implement the
6551 <tt>va_arg</tt> macro in C.</p>
6554 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
6555 argument. It returns a value of the specified argument type and increments
6556 the <tt>va_list</tt> to point to the next argument. The actual type
6557 of <tt>va_list</tt> is target specific.</p>
6560 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6561 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6562 to the next argument. For more information, see the variable argument
6563 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6565 <p>It is legal for this instruction to be called in a function which does not
6566 take a variable number of arguments, for example, the <tt>vfprintf</tt>
6569 <p><tt>va_arg</tt> is an LLVM instruction instead of
6570 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6574 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6576 <p>Note that the code generator does not yet fully support va_arg on many
6577 targets. Also, it does not currently support va_arg with aggregate types on
6582 <!-- _______________________________________________________________________ -->
6584 <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6591 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
6592 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
6594 <clause> := catch <type> <value>
6595 <clause> := filter <array constant type> <array constant>
6599 <p>The '<tt>landingpad</tt>' instruction is used by
6600 <a href="ExceptionHandling.html#overview">LLVM's exception handling
6601 system</a> to specify that a basic block is a landing pad — one where
6602 the exception lands, and corresponds to the code found in the
6603 <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6604 defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6605 re-entry to the function. The <tt>resultval</tt> has the
6606 type <tt>resultty</tt>.</p>
6609 <p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6610 function associated with the unwinding mechanism. The optional
6611 <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6613 <p>A <tt>clause</tt> begins with the clause type — <tt>catch</tt>
6614 or <tt>filter</tt> — and contains the global variable representing the
6615 "type" that may be caught or filtered respectively. Unlike the
6616 <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6617 its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6618 throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6619 one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6622 <p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6623 personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6624 therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6625 calling conventions, how the personality function results are represented in
6626 LLVM IR is target specific.</p>
6628 <p>The clauses are applied in order from top to bottom. If two
6629 <tt>landingpad</tt> instructions are merged together through inlining, the
6630 clauses from the calling function are appended to the list of clauses.
6631 When the call stack is being unwound due to an exception being thrown, the
6632 exception is compared against each <tt>clause</tt> in turn. If it doesn't
6633 match any of the clauses, and the <tt>cleanup</tt> flag is not set, then
6634 unwinding continues further up the call stack.</p>
6636 <p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6639 <li>A landing pad block is a basic block which is the unwind destination of an
6640 '<tt>invoke</tt>' instruction.</li>
6641 <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6642 first non-PHI instruction.</li>
6643 <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6645 <li>A basic block that is not a landing pad block may not include a
6646 '<tt>landingpad</tt>' instruction.</li>
6647 <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6648 personality function.</li>
6653 ;; A landing pad which can catch an integer.
6654 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6656 ;; A landing pad that is a cleanup.
6657 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6659 ;; A landing pad which can catch an integer and can only throw a double.
6660 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6662 filter [1 x i8**] [@_ZTId]
6671 <!-- *********************************************************************** -->
6672 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
6673 <!-- *********************************************************************** -->
6677 <p>LLVM supports the notion of an "intrinsic function". These functions have
6678 well known names and semantics and are required to follow certain
6679 restrictions. Overall, these intrinsics represent an extension mechanism for
6680 the LLVM language that does not require changing all of the transformations
6681 in LLVM when adding to the language (or the bitcode reader/writer, the
6682 parser, etc...).</p>
6684 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6685 prefix is reserved in LLVM for intrinsic names; thus, function names may not
6686 begin with this prefix. Intrinsic functions must always be external
6687 functions: you cannot define the body of intrinsic functions. Intrinsic
6688 functions may only be used in call or invoke instructions: it is illegal to
6689 take the address of an intrinsic function. Additionally, because intrinsic
6690 functions are part of the LLVM language, it is required if any are added that
6691 they be documented here.</p>
6693 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6694 family of functions that perform the same operation but on different data
6695 types. Because LLVM can represent over 8 million different integer types,
6696 overloading is used commonly to allow an intrinsic function to operate on any
6697 integer type. One or more of the argument types or the result type can be
6698 overloaded to accept any integer type. Argument types may also be defined as
6699 exactly matching a previous argument's type or the result type. This allows
6700 an intrinsic function which accepts multiple arguments, but needs all of them
6701 to be of the same type, to only be overloaded with respect to a single
6702 argument or the result.</p>
6704 <p>Overloaded intrinsics will have the names of its overloaded argument types
6705 encoded into its function name, each preceded by a period. Only those types
6706 which are overloaded result in a name suffix. Arguments whose type is matched
6707 against another type do not. For example, the <tt>llvm.ctpop</tt> function
6708 can take an integer of any width and returns an integer of exactly the same
6709 integer width. This leads to a family of functions such as
6710 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6711 %val)</tt>. Only one type, the return type, is overloaded, and only one type
6712 suffix is required. Because the argument's type is matched against the return
6713 type, it does not require its own name suffix.</p>
6715 <p>To learn how to add an intrinsic function, please see the
6716 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6718 <!-- ======================================================================= -->
6720 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6725 <p>Variable argument support is defined in LLVM with
6726 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6727 intrinsic functions. These functions are related to the similarly named
6728 macros defined in the <tt><stdarg.h></tt> header file.</p>
6730 <p>All of these functions operate on arguments that use a target-specific value
6731 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6732 not define what this type is, so all transformations should be prepared to
6733 handle these functions regardless of the type used.</p>
6735 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6736 instruction and the variable argument handling intrinsic functions are
6739 <pre class="doc_code">
6740 define i32 @test(i32 %X, ...) {
6741 ; Initialize variable argument processing
6743 %ap2 = bitcast i8** %ap to i8*
6744 call void @llvm.va_start(i8* %ap2)
6746 ; Read a single integer argument
6747 %tmp = va_arg i8** %ap, i32
6749 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6751 %aq2 = bitcast i8** %aq to i8*
6752 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6753 call void @llvm.va_end(i8* %aq2)
6755 ; Stop processing of arguments.
6756 call void @llvm.va_end(i8* %ap2)
6760 declare void @llvm.va_start(i8*)
6761 declare void @llvm.va_copy(i8*, i8*)
6762 declare void @llvm.va_end(i8*)
6765 <!-- _______________________________________________________________________ -->
6767 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6775 declare void %llvm.va_start(i8* <arglist>)
6779 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6780 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6783 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6786 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6787 macro available in C. In a target-dependent way, it initializes
6788 the <tt>va_list</tt> element to which the argument points, so that the next
6789 call to <tt>va_arg</tt> will produce the first variable argument passed to
6790 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6791 need to know the last argument of the function as the compiler can figure
6796 <!-- _______________________________________________________________________ -->
6798 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6805 declare void @llvm.va_end(i8* <arglist>)
6809 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6810 which has been initialized previously
6811 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6812 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6815 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6818 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6819 macro available in C. In a target-dependent way, it destroys
6820 the <tt>va_list</tt> element to which the argument points. Calls
6821 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6822 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6823 with calls to <tt>llvm.va_end</tt>.</p>
6827 <!-- _______________________________________________________________________ -->
6829 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6836 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6840 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6841 from the source argument list to the destination argument list.</p>
6844 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6845 The second argument is a pointer to a <tt>va_list</tt> element to copy
6849 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6850 macro available in C. In a target-dependent way, it copies the
6851 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6852 element. This intrinsic is necessary because
6853 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6854 arbitrarily complex and require, for example, memory allocation.</p>
6860 <!-- ======================================================================= -->
6862 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6867 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6868 Collection</a> (GC) requires the implementation and generation of these
6869 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6870 roots on the stack</a>, as well as garbage collector implementations that
6871 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6872 barriers. Front-ends for type-safe garbage collected languages should generate
6873 these intrinsics to make use of the LLVM garbage collectors. For more details,
6874 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6877 <p>The garbage collection intrinsics only operate on objects in the generic
6878 address space (address space zero).</p>
6880 <!-- _______________________________________________________________________ -->
6882 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6889 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6893 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6894 the code generator, and allows some metadata to be associated with it.</p>
6897 <p>The first argument specifies the address of a stack object that contains the
6898 root pointer. The second pointer (which must be either a constant or a
6899 global value address) contains the meta-data to be associated with the
6903 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6904 location. At compile-time, the code generator generates information to allow
6905 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6906 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6911 <!-- _______________________________________________________________________ -->
6913 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6920 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6924 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6925 locations, allowing garbage collector implementations that require read
6929 <p>The second argument is the address to read from, which should be an address
6930 allocated from the garbage collector. The first object is a pointer to the
6931 start of the referenced object, if needed by the language runtime (otherwise
6935 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6936 instruction, but may be replaced with substantially more complex code by the
6937 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6938 may only be used in a function which <a href="#gc">specifies a GC
6943 <!-- _______________________________________________________________________ -->
6945 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6952 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6956 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6957 locations, allowing garbage collector implementations that require write
6958 barriers (such as generational or reference counting collectors).</p>
6961 <p>The first argument is the reference to store, the second is the start of the
6962 object to store it to, and the third is the address of the field of Obj to
6963 store to. If the runtime does not require a pointer to the object, Obj may
6967 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6968 instruction, but may be replaced with substantially more complex code by the
6969 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6970 may only be used in a function which <a href="#gc">specifies a GC
6977 <!-- ======================================================================= -->
6979 <a name="int_codegen">Code Generator Intrinsics</a>
6984 <p>These intrinsics are provided by LLVM to expose special features that may
6985 only be implemented with code generator support.</p>
6987 <!-- _______________________________________________________________________ -->
6989 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6996 declare i8 *@llvm.returnaddress(i32 <level>)
7000 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
7001 target-specific value indicating the return address of the current function
7002 or one of its callers.</p>
7005 <p>The argument to this intrinsic indicates which function to return the address
7006 for. Zero indicates the calling function, one indicates its caller, etc.
7007 The argument is <b>required</b> to be a constant integer value.</p>
7010 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
7011 indicating the return address of the specified call frame, or zero if it
7012 cannot be identified. The value returned by this intrinsic is likely to be
7013 incorrect or 0 for arguments other than zero, so it should only be used for
7014 debugging purposes.</p>
7016 <p>Note that calling this intrinsic does not prevent function inlining or other
7017 aggressive transformations, so the value returned may not be that of the
7018 obvious source-language caller.</p>
7022 <!-- _______________________________________________________________________ -->
7024 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
7031 declare i8* @llvm.frameaddress(i32 <level>)
7035 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
7036 target-specific frame pointer value for the specified stack frame.</p>
7039 <p>The argument to this intrinsic indicates which function to return the frame
7040 pointer for. Zero indicates the calling function, one indicates its caller,
7041 etc. The argument is <b>required</b> to be a constant integer value.</p>
7044 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
7045 indicating the frame address of the specified call frame, or zero if it
7046 cannot be identified. The value returned by this intrinsic is likely to be
7047 incorrect or 0 for arguments other than zero, so it should only be used for
7048 debugging purposes.</p>
7050 <p>Note that calling this intrinsic does not prevent function inlining or other
7051 aggressive transformations, so the value returned may not be that of the
7052 obvious source-language caller.</p>
7056 <!-- _______________________________________________________________________ -->
7058 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
7065 declare i8* @llvm.stacksave()
7069 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
7070 of the function stack, for use
7071 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
7072 useful for implementing language features like scoped automatic variable
7073 sized arrays in C99.</p>
7076 <p>This intrinsic returns a opaque pointer value that can be passed
7077 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
7078 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
7079 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
7080 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
7081 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
7082 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
7086 <!-- _______________________________________________________________________ -->
7088 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
7095 declare void @llvm.stackrestore(i8* %ptr)
7099 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
7100 the function stack to the state it was in when the
7101 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
7102 executed. This is useful for implementing language features like scoped
7103 automatic variable sized arrays in C99.</p>
7106 <p>See the description
7107 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
7111 <!-- _______________________________________________________________________ -->
7113 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
7120 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7124 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
7125 insert a prefetch instruction if supported; otherwise, it is a noop.
7126 Prefetches have no effect on the behavior of the program but can change its
7127 performance characteristics.</p>
7130 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
7131 specifier determining if the fetch should be for a read (0) or write (1),
7132 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
7133 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
7134 specifies whether the prefetch is performed on the data (1) or instruction (0)
7135 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
7136 must be constant integers.</p>
7139 <p>This intrinsic does not modify the behavior of the program. In particular,
7140 prefetches cannot trap and do not produce a value. On targets that support
7141 this intrinsic, the prefetch can provide hints to the processor cache for
7142 better performance.</p>
7146 <!-- _______________________________________________________________________ -->
7148 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
7155 declare void @llvm.pcmarker(i32 <id>)
7159 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
7160 Counter (PC) in a region of code to simulators and other tools. The method
7161 is target specific, but it is expected that the marker will use exported
7162 symbols to transmit the PC of the marker. The marker makes no guarantees
7163 that it will remain with any specific instruction after optimizations. It is
7164 possible that the presence of a marker will inhibit optimizations. The
7165 intended use is to be inserted after optimizations to allow correlations of
7166 simulation runs.</p>
7169 <p><tt>id</tt> is a numerical id identifying the marker.</p>
7172 <p>This intrinsic does not modify the behavior of the program. Backends that do
7173 not support this intrinsic may ignore it.</p>
7177 <!-- _______________________________________________________________________ -->
7179 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
7186 declare i64 @llvm.readcyclecounter()
7190 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
7191 counter register (or similar low latency, high accuracy clocks) on those
7192 targets that support it. On X86, it should map to RDTSC. On Alpha, it
7193 should map to RPCC. As the backing counters overflow quickly (on the order
7194 of 9 seconds on alpha), this should only be used for small timings.</p>
7197 <p>When directly supported, reading the cycle counter should not modify any
7198 memory. Implementations are allowed to either return a application specific
7199 value or a system wide value. On backends without support, this is lowered
7200 to a constant 0.</p>
7206 <!-- ======================================================================= -->
7208 <a name="int_libc">Standard C Library Intrinsics</a>
7213 <p>LLVM provides intrinsics for a few important standard C library functions.
7214 These intrinsics allow source-language front-ends to pass information about
7215 the alignment of the pointer arguments to the code generator, providing
7216 opportunity for more efficient code generation.</p>
7218 <!-- _______________________________________________________________________ -->
7220 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
7226 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
7227 integer bit width and for different address spaces. Not all targets support
7228 all bit widths however.</p>
7231 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7232 i32 <len>, i32 <align>, i1 <isvolatile>)
7233 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7234 i64 <len>, i32 <align>, i1 <isvolatile>)
7238 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7239 source location to the destination location.</p>
7241 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
7242 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7243 and the pointers can be in specified address spaces.</p>
7247 <p>The first argument is a pointer to the destination, the second is a pointer
7248 to the source. The third argument is an integer argument specifying the
7249 number of bytes to copy, the fourth argument is the alignment of the
7250 source and destination locations, and the fifth is a boolean indicating a
7251 volatile access.</p>
7253 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7254 then the caller guarantees that both the source and destination pointers are
7255 aligned to that boundary.</p>
7257 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7258 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
7259 The detailed access behavior is not very cleanly specified and it is unwise
7260 to depend on it.</p>
7264 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7265 source location to the destination location, which are not allowed to
7266 overlap. It copies "len" bytes of memory over. If the argument is known to
7267 be aligned to some boundary, this can be specified as the fourth argument,
7268 otherwise it should be set to 0 or 1.</p>
7272 <!-- _______________________________________________________________________ -->
7274 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
7280 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
7281 width and for different address space. Not all targets support all bit
7285 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
7286 i32 <len>, i32 <align>, i1 <isvolatile>)
7287 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
7288 i64 <len>, i32 <align>, i1 <isvolatile>)
7292 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
7293 source location to the destination location. It is similar to the
7294 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
7297 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
7298 intrinsics do not return a value, takes extra alignment/isvolatile arguments
7299 and the pointers can be in specified address spaces.</p>
7303 <p>The first argument is a pointer to the destination, the second is a pointer
7304 to the source. The third argument is an integer argument specifying the
7305 number of bytes to copy, the fourth argument is the alignment of the
7306 source and destination locations, and the fifth is a boolean indicating a
7307 volatile access.</p>
7309 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7310 then the caller guarantees that the source and destination pointers are
7311 aligned to that boundary.</p>
7313 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7314 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
7315 The detailed access behavior is not very cleanly specified and it is unwise
7316 to depend on it.</p>
7320 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
7321 source location to the destination location, which may overlap. It copies
7322 "len" bytes of memory over. If the argument is known to be aligned to some
7323 boundary, this can be specified as the fourth argument, otherwise it should
7324 be set to 0 or 1.</p>
7328 <!-- _______________________________________________________________________ -->
7330 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
7336 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
7337 width and for different address spaces. However, not all targets support all
7341 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
7342 i32 <len>, i32 <align>, i1 <isvolatile>)
7343 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
7344 i64 <len>, i32 <align>, i1 <isvolatile>)
7348 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
7349 particular byte value.</p>
7351 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
7352 intrinsic does not return a value and takes extra alignment/volatile
7353 arguments. Also, the destination can be in an arbitrary address space.</p>
7356 <p>The first argument is a pointer to the destination to fill, the second is the
7357 byte value with which to fill it, the third argument is an integer argument
7358 specifying the number of bytes to fill, and the fourth argument is the known
7359 alignment of the destination location.</p>
7361 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7362 then the caller guarantees that the destination pointer is aligned to that
7365 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7366 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
7367 The detailed access behavior is not very cleanly specified and it is unwise
7368 to depend on it.</p>
7371 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
7372 at the destination location. If the argument is known to be aligned to some
7373 boundary, this can be specified as the fourth argument, otherwise it should
7374 be set to 0 or 1.</p>
7378 <!-- _______________________________________________________________________ -->
7380 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
7386 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
7387 floating point or vector of floating point type. Not all targets support all
7391 declare float @llvm.sqrt.f32(float %Val)
7392 declare double @llvm.sqrt.f64(double %Val)
7393 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
7394 declare fp128 @llvm.sqrt.f128(fp128 %Val)
7395 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7399 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
7400 returning the same value as the libm '<tt>sqrt</tt>' functions would.
7401 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
7402 behavior for negative numbers other than -0.0 (which allows for better
7403 optimization, because there is no need to worry about errno being
7404 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
7407 <p>The argument and return value are floating point numbers of the same
7411 <p>This function returns the sqrt of the specified operand if it is a
7412 nonnegative floating point number.</p>
7416 <!-- _______________________________________________________________________ -->
7418 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
7424 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
7425 floating point or vector of floating point type. Not all targets support all
7429 declare float @llvm.powi.f32(float %Val, i32 %power)
7430 declare double @llvm.powi.f64(double %Val, i32 %power)
7431 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
7432 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
7433 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
7437 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
7438 specified (positive or negative) power. The order of evaluation of
7439 multiplications is not defined. When a vector of floating point type is
7440 used, the second argument remains a scalar integer value.</p>
7443 <p>The second argument is an integer power, and the first is a value to raise to
7447 <p>This function returns the first value raised to the second power with an
7448 unspecified sequence of rounding operations.</p>
7452 <!-- _______________________________________________________________________ -->
7454 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
7460 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
7461 floating point or vector of floating point type. Not all targets support all
7465 declare float @llvm.sin.f32(float %Val)
7466 declare double @llvm.sin.f64(double %Val)
7467 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
7468 declare fp128 @llvm.sin.f128(fp128 %Val)
7469 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
7473 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
7476 <p>The argument and return value are floating point numbers of the same
7480 <p>This function returns the sine of the specified operand, returning the same
7481 values as the libm <tt>sin</tt> functions would, and handles error conditions
7482 in the same way.</p>
7486 <!-- _______________________________________________________________________ -->
7488 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
7494 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
7495 floating point or vector of floating point type. Not all targets support all
7499 declare float @llvm.cos.f32(float %Val)
7500 declare double @llvm.cos.f64(double %Val)
7501 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
7502 declare fp128 @llvm.cos.f128(fp128 %Val)
7503 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
7507 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
7510 <p>The argument and return value are floating point numbers of the same
7514 <p>This function returns the cosine of the specified operand, returning the same
7515 values as the libm <tt>cos</tt> functions would, and handles error conditions
7516 in the same way.</p>
7520 <!-- _______________________________________________________________________ -->
7522 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
7528 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
7529 floating point or vector of floating point type. Not all targets support all
7533 declare float @llvm.pow.f32(float %Val, float %Power)
7534 declare double @llvm.pow.f64(double %Val, double %Power)
7535 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
7536 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
7537 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
7541 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
7542 specified (positive or negative) power.</p>
7545 <p>The second argument is a floating point power, and the first is a value to
7546 raise to that power.</p>
7549 <p>This function returns the first value raised to the second power, returning
7550 the same values as the libm <tt>pow</tt> functions would, and handles error
7551 conditions in the same way.</p>
7555 <!-- _______________________________________________________________________ -->
7557 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7563 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7564 floating point or vector of floating point type. Not all targets support all
7568 declare float @llvm.exp.f32(float %Val)
7569 declare double @llvm.exp.f64(double %Val)
7570 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
7571 declare fp128 @llvm.exp.f128(fp128 %Val)
7572 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
7576 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
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>exp</tt> functions
7584 would, and handles error conditions in the same way.</p>
7588 <!-- _______________________________________________________________________ -->
7590 <a name="int_exp2">'<tt>llvm.exp2.*</tt>' Intrinsic</a>
7596 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp2</tt> on any
7597 floating point or vector of floating point type. Not all targets support all
7601 declare float @llvm.exp2.f32(float %Val)
7602 declare double @llvm.exp2.f64(double %Val)
7603 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
7604 declare fp128 @llvm.exp2.f128(fp128 %Val)
7605 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
7609 <p>The '<tt>llvm.exp2.*</tt>' intrinsics perform the exp2 function.</p>
7612 <p>The argument and return value are floating point numbers of the same
7616 <p>This function returns the same values as the libm <tt>exp2</tt> functions
7617 would, and handles error conditions in the same way.</p>
7621 <!-- _______________________________________________________________________ -->
7623 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7629 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7630 floating point or vector of floating point type. Not all targets support all
7634 declare float @llvm.log.f32(float %Val)
7635 declare double @llvm.log.f64(double %Val)
7636 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
7637 declare fp128 @llvm.log.f128(fp128 %Val)
7638 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
7642 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7645 <p>The argument and return value are floating point numbers of the same
7649 <p>This function returns the same values as the libm <tt>log</tt> functions
7650 would, and handles error conditions in the same way.</p>
7654 <!-- _______________________________________________________________________ -->
7656 <a name="int_log10">'<tt>llvm.log10.*</tt>' Intrinsic</a>
7662 <p>This is an overloaded intrinsic. You can use <tt>llvm.log10</tt> on any
7663 floating point or vector of floating point type. Not all targets support all
7667 declare float @llvm.log10.f32(float %Val)
7668 declare double @llvm.log10.f64(double %Val)
7669 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
7670 declare fp128 @llvm.log10.f128(fp128 %Val)
7671 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
7675 <p>The '<tt>llvm.log10.*</tt>' intrinsics perform the log10 function.</p>
7678 <p>The argument and return value are floating point numbers of the same
7682 <p>This function returns the same values as the libm <tt>log10</tt> functions
7683 would, and handles error conditions in the same way.</p>
7687 <!-- _______________________________________________________________________ -->
7689 <a name="int_log2">'<tt>llvm.log2.*</tt>' Intrinsic</a>
7695 <p>This is an overloaded intrinsic. You can use <tt>llvm.log2</tt> on any
7696 floating point or vector of floating point type. Not all targets support all
7700 declare float @llvm.log2.f32(float %Val)
7701 declare double @llvm.log2.f64(double %Val)
7702 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
7703 declare fp128 @llvm.log2.f128(fp128 %Val)
7704 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
7708 <p>The '<tt>llvm.log2.*</tt>' intrinsics perform the log2 function.</p>
7711 <p>The argument and return value are floating point numbers of the same
7715 <p>This function returns the same values as the libm <tt>log2</tt> functions
7716 would, and handles error conditions in the same way.</p>
7720 <!-- _______________________________________________________________________ -->
7722 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7728 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7729 floating point or vector of floating point type. Not all targets support all
7733 declare float @llvm.fma.f32(float %a, float %b, float %c)
7734 declare double @llvm.fma.f64(double %a, double %b, double %c)
7735 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7736 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7737 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7741 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7745 <p>The argument and return value are floating point numbers of the same
7749 <p>This function returns the same values as the libm <tt>fma</tt> functions
7754 <!-- _______________________________________________________________________ -->
7756 <a name="int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a>
7762 <p>This is an overloaded intrinsic. You can use <tt>llvm.fabs</tt> on any
7763 floating point or vector of floating point type. Not all targets support all
7767 declare float @llvm.fabs.f32(float %Val)
7768 declare double @llvm.fabs.f64(double %Val)
7769 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
7770 declare fp128 @llvm.fabs.f128(fp128 %Val)
7771 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
7775 <p>The '<tt>llvm.fabs.*</tt>' intrinsics return the absolute value of
7779 <p>The argument and return value are floating point numbers of the same
7783 <p>This function returns the same values as the libm <tt>fabs</tt> functions
7784 would, and handles error conditions in the same way.</p>
7788 <!-- _______________________________________________________________________ -->
7790 <a name="int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a>
7796 <p>This is an overloaded intrinsic. You can use <tt>llvm.floor</tt> on any
7797 floating point or vector of floating point type. Not all targets support all
7801 declare float @llvm.floor.f32(float %Val)
7802 declare double @llvm.floor.f64(double %Val)
7803 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
7804 declare fp128 @llvm.floor.f128(fp128 %Val)
7805 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
7809 <p>The '<tt>llvm.floor.*</tt>' intrinsics return the floor of
7813 <p>The argument and return value are floating point numbers of the same
7817 <p>This function returns the same values as the libm <tt>floor</tt> functions
7818 would, and handles error conditions in the same way.</p>
7822 <!-- _______________________________________________________________________ -->
7824 <a name="int_ceil">'<tt>llvm.ceil.*</tt>' Intrinsic</a>
7830 <p>This is an overloaded intrinsic. You can use <tt>llvm.ceil</tt> on any
7831 floating point or vector of floating point type. Not all targets support all
7835 declare float @llvm.ceil.f32(float %Val)
7836 declare double @llvm.ceil.f64(double %Val)
7837 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
7838 declare fp128 @llvm.ceil.f128(fp128 %Val)
7839 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
7843 <p>The '<tt>llvm.ceil.*</tt>' intrinsics return the ceiling of
7847 <p>The argument and return value are floating point numbers of the same
7851 <p>This function returns the same values as the libm <tt>ceil</tt> functions
7852 would, and handles error conditions in the same way.</p>
7856 <!-- _______________________________________________________________________ -->
7858 <a name="int_trunc">'<tt>llvm.trunc.*</tt>' Intrinsic</a>
7864 <p>This is an overloaded intrinsic. You can use <tt>llvm.trunc</tt> on any
7865 floating point or vector of floating point type. Not all targets support all
7869 declare float @llvm.trunc.f32(float %Val)
7870 declare double @llvm.trunc.f64(double %Val)
7871 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
7872 declare fp128 @llvm.trunc.f128(fp128 %Val)
7873 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
7877 <p>The '<tt>llvm.trunc.*</tt>' intrinsics returns the operand rounded to the
7878 nearest integer not larger in magnitude than the operand.</p>
7881 <p>The argument and return value are floating point numbers of the same
7885 <p>This function returns the same values as the libm <tt>trunc</tt> functions
7886 would, and handles error conditions in the same way.</p>
7890 <!-- _______________________________________________________________________ -->
7892 <a name="int_rint">'<tt>llvm.rint.*</tt>' Intrinsic</a>
7898 <p>This is an overloaded intrinsic. You can use <tt>llvm.rint</tt> on any
7899 floating point or vector of floating point type. Not all targets support all
7903 declare float @llvm.rint.f32(float %Val)
7904 declare double @llvm.rint.f64(double %Val)
7905 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
7906 declare fp128 @llvm.rint.f128(fp128 %Val)
7907 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
7911 <p>The '<tt>llvm.rint.*</tt>' intrinsics returns the operand rounded to the
7912 nearest integer. It may raise an inexact floating-point exception if the
7913 operand isn't an integer.</p>
7916 <p>The argument and return value are floating point numbers of the same
7920 <p>This function returns the same values as the libm <tt>rint</tt> functions
7921 would, and handles error conditions in the same way.</p>
7925 <!-- _______________________________________________________________________ -->
7927 <a name="int_nearbyint">'<tt>llvm.nearbyint.*</tt>' Intrinsic</a>
7933 <p>This is an overloaded intrinsic. You can use <tt>llvm.nearbyint</tt> on any
7934 floating point or vector of floating point type. Not all targets support all
7938 declare float @llvm.nearbyint.f32(float %Val)
7939 declare double @llvm.nearbyint.f64(double %Val)
7940 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
7941 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
7942 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
7946 <p>The '<tt>llvm.nearbyint.*</tt>' intrinsics returns the operand rounded to the
7947 nearest integer.</p>
7950 <p>The argument and return value are floating point numbers of the same
7954 <p>This function returns the same values as the libm <tt>nearbyint</tt>
7955 functions would, and handles error conditions in the same way.</p>
7961 <!-- ======================================================================= -->
7963 <a name="int_manip">Bit Manipulation Intrinsics</a>
7968 <p>LLVM provides intrinsics for a few important bit manipulation operations.
7969 These allow efficient code generation for some algorithms.</p>
7971 <!-- _______________________________________________________________________ -->
7973 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7979 <p>This is an overloaded intrinsic function. You can use bswap on any integer
7980 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7983 declare i16 @llvm.bswap.i16(i16 <id>)
7984 declare i32 @llvm.bswap.i32(i32 <id>)
7985 declare i64 @llvm.bswap.i64(i64 <id>)
7989 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7990 values with an even number of bytes (positive multiple of 16 bits). These
7991 are useful for performing operations on data that is not in the target's
7992 native byte order.</p>
7995 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7996 and low byte of the input i16 swapped. Similarly,
7997 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7998 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7999 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
8000 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
8001 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
8002 more, respectively).</p>
8006 <!-- _______________________________________________________________________ -->
8008 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
8014 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
8015 width, or on any vector with integer elements. Not all targets support all
8016 bit widths or vector types, however.</p>
8019 declare i8 @llvm.ctpop.i8(i8 <src>)
8020 declare i16 @llvm.ctpop.i16(i16 <src>)
8021 declare i32 @llvm.ctpop.i32(i32 <src>)
8022 declare i64 @llvm.ctpop.i64(i64 <src>)
8023 declare i256 @llvm.ctpop.i256(i256 <src>)
8024 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
8028 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
8032 <p>The only argument is the value to be counted. The argument may be of any
8033 integer type, or a vector with integer elements.
8034 The return type must match the argument type.</p>
8037 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
8038 element of a vector.</p>
8042 <!-- _______________________________________________________________________ -->
8044 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
8050 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
8051 integer bit width, or any vector whose elements are integers. Not all
8052 targets support all bit widths or vector types, however.</p>
8055 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
8056 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
8057 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
8058 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
8059 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
8060 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8064 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
8065 leading zeros in a variable.</p>
8068 <p>The first argument is the value to be counted. This argument may be of any
8069 integer type, or a vectory with integer element type. The return type
8070 must match the first argument type.</p>
8072 <p>The second argument must be a constant and is a flag to indicate whether the
8073 intrinsic should ensure that a zero as the first argument produces a defined
8074 result. Historically some architectures did not provide a defined result for
8075 zero values as efficiently, and many algorithms are now predicated on
8076 avoiding zero-value inputs.</p>
8079 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
8080 zeros in a variable, or within each element of the vector.
8081 If <tt>src == 0</tt> then the result is the size in bits of the type of
8082 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
8083 For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
8087 <!-- _______________________________________________________________________ -->
8089 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
8095 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
8096 integer bit width, or any vector of integer elements. Not all targets
8097 support all bit widths or vector types, however.</p>
8100 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
8101 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
8102 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
8103 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
8104 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
8105 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
8109 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
8113 <p>The first argument is the value to be counted. This argument may be of any
8114 integer type, or a vectory with integer element type. The return type
8115 must match the first argument type.</p>
8117 <p>The second argument must be a constant and is a flag to indicate whether the
8118 intrinsic should ensure that a zero as the first argument produces a defined
8119 result. Historically some architectures did not provide a defined result for
8120 zero values as efficiently, and many algorithms are now predicated on
8121 avoiding zero-value inputs.</p>
8124 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
8125 zeros in a variable, or within each element of a vector.
8126 If <tt>src == 0</tt> then the result is the size in bits of the type of
8127 <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
8128 For example, <tt>llvm.cttz(2) = 1</tt>.</p>
8134 <!-- ======================================================================= -->
8136 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
8141 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
8143 <!-- _______________________________________________________________________ -->
8145 <a name="int_sadd_overflow">
8146 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
8153 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
8154 on any integer bit width.</p>
8157 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
8158 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8159 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
8163 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
8164 a signed addition of the two arguments, and indicate whether an overflow
8165 occurred during the signed summation.</p>
8168 <p>The arguments (%a and %b) and the first element of the result structure may
8169 be of integer types of any bit width, but they must have the same bit
8170 width. The second element of the result structure must be of
8171 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8172 undergo signed addition.</p>
8175 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
8176 a signed addition of the two variables. They return a structure — the
8177 first element of which is the signed summation, and the second element of
8178 which is a bit specifying if the signed summation resulted in an
8183 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
8184 %sum = extractvalue {i32, i1} %res, 0
8185 %obit = extractvalue {i32, i1} %res, 1
8186 br i1 %obit, label %overflow, label %normal
8191 <!-- _______________________________________________________________________ -->
8193 <a name="int_uadd_overflow">
8194 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
8201 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
8202 on any integer bit width.</p>
8205 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
8206 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8207 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
8211 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
8212 an unsigned addition of the two arguments, and indicate whether a carry
8213 occurred during the unsigned summation.</p>
8216 <p>The arguments (%a and %b) and the first element of the result structure may
8217 be of integer types of any bit width, but they must have the same bit
8218 width. The second element of the result structure must be of
8219 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8220 undergo unsigned addition.</p>
8223 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
8224 an unsigned addition of the two arguments. They return a structure —
8225 the first element of which is the sum, and the second element of which is a
8226 bit specifying if the unsigned summation resulted in a carry.</p>
8230 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
8231 %sum = extractvalue {i32, i1} %res, 0
8232 %obit = extractvalue {i32, i1} %res, 1
8233 br i1 %obit, label %carry, label %normal
8238 <!-- _______________________________________________________________________ -->
8240 <a name="int_ssub_overflow">
8241 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
8248 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
8249 on any integer bit width.</p>
8252 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
8253 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8254 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
8258 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
8259 a signed subtraction of the two arguments, and indicate whether an overflow
8260 occurred during the signed subtraction.</p>
8263 <p>The arguments (%a and %b) and the first element of the result structure may
8264 be of integer types of any bit width, but they must have the same bit
8265 width. The second element of the result structure must be of
8266 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8267 undergo signed subtraction.</p>
8270 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
8271 a signed subtraction of the two arguments. They return a structure —
8272 the first element of which is the subtraction, and the second element of
8273 which is a bit specifying if the signed subtraction resulted in an
8278 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
8279 %sum = extractvalue {i32, i1} %res, 0
8280 %obit = extractvalue {i32, i1} %res, 1
8281 br i1 %obit, label %overflow, label %normal
8286 <!-- _______________________________________________________________________ -->
8288 <a name="int_usub_overflow">
8289 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
8296 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
8297 on any integer bit width.</p>
8300 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
8301 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8302 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
8306 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
8307 an unsigned subtraction of the two arguments, and indicate whether an
8308 overflow occurred during the unsigned subtraction.</p>
8311 <p>The arguments (%a and %b) and the first element of the result structure may
8312 be of integer types of any bit width, but they must have the same bit
8313 width. The second element of the result structure must be of
8314 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8315 undergo unsigned subtraction.</p>
8318 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
8319 an unsigned subtraction of the two arguments. They return a structure —
8320 the first element of which is the subtraction, and the second element of
8321 which is a bit specifying if the unsigned subtraction resulted in an
8326 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
8327 %sum = extractvalue {i32, i1} %res, 0
8328 %obit = extractvalue {i32, i1} %res, 1
8329 br i1 %obit, label %overflow, label %normal
8334 <!-- _______________________________________________________________________ -->
8336 <a name="int_smul_overflow">
8337 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
8344 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
8345 on any integer bit width.</p>
8348 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
8349 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8350 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
8355 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
8356 a signed multiplication of the two arguments, and indicate whether an
8357 overflow occurred during the signed multiplication.</p>
8360 <p>The arguments (%a and %b) and the first element of the result structure may
8361 be of integer types of any bit width, but they must have the same bit
8362 width. The second element of the result structure must be of
8363 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8364 undergo signed multiplication.</p>
8367 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
8368 a signed multiplication of the two arguments. They return a structure —
8369 the first element of which is the multiplication, and the second element of
8370 which is a bit specifying if the signed multiplication resulted in an
8375 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8376 %sum = extractvalue {i32, i1} %res, 0
8377 %obit = extractvalue {i32, i1} %res, 1
8378 br i1 %obit, label %overflow, label %normal
8383 <!-- _______________________________________________________________________ -->
8385 <a name="int_umul_overflow">
8386 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
8393 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
8394 on any integer bit width.</p>
8397 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8398 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8399 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8403 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
8404 a unsigned multiplication of the two arguments, and indicate whether an
8405 overflow occurred during the unsigned multiplication.</p>
8408 <p>The arguments (%a and %b) and the first element of the result structure may
8409 be of integer types of any bit width, but they must have the same bit
8410 width. The second element of the result structure must be of
8411 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8412 undergo unsigned multiplication.</p>
8415 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
8416 an unsigned multiplication of the two arguments. They return a structure
8417 — the first element of which is the multiplication, and the second
8418 element of which is a bit specifying if the unsigned multiplication resulted
8423 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8424 %sum = extractvalue {i32, i1} %res, 0
8425 %obit = extractvalue {i32, i1} %res, 1
8426 br i1 %obit, label %overflow, label %normal
8433 <!-- ======================================================================= -->
8435 <a name="spec_arithmetic">Specialised Arithmetic Intrinsics</a>
8438 <!-- _______________________________________________________________________ -->
8441 <a name="fmuladd">'<tt>llvm.fmuladd.*</tt>' Intrinsic</a>
8448 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8449 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8453 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsic functions represent multiply-add
8454 expressions that can be fused if the code generator determines that the fused
8455 expression would be legal and efficient.</p>
8458 <p>The '<tt>llvm.fmuladd.*</tt>' intrinsics each take three arguments: two
8459 multiplicands, a and b, and an addend c.</p>
8462 <p>The expression:</p>
8464 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8466 <p>is equivalent to the expression a * b + c, except that rounding will not be
8467 performed between the multiplication and addition steps if the code generator
8468 fuses the operations. Fusion is not guaranteed, even if the target platform
8469 supports it. If a fused multiply-add is required the corresponding llvm.fma.*
8470 intrinsic function should be used instead.</p>
8474 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8479 <!-- ======================================================================= -->
8481 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
8486 <p>For most target platforms, half precision floating point is a storage-only
8487 format. This means that it is
8488 a dense encoding (in memory) but does not support computation in the
8491 <p>This means that code must first load the half-precision floating point
8492 value as an i16, then convert it to float with <a
8493 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
8494 Computation can then be performed on the float value (including extending to
8495 double etc). To store the value back to memory, it is first converted to
8496 float if needed, then converted to i16 with
8497 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
8498 storing as an i16 value.</p>
8500 <!-- _______________________________________________________________________ -->
8502 <a name="int_convert_to_fp16">
8503 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
8511 declare i16 @llvm.convert.to.fp16(f32 %a)
8515 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8516 a conversion from single precision floating point format to half precision
8517 floating point format.</p>
8520 <p>The intrinsic function contains single argument - the value to be
8524 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8525 a conversion from single precision floating point format to half precision
8526 floating point format. The return value is an <tt>i16</tt> which
8527 contains the converted number.</p>
8531 %res = call i16 @llvm.convert.to.fp16(f32 %a)
8532 store i16 %res, i16* @x, align 2
8537 <!-- _______________________________________________________________________ -->
8539 <a name="int_convert_from_fp16">
8540 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
8548 declare f32 @llvm.convert.from.fp16(i16 %a)
8552 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
8553 a conversion from half precision floating point format to single precision
8554 floating point format.</p>
8557 <p>The intrinsic function contains single argument - the value to be
8561 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
8562 conversion from half single precision floating point format to single
8563 precision floating point format. The input half-float value is represented by
8564 an <tt>i16</tt> value.</p>
8568 %a = load i16* @x, align 2
8569 %res = call f32 @llvm.convert.from.fp16(i16 %a)
8576 <!-- ======================================================================= -->
8578 <a name="int_debugger">Debugger Intrinsics</a>
8583 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
8584 prefix), are described in
8585 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
8586 Level Debugging</a> document.</p>
8590 <!-- ======================================================================= -->
8592 <a name="int_eh">Exception Handling Intrinsics</a>
8597 <p>The LLVM exception handling intrinsics (which all start with
8598 <tt>llvm.eh.</tt> prefix), are described in
8599 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
8600 Handling</a> document.</p>
8604 <!-- ======================================================================= -->
8606 <a name="int_trampoline">Trampoline Intrinsics</a>
8611 <p>These intrinsics make it possible to excise one parameter, marked with
8612 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
8613 The result is a callable
8614 function pointer lacking the nest parameter - the caller does not need to
8615 provide a value for it. Instead, the value to use is stored in advance in a
8616 "trampoline", a block of memory usually allocated on the stack, which also
8617 contains code to splice the nest value into the argument list. This is used
8618 to implement the GCC nested function address extension.</p>
8620 <p>For example, if the function is
8621 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
8622 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
8625 <pre class="doc_code">
8626 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8627 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8628 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8629 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8630 %fp = bitcast i8* %p to i32 (i32, i32)*
8633 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
8634 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
8636 <!-- _______________________________________________________________________ -->
8639 '<tt>llvm.init.trampoline</tt>' Intrinsic
8647 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
8651 <p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
8652 turning it into a trampoline.</p>
8655 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
8656 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
8657 sufficiently aligned block of memory; this memory is written to by the
8658 intrinsic. Note that the size and the alignment are target-specific - LLVM
8659 currently provides no portable way of determining them, so a front-end that
8660 generates this intrinsic needs to have some target-specific knowledge.
8661 The <tt>func</tt> argument must hold a function bitcast to
8662 an <tt>i8*</tt>.</p>
8665 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
8666 dependent code, turning it into a function. Then <tt>tramp</tt> needs to be
8667 passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
8668 which can be <a href="#int_trampoline">bitcast (to a new function) and
8669 called</a>. The new function's signature is the same as that of
8670 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
8671 removed. At most one such <tt>nest</tt> argument is allowed, and it must be of
8672 pointer type. Calling the new function is equivalent to calling <tt>func</tt>
8673 with the same argument list, but with <tt>nval</tt> used for the missing
8674 <tt>nest</tt> argument. If, after calling <tt>llvm.init.trampoline</tt>, the
8675 memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
8676 to the returned function pointer is undefined.</p>
8679 <!-- _______________________________________________________________________ -->
8682 '<tt>llvm.adjust.trampoline</tt>' Intrinsic
8690 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
8694 <p>This performs any required machine-specific adjustment to the address of a
8695 trampoline (passed as <tt>tramp</tt>).</p>
8698 <p><tt>tramp</tt> must point to a block of memory which already has trampoline code
8699 filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
8703 <p>On some architectures the address of the code to be executed needs to be
8704 different to the address where the trampoline is actually stored. This
8705 intrinsic returns the executable address corresponding to <tt>tramp</tt>
8706 after performing the required machine specific adjustments.
8707 The pointer returned can then be <a href="#int_trampoline"> bitcast and
8715 <!-- ======================================================================= -->
8717 <a name="int_memorymarkers">Memory Use Markers</a>
8722 <p>This class of intrinsics exists to information about the lifetime of memory
8723 objects and ranges where variables are immutable.</p>
8725 <!-- _______________________________________________________________________ -->
8727 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8734 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8738 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8739 object's lifetime.</p>
8742 <p>The first argument is a constant integer representing the size of the
8743 object, or -1 if it is variable sized. The second argument is a pointer to
8747 <p>This intrinsic indicates that before this point in the code, the value of the
8748 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8749 never be used and has an undefined value. A load from the pointer that
8750 precedes this intrinsic can be replaced with
8751 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8755 <!-- _______________________________________________________________________ -->
8757 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8764 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8768 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8769 object's lifetime.</p>
8772 <p>The first argument is a constant integer representing the size of the
8773 object, or -1 if it is variable sized. The second argument is a pointer to
8777 <p>This intrinsic indicates that after this point in the code, the value of the
8778 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8779 never be used and has an undefined value. Any stores into the memory object
8780 following this intrinsic may be removed as dead.
8784 <!-- _______________________________________________________________________ -->
8786 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8793 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8797 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8798 a memory object will not change.</p>
8801 <p>The first argument is a constant integer representing the size of the
8802 object, or -1 if it is variable sized. The second argument is a pointer to
8806 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8807 the return value, the referenced memory location is constant and
8812 <!-- _______________________________________________________________________ -->
8814 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8821 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8825 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8826 a memory object are mutable.</p>
8829 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8830 The second argument is a constant integer representing the size of the
8831 object, or -1 if it is variable sized and the third argument is a pointer
8835 <p>This intrinsic indicates that the memory is mutable again.</p>
8841 <!-- ======================================================================= -->
8843 <a name="int_general">General Intrinsics</a>
8848 <p>This class of intrinsics is designed to be generic and has no specific
8851 <!-- _______________________________________________________________________ -->
8853 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8860 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8864 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8867 <p>The first argument is a pointer to a value, the second is a pointer to a
8868 global string, the third is a pointer to a global string which is the source
8869 file name, and the last argument is the line number.</p>
8872 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8873 This can be useful for special purpose optimizations that want to look for
8874 these annotations. These have no other defined use; they are ignored by code
8875 generation and optimization.</p>
8879 <!-- _______________________________________________________________________ -->
8881 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8887 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8888 any integer bit width.</p>
8891 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8892 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8893 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8894 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8895 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8899 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8902 <p>The first argument is an integer value (result of some expression), the
8903 second is a pointer to a global string, the third is a pointer to a global
8904 string which is the source file name, and the last argument is the line
8905 number. It returns the value of the first argument.</p>
8908 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8909 arbitrary strings. This can be useful for special purpose optimizations that
8910 want to look for these annotations. These have no other defined use; they
8911 are ignored by code generation and optimization.</p>
8915 <!-- _______________________________________________________________________ -->
8917 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8924 declare void @llvm.trap() noreturn nounwind
8928 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8934 <p>This intrinsic is lowered to the target dependent trap instruction. If the
8935 target does not have a trap instruction, this intrinsic will be lowered to
8936 a call of the <tt>abort()</tt> function.</p>
8940 <!-- _______________________________________________________________________ -->
8942 <a name="int_debugtrap">'<tt>llvm.debugtrap</tt>' Intrinsic</a>
8949 declare void @llvm.debugtrap() nounwind
8953 <p>The '<tt>llvm.debugtrap</tt>' intrinsic.</p>
8959 <p>This intrinsic is lowered to code which is intended to cause an execution
8960 trap with the intention of requesting the attention of a debugger.</p>
8964 <!-- _______________________________________________________________________ -->
8966 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8973 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8977 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8978 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8979 ensure that it is placed on the stack before local variables.</p>
8982 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8983 arguments. The first argument is the value loaded from the stack
8984 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8985 that has enough space to hold the value of the guard.</p>
8988 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8989 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8990 stack. This is to ensure that if a local variable on the stack is
8991 overwritten, it will destroy the value of the guard. When the function exits,
8992 the guard on the stack is checked against the original guard. If they are
8993 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8998 <!-- _______________________________________________________________________ -->
9000 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
9007 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
9008 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
9012 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
9013 the optimizers to determine at compile time whether a) an operation (like
9014 memcpy) will overflow a buffer that corresponds to an object, or b) that a
9015 runtime check for overflow isn't necessary. An object in this context means
9016 an allocation of a specific class, structure, array, or other object.</p>
9019 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
9020 argument is a pointer to or into the <tt>object</tt>. The second argument
9021 is a boolean and determines whether <tt>llvm.objectsize</tt> returns 0 (if
9022 true) or -1 (if false) when the object size is unknown.
9023 The second argument only accepts constants.</p>
9026 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to a constant representing
9027 the size of the object concerned. If the size cannot be determined at compile
9028 time, <tt>llvm.objectsize</tt> returns <tt>i32/i64 -1 or 0</tt>
9029 (depending on the <tt>min</tt> argument).</p>
9032 <!-- _______________________________________________________________________ -->
9034 <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a>
9041 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
9042 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
9046 <p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the
9047 most probable) value of <tt>val</tt>, which can be used by optimizers.</p>
9050 <p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first
9051 argument is a value. The second argument is an expected value, this needs to
9052 be a constant value, variables are not allowed.</p>
9055 <p>This intrinsic is lowered to the <tt>val</tt>.</p>
9058 <!-- _______________________________________________________________________ -->
9060 <a name="int_donothing">'<tt>llvm.donothing</tt>' Intrinsic</a>
9067 declare void @llvm.donothing() nounwind readnone
9071 <p>The <tt>llvm.donothing</tt> intrinsic doesn't perform any operation. It's the
9072 only intrinsic that can be called with an invoke instruction.</p>
9078 <p>This intrinsic does nothing, and it's removed by optimizers and ignored by
9085 <!-- *********************************************************************** -->
9088 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
9089 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
9090 <a href="http://validator.w3.org/check/referer"><img
9091 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
9093 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
9094 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
9095 Last modified: $Date$