1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <h1>LLVM Language Reference Manual</h1>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a>
25 <li><a href="#linkage_private">'<tt>private</tt>' Linkage</a></li>
26 <li><a href="#linkage_linker_private">'<tt>linker_private</tt>' Linkage</a></li>
27 <li><a href="#linkage_linker_private_weak">'<tt>linker_private_weak</tt>' Linkage</a></li>
28 <li><a href="#linkage_linker_private_weak_def_auto">'<tt>linker_private_weak_def_auto</tt>' Linkage</a></li>
29 <li><a href="#linkage_internal">'<tt>internal</tt>' Linkage</a></li>
30 <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
31 <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
32 <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
33 <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
34 <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
35 <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
36 <li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
37 <li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
38 <li><a href="#linkage_external">'<tt>externally visible</tt>' Linkage</a></li>
39 <li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
40 <li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
43 <li><a href="#callingconv">Calling Conventions</a></li>
44 <li><a href="#namedtypes">Named Types</a></li>
45 <li><a href="#globalvars">Global Variables</a></li>
46 <li><a href="#functionstructure">Functions</a></li>
47 <li><a href="#aliasstructure">Aliases</a></li>
48 <li><a href="#namedmetadatastructure">Named Metadata</a></li>
49 <li><a href="#paramattrs">Parameter Attributes</a></li>
50 <li><a href="#fnattrs">Function Attributes</a></li>
51 <li><a href="#gc">Garbage Collector Names</a></li>
52 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
53 <li><a href="#datalayout">Data Layout</a></li>
54 <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
55 <li><a href="#volatile">Volatile Memory Accesses</a></li>
56 <li><a href="#memmodel">Memory Model for Concurrent Operations</a></li>
57 <li><a href="#ordering">Atomic Memory Ordering Constraints</a></li>
60 <li><a href="#typesystem">Type System</a>
62 <li><a href="#t_classifications">Type Classifications</a></li>
63 <li><a href="#t_primitive">Primitive Types</a>
65 <li><a href="#t_integer">Integer Type</a></li>
66 <li><a href="#t_floating">Floating Point Types</a></li>
67 <li><a href="#t_x86mmx">X86mmx Type</a></li>
68 <li><a href="#t_void">Void Type</a></li>
69 <li><a href="#t_label">Label Type</a></li>
70 <li><a href="#t_metadata">Metadata Type</a></li>
73 <li><a href="#t_derived">Derived Types</a>
75 <li><a href="#t_aggregate">Aggregate Types</a>
77 <li><a href="#t_array">Array Type</a></li>
78 <li><a href="#t_struct">Structure Type</a></li>
79 <li><a href="#t_opaque">Opaque Structure Types</a></li>
80 <li><a href="#t_vector">Vector Type</a></li>
83 <li><a href="#t_function">Function Type</a></li>
84 <li><a href="#t_pointer">Pointer Type</a></li>
89 <li><a href="#constants">Constants</a>
91 <li><a href="#simpleconstants">Simple Constants</a></li>
92 <li><a href="#complexconstants">Complex Constants</a></li>
93 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
94 <li><a href="#undefvalues">Undefined Values</a></li>
95 <li><a href="#trapvalues">Trap Values</a></li>
96 <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
97 <li><a href="#constantexprs">Constant Expressions</a></li>
100 <li><a href="#othervalues">Other Values</a>
102 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
103 <li><a href="#metadata">Metadata Nodes and Metadata Strings</a></li>
106 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
108 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
109 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
110 Global Variable</a></li>
111 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
112 Global Variable</a></li>
113 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
114 Global Variable</a></li>
117 <li><a href="#instref">Instruction Reference</a>
119 <li><a href="#terminators">Terminator Instructions</a>
121 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
122 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
123 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
124 <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
125 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
126 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
127 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
130 <li><a href="#binaryops">Binary Operations</a>
132 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
133 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
134 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
135 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
136 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
137 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
138 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
139 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
140 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
141 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
142 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
143 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
146 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
148 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
149 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
150 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
151 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
152 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
153 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
156 <li><a href="#vectorops">Vector Operations</a>
158 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
159 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
160 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
163 <li><a href="#aggregateops">Aggregate Operations</a>
165 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
166 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
169 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
171 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
172 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
173 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
174 <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
175 <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
176 <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
177 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
180 <li><a href="#convertops">Conversion Operations</a>
182 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
183 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
184 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
185 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
186 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
187 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
188 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
189 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
190 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
191 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
192 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
193 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
196 <li><a href="#otherops">Other Operations</a>
198 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
199 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
200 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
201 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
202 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
203 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
208 <li><a href="#intrinsics">Intrinsic Functions</a>
210 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
212 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
213 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
214 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
217 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
219 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
220 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
221 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
224 <li><a href="#int_codegen">Code Generator Intrinsics</a>
226 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
227 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
228 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
229 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
230 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
231 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
232 <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
235 <li><a href="#int_libc">Standard C Library Intrinsics</a>
237 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
238 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
239 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
240 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
241 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
242 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
243 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
244 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
245 <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
246 <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
247 <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
250 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
252 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
253 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
254 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
255 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
258 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
260 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
261 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
262 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
263 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
264 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
265 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
268 <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
270 <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
271 <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
274 <li><a href="#int_debugger">Debugger intrinsics</a></li>
275 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
276 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
278 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
281 <li><a href="#int_atomics">Atomic intrinsics</a>
283 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
284 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
285 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
286 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
287 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
288 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
289 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
290 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
291 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
292 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
293 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
294 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
295 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
298 <li><a href="#int_memorymarkers">Memory Use Markers</a>
300 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
301 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
302 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
303 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
306 <li><a href="#int_general">General intrinsics</a>
308 <li><a href="#int_var_annotation">
309 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
310 <li><a href="#int_annotation">
311 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
312 <li><a href="#int_trap">
313 '<tt>llvm.trap</tt>' Intrinsic</a></li>
314 <li><a href="#int_stackprotector">
315 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
316 <li><a href="#int_objectsize">
317 '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
324 <div class="doc_author">
325 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
326 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
329 <!-- *********************************************************************** -->
330 <h2><a name="abstract">Abstract</a></h2>
331 <!-- *********************************************************************** -->
335 <p>This document is a reference manual for the LLVM assembly language. LLVM is
336 a Static Single Assignment (SSA) based representation that provides type
337 safety, low-level operations, flexibility, and the capability of representing
338 'all' high-level languages cleanly. It is the common code representation
339 used throughout all phases of the LLVM compilation strategy.</p>
343 <!-- *********************************************************************** -->
344 <h2><a name="introduction">Introduction</a></h2>
345 <!-- *********************************************************************** -->
349 <p>The LLVM code representation is designed to be used in three different forms:
350 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
351 for fast loading by a Just-In-Time compiler), and as a human readable
352 assembly language representation. This allows LLVM to provide a powerful
353 intermediate representation for efficient compiler transformations and
354 analysis, while providing a natural means to debug and visualize the
355 transformations. The three different forms of LLVM are all equivalent. This
356 document describes the human readable representation and notation.</p>
358 <p>The LLVM representation aims to be light-weight and low-level while being
359 expressive, typed, and extensible at the same time. It aims to be a
360 "universal IR" of sorts, by being at a low enough level that high-level ideas
361 may be cleanly mapped to it (similar to how microprocessors are "universal
362 IR's", allowing many source languages to be mapped to them). By providing
363 type information, LLVM can be used as the target of optimizations: for
364 example, through pointer analysis, it can be proven that a C automatic
365 variable is never accessed outside of the current function, allowing it to
366 be promoted to a simple SSA value instead of a memory location.</p>
368 <!-- _______________________________________________________________________ -->
370 <a name="wellformed">Well-Formedness</a>
375 <p>It is important to note that this document describes 'well formed' LLVM
376 assembly language. There is a difference between what the parser accepts and
377 what is considered 'well formed'. For example, the following instruction is
378 syntactically okay, but not well formed:</p>
380 <pre class="doc_code">
381 %x = <a href="#i_add">add</a> i32 1, %x
384 <p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
385 LLVM infrastructure provides a verification pass that may be used to verify
386 that an LLVM module is well formed. This pass is automatically run by the
387 parser after parsing input assembly and by the optimizer before it outputs
388 bitcode. The violations pointed out by the verifier pass indicate bugs in
389 transformation passes or input to the parser.</p>
395 <!-- Describe the typesetting conventions here. -->
397 <!-- *********************************************************************** -->
398 <h2><a name="identifiers">Identifiers</a></h2>
399 <!-- *********************************************************************** -->
403 <p>LLVM identifiers come in two basic types: global and local. Global
404 identifiers (functions, global variables) begin with the <tt>'@'</tt>
405 character. Local identifiers (register names, types) begin with
406 the <tt>'%'</tt> character. Additionally, there are three different formats
407 for identifiers, for different purposes:</p>
410 <li>Named values are represented as a string of characters with their prefix.
411 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
412 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
413 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
414 other characters in their names can be surrounded with quotes. Special
415 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
416 ASCII code for the character in hexadecimal. In this way, any character
417 can be used in a name value, even quotes themselves.</li>
419 <li>Unnamed values are represented as an unsigned numeric value with their
420 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
422 <li>Constants, which are described in a <a href="#constants">section about
423 constants</a>, below.</li>
426 <p>LLVM requires that values start with a prefix for two reasons: Compilers
427 don't need to worry about name clashes with reserved words, and the set of
428 reserved words may be expanded in the future without penalty. Additionally,
429 unnamed identifiers allow a compiler to quickly come up with a temporary
430 variable without having to avoid symbol table conflicts.</p>
432 <p>Reserved words in LLVM are very similar to reserved words in other
433 languages. There are keywords for different opcodes
434 ('<tt><a href="#i_add">add</a></tt>',
435 '<tt><a href="#i_bitcast">bitcast</a></tt>',
436 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
437 ('<tt><a href="#t_void">void</a></tt>',
438 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
439 reserved words cannot conflict with variable names, because none of them
440 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
442 <p>Here is an example of LLVM code to multiply the integer variable
443 '<tt>%X</tt>' by 8:</p>
447 <pre class="doc_code">
448 %result = <a href="#i_mul">mul</a> i32 %X, 8
451 <p>After strength reduction:</p>
453 <pre class="doc_code">
454 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
457 <p>And the hard way:</p>
459 <pre class="doc_code">
460 %0 = <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
461 %1 = <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
462 %result = <a href="#i_add">add</a> i32 %1, %1
465 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
466 lexical features of LLVM:</p>
469 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
472 <li>Unnamed temporaries are created when the result of a computation is not
473 assigned to a named value.</li>
475 <li>Unnamed temporaries are numbered sequentially</li>
478 <p>It also shows a convention that we follow in this document. When
479 demonstrating instructions, we will follow an instruction with a comment that
480 defines the type and name of value produced. Comments are shown in italic
485 <!-- *********************************************************************** -->
486 <h2><a name="highlevel">High Level Structure</a></h2>
487 <!-- *********************************************************************** -->
489 <!-- ======================================================================= -->
491 <a name="modulestructure">Module Structure</a>
496 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
497 of the input programs. Each module consists of functions, global variables,
498 and symbol table entries. Modules may be combined together with the LLVM
499 linker, which merges function (and global variable) definitions, resolves
500 forward declarations, and merges symbol table entries. Here is an example of
501 the "hello world" module:</p>
503 <pre class="doc_code">
504 <i>; Declare the string constant as a global constant.</i>
505 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
507 <i>; External declaration of the puts function</i>
508 <a href="#functionstructure">declare</a> i32 @puts(i8*) <i>; i32 (i8*)* </i>
510 <i>; Definition of main function</i>
511 define i32 @main() { <i>; i32()* </i>
512 <i>; Convert [13 x i8]* to i8 *...</i>
513 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8*</i>
515 <i>; Call puts function to write out the string to stdout.</i>
516 <a href="#i_call">call</a> i32 @puts(i8* %cast210) <i>; i32</i>
517 <a href="#i_ret">ret</a> i32 0
520 <i>; Named metadata</i>
521 !1 = metadata !{i32 41}
525 <p>This example is made up of a <a href="#globalvars">global variable</a> named
526 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function,
527 a <a href="#functionstructure">function definition</a> for
528 "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
531 <p>In general, a module is made up of a list of global values, where both
532 functions and global variables are global values. Global values are
533 represented by a pointer to a memory location (in this case, a pointer to an
534 array of char, and a pointer to a function), and have one of the
535 following <a href="#linkage">linkage types</a>.</p>
539 <!-- ======================================================================= -->
541 <a name="linkage">Linkage Types</a>
546 <p>All Global Variables and Functions have one of the following types of
550 <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
551 <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
552 by objects in the current module. In particular, linking code into a
553 module with an private global value may cause the private to be renamed as
554 necessary to avoid collisions. Because the symbol is private to the
555 module, all references can be updated. This doesn't show up in any symbol
556 table in the object file.</dd>
558 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
559 <dd>Similar to <tt>private</tt>, but the symbol is passed through the
560 assembler and evaluated by the linker. Unlike normal strong symbols, they
561 are removed by the linker from the final linked image (executable or
562 dynamic library).</dd>
564 <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
565 <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
566 <tt>linker_private_weak</tt> symbols are subject to coalescing by the
567 linker. The symbols are removed by the linker from the final linked image
568 (executable or dynamic library).</dd>
570 <dt><tt><b><a name="linkage_linker_private_weak_def_auto">linker_private_weak_def_auto</a></b></tt></dt>
571 <dd>Similar to "<tt>linker_private_weak</tt>", but it's known that the address
572 of the object is not taken. For instance, functions that had an inline
573 definition, but the compiler decided not to inline it. Note,
574 unlike <tt>linker_private</tt> and <tt>linker_private_weak</tt>,
575 <tt>linker_private_weak_def_auto</tt> may have only <tt>default</tt>
576 visibility. The symbols are removed by the linker from the final linked
577 image (executable or dynamic library).</dd>
579 <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
580 <dd>Similar to private, but the value shows as a local symbol
581 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
582 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
584 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
585 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
586 into the object file corresponding to the LLVM module. They exist to
587 allow inlining and other optimizations to take place given knowledge of
588 the definition of the global, which is known to be somewhere outside the
589 module. Globals with <tt>available_externally</tt> linkage are allowed to
590 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
591 This linkage type is only allowed on definitions, not declarations.</dd>
593 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
594 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
595 the same name when linkage occurs. This can be used to implement
596 some forms of inline functions, templates, or other code which must be
597 generated in each translation unit that uses it, but where the body may
598 be overridden with a more definitive definition later. Unreferenced
599 <tt>linkonce</tt> globals are allowed to be discarded. Note that
600 <tt>linkonce</tt> linkage does not actually allow the optimizer to
601 inline the body of this function into callers because it doesn't know if
602 this definition of the function is the definitive definition within the
603 program or whether it will be overridden by a stronger definition.
604 To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
607 <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
608 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
609 <tt>linkonce</tt> linkage, except that unreferenced globals with
610 <tt>weak</tt> linkage may not be discarded. This is used for globals that
611 are declared "weak" in C source code.</dd>
613 <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
614 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
615 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
617 Symbols with "<tt>common</tt>" linkage are merged in the same way as
618 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
619 <tt>common</tt> symbols may not have an explicit section,
620 must have a zero initializer, and may not be marked '<a
621 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
622 have common linkage.</dd>
625 <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
626 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
627 pointer to array type. When two global variables with appending linkage
628 are linked together, the two global arrays are appended together. This is
629 the LLVM, typesafe, equivalent of having the system linker append together
630 "sections" with identical names when .o files are linked.</dd>
632 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
633 <dd>The semantics of this linkage follow the ELF object file model: the symbol
634 is weak until linked, if not linked, the symbol becomes null instead of
635 being an undefined reference.</dd>
637 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
638 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
639 <dd>Some languages allow differing globals to be merged, such as two functions
640 with different semantics. Other languages, such as <tt>C++</tt>, ensure
641 that only equivalent globals are ever merged (the "one definition rule"
642 — "ODR"). Such languages can use the <tt>linkonce_odr</tt>
643 and <tt>weak_odr</tt> linkage types to indicate that the global will only
644 be merged with equivalent globals. These linkage types are otherwise the
645 same as their non-<tt>odr</tt> versions.</dd>
647 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
648 <dd>If none of the above identifiers are used, the global is externally
649 visible, meaning that it participates in linkage and can be used to
650 resolve external symbol references.</dd>
653 <p>The next two types of linkage are targeted for Microsoft Windows platform
654 only. They are designed to support importing (exporting) symbols from (to)
655 DLLs (Dynamic Link Libraries).</p>
658 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
659 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
660 or variable via a global pointer to a pointer that is set up by the DLL
661 exporting the symbol. On Microsoft Windows targets, the pointer name is
662 formed by combining <code>__imp_</code> and the function or variable
665 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
666 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
667 pointer to a pointer in a DLL, so that it can be referenced with the
668 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
669 name is formed by combining <code>__imp_</code> and the function or
673 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
674 another module defined a "<tt>.LC0</tt>" variable and was linked with this
675 one, one of the two would be renamed, preventing a collision. Since
676 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
677 declarations), they are accessible outside of the current module.</p>
679 <p>It is illegal for a function <i>declaration</i> to have any linkage type
680 other than "externally visible", <tt>dllimport</tt>
681 or <tt>extern_weak</tt>.</p>
683 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
684 or <tt>weak_odr</tt> linkages.</p>
688 <!-- ======================================================================= -->
690 <a name="callingconv">Calling Conventions</a>
695 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
696 and <a href="#i_invoke">invokes</a> can all have an optional calling
697 convention specified for the call. The calling convention of any pair of
698 dynamic caller/callee must match, or the behavior of the program is
699 undefined. The following calling conventions are supported by LLVM, and more
700 may be added in the future:</p>
703 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
704 <dd>This calling convention (the default if no other calling convention is
705 specified) matches the target C calling conventions. This calling
706 convention supports varargs function calls and tolerates some mismatch in
707 the declared prototype and implemented declaration of the function (as
710 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
711 <dd>This calling convention attempts to make calls as fast as possible
712 (e.g. by passing things in registers). This calling convention allows the
713 target to use whatever tricks it wants to produce fast code for the
714 target, without having to conform to an externally specified ABI
715 (Application Binary Interface).
716 <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
717 when this or the GHC convention is used.</a> This calling convention
718 does not support varargs and requires the prototype of all callees to
719 exactly match the prototype of the function definition.</dd>
721 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
722 <dd>This calling convention attempts to make code in the caller as efficient
723 as possible under the assumption that the call is not commonly executed.
724 As such, these calls often preserve all registers so that the call does
725 not break any live ranges in the caller side. This calling convention
726 does not support varargs and requires the prototype of all callees to
727 exactly match the prototype of the function definition.</dd>
729 <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
730 <dd>This calling convention has been implemented specifically for use by the
731 <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
732 It passes everything in registers, going to extremes to achieve this by
733 disabling callee save registers. This calling convention should not be
734 used lightly but only for specific situations such as an alternative to
735 the <em>register pinning</em> performance technique often used when
736 implementing functional programming languages.At the moment only X86
737 supports this convention and it has the following limitations:
739 <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
740 floating point types are supported.</li>
741 <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
742 6 floating point parameters.</li>
744 This calling convention supports
745 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
746 requires both the caller and callee are using it.
749 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
750 <dd>Any calling convention may be specified by number, allowing
751 target-specific calling conventions to be used. Target specific calling
752 conventions start at 64.</dd>
755 <p>More calling conventions can be added/defined on an as-needed basis, to
756 support Pascal conventions or any other well-known target-independent
761 <!-- ======================================================================= -->
763 <a name="visibility">Visibility Styles</a>
768 <p>All Global Variables and Functions have one of the following visibility
772 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
773 <dd>On targets that use the ELF object file format, default visibility means
774 that the declaration is visible to other modules and, in shared libraries,
775 means that the declared entity may be overridden. On Darwin, default
776 visibility means that the declaration is visible to other modules. Default
777 visibility corresponds to "external linkage" in the language.</dd>
779 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
780 <dd>Two declarations of an object with hidden visibility refer to the same
781 object if they are in the same shared object. Usually, hidden visibility
782 indicates that the symbol will not be placed into the dynamic symbol
783 table, so no other module (executable or shared library) can reference it
786 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
787 <dd>On ELF, protected visibility indicates that the symbol will be placed in
788 the dynamic symbol table, but that references within the defining module
789 will bind to the local symbol. That is, the symbol cannot be overridden by
795 <!-- ======================================================================= -->
797 <a name="namedtypes">Named Types</a>
802 <p>LLVM IR allows you to specify name aliases for certain types. This can make
803 it easier to read the IR and make the IR more condensed (particularly when
804 recursive types are involved). An example of a name specification is:</p>
806 <pre class="doc_code">
807 %mytype = type { %mytype*, i32 }
810 <p>You may give a name to any <a href="#typesystem">type</a> except
811 "<a href="#t_void">void</a>". Type name aliases may be used anywhere a type
812 is expected with the syntax "%mytype".</p>
814 <p>Note that type names are aliases for the structural type that they indicate,
815 and that you can therefore specify multiple names for the same type. This
816 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
817 uses structural typing, the name is not part of the type. When printing out
818 LLVM IR, the printer will pick <em>one name</em> to render all types of a
819 particular shape. This means that if you have code where two different
820 source types end up having the same LLVM type, that the dumper will sometimes
821 print the "wrong" or unexpected type. This is an important design point and
822 isn't going to change.</p>
826 <!-- ======================================================================= -->
828 <a name="globalvars">Global Variables</a>
833 <p>Global variables define regions of memory allocated at compilation time
834 instead of run-time. Global variables may optionally be initialized, may
835 have an explicit section to be placed in, and may have an optional explicit
836 alignment specified. A variable may be defined as "thread_local", which
837 means that it will not be shared by threads (each thread will have a
838 separated copy of the variable). A variable may be defined as a global
839 "constant," which indicates that the contents of the variable
840 will <b>never</b> be modified (enabling better optimization, allowing the
841 global data to be placed in the read-only section of an executable, etc).
842 Note that variables that need runtime initialization cannot be marked
843 "constant" as there is a store to the variable.</p>
845 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
846 constant, even if the final definition of the global is not. This capability
847 can be used to enable slightly better optimization of the program, but
848 requires the language definition to guarantee that optimizations based on the
849 'constantness' are valid for the translation units that do not include the
852 <p>As SSA values, global variables define pointer values that are in scope
853 (i.e. they dominate) all basic blocks in the program. Global variables
854 always define a pointer to their "content" type because they describe a
855 region of memory, and all memory objects in LLVM are accessed through
858 <p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
859 that the address is not significant, only the content. Constants marked
860 like this can be merged with other constants if they have the same
861 initializer. Note that a constant with significant address <em>can</em>
862 be merged with a <tt>unnamed_addr</tt> constant, the result being a
863 constant whose address is significant.</p>
865 <p>A global variable may be declared to reside in a target-specific numbered
866 address space. For targets that support them, address spaces may affect how
867 optimizations are performed and/or what target instructions are used to
868 access the variable. The default address space is zero. The address space
869 qualifier must precede any other attributes.</p>
871 <p>LLVM allows an explicit section to be specified for globals. If the target
872 supports it, it will emit globals to the section specified.</p>
874 <p>An explicit alignment may be specified for a global, which must be a power
875 of 2. If not present, or if the alignment is set to zero, the alignment of
876 the global is set by the target to whatever it feels convenient. If an
877 explicit alignment is specified, the global is forced to have exactly that
878 alignment. Targets and optimizers are not allowed to over-align the global
879 if the global has an assigned section. In this case, the extra alignment
880 could be observable: for example, code could assume that the globals are
881 densely packed in their section and try to iterate over them as an array,
882 alignment padding would break this iteration.</p>
884 <p>For example, the following defines a global in a numbered address space with
885 an initializer, section, and alignment:</p>
887 <pre class="doc_code">
888 @G = addrspace(5) constant float 1.0, section "foo", align 4
894 <!-- ======================================================================= -->
896 <a name="functionstructure">Functions</a>
901 <p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
902 optional <a href="#linkage">linkage type</a>, an optional
903 <a href="#visibility">visibility style</a>, an optional
904 <a href="#callingconv">calling convention</a>,
905 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
906 <a href="#paramattrs">parameter attribute</a> for the return type, a function
907 name, a (possibly empty) argument list (each with optional
908 <a href="#paramattrs">parameter attributes</a>), optional
909 <a href="#fnattrs">function attributes</a>, an optional section, an optional
910 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
911 curly brace, a list of basic blocks, and a closing curly brace.</p>
913 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
914 optional <a href="#linkage">linkage type</a>, an optional
915 <a href="#visibility">visibility style</a>, an optional
916 <a href="#callingconv">calling convention</a>,
917 an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
918 <a href="#paramattrs">parameter attribute</a> for the return type, a function
919 name, a possibly empty list of arguments, an optional alignment, and an
920 optional <a href="#gc">garbage collector name</a>.</p>
922 <p>A function definition contains a list of basic blocks, forming the CFG
923 (Control Flow Graph) for the function. Each basic block may optionally start
924 with a label (giving the basic block a symbol table entry), contains a list
925 of instructions, and ends with a <a href="#terminators">terminator</a>
926 instruction (such as a branch or function return).</p>
928 <p>The first basic block in a function is special in two ways: it is immediately
929 executed on entrance to the function, and it is not allowed to have
930 predecessor basic blocks (i.e. there can not be any branches to the entry
931 block of a function). Because the block can have no predecessors, it also
932 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
934 <p>LLVM allows an explicit section to be specified for functions. If the target
935 supports it, it will emit functions to the section specified.</p>
937 <p>An explicit alignment may be specified for a function. If not present, or if
938 the alignment is set to zero, the alignment of the function is set by the
939 target to whatever it feels convenient. If an explicit alignment is
940 specified, the function is forced to have at least that much alignment. All
941 alignments must be a power of 2.</p>
943 <p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
944 be significant and two identical functions can be merged</p>.
947 <pre class="doc_code">
948 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
949 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
950 <ResultType> @<FunctionName> ([argument list])
951 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
952 [<a href="#gc">gc</a>] { ... }
957 <!-- ======================================================================= -->
959 <a name="aliasstructure">Aliases</a>
964 <p>Aliases act as "second name" for the aliasee value (which can be either
965 function, global variable, another alias or bitcast of global value). Aliases
966 may have an optional <a href="#linkage">linkage type</a>, and an
967 optional <a href="#visibility">visibility style</a>.</p>
970 <pre class="doc_code">
971 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
976 <!-- ======================================================================= -->
978 <a name="namedmetadatastructure">Named Metadata</a>
983 <p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
984 nodes</a> (but not metadata strings) are the only valid operands for
985 a named metadata.</p>
988 <pre class="doc_code">
989 ; Some unnamed metadata nodes, which are referenced by the named metadata.
990 !0 = metadata !{metadata !"zero"}
991 !1 = metadata !{metadata !"one"}
992 !2 = metadata !{metadata !"two"}
994 !name = !{!0, !1, !2}
999 <!-- ======================================================================= -->
1001 <a name="paramattrs">Parameter Attributes</a>
1006 <p>The return type and each parameter of a function type may have a set of
1007 <i>parameter attributes</i> associated with them. Parameter attributes are
1008 used to communicate additional information about the result or parameters of
1009 a function. Parameter attributes are considered to be part of the function,
1010 not of the function type, so functions with different parameter attributes
1011 can have the same function type.</p>
1013 <p>Parameter attributes are simple keywords that follow the type specified. If
1014 multiple parameter attributes are needed, they are space separated. For
1017 <pre class="doc_code">
1018 declare i32 @printf(i8* noalias nocapture, ...)
1019 declare i32 @atoi(i8 zeroext)
1020 declare signext i8 @returns_signed_char()
1023 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
1024 <tt>readonly</tt>) come immediately after the argument list.</p>
1026 <p>Currently, only the following parameter attributes are defined:</p>
1029 <dt><tt><b>zeroext</b></tt></dt>
1030 <dd>This indicates to the code generator that the parameter or return value
1031 should be zero-extended to the extent required by the target's ABI (which
1032 is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1033 parameter) or the callee (for a return value).</dd>
1035 <dt><tt><b>signext</b></tt></dt>
1036 <dd>This indicates to the code generator that the parameter or return value
1037 should be sign-extended to the extent required by the target's ABI (which
1038 is usually 32-bits) by the caller (for a parameter) or the callee (for a
1041 <dt><tt><b>inreg</b></tt></dt>
1042 <dd>This indicates that this parameter or return value should be treated in a
1043 special target-dependent fashion during while emitting code for a function
1044 call or return (usually, by putting it in a register as opposed to memory,
1045 though some targets use it to distinguish between two different kinds of
1046 registers). Use of this attribute is target-specific.</dd>
1048 <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1049 <dd><p>This indicates that the pointer parameter should really be passed by
1050 value to the function. The attribute implies that a hidden copy of the
1052 is made between the caller and the callee, so the callee is unable to
1053 modify the value in the callee. This attribute is only valid on LLVM
1054 pointer arguments. It is generally used to pass structs and arrays by
1055 value, but is also valid on pointers to scalars. The copy is considered
1056 to belong to the caller not the callee (for example,
1057 <tt><a href="#readonly">readonly</a></tt> functions should not write to
1058 <tt>byval</tt> parameters). This is not a valid attribute for return
1061 <p>The byval attribute also supports specifying an alignment with
1062 the align attribute. It indicates the alignment of the stack slot to
1063 form and the known alignment of the pointer specified to the call site. If
1064 the alignment is not specified, then the code generator makes a
1065 target-specific assumption.</p></dd>
1067 <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1068 <dd>This indicates that the pointer parameter specifies the address of a
1069 structure that is the return value of the function in the source program.
1070 This pointer must be guaranteed by the caller to be valid: loads and
1071 stores to the structure may be assumed by the callee to not to trap. This
1072 may only be applied to the first parameter. This is not a valid attribute
1073 for return values. </dd>
1075 <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1076 <dd>This indicates that pointer values
1077 <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1078 value do not alias pointer values which are not <i>based</i> on it,
1079 ignoring certain "irrelevant" dependencies.
1080 For a call to the parent function, dependencies between memory
1081 references from before or after the call and from those during the call
1082 are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1083 return value used in that call.
1084 The caller shares the responsibility with the callee for ensuring that
1085 these requirements are met.
1086 For further details, please see the discussion of the NoAlias response in
1087 <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1089 Note that this definition of <tt>noalias</tt> is intentionally
1090 similar to the definition of <tt>restrict</tt> in C99 for function
1091 arguments, though it is slightly weaker.
1093 For function return values, C99's <tt>restrict</tt> is not meaningful,
1094 while LLVM's <tt>noalias</tt> is.
1097 <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1098 <dd>This indicates that the callee does not make any copies of the pointer
1099 that outlive the callee itself. This is not a valid attribute for return
1102 <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1103 <dd>This indicates that the pointer parameter can be excised using the
1104 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1105 attribute for return values.</dd>
1110 <!-- ======================================================================= -->
1112 <a name="gc">Garbage Collector Names</a>
1117 <p>Each function may specify a garbage collector name, which is simply a
1120 <pre class="doc_code">
1121 define void @f() gc "name" { ... }
1124 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1125 collector which will cause the compiler to alter its output in order to
1126 support the named garbage collection algorithm.</p>
1130 <!-- ======================================================================= -->
1132 <a name="fnattrs">Function Attributes</a>
1137 <p>Function attributes are set to communicate additional information about a
1138 function. Function attributes are considered to be part of the function, not
1139 of the function type, so functions with different parameter attributes can
1140 have the same function type.</p>
1142 <p>Function attributes are simple keywords that follow the type specified. If
1143 multiple attributes are needed, they are space separated. For example:</p>
1145 <pre class="doc_code">
1146 define void @f() noinline { ... }
1147 define void @f() alwaysinline { ... }
1148 define void @f() alwaysinline optsize { ... }
1149 define void @f() optsize { ... }
1153 <dt><tt><b>alignstack(<<em>n</em>>)</b></tt></dt>
1154 <dd>This attribute indicates that, when emitting the prologue and epilogue,
1155 the backend should forcibly align the stack pointer. Specify the
1156 desired alignment, which must be a power of two, in parentheses.
1158 <dt><tt><b>alwaysinline</b></tt></dt>
1159 <dd>This attribute indicates that the inliner should attempt to inline this
1160 function into callers whenever possible, ignoring any active inlining size
1161 threshold for this caller.</dd>
1163 <dt><tt><b>hotpatch</b></tt></dt>
1164 <dd>This attribute indicates that the function should be 'hotpatchable',
1165 meaning the function can be patched and/or hooked even while it is
1166 loaded into memory. On x86, the function prologue will be preceded
1167 by six bytes of padding and will begin with a two-byte instruction.
1168 Most of the functions in the Windows system DLLs in Windows XP SP2 or
1169 higher were compiled in this fashion.</dd>
1171 <dt><tt><b>nonlazybind</b></tt></dt>
1172 <dd>This attribute suppresses lazy symbol binding for the function. This
1173 may make calls to the function faster, at the cost of extra program
1174 startup time if the function is not called during program startup.</dd>
1176 <dt><tt><b>inlinehint</b></tt></dt>
1177 <dd>This attribute indicates that the source code contained a hint that inlining
1178 this function is desirable (such as the "inline" keyword in C/C++). It
1179 is just a hint; it imposes no requirements on the inliner.</dd>
1181 <dt><tt><b>naked</b></tt></dt>
1182 <dd>This attribute disables prologue / epilogue emission for the function.
1183 This can have very system-specific consequences.</dd>
1185 <dt><tt><b>noimplicitfloat</b></tt></dt>
1186 <dd>This attributes disables implicit floating point instructions.</dd>
1188 <dt><tt><b>noinline</b></tt></dt>
1189 <dd>This attribute indicates that the inliner should never inline this
1190 function in any situation. This attribute may not be used together with
1191 the <tt>alwaysinline</tt> attribute.</dd>
1193 <dt><tt><b>noredzone</b></tt></dt>
1194 <dd>This attribute indicates that the code generator should not use a red
1195 zone, even if the target-specific ABI normally permits it.</dd>
1197 <dt><tt><b>noreturn</b></tt></dt>
1198 <dd>This function attribute indicates that the function never returns
1199 normally. This produces undefined behavior at runtime if the function
1200 ever does dynamically return.</dd>
1202 <dt><tt><b>nounwind</b></tt></dt>
1203 <dd>This function attribute indicates that the function never returns with an
1204 unwind or exceptional control flow. If the function does unwind, its
1205 runtime behavior is undefined.</dd>
1207 <dt><tt><b>optsize</b></tt></dt>
1208 <dd>This attribute suggests that optimization passes and code generator passes
1209 make choices that keep the code size of this function low, and otherwise
1210 do optimizations specifically to reduce code size.</dd>
1212 <dt><tt><b>readnone</b></tt></dt>
1213 <dd>This attribute indicates that the function computes its result (or decides
1214 to unwind an exception) based strictly on its arguments, without
1215 dereferencing any pointer arguments or otherwise accessing any mutable
1216 state (e.g. memory, control registers, etc) visible to caller functions.
1217 It does not write through any pointer arguments
1218 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1219 changes any state visible to callers. This means that it cannot unwind
1220 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1221 could use the <tt>unwind</tt> instruction.</dd>
1223 <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1224 <dd>This attribute indicates that the function does not write through any
1225 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1226 arguments) or otherwise modify any state (e.g. memory, control registers,
1227 etc) visible to caller functions. It may dereference pointer arguments
1228 and read state that may be set in the caller. A readonly function always
1229 returns the same value (or unwinds an exception identically) when called
1230 with the same set of arguments and global state. It cannot unwind an
1231 exception by calling the <tt>C++</tt> exception throwing methods, but may
1232 use the <tt>unwind</tt> instruction.</dd>
1234 <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1235 <dd>This attribute indicates that the function should emit a stack smashing
1236 protector. It is in the form of a "canary"—a random value placed on
1237 the stack before the local variables that's checked upon return from the
1238 function to see if it has been overwritten. A heuristic is used to
1239 determine if a function needs stack protectors or not.<br>
1241 If a function that has an <tt>ssp</tt> attribute is inlined into a
1242 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1243 function will have an <tt>ssp</tt> attribute.</dd>
1245 <dt><tt><b>sspreq</b></tt></dt>
1246 <dd>This attribute indicates that the function should <em>always</em> emit a
1247 stack smashing protector. This overrides
1248 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1250 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1251 function that doesn't have an <tt>sspreq</tt> attribute or which has
1252 an <tt>ssp</tt> attribute, then the resulting function will have
1253 an <tt>sspreq</tt> attribute.</dd>
1255 <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1256 <dd>This attribute indicates that the ABI being targeted requires that
1257 an unwind table entry be produce for this function even if we can
1258 show that no exceptions passes by it. This is normally the case for
1259 the ELF x86-64 abi, but it can be disabled for some compilation
1266 <!-- ======================================================================= -->
1268 <a name="moduleasm">Module-Level Inline Assembly</a>
1273 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1274 the GCC "file scope inline asm" blocks. These blocks are internally
1275 concatenated by LLVM and treated as a single unit, but may be separated in
1276 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1278 <pre class="doc_code">
1279 module asm "inline asm code goes here"
1280 module asm "more can go here"
1283 <p>The strings can contain any character by escaping non-printable characters.
1284 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1287 <p>The inline asm code is simply printed to the machine code .s file when
1288 assembly code is generated.</p>
1292 <!-- ======================================================================= -->
1294 <a name="datalayout">Data Layout</a>
1299 <p>A module may specify a target specific data layout string that specifies how
1300 data is to be laid out in memory. The syntax for the data layout is
1303 <pre class="doc_code">
1304 target datalayout = "<i>layout specification</i>"
1307 <p>The <i>layout specification</i> consists of a list of specifications
1308 separated by the minus sign character ('-'). Each specification starts with
1309 a letter and may include other information after the letter to define some
1310 aspect of the data layout. The specifications accepted are as follows:</p>
1314 <dd>Specifies that the target lays out data in big-endian form. That is, the
1315 bits with the most significance have the lowest address location.</dd>
1318 <dd>Specifies that the target lays out data in little-endian form. That is,
1319 the bits with the least significance have the lowest address
1322 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1323 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1324 <i>preferred</i> alignments. All sizes are in bits. Specifying
1325 the <i>pref</i> alignment is optional. If omitted, the
1326 preceding <tt>:</tt> should be omitted too.</dd>
1328 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1329 <dd>This specifies the alignment for an integer type of a given bit
1330 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1332 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1333 <dd>This specifies the alignment for a vector type of a given bit
1336 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1337 <dd>This specifies the alignment for a floating point type of a given bit
1338 <i>size</i>. Only values of <i>size</i> that are supported by the target
1339 will work. 32 (float) and 64 (double) are supported on all targets;
1340 80 or 128 (different flavors of long double) are also supported on some
1343 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1344 <dd>This specifies the alignment for an aggregate type of a given bit
1347 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1348 <dd>This specifies the alignment for a stack object of a given bit
1351 <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1352 <dd>This specifies a set of native integer widths for the target CPU
1353 in bits. For example, it might contain "n32" for 32-bit PowerPC,
1354 "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64. Elements of
1355 this set are considered to support most general arithmetic
1356 operations efficiently.</dd>
1359 <p>When constructing the data layout for a given target, LLVM starts with a
1360 default set of specifications which are then (possibly) overridden by the
1361 specifications in the <tt>datalayout</tt> keyword. The default specifications
1362 are given in this list:</p>
1365 <li><tt>E</tt> - big endian</li>
1366 <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1367 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1368 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1369 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1370 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1371 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1372 alignment of 64-bits</li>
1373 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1374 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1375 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1376 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1377 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1378 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1381 <p>When LLVM is determining the alignment for a given type, it uses the
1382 following rules:</p>
1385 <li>If the type sought is an exact match for one of the specifications, that
1386 specification is used.</li>
1388 <li>If no match is found, and the type sought is an integer type, then the
1389 smallest integer type that is larger than the bitwidth of the sought type
1390 is used. If none of the specifications are larger than the bitwidth then
1391 the the largest integer type is used. For example, given the default
1392 specifications above, the i7 type will use the alignment of i8 (next
1393 largest) while both i65 and i256 will use the alignment of i64 (largest
1396 <li>If no match is found, and the type sought is a vector type, then the
1397 largest vector type that is smaller than the sought vector type will be
1398 used as a fall back. This happens because <128 x double> can be
1399 implemented in terms of 64 <2 x double>, for example.</li>
1404 <!-- ======================================================================= -->
1406 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1411 <p>Any memory access must be done through a pointer value associated
1412 with an address range of the memory access, otherwise the behavior
1413 is undefined. Pointer values are associated with address ranges
1414 according to the following rules:</p>
1417 <li>A pointer value is associated with the addresses associated with
1418 any value it is <i>based</i> on.
1419 <li>An address of a global variable is associated with the address
1420 range of the variable's storage.</li>
1421 <li>The result value of an allocation instruction is associated with
1422 the address range of the allocated storage.</li>
1423 <li>A null pointer in the default address-space is associated with
1425 <li>An integer constant other than zero or a pointer value returned
1426 from a function not defined within LLVM may be associated with address
1427 ranges allocated through mechanisms other than those provided by
1428 LLVM. Such ranges shall not overlap with any ranges of addresses
1429 allocated by mechanisms provided by LLVM.</li>
1432 <p>A pointer value is <i>based</i> on another pointer value according
1433 to the following rules:</p>
1436 <li>A pointer value formed from a
1437 <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1438 is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1439 <li>The result value of a
1440 <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1441 of the <tt>bitcast</tt>.</li>
1442 <li>A pointer value formed by an
1443 <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1444 pointer values that contribute (directly or indirectly) to the
1445 computation of the pointer's value.</li>
1446 <li>The "<i>based</i> on" relationship is transitive.</li>
1449 <p>Note that this definition of <i>"based"</i> is intentionally
1450 similar to the definition of <i>"based"</i> in C99, though it is
1451 slightly weaker.</p>
1453 <p>LLVM IR does not associate types with memory. The result type of a
1454 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1455 alignment of the memory from which to load, as well as the
1456 interpretation of the value. The first operand type of a
1457 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1458 and alignment of the store.</p>
1460 <p>Consequently, type-based alias analysis, aka TBAA, aka
1461 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1462 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1463 additional information which specialized optimization passes may use
1464 to implement type-based alias analysis.</p>
1468 <!-- ======================================================================= -->
1470 <a name="volatile">Volatile Memory Accesses</a>
1475 <p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1476 href="#i_store"><tt>store</tt></a>s, and <a
1477 href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1478 The optimizers must not change the number of volatile operations or change their
1479 order of execution relative to other volatile operations. The optimizers
1480 <i>may</i> change the order of volatile operations relative to non-volatile
1481 operations. This is not Java's "volatile" and has no cross-thread
1482 synchronization behavior.</p>
1486 <!-- ======================================================================= -->
1488 <a name="memmodel">Memory Model for Concurrent Operations</a>
1493 <p>The LLVM IR does not define any way to start parallel threads of execution
1494 or to register signal handlers. Nonetheless, there are platform-specific
1495 ways to create them, and we define LLVM IR's behavior in their presence. This
1496 model is inspired by the C++0x memory model.</p>
1498 <p>We define a <i>happens-before</i> partial order as the least partial order
1501 <li>Is a superset of single-thread program order, and</li>
1502 <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1503 <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1504 by platform-specific techniques, like pthread locks, thread
1505 creation, thread joining, etc., and by atomic instructions.
1506 (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1510 <p>Note that program order does not introduce <i>happens-before</i> edges
1511 between a thread and signals executing inside that thread.</p>
1513 <p>Every (defined) read operation (load instructions, memcpy, atomic
1514 loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1515 (defined) write operations (store instructions, atomic
1516 stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1517 initialized globals are considered to have a write of the initializer which is
1518 atomic and happens before any other read or write of the memory in question.
1519 For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1520 any write to the same byte, except:</p>
1523 <li>If <var>write<sub>1</sub></var> happens before
1524 <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1525 before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1526 does not see <var>write<sub>1</sub></var>.
1527 <li>If <var>R<sub>byte</sub></var> happens before <var>write<sub>3</var>,
1528 then <var>R<sub>byte</sub></var> does not see
1529 <var>write<sub>3</sub></var>.
1532 <p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1534 <li>If there is no write to the same byte that happens before
1535 <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1536 <tt>undef</tt> for that byte.
1537 <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1538 <var>R<sub>byte</sub></var> returns the value written by that
1540 <li>Otherwise, if <var>R</var> is atomic, and all the writes
1541 <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1542 values written. See the <a href="#ordering">Atomic Memory Ordering
1543 Constraints</a> section for additional constraints on how the choice
1545 <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1548 <p><var>R</var> returns the value composed of the series of bytes it read.
1549 This implies that some bytes within the value may be <tt>undef</tt>
1550 <b>without</b> the entire value being <tt>undef</tt>. Note that this only
1551 defines the semantics of the operation; it doesn't mean that targets will
1552 emit more than one instruction to read the series of bytes.</p>
1554 <p>Note that in cases where none of the atomic intrinsics are used, this model
1555 places only one restriction on IR transformations on top of what is required
1556 for single-threaded execution: introducing a store to a byte which might not
1557 otherwise be stored to can introduce undefined behavior. (Specifically, in
1558 the case where another thread might write to and read from an address,
1559 introducing a store can change a load that may see exactly one write into
1560 a load that may see multiple writes.)</p>
1562 <!-- FIXME: This model assumes all targets where concurrency is relevant have
1563 a byte-size store which doesn't affect adjacent bytes. As far as I can tell,
1564 none of the backends currently in the tree fall into this category; however,
1565 there might be targets which care. If there are, we want a paragraph
1568 Targets may specify that stores narrower than a certain width are not
1569 available; on such a target, for the purposes of this model, treat any
1570 non-atomic write with an alignment or width less than the minimum width
1571 as if it writes to the relevant surrounding bytes.
1576 <!-- ======================================================================= -->
1577 <div class="doc_subsection">
1578 <a name="ordering">Atomic Memory Ordering Constraints</a>
1581 <div class="doc_text">
1583 <p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1584 <a href="#i_atomicrmw"><code>atomicrmw</code></a>, and
1585 <a href="#i_fence"><code>fence</code></a>) take an ordering parameter
1586 that determines which other atomic instructions on the same address they
1587 <i>synchronize with</i>. These semantics are borrowed from Java and C++0x,
1588 but are somewhat more colloquial. If these descriptions aren't precise enough,
1589 check those specs. <a href="#i_fence"><code>fence</code></a> instructions
1590 treat these orderings somewhat differently since they don't take an address.
1591 See that instruction's documentation for details.</p>
1593 <!-- FIXME Note atomic load+store here once those get added. -->
1596 <!-- FIXME: unordered is intended to be used for atomic load and store;
1597 it isn't allowed for any instruction yet. -->
1598 <dt><code>unordered</code></dt>
1599 <dd>The set of values that can be read is governed by the happens-before
1600 partial order. A value cannot be read unless some operation wrote it.
1601 This is intended to provide a guarantee strong enough to model Java's
1602 non-volatile shared variables. This ordering cannot be specified for
1603 read-modify-write operations; it is not strong enough to make them atomic
1604 in any interesting way.</dd>
1605 <dt><code>monotonic</code></dt>
1606 <dd>In addition to the guarantees of <code>unordered</code>, there is a single
1607 total order for modifications by <code>monotonic</code> operations on each
1608 address. All modification orders must be compatible with the happens-before
1609 order. There is no guarantee that the modification orders can be combined to
1610 a global total order for the whole program (and this often will not be
1611 possible). The read in an atomic read-modify-write operation
1612 (<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1613 <a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1614 reads the value in the modification order immediately before the value it
1615 writes. If one atomic read happens before another atomic read of the same
1616 address, the later read must see the same value or a later value in the
1617 address's modification order. This disallows reordering of
1618 <code>monotonic</code> (or stronger) operations on the same address. If an
1619 address is written <code>monotonic</code>ally by one thread, and other threads
1620 <code>monotonic</code>ally read that address repeatedly, the other threads must
1621 eventually see the write. This is intended to model C++'s relaxed atomic
1623 <dt><code>acquire</code></dt>
1624 <dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1625 reads a value written by a <code>release</code> atomic operation, it
1626 <i>synchronizes-with</i> that operation.</dd>
1627 <dt><code>release</code></dt>
1628 <dd>In addition to the guarantees of <code>monotonic</code>,
1629 a <i>synchronizes-with</i> edge may be formed by an <code>acquire</code>
1631 <dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1632 <code>acquire</code> and <code>release</code> operation on its address.</dd>
1633 <dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1634 <dd>In addition to the guarantees of <code>acq_rel</code>
1635 (<code>acquire</code> for an operation which only reads, <code>release</code>
1636 for an operation which only writes), there is a global total order on all
1637 sequentially-consistent operations on all addresses, which is consistent with
1638 the <i>happens-before</i> partial order and with the modification orders of
1639 all the affected addresses. Each sequentially-consistent read sees the last
1640 preceding write to the same address in this global order. This is intended
1641 to model C++'s sequentially-consistent atomic variables and Java's volatile
1642 shared variables.</dd>
1645 <p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1646 it only <i>synchronizes with</i> or participates in modification and seq_cst
1647 total orderings with other operations running in the same thread (for example,
1648 in signal handlers).</p>
1654 <!-- *********************************************************************** -->
1655 <h2><a name="typesystem">Type System</a></h2>
1656 <!-- *********************************************************************** -->
1660 <p>The LLVM type system is one of the most important features of the
1661 intermediate representation. Being typed enables a number of optimizations
1662 to be performed on the intermediate representation directly, without having
1663 to do extra analyses on the side before the transformation. A strong type
1664 system makes it easier to read the generated code and enables novel analyses
1665 and transformations that are not feasible to perform on normal three address
1666 code representations.</p>
1668 <!-- ======================================================================= -->
1670 <a name="t_classifications">Type Classifications</a>
1675 <p>The types fall into a few useful classifications:</p>
1677 <table border="1" cellspacing="0" cellpadding="4">
1679 <tr><th>Classification</th><th>Types</th></tr>
1681 <td><a href="#t_integer">integer</a></td>
1682 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1685 <td><a href="#t_floating">floating point</a></td>
1686 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1689 <td><a name="t_firstclass">first class</a></td>
1690 <td><a href="#t_integer">integer</a>,
1691 <a href="#t_floating">floating point</a>,
1692 <a href="#t_pointer">pointer</a>,
1693 <a href="#t_vector">vector</a>,
1694 <a href="#t_struct">structure</a>,
1695 <a href="#t_array">array</a>,
1696 <a href="#t_label">label</a>,
1697 <a href="#t_metadata">metadata</a>.
1701 <td><a href="#t_primitive">primitive</a></td>
1702 <td><a href="#t_label">label</a>,
1703 <a href="#t_void">void</a>,
1704 <a href="#t_integer">integer</a>,
1705 <a href="#t_floating">floating point</a>,
1706 <a href="#t_x86mmx">x86mmx</a>,
1707 <a href="#t_metadata">metadata</a>.</td>
1710 <td><a href="#t_derived">derived</a></td>
1711 <td><a href="#t_array">array</a>,
1712 <a href="#t_function">function</a>,
1713 <a href="#t_pointer">pointer</a>,
1714 <a href="#t_struct">structure</a>,
1715 <a href="#t_vector">vector</a>,
1716 <a href="#t_opaque">opaque</a>.
1722 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1723 important. Values of these types are the only ones which can be produced by
1728 <!-- ======================================================================= -->
1730 <a name="t_primitive">Primitive Types</a>
1735 <p>The primitive types are the fundamental building blocks of the LLVM
1738 <!-- _______________________________________________________________________ -->
1740 <a name="t_integer">Integer Type</a>
1746 <p>The integer type is a very simple type that simply specifies an arbitrary
1747 bit width for the integer type desired. Any bit width from 1 bit to
1748 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1755 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1759 <table class="layout">
1761 <td class="left"><tt>i1</tt></td>
1762 <td class="left">a single-bit integer.</td>
1765 <td class="left"><tt>i32</tt></td>
1766 <td class="left">a 32-bit integer.</td>
1769 <td class="left"><tt>i1942652</tt></td>
1770 <td class="left">a really big integer of over 1 million bits.</td>
1776 <!-- _______________________________________________________________________ -->
1778 <a name="t_floating">Floating Point Types</a>
1785 <tr><th>Type</th><th>Description</th></tr>
1786 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1787 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1788 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1789 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1790 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1796 <!-- _______________________________________________________________________ -->
1798 <a name="t_x86mmx">X86mmx Type</a>
1804 <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>
1813 <!-- _______________________________________________________________________ -->
1815 <a name="t_void">Void Type</a>
1821 <p>The void type does not represent any value and has no size.</p>
1830 <!-- _______________________________________________________________________ -->
1832 <a name="t_label">Label Type</a>
1838 <p>The label type represents code labels.</p>
1847 <!-- _______________________________________________________________________ -->
1849 <a name="t_metadata">Metadata Type</a>
1855 <p>The metadata type represents embedded metadata. No derived types may be
1856 created from metadata except for <a href="#t_function">function</a>
1868 <!-- ======================================================================= -->
1870 <a name="t_derived">Derived Types</a>
1875 <p>The real power in LLVM comes from the derived types in the system. This is
1876 what allows a programmer to represent arrays, functions, pointers, and other
1877 useful types. Each of these types contain one or more element types which
1878 may be a primitive type, or another derived type. For example, it is
1879 possible to have a two dimensional array, using an array as the element type
1880 of another array.</p>
1885 <!-- _______________________________________________________________________ -->
1887 <a name="t_aggregate">Aggregate Types</a>
1892 <p>Aggregate Types are a subset of derived types that can contain multiple
1893 member types. <a href="#t_array">Arrays</a>,
1894 <a href="#t_struct">structs</a>, and <a href="#t_vector">vectors</a> are
1895 aggregate types.</p>
1899 <!-- _______________________________________________________________________ -->
1901 <a name="t_array">Array Type</a>
1907 <p>The array type is a very simple derived type that arranges elements
1908 sequentially in memory. The array type requires a size (number of elements)
1909 and an underlying data type.</p>
1913 [<# elements> x <elementtype>]
1916 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1917 be any type with a size.</p>
1920 <table class="layout">
1922 <td class="left"><tt>[40 x i32]</tt></td>
1923 <td class="left">Array of 40 32-bit integer values.</td>
1926 <td class="left"><tt>[41 x i32]</tt></td>
1927 <td class="left">Array of 41 32-bit integer values.</td>
1930 <td class="left"><tt>[4 x i8]</tt></td>
1931 <td class="left">Array of 4 8-bit integer values.</td>
1934 <p>Here are some examples of multidimensional arrays:</p>
1935 <table class="layout">
1937 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1938 <td class="left">3x4 array of 32-bit integer values.</td>
1941 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1942 <td class="left">12x10 array of single precision floating point values.</td>
1945 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1946 <td class="left">2x3x4 array of 16-bit integer values.</td>
1950 <p>There is no restriction on indexing beyond the end of the array implied by
1951 a static type (though there are restrictions on indexing beyond the bounds
1952 of an allocated object in some cases). This means that single-dimension
1953 'variable sized array' addressing can be implemented in LLVM with a zero
1954 length array type. An implementation of 'pascal style arrays' in LLVM could
1955 use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1959 <!-- _______________________________________________________________________ -->
1961 <a name="t_function">Function Type</a>
1967 <p>The function type can be thought of as a function signature. It consists of
1968 a return type and a list of formal parameter types. The return type of a
1969 function type is a first class type or a void type.</p>
1973 <returntype> (<parameter list>)
1976 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1977 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1978 which indicates that the function takes a variable number of arguments.
1979 Variable argument functions can access their arguments with
1980 the <a href="#int_varargs">variable argument handling intrinsic</a>
1981 functions. '<tt><returntype></tt>' is any type except
1982 <a href="#t_label">label</a>.</p>
1985 <table class="layout">
1987 <td class="left"><tt>i32 (i32)</tt></td>
1988 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1990 </tr><tr class="layout">
1991 <td class="left"><tt>float (i16, i32 *) *
1993 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1994 an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
1995 returning <tt>float</tt>.
1997 </tr><tr class="layout">
1998 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1999 <td class="left">A vararg function that takes at least one
2000 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2001 which returns an integer. This is the signature for <tt>printf</tt> in
2004 </tr><tr class="layout">
2005 <td class="left"><tt>{i32, i32} (i32)</tt></td>
2006 <td class="left">A function taking an <tt>i32</tt>, returning a
2007 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2014 <!-- _______________________________________________________________________ -->
2016 <a name="t_struct">Structure Type</a>
2022 <p>The structure type is used to represent a collection of data members together
2023 in memory. The elements of a structure may be any type that has a size.</p>
2025 <p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2026 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2027 with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2028 Structures in registers are accessed using the
2029 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2030 '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2032 <p>Structures may optionally be "packed" structures, which indicate that the
2033 alignment of the struct is one byte, and that there is no padding between
2034 the elements. In non-packed structs, padding between field types is defined
2035 by the target data string to match the underlying processor.</p>
2037 <p>Structures can either be "anonymous" or "named". An anonymous structure is
2038 defined inline with other types (e.g. <tt>{i32, i32}*</tt>) and a named types
2039 are always defined at the top level with a name. Anonmyous types are uniqued
2040 by their contents and can never be recursive since there is no way to write
2041 one. Named types can be recursive.
2046 %T1 = type { <type list> } <i>; Named normal struct type</i>
2047 %T2 = type <{ <type list> }> <i>; Named packed struct type</i>
2051 <table class="layout">
2053 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2054 <td class="left">A triple of three <tt>i32</tt> values</td>
2057 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
2058 <td class="left">A pair, where the first element is a <tt>float</tt> and the
2059 second element is a <a href="#t_pointer">pointer</a> to a
2060 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2061 an <tt>i32</tt>.</td>
2064 <td class="left"><tt><{ i8, i32 }></tt></td>
2065 <td class="left">A packed struct known to be 5 bytes in size.</td>
2071 <!-- _______________________________________________________________________ -->
2073 <a name="t_opaque">Opaque Structure Types</a>
2079 <p>Opaque structure types are used to represent named structure types that do
2080 not have a body specified. This corresponds (for example) to the C notion of
2081 a forward declared structure.</p>
2090 <table class="layout">
2092 <td class="left"><tt>opaque</tt></td>
2093 <td class="left">An opaque type.</td>
2101 <!-- _______________________________________________________________________ -->
2103 <a name="t_pointer">Pointer Type</a>
2109 <p>The pointer type is used to specify memory locations.
2110 Pointers are commonly used to reference objects in memory.</p>
2112 <p>Pointer types may have an optional address space attribute defining the
2113 numbered address space where the pointed-to object resides. The default
2114 address space is number zero. The semantics of non-zero address
2115 spaces are target-specific.</p>
2117 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2118 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
2126 <table class="layout">
2128 <td class="left"><tt>[4 x i32]*</tt></td>
2129 <td class="left">A <a href="#t_pointer">pointer</a> to <a
2130 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2133 <td class="left"><tt>i32 (i32*) *</tt></td>
2134 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2135 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2139 <td class="left"><tt>i32 addrspace(5)*</tt></td>
2140 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2141 that resides in address space #5.</td>
2147 <!-- _______________________________________________________________________ -->
2149 <a name="t_vector">Vector Type</a>
2155 <p>A vector type is a simple derived type that represents a vector of elements.
2156 Vector types are used when multiple primitive data are operated in parallel
2157 using a single instruction (SIMD). A vector type requires a size (number of
2158 elements) and an underlying primitive data type. Vector types are considered
2159 <a href="#t_firstclass">first class</a>.</p>
2163 < <# elements> x <elementtype> >
2166 <p>The number of elements is a constant integer value larger than 0; elementtype
2167 may be any integer or floating point type. Vectors of size zero are not
2168 allowed, and pointers are not allowed as the element type.</p>
2171 <table class="layout">
2173 <td class="left"><tt><4 x i32></tt></td>
2174 <td class="left">Vector of 4 32-bit integer values.</td>
2177 <td class="left"><tt><8 x float></tt></td>
2178 <td class="left">Vector of 8 32-bit floating-point values.</td>
2181 <td class="left"><tt><2 x i64></tt></td>
2182 <td class="left">Vector of 2 64-bit integer values.</td>
2188 <!-- *********************************************************************** -->
2189 <h2><a name="constants">Constants</a></h2>
2190 <!-- *********************************************************************** -->
2194 <p>LLVM has several different basic types of constants. This section describes
2195 them all and their syntax.</p>
2197 <!-- ======================================================================= -->
2199 <a name="simpleconstants">Simple Constants</a>
2205 <dt><b>Boolean constants</b></dt>
2206 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2207 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2209 <dt><b>Integer constants</b></dt>
2210 <dd>Standard integers (such as '4') are constants of
2211 the <a href="#t_integer">integer</a> type. Negative numbers may be used
2212 with integer types.</dd>
2214 <dt><b>Floating point constants</b></dt>
2215 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2216 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2217 notation (see below). The assembler requires the exact decimal value of a
2218 floating-point constant. For example, the assembler accepts 1.25 but
2219 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
2220 constants must have a <a href="#t_floating">floating point</a> type. </dd>
2222 <dt><b>Null pointer constants</b></dt>
2223 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2224 and must be of <a href="#t_pointer">pointer type</a>.</dd>
2227 <p>The one non-intuitive notation for constants is the hexadecimal form of
2228 floating point constants. For example, the form '<tt>double
2229 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2230 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
2231 constants are required (and the only time that they are generated by the
2232 disassembler) is when a floating point constant must be emitted but it cannot
2233 be represented as a decimal floating point number in a reasonable number of
2234 digits. For example, NaN's, infinities, and other special values are
2235 represented in their IEEE hexadecimal format so that assembly and disassembly
2236 do not cause any bits to change in the constants.</p>
2238 <p>When using the hexadecimal form, constants of types float and double are
2239 represented using the 16-digit form shown above (which matches the IEEE754
2240 representation for double); float values must, however, be exactly
2241 representable as IEE754 single precision. Hexadecimal format is always used
2242 for long double, and there are three forms of long double. The 80-bit format
2243 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2244 The 128-bit format used by PowerPC (two adjacent doubles) is represented
2245 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
2246 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2247 currently supported target uses this format. Long doubles will only work if
2248 they match the long double format on your target. All hexadecimal formats
2249 are big-endian (sign bit at the left).</p>
2251 <p>There are no constants of type x86mmx.</p>
2254 <!-- ======================================================================= -->
2256 <a name="aggregateconstants"></a> <!-- old anchor -->
2257 <a name="complexconstants">Complex Constants</a>
2262 <p>Complex constants are a (potentially recursive) combination of simple
2263 constants and smaller complex constants.</p>
2266 <dt><b>Structure constants</b></dt>
2267 <dd>Structure constants are represented with notation similar to structure
2268 type definitions (a comma separated list of elements, surrounded by braces
2269 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2270 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2271 Structure constants must have <a href="#t_struct">structure type</a>, and
2272 the number and types of elements must match those specified by the
2275 <dt><b>Array constants</b></dt>
2276 <dd>Array constants are represented with notation similar to array type
2277 definitions (a comma separated list of elements, surrounded by square
2278 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
2279 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
2280 the number and types of elements must match those specified by the
2283 <dt><b>Vector constants</b></dt>
2284 <dd>Vector constants are represented with notation similar to vector type
2285 definitions (a comma separated list of elements, surrounded by
2286 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
2287 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
2288 have <a href="#t_vector">vector type</a>, and the number and types of
2289 elements must match those specified by the type.</dd>
2291 <dt><b>Zero initialization</b></dt>
2292 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2293 value to zero of <em>any</em> type, including scalar and
2294 <a href="#t_aggregate">aggregate</a> types.
2295 This is often used to avoid having to print large zero initializers
2296 (e.g. for large arrays) and is always exactly equivalent to using explicit
2297 zero initializers.</dd>
2299 <dt><b>Metadata node</b></dt>
2300 <dd>A metadata node is a structure-like constant with
2301 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
2302 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
2303 be interpreted as part of the instruction stream, metadata is a place to
2304 attach additional information such as debug info.</dd>
2309 <!-- ======================================================================= -->
2311 <a name="globalconstants">Global Variable and Function Addresses</a>
2316 <p>The addresses of <a href="#globalvars">global variables</a>
2317 and <a href="#functionstructure">functions</a> are always implicitly valid
2318 (link-time) constants. These constants are explicitly referenced when
2319 the <a href="#identifiers">identifier for the global</a> is used and always
2320 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2321 legal LLVM file:</p>
2323 <pre class="doc_code">
2326 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2331 <!-- ======================================================================= -->
2333 <a name="undefvalues">Undefined Values</a>
2338 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2339 indicates that the user of the value may receive an unspecified bit-pattern.
2340 Undefined values may be of any type (other than '<tt>label</tt>'
2341 or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2343 <p>Undefined values are useful because they indicate to the compiler that the
2344 program is well defined no matter what value is used. This gives the
2345 compiler more freedom to optimize. Here are some examples of (potentially
2346 surprising) transformations that are valid (in pseudo IR):</p>
2349 <pre class="doc_code">
2359 <p>This is safe because all of the output bits are affected by the undef bits.
2360 Any output bit can have a zero or one depending on the input bits.</p>
2362 <pre class="doc_code">
2373 <p>These logical operations have bits that are not always affected by the input.
2374 For example, if <tt>%X</tt> has a zero bit, then the output of the
2375 '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2376 the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2377 optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2378 However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2379 0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2380 all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2381 set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2383 <pre class="doc_code">
2384 %A = select undef, %X, %Y
2385 %B = select undef, 42, %Y
2386 %C = select %X, %Y, undef
2397 <p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2398 branch) conditions can go <em>either way</em>, but they have to come from one
2399 of the two operands. In the <tt>%A</tt> example, if <tt>%X</tt> and
2400 <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2401 have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2402 optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2403 same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2406 <pre class="doc_code">
2407 %A = xor undef, undef
2425 <p>This example points out that two '<tt>undef</tt>' operands are not
2426 necessarily the same. This can be surprising to people (and also matches C
2427 semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2428 if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2429 short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2430 its value over its "live range". This is true because the variable doesn't
2431 actually <em>have a live range</em>. Instead, the value is logically read
2432 from arbitrary registers that happen to be around when needed, so the value
2433 is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2434 need to have the same semantics or the core LLVM "replace all uses with"
2435 concept would not hold.</p>
2437 <pre class="doc_code">
2445 <p>These examples show the crucial difference between an <em>undefined
2446 value</em> and <em>undefined behavior</em>. An undefined value (like
2447 '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2448 the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2449 the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2450 defined on SNaN's. However, in the second example, we can make a more
2451 aggressive assumption: because the <tt>undef</tt> is allowed to be an
2452 arbitrary value, we are allowed to assume that it could be zero. Since a
2453 divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2454 the operation does not execute at all. This allows us to delete the divide and
2455 all code after it. Because the undefined operation "can't happen", the
2456 optimizer can assume that it occurs in dead code.</p>
2458 <pre class="doc_code">
2459 a: store undef -> %X
2460 b: store %X -> undef
2466 <p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2467 undefined value can be assumed to not have any effect; we can assume that the
2468 value is overwritten with bits that happen to match what was already there.
2469 However, a store <em>to</em> an undefined location could clobber arbitrary
2470 memory, therefore, it has undefined behavior.</p>
2474 <!-- ======================================================================= -->
2476 <a name="trapvalues">Trap Values</a>
2481 <p>Trap values are similar to <a href="#undefvalues">undef values</a>, however
2482 instead of representing an unspecified bit pattern, they represent the
2483 fact that an instruction or constant expression which cannot evoke side
2484 effects has nevertheless detected a condition which results in undefined
2487 <p>There is currently no way of representing a trap value in the IR; they
2488 only exist when produced by operations such as
2489 <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2491 <p>Trap value behavior is defined in terms of value <i>dependence</i>:</p>
2494 <li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2495 their operands.</li>
2497 <li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2498 to their dynamic predecessor basic block.</li>
2500 <li>Function arguments depend on the corresponding actual argument values in
2501 the dynamic callers of their functions.</li>
2503 <li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2504 <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2505 control back to them.</li>
2507 <li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2508 <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_unwind"><tt>unwind</tt></a>,
2509 or exception-throwing call instructions that dynamically transfer control
2512 <li>Non-volatile loads and stores depend on the most recent stores to all of the
2513 referenced memory addresses, following the order in the IR
2514 (including loads and stores implied by intrinsics such as
2515 <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2517 <!-- TODO: In the case of multiple threads, this only applies if the store
2518 "happens-before" the load or store. -->
2520 <!-- TODO: floating-point exception state -->
2522 <li>An instruction with externally visible side effects depends on the most
2523 recent preceding instruction with externally visible side effects, following
2524 the order in the IR. (This includes
2525 <a href="#volatile">volatile operations</a>.)</li>
2527 <li>An instruction <i>control-depends</i> on a
2528 <a href="#terminators">terminator instruction</a>
2529 if the terminator instruction has multiple successors and the instruction
2530 is always executed when control transfers to one of the successors, and
2531 may not be executed when control is transferred to another.</li>
2533 <li>Additionally, an instruction also <i>control-depends</i> on a terminator
2534 instruction if the set of instructions it otherwise depends on would be
2535 different if the terminator had transferred control to a different
2538 <li>Dependence is transitive.</li>
2542 <p>Whenever a trap value is generated, all values which depend on it evaluate
2543 to trap. If they have side effects, the evoke their side effects as if each
2544 operand with a trap value were undef. If they have externally-visible side
2545 effects, the behavior is undefined.</p>
2547 <p>Here are some examples:</p>
2549 <pre class="doc_code">
2551 %trap = sub nuw i32 0, 1 ; Results in a trap value.
2552 %still_trap = and i32 %trap, 0 ; Whereas (and i32 undef, 0) would return 0.
2553 %trap_yet_again = getelementptr i32* @h, i32 %still_trap
2554 store i32 0, i32* %trap_yet_again ; undefined behavior
2556 store i32 %trap, i32* @g ; Trap value conceptually stored to memory.
2557 %trap2 = load i32* @g ; Returns a trap value, not just undef.
2559 volatile store i32 %trap, i32* @g ; External observation; undefined behavior.
2561 %narrowaddr = bitcast i32* @g to i16*
2562 %wideaddr = bitcast i32* @g to i64*
2563 %trap3 = load i16* %narrowaddr ; Returns a trap value.
2564 %trap4 = load i64* %wideaddr ; Returns a trap value.
2566 %cmp = icmp slt i32 %trap, 0 ; Returns a trap value.
2567 br i1 %cmp, label %true, label %end ; Branch to either destination.
2570 volatile store i32 0, i32* @g ; This is control-dependent on %cmp, so
2571 ; it has undefined behavior.
2575 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2576 ; Both edges into this PHI are
2577 ; control-dependent on %cmp, so this
2578 ; always results in a trap value.
2580 volatile store i32 0, i32* @g ; This would depend on the store in %true
2581 ; if %cmp is true, or the store in %entry
2582 ; otherwise, so this is undefined behavior.
2584 br i1 %cmp, label %second_true, label %second_end
2585 ; The same branch again, but this time the
2586 ; true block doesn't have side effects.
2593 volatile store i32 0, i32* @g ; This time, the instruction always depends
2594 ; on the store in %end. Also, it is
2595 ; control-equivalent to %end, so this is
2596 ; well-defined (again, ignoring earlier
2597 ; undefined behavior in this example).
2602 <!-- ======================================================================= -->
2604 <a name="blockaddress">Addresses of Basic Blocks</a>
2609 <p><b><tt>blockaddress(@function, %block)</tt></b></p>
2611 <p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2612 basic block in the specified function, and always has an i8* type. Taking
2613 the address of the entry block is illegal.</p>
2615 <p>This value only has defined behavior when used as an operand to the
2616 '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2617 comparisons against null. Pointer equality tests between labels addresses
2618 results in undefined behavior — though, again, comparison against null
2619 is ok, and no label is equal to the null pointer. This may be passed around
2620 as an opaque pointer sized value as long as the bits are not inspected. This
2621 allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2622 long as the original value is reconstituted before the <tt>indirectbr</tt>
2625 <p>Finally, some targets may provide defined semantics when using the value as
2626 the operand to an inline assembly, but that is target specific.</p>
2631 <!-- ======================================================================= -->
2633 <a name="constantexprs">Constant Expressions</a>
2638 <p>Constant expressions are used to allow expressions involving other constants
2639 to be used as constants. Constant expressions may be of
2640 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2641 operation that does not have side effects (e.g. load and call are not
2642 supported). The following is the syntax for constant expressions:</p>
2645 <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2646 <dd>Truncate a constant to another type. The bit size of CST must be larger
2647 than the bit size of TYPE. Both types must be integers.</dd>
2649 <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2650 <dd>Zero extend a constant to another type. The bit size of CST must be
2651 smaller than the bit size of TYPE. Both types must be integers.</dd>
2653 <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2654 <dd>Sign extend a constant to another type. The bit size of CST must be
2655 smaller than the bit size of TYPE. Both types must be integers.</dd>
2657 <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2658 <dd>Truncate a floating point constant to another floating point type. The
2659 size of CST must be larger than the size of TYPE. Both types must be
2660 floating point.</dd>
2662 <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2663 <dd>Floating point extend a constant to another type. The size of CST must be
2664 smaller or equal to the size of TYPE. Both types must be floating
2667 <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2668 <dd>Convert a floating point constant to the corresponding unsigned integer
2669 constant. TYPE must be a scalar or vector integer type. CST must be of
2670 scalar or vector floating point type. Both CST and TYPE must be scalars,
2671 or vectors of the same number of elements. If the value won't fit in the
2672 integer type, the results are undefined.</dd>
2674 <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2675 <dd>Convert a floating point constant to the corresponding signed integer
2676 constant. TYPE must be a scalar or vector integer type. CST must be of
2677 scalar or vector floating point type. Both CST and TYPE must be scalars,
2678 or vectors of the same number of elements. If the value won't fit in the
2679 integer type, the results are undefined.</dd>
2681 <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2682 <dd>Convert an unsigned integer constant to the corresponding floating point
2683 constant. TYPE must be a scalar or vector floating point type. CST must be
2684 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2685 vectors of the same number of elements. If the value won't fit in the
2686 floating point type, the results are undefined.</dd>
2688 <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2689 <dd>Convert a signed integer constant to the corresponding floating point
2690 constant. TYPE must be a scalar or vector floating point type. CST must be
2691 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2692 vectors of the same number of elements. If the value won't fit in the
2693 floating point type, the results are undefined.</dd>
2695 <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2696 <dd>Convert a pointer typed constant to the corresponding integer constant
2697 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2698 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2699 make it fit in <tt>TYPE</tt>.</dd>
2701 <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2702 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2703 type. CST must be of integer type. The CST value is zero extended,
2704 truncated, or unchanged to make it fit in a pointer size. This one is
2705 <i>really</i> dangerous!</dd>
2707 <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2708 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2709 are the same as those for the <a href="#i_bitcast">bitcast
2710 instruction</a>.</dd>
2712 <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2713 <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2714 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2715 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2716 instruction, the index list may have zero or more indexes, which are
2717 required to make sense for the type of "CSTPTR".</dd>
2719 <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2720 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2722 <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2723 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2725 <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2726 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2728 <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2729 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2732 <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2733 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2736 <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2737 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2740 <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2741 <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2742 constants. The index list is interpreted in a similar manner as indices in
2743 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2744 index value must be specified.</dd>
2746 <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2747 <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2748 constants. The index list is interpreted in a similar manner as indices in
2749 a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2750 index value must be specified.</dd>
2752 <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2753 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2754 be any of the <a href="#binaryops">binary</a>
2755 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2756 on operands are the same as those for the corresponding instruction
2757 (e.g. no bitwise operations on floating point values are allowed).</dd>
2764 <!-- *********************************************************************** -->
2765 <h2><a name="othervalues">Other Values</a></h2>
2766 <!-- *********************************************************************** -->
2768 <!-- ======================================================================= -->
2770 <a name="inlineasm">Inline Assembler Expressions</a>
2775 <p>LLVM supports inline assembler expressions (as opposed
2776 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2777 a special value. This value represents the inline assembler as a string
2778 (containing the instructions to emit), a list of operand constraints (stored
2779 as a string), a flag that indicates whether or not the inline asm
2780 expression has side effects, and a flag indicating whether the function
2781 containing the asm needs to align its stack conservatively. An example
2782 inline assembler expression is:</p>
2784 <pre class="doc_code">
2785 i32 (i32) asm "bswap $0", "=r,r"
2788 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2789 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2792 <pre class="doc_code">
2793 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2796 <p>Inline asms with side effects not visible in the constraint list must be
2797 marked as having side effects. This is done through the use of the
2798 '<tt>sideeffect</tt>' keyword, like so:</p>
2800 <pre class="doc_code">
2801 call void asm sideeffect "eieio", ""()
2804 <p>In some cases inline asms will contain code that will not work unless the
2805 stack is aligned in some way, such as calls or SSE instructions on x86,
2806 yet will not contain code that does that alignment within the asm.
2807 The compiler should make conservative assumptions about what the asm might
2808 contain and should generate its usual stack alignment code in the prologue
2809 if the '<tt>alignstack</tt>' keyword is present:</p>
2811 <pre class="doc_code">
2812 call void asm alignstack "eieio", ""()
2815 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2818 <p>TODO: The format of the asm and constraints string still need to be
2819 documented here. Constraints on what can be done (e.g. duplication, moving,
2820 etc need to be documented). This is probably best done by reference to
2821 another document that covers inline asm from a holistic perspective.</p>
2824 <a name="inlineasm_md">Inline Asm Metadata</a>
2829 <p>The call instructions that wrap inline asm nodes may have a "!srcloc" MDNode
2830 attached to it that contains a list of constant integers. If present, the
2831 code generator will use the integer as the location cookie value when report
2832 errors through the LLVMContext error reporting mechanisms. This allows a
2833 front-end to correlate backend errors that occur with inline asm back to the
2834 source code that produced it. For example:</p>
2836 <pre class="doc_code">
2837 call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2839 !42 = !{ i32 1234567 }
2842 <p>It is up to the front-end to make sense of the magic numbers it places in the
2843 IR. If the MDNode contains multiple constants, the code generator will use
2844 the one that corresponds to the line of the asm that the error occurs on.</p>
2850 <!-- ======================================================================= -->
2852 <a name="metadata">Metadata Nodes and Metadata Strings</a>
2857 <p>LLVM IR allows metadata to be attached to instructions in the program that
2858 can convey extra information about the code to the optimizers and code
2859 generator. One example application of metadata is source-level debug
2860 information. There are two metadata primitives: strings and nodes. All
2861 metadata has the <tt>metadata</tt> type and is identified in syntax by a
2862 preceding exclamation point ('<tt>!</tt>').</p>
2864 <p>A metadata string is a string surrounded by double quotes. It can contain
2865 any character by escaping non-printable characters with "\xx" where "xx" is
2866 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2868 <p>Metadata nodes are represented with notation similar to structure constants
2869 (a comma separated list of elements, surrounded by braces and preceded by an
2870 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2871 10}</tt>". Metadata nodes can have any values as their operand.</p>
2873 <p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2874 metadata nodes, which can be looked up in the module symbol table. For
2875 example: "<tt>!foo = metadata !{!4, !3}</tt>".
2877 <p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2878 function is using two metadata arguments.</p>
2880 <div class="doc_code">
2882 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2886 <p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2887 attached with <tt>add</tt> instruction using <tt>!dbg</tt> identifier.</p>
2889 <div class="doc_code">
2891 %indvar.next = add i64 %indvar, 1, !dbg !21
2899 <!-- *********************************************************************** -->
2901 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2903 <!-- *********************************************************************** -->
2905 <p>LLVM has a number of "magic" global variables that contain data that affect
2906 code generation or other IR semantics. These are documented here. All globals
2907 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2908 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2911 <!-- ======================================================================= -->
2913 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2918 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2919 href="#linkage_appending">appending linkage</a>. This array contains a list of
2920 pointers to global variables and functions which may optionally have a pointer
2921 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2927 @llvm.used = appending global [2 x i8*] [
2929 i8* bitcast (i32* @Y to i8*)
2930 ], section "llvm.metadata"
2933 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2934 compiler, assembler, and linker are required to treat the symbol as if there is
2935 a reference to the global that it cannot see. For example, if a variable has
2936 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2937 list, it cannot be deleted. This is commonly used to represent references from
2938 inline asms and other things the compiler cannot "see", and corresponds to
2939 "attribute((used))" in GNU C.</p>
2941 <p>On some targets, the code generator must emit a directive to the assembler or
2942 object file to prevent the assembler and linker from molesting the symbol.</p>
2946 <!-- ======================================================================= -->
2948 <a name="intg_compiler_used">
2949 The '<tt>llvm.compiler.used</tt>' Global Variable
2955 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2956 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2957 touching the symbol. On targets that support it, this allows an intelligent
2958 linker to optimize references to the symbol without being impeded as it would be
2959 by <tt>@llvm.used</tt>.</p>
2961 <p>This is a rare construct that should only be used in rare circumstances, and
2962 should not be exposed to source languages.</p>
2966 <!-- ======================================================================= -->
2968 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2973 %0 = type { i32, void ()* }
2974 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
2976 <p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor functions and associated priorities. The functions referenced by this array will be called in ascending order of priority (i.e. lowest first) when the module is loaded. The order of functions with the same priority is not defined.
2981 <!-- ======================================================================= -->
2983 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
2988 %0 = type { i32, void ()* }
2989 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
2992 <p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions and associated priorities. The functions referenced by this array will be called in descending order of priority (i.e. highest first) when the module is loaded. The order of functions with the same priority is not defined.
2999 <!-- *********************************************************************** -->
3000 <h2><a name="instref">Instruction Reference</a></h2>
3001 <!-- *********************************************************************** -->
3005 <p>The LLVM instruction set consists of several different classifications of
3006 instructions: <a href="#terminators">terminator
3007 instructions</a>, <a href="#binaryops">binary instructions</a>,
3008 <a href="#bitwiseops">bitwise binary instructions</a>,
3009 <a href="#memoryops">memory instructions</a>, and
3010 <a href="#otherops">other instructions</a>.</p>
3012 <!-- ======================================================================= -->
3014 <a name="terminators">Terminator Instructions</a>
3019 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3020 in a program ends with a "Terminator" instruction, which indicates which
3021 block should be executed after the current block is finished. These
3022 terminator instructions typically yield a '<tt>void</tt>' value: they produce
3023 control flow, not values (the one exception being the
3024 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3026 <p>There are seven different terminator instructions: the
3027 '<a href="#i_ret"><tt>ret</tt></a>' instruction, the
3028 '<a href="#i_br"><tt>br</tt></a>' instruction, the
3029 '<a href="#i_switch"><tt>switch</tt></a>' instruction, the
3030 '<a href="#i_indirectbr">'<tt>indirectbr</tt></a>' Instruction, the
3031 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the
3032 '<a href="#i_unwind"><tt>unwind</tt></a>' instruction, and the
3033 '<a href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
3035 <!-- _______________________________________________________________________ -->
3037 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3044 ret <type> <value> <i>; Return a value from a non-void function</i>
3045 ret void <i>; Return from void function</i>
3049 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3050 a value) from a function back to the caller.</p>
3052 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3053 value and then causes control flow, and one that just causes control flow to
3057 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3058 return value. The type of the return value must be a
3059 '<a href="#t_firstclass">first class</a>' type.</p>
3061 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
3062 non-void return type and contains a '<tt>ret</tt>' instruction with no return
3063 value or a return value with a type that does not match its type, or if it
3064 has a void return type and contains a '<tt>ret</tt>' instruction with a
3068 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3069 the calling function's context. If the caller is a
3070 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3071 instruction after the call. If the caller was an
3072 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3073 the beginning of the "normal" destination block. If the instruction returns
3074 a value, that value shall set the call or invoke instruction's return
3079 ret i32 5 <i>; Return an integer value of 5</i>
3080 ret void <i>; Return from a void function</i>
3081 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3085 <!-- _______________________________________________________________________ -->
3087 <a name="i_br">'<tt>br</tt>' Instruction</a>
3094 br i1 <cond>, label <iftrue>, label <iffalse>
3095 br label <dest> <i>; Unconditional branch</i>
3099 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3100 different basic block in the current function. There are two forms of this
3101 instruction, corresponding to a conditional branch and an unconditional
3105 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3106 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
3107 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3111 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3112 argument is evaluated. If the value is <tt>true</tt>, control flows to the
3113 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
3114 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3119 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3120 br i1 %cond, label %IfEqual, label %IfUnequal
3122 <a href="#i_ret">ret</a> i32 1
3124 <a href="#i_ret">ret</a> i32 0
3129 <!-- _______________________________________________________________________ -->
3131 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3138 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
3142 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3143 several different places. It is a generalization of the '<tt>br</tt>'
3144 instruction, allowing a branch to occur to one of many possible
3148 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3149 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3150 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3151 The table is not allowed to contain duplicate constant entries.</p>
3154 <p>The <tt>switch</tt> instruction specifies a table of values and
3155 destinations. When the '<tt>switch</tt>' instruction is executed, this table
3156 is searched for the given value. If the value is found, control flow is
3157 transferred to the corresponding destination; otherwise, control flow is
3158 transferred to the default destination.</p>
3160 <h5>Implementation:</h5>
3161 <p>Depending on properties of the target machine and the particular
3162 <tt>switch</tt> instruction, this instruction may be code generated in
3163 different ways. For example, it could be generated as a series of chained
3164 conditional branches or with a lookup table.</p>
3168 <i>; Emulate a conditional br instruction</i>
3169 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3170 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3172 <i>; Emulate an unconditional br instruction</i>
3173 switch i32 0, label %dest [ ]
3175 <i>; Implement a jump table:</i>
3176 switch i32 %val, label %otherwise [ i32 0, label %onzero
3178 i32 2, label %ontwo ]
3184 <!-- _______________________________________________________________________ -->
3186 <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3193 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
3198 <p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3199 within the current function, whose address is specified by
3200 "<tt>address</tt>". Address must be derived from a <a
3201 href="#blockaddress">blockaddress</a> constant.</p>
3205 <p>The '<tt>address</tt>' argument is the address of the label to jump to. The
3206 rest of the arguments indicate the full set of possible destinations that the
3207 address may point to. Blocks are allowed to occur multiple times in the
3208 destination list, though this isn't particularly useful.</p>
3210 <p>This destination list is required so that dataflow analysis has an accurate
3211 understanding of the CFG.</p>
3215 <p>Control transfers to the block specified in the address argument. All
3216 possible destination blocks must be listed in the label list, otherwise this
3217 instruction has undefined behavior. This implies that jumps to labels
3218 defined in other functions have undefined behavior as well.</p>
3220 <h5>Implementation:</h5>
3222 <p>This is typically implemented with a jump through a register.</p>
3226 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3232 <!-- _______________________________________________________________________ -->
3234 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3241 <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>]
3242 to label <normal label> unwind label <exception label>
3246 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3247 function, with the possibility of control flow transfer to either the
3248 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
3249 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3250 control flow will return to the "normal" label. If the callee (or any
3251 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
3252 instruction, control is interrupted and continued at the dynamically nearest
3253 "exception" label.</p>
3256 <p>This instruction requires several arguments:</p>
3259 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3260 convention</a> the call should use. If none is specified, the call
3261 defaults to using C calling conventions.</li>
3263 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3264 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3265 '<tt>inreg</tt>' attributes are valid here.</li>
3267 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3268 function value being invoked. In most cases, this is a direct function
3269 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3270 off an arbitrary pointer to function value.</li>
3272 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3273 function to be invoked. </li>
3275 <li>'<tt>function args</tt>': argument list whose types match the function
3276 signature argument types and parameter attributes. All arguments must be
3277 of <a href="#t_firstclass">first class</a> type. If the function
3278 signature indicates the function accepts a variable number of arguments,
3279 the extra arguments can be specified.</li>
3281 <li>'<tt>normal label</tt>': the label reached when the called function
3282 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3284 <li>'<tt>exception label</tt>': the label reached when a callee returns with
3285 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
3287 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3288 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3289 '<tt>readnone</tt>' attributes are valid here.</li>
3293 <p>This instruction is designed to operate as a standard
3294 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
3295 primary difference is that it establishes an association with a label, which
3296 is used by the runtime library to unwind the stack.</p>
3298 <p>This instruction is used in languages with destructors to ensure that proper
3299 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3300 exception. Additionally, this is important for implementation of
3301 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3303 <p>For the purposes of the SSA form, the definition of the value returned by the
3304 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3305 block to the "normal" label. If the callee unwinds then no return value is
3308 <p>Note that the code generator does not yet completely support unwind, and
3309 that the invoke/unwind semantics are likely to change in future versions.</p>
3313 %retval = invoke i32 @Test(i32 15) to label %Continue
3314 unwind label %TestCleanup <i>; {i32}:retval set</i>
3315 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3316 unwind label %TestCleanup <i>; {i32}:retval set</i>
3321 <!-- _______________________________________________________________________ -->
3324 <a name="i_unwind">'<tt>unwind</tt>' Instruction</a>
3335 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
3336 at the first callee in the dynamic call stack which used
3337 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
3338 This is primarily used to implement exception handling.</p>
3341 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
3342 immediately halt. The dynamic call stack is then searched for the
3343 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
3344 Once found, execution continues at the "exceptional" destination block
3345 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
3346 instruction in the dynamic call chain, undefined behavior results.</p>
3348 <p>Note that the code generator does not yet completely support unwind, and
3349 that the invoke/unwind semantics are likely to change in future versions.</p>
3353 <!-- _______________________________________________________________________ -->
3356 <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3367 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
3368 instruction is used to inform the optimizer that a particular portion of the
3369 code is not reachable. This can be used to indicate that the code after a
3370 no-return function cannot be reached, and other facts.</p>
3373 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3379 <!-- ======================================================================= -->
3381 <a name="binaryops">Binary Operations</a>
3386 <p>Binary operators are used to do most of the computation in a program. They
3387 require two operands of the same type, execute an operation on them, and
3388 produce a single value. The operands might represent multiple data, as is
3389 the case with the <a href="#t_vector">vector</a> data type. The result value
3390 has the same type as its operands.</p>
3392 <p>There are several different binary operators:</p>
3394 <!-- _______________________________________________________________________ -->
3396 <a name="i_add">'<tt>add</tt>' Instruction</a>
3403 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3404 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3405 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3406 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3410 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3413 <p>The two arguments to the '<tt>add</tt>' instruction must
3414 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3415 integer values. Both arguments must have identical types.</p>
3418 <p>The value produced is the integer sum of the two operands.</p>
3420 <p>If the sum has unsigned overflow, the result returned is the mathematical
3421 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3423 <p>Because LLVM integers use a two's complement representation, this instruction
3424 is appropriate for both signed and unsigned integers.</p>
3426 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3427 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3428 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3429 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3430 respectively, occurs.</p>
3434 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
3439 <!-- _______________________________________________________________________ -->
3441 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3448 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3452 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3455 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
3456 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3457 floating point values. Both arguments must have identical types.</p>
3460 <p>The value produced is the floating point sum of the two operands.</p>
3464 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
3469 <!-- _______________________________________________________________________ -->
3471 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3478 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3479 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3480 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3481 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3485 <p>The '<tt>sub</tt>' instruction returns the difference of its two
3488 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
3489 '<tt>neg</tt>' instruction present in most other intermediate
3490 representations.</p>
3493 <p>The two arguments to the '<tt>sub</tt>' instruction must
3494 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3495 integer values. Both arguments must have identical types.</p>
3498 <p>The value produced is the integer difference of the two operands.</p>
3500 <p>If the difference has unsigned overflow, the result returned is the
3501 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3504 <p>Because LLVM integers use a two's complement representation, this instruction
3505 is appropriate for both signed and unsigned integers.</p>
3507 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3508 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3509 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3510 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3511 respectively, occurs.</p>
3515 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
3516 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
3521 <!-- _______________________________________________________________________ -->
3523 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3530 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3534 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
3537 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3538 '<tt>fneg</tt>' instruction present in most other intermediate
3539 representations.</p>
3542 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
3543 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3544 floating point values. Both arguments must have identical types.</p>
3547 <p>The value produced is the floating point difference of the two operands.</p>
3551 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
3552 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
3557 <!-- _______________________________________________________________________ -->
3559 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
3566 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3567 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3568 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3569 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3573 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3576 <p>The two arguments to the '<tt>mul</tt>' instruction must
3577 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3578 integer values. Both arguments must have identical types.</p>
3581 <p>The value produced is the integer product of the two operands.</p>
3583 <p>If the result of the multiplication has unsigned overflow, the result
3584 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3585 width of the result.</p>
3587 <p>Because LLVM integers use a two's complement representation, and the result
3588 is the same width as the operands, this instruction returns the correct
3589 result for both signed and unsigned integers. If a full product
3590 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3591 be sign-extended or zero-extended as appropriate to the width of the full
3594 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3595 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3596 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3597 is a <a href="#trapvalues">trap value</a> if unsigned and/or signed overflow,
3598 respectively, occurs.</p>
3602 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3607 <!-- _______________________________________________________________________ -->
3609 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3616 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3620 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3623 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3624 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3625 floating point values. Both arguments must have identical types.</p>
3628 <p>The value produced is the floating point product of the two operands.</p>
3632 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3637 <!-- _______________________________________________________________________ -->
3639 <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
3646 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3647 <result> = udiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3651 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3654 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3655 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3656 values. Both arguments must have identical types.</p>
3659 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3661 <p>Note that unsigned integer division and signed integer division are distinct
3662 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3664 <p>Division by zero leads to undefined behavior.</p>
3666 <p>If the <tt>exact</tt> keyword is present, the result value of the
3667 <tt>udiv</tt> is a <a href="#trapvalues">trap value</a> if %op1 is not a
3668 multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
3673 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3678 <!-- _______________________________________________________________________ -->
3680 <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
3687 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3688 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3692 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3695 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3696 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3697 values. Both arguments must have identical types.</p>
3700 <p>The value produced is the signed integer quotient of the two operands rounded
3703 <p>Note that signed integer division and unsigned integer division are distinct
3704 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3706 <p>Division by zero leads to undefined behavior. Overflow also leads to
3707 undefined behavior; this is a rare case, but can occur, for example, by doing
3708 a 32-bit division of -2147483648 by -1.</p>
3710 <p>If the <tt>exact</tt> keyword is present, the result value of the
3711 <tt>sdiv</tt> is a <a href="#trapvalues">trap value</a> if the result would
3716 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3721 <!-- _______________________________________________________________________ -->
3723 <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
3730 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3734 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3737 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3738 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3739 floating point values. Both arguments must have identical types.</p>
3742 <p>The value produced is the floating point quotient of the two operands.</p>
3746 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3751 <!-- _______________________________________________________________________ -->
3753 <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3760 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3764 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3765 division of its two arguments.</p>
3768 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3769 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3770 values. Both arguments must have identical types.</p>
3773 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3774 This instruction always performs an unsigned division to get the
3777 <p>Note that unsigned integer remainder and signed integer remainder are
3778 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3780 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3784 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3789 <!-- _______________________________________________________________________ -->
3791 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3798 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3802 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3803 division of its two operands. This instruction can also take
3804 <a href="#t_vector">vector</a> versions of the values in which case the
3805 elements must be integers.</p>
3808 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3809 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3810 values. Both arguments must have identical types.</p>
3813 <p>This instruction returns the <i>remainder</i> of a division (where the result
3814 is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
3815 <i>modulo</i> operator (where the result is either zero or has the same sign
3816 as the divisor, <tt>op2</tt>) of a value.
3817 For more information about the difference,
3818 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3819 Math Forum</a>. For a table of how this is implemented in various languages,
3820 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3821 Wikipedia: modulo operation</a>.</p>
3823 <p>Note that signed integer remainder and unsigned integer remainder are
3824 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3826 <p>Taking the remainder of a division by zero leads to undefined behavior.
3827 Overflow also leads to undefined behavior; this is a rare case, but can
3828 occur, for example, by taking the remainder of a 32-bit division of
3829 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3830 lets srem be implemented using instructions that return both the result of
3831 the division and the remainder.)</p>
3835 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3840 <!-- _______________________________________________________________________ -->
3842 <a name="i_frem">'<tt>frem</tt>' Instruction</a>
3849 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3853 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3854 its two operands.</p>
3857 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3858 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3859 floating point values. Both arguments must have identical types.</p>
3862 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3863 has the same sign as the dividend.</p>
3867 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3874 <!-- ======================================================================= -->
3876 <a name="bitwiseops">Bitwise Binary Operations</a>
3881 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3882 program. They are generally very efficient instructions and can commonly be
3883 strength reduced from other instructions. They require two operands of the
3884 same type, execute an operation on them, and produce a single value. The
3885 resulting value is the same type as its operands.</p>
3887 <!-- _______________________________________________________________________ -->
3889 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
3896 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3897 <result> = shl nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3898 <result> = shl nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3899 <result> = shl nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3903 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3904 a specified number of bits.</p>
3907 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3908 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3909 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3912 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3913 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3914 is (statically or dynamically) negative or equal to or larger than the number
3915 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3916 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3917 shift amount in <tt>op2</tt>.</p>
3919 <p>If the <tt>nuw</tt> keyword is present, then the shift produces a
3920 <a href="#trapvalues">trap value</a> if it shifts out any non-zero bits. If
3921 the <tt>nsw</tt> keyword is present, then the shift produces a
3922 <a href="#trapvalues">trap value</a> if it shifts out any bits that disagree
3923 with the resultant sign bit. As such, NUW/NSW have the same semantics as
3924 they would if the shift were expressed as a mul instruction with the same
3925 nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
3929 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3930 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3931 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3932 <result> = shl i32 1, 32 <i>; undefined</i>
3933 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3938 <!-- _______________________________________________________________________ -->
3940 <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
3947 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3948 <result> = lshr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3952 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
3953 operand shifted to the right a specified number of bits with zero fill.</p>
3956 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
3957 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3958 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3961 <p>This instruction always performs a logical shift right operation. The most
3962 significant bits of the result will be filled with zero bits after the shift.
3963 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
3964 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3965 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3966 shift amount in <tt>op2</tt>.</p>
3968 <p>If the <tt>exact</tt> keyword is present, the result value of the
3969 <tt>lshr</tt> is a <a href="#trapvalues">trap value</a> if any of the bits
3970 shifted out are non-zero.</p>
3975 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
3976 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
3977 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
3978 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
3979 <result> = lshr i32 1, 32 <i>; undefined</i>
3980 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3985 <!-- _______________________________________________________________________ -->
3987 <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
3994 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3995 <result> = ashr exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3999 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4000 operand shifted to the right a specified number of bits with sign
4004 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4005 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4006 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4009 <p>This instruction always performs an arithmetic shift right operation, The
4010 most significant bits of the result will be filled with the sign bit
4011 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
4012 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4013 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4014 the corresponding shift amount in <tt>op2</tt>.</p>
4016 <p>If the <tt>exact</tt> keyword is present, the result value of the
4017 <tt>ashr</tt> is a <a href="#trapvalues">trap value</a> if any of the bits
4018 shifted out are non-zero.</p>
4022 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
4023 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
4024 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
4025 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
4026 <result> = ashr i32 1, 32 <i>; undefined</i>
4027 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
4032 <!-- _______________________________________________________________________ -->
4034 <a name="i_and">'<tt>and</tt>' Instruction</a>
4041 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4045 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4049 <p>The two arguments to the '<tt>and</tt>' instruction must be
4050 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4051 values. Both arguments must have identical types.</p>
4054 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4056 <table border="1" cellspacing="0" cellpadding="4">
4088 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
4089 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
4090 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
4093 <!-- _______________________________________________________________________ -->
4095 <a name="i_or">'<tt>or</tt>' Instruction</a>
4102 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4106 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4110 <p>The two arguments to the '<tt>or</tt>' instruction must be
4111 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4112 values. Both arguments must have identical types.</p>
4115 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4117 <table border="1" cellspacing="0" cellpadding="4">
4149 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
4150 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
4151 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
4156 <!-- _______________________________________________________________________ -->
4158 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4165 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4169 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4170 its two operands. The <tt>xor</tt> is used to implement the "one's
4171 complement" operation, which is the "~" operator in C.</p>
4174 <p>The two arguments to the '<tt>xor</tt>' instruction must be
4175 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4176 values. Both arguments must have identical types.</p>
4179 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4181 <table border="1" cellspacing="0" cellpadding="4">
4213 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
4214 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
4215 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
4216 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
4223 <!-- ======================================================================= -->
4225 <a name="vectorops">Vector Operations</a>
4230 <p>LLVM supports several instructions to represent vector operations in a
4231 target-independent manner. These instructions cover the element-access and
4232 vector-specific operations needed to process vectors effectively. While LLVM
4233 does directly support these vector operations, many sophisticated algorithms
4234 will want to use target-specific intrinsics to take full advantage of a
4235 specific target.</p>
4237 <!-- _______________________________________________________________________ -->
4239 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4246 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
4250 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4251 from a vector at a specified index.</p>
4255 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4256 of <a href="#t_vector">vector</a> type. The second operand is an index
4257 indicating the position from which to extract the element. The index may be
4261 <p>The result is a scalar of the same type as the element type of
4262 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
4263 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4264 results are undefined.</p>
4268 <result> = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
4273 <!-- _______________________________________________________________________ -->
4275 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4282 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
4286 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4287 vector at a specified index.</p>
4290 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4291 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
4292 whose type must equal the element type of the first operand. The third
4293 operand is an index indicating the position at which to insert the value.
4294 The index may be a variable.</p>
4297 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
4298 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4299 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4300 results are undefined.</p>
4304 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
4309 <!-- _______________________________________________________________________ -->
4311 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4318 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
4322 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4323 from two input vectors, returning a vector with the same element type as the
4324 input and length that is the same as the shuffle mask.</p>
4327 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4328 with types that match each other. The third argument is a shuffle mask whose
4329 element type is always 'i32'. The result of the instruction is a vector
4330 whose length is the same as the shuffle mask and whose element type is the
4331 same as the element type of the first two operands.</p>
4333 <p>The shuffle mask operand is required to be a constant vector with either
4334 constant integer or undef values.</p>
4337 <p>The elements of the two input vectors are numbered from left to right across
4338 both of the vectors. The shuffle mask operand specifies, for each element of
4339 the result vector, which element of the two input vectors the result element
4340 gets. The element selector may be undef (meaning "don't care") and the
4341 second operand may be undef if performing a shuffle from only one vector.</p>
4345 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4346 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
4347 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
4348 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
4349 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
4350 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
4351 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
4352 <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>
4359 <!-- ======================================================================= -->
4361 <a name="aggregateops">Aggregate Operations</a>
4366 <p>LLVM supports several instructions for working with
4367 <a href="#t_aggregate">aggregate</a> values.</p>
4369 <!-- _______________________________________________________________________ -->
4371 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4378 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
4382 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4383 from an <a href="#t_aggregate">aggregate</a> value.</p>
4386 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4387 of <a href="#t_struct">struct</a> or
4388 <a href="#t_array">array</a> type. The operands are constant indices to
4389 specify which value to extract in a similar manner as indices in a
4390 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4391 <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4393 <li>Since the value being indexed is not a pointer, the first index is
4394 omitted and assumed to be zero.</li>
4395 <li>At least one index must be specified.</li>
4396 <li>Not only struct indices but also array indices must be in
4401 <p>The result is the value at the position in the aggregate specified by the
4406 <result> = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
4411 <!-- _______________________________________________________________________ -->
4413 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4420 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* <i>; yields <aggregate type></i>
4424 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4425 in an <a href="#t_aggregate">aggregate</a> value.</p>
4428 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4429 of <a href="#t_struct">struct</a> or
4430 <a href="#t_array">array</a> type. The second operand is a first-class
4431 value to insert. The following operands are constant indices indicating
4432 the position at which to insert the value in a similar manner as indices in a
4433 '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction. The
4434 value to insert must have the same type as the value identified by the
4438 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
4439 that of <tt>val</tt> except that the value at the position specified by the
4440 indices is that of <tt>elt</tt>.</p>
4444 %agg1 = insertvalue {i32, float} undef, i32 1, 0 <i>; yields {i32 1, float undef}</i>
4445 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 <i>; yields {i32 1, float %val}</i>
4446 %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0 <i>; yields {i32 1, float %val}</i>
4453 <!-- ======================================================================= -->
4455 <a name="memoryops">Memory Access and Addressing Operations</a>
4460 <p>A key design point of an SSA-based representation is how it represents
4461 memory. In LLVM, no memory locations are in SSA form, which makes things
4462 very simple. This section describes how to read, write, and allocate
4465 <!-- _______________________________________________________________________ -->
4467 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4474 <result> = alloca <type>[, <ty> <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
4478 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4479 currently executing function, to be automatically released when this function
4480 returns to its caller. The object is always allocated in the generic address
4481 space (address space zero).</p>
4484 <p>The '<tt>alloca</tt>' instruction
4485 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
4486 runtime stack, returning a pointer of the appropriate type to the program.
4487 If "NumElements" is specified, it is the number of elements allocated,
4488 otherwise "NumElements" is defaulted to be one. If a constant alignment is
4489 specified, the value result of the allocation is guaranteed to be aligned to
4490 at least that boundary. If not specified, or if zero, the target can choose
4491 to align the allocation on any convenient boundary compatible with the
4494 <p>'<tt>type</tt>' may be any sized type.</p>
4497 <p>Memory is allocated; a pointer is returned. The operation is undefined if
4498 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
4499 memory is automatically released when the function returns. The
4500 '<tt>alloca</tt>' instruction is commonly used to represent automatic
4501 variables that must have an address available. When the function returns
4502 (either with the <tt><a href="#i_ret">ret</a></tt>
4503 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
4504 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
4508 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
4509 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
4510 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
4511 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
4516 <!-- _______________________________________________________________________ -->
4518 <a name="i_load">'<tt>load</tt>' Instruction</a>
4525 <result> = load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4526 <result> = volatile load <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>]
4527 !<index> = !{ i32 1 }
4531 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4534 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4535 from which to load. The pointer must point to
4536 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
4537 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
4538 number or order of execution of this <tt>load</tt> with other <a
4539 href="#volatile">volatile operations</a>.</p>
4541 <p>The optional constant <tt>align</tt> argument specifies the alignment of the
4542 operation (that is, the alignment of the memory address). A value of 0 or an
4543 omitted <tt>align</tt> argument means that the operation has the preferential
4544 alignment for the target. It is the responsibility of the code emitter to
4545 ensure that the alignment information is correct. Overestimating the
4546 alignment results in undefined behavior. Underestimating the alignment may
4547 produce less efficient code. An alignment of 1 is always safe.</p>
4549 <p>The optional <tt>!nontemporal</tt> metadata must reference a single
4550 metatadata name <index> corresponding to a metadata node with
4551 one <tt>i32</tt> entry of value 1. The existence of
4552 the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
4553 and code generator that this load is not expected to be reused in the cache.
4554 The code generator may select special instructions to save cache bandwidth,
4555 such as the <tt>MOVNT</tt> instruction on x86.</p>
4558 <p>The location of memory pointed to is loaded. If the value being loaded is of
4559 scalar type then the number of bytes read does not exceed the minimum number
4560 of bytes needed to hold all bits of the type. For example, loading an
4561 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
4562 <tt>i20</tt> with a size that is not an integral number of bytes, the result
4563 is undefined if the value was not originally written using a store of the
4568 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4569 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
4570 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
4575 <!-- _______________________________________________________________________ -->
4577 <a name="i_store">'<tt>store</tt>' Instruction</a>
4584 store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4585 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] <i>; yields {void}</i>
4589 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4592 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4593 and an address at which to store it. The type of the
4594 '<tt><pointer></tt>' operand must be a pointer to
4595 the <a href="#t_firstclass">first class</a> type of the
4596 '<tt><value></tt>' operand. If the <tt>store</tt> is marked as
4597 <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
4598 order of execution of this <tt>store</tt> with other <a
4599 href="#volatile">volatile operations</a>.</p>
4601 <p>The optional constant "align" argument specifies the alignment of the
4602 operation (that is, the alignment of the memory address). A value of 0 or an
4603 omitted "align" argument means that the operation has the preferential
4604 alignment for the target. It is the responsibility of the code emitter to
4605 ensure that the alignment information is correct. Overestimating the
4606 alignment results in an undefined behavior. Underestimating the alignment may
4607 produce less efficient code. An alignment of 1 is always safe.</p>
4609 <p>The optional !nontemporal metadata must reference a single metatadata
4610 name <index> corresponding to a metadata node with one i32 entry of
4611 value 1. The existence of the !nontemporal metatadata on the
4612 instruction tells the optimizer and code generator that this load is
4613 not expected to be reused in the cache. The code generator may
4614 select special instructions to save cache bandwidth, such as the
4615 MOVNT instruction on x86.</p>
4619 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4620 location specified by the '<tt><pointer></tt>' operand. If
4621 '<tt><value></tt>' is of scalar type then the number of bytes written
4622 does not exceed the minimum number of bytes needed to hold all bits of the
4623 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4624 writing a value of a type like <tt>i20</tt> with a size that is not an
4625 integral number of bytes, it is unspecified what happens to the extra bits
4626 that do not belong to the type, but they will typically be overwritten.</p>
4630 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4631 store i32 3, i32* %ptr <i>; yields {void}</i>
4632 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4637 <!-- _______________________________________________________________________ -->
4638 <div class="doc_subsubsection"> <a name="i_fence">'<tt>fence</tt>'
4639 Instruction</a> </div>
4641 <div class="doc_text">
4645 fence [singlethread] <ordering> <i>; yields {void}</i>
4649 <p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
4650 between operations.</p>
4652 <h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
4653 href="#ordering">ordering</a> argument which defines what
4654 <i>synchronizes-with</i> edges they add. They can only be given
4655 <code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
4656 <code>seq_cst</code> orderings.</p>
4659 <p>A fence <var>A</var> which has (at least) <code>release</code> ordering
4660 semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
4661 <code>acquire</code> ordering semantics if and only if there exist atomic
4662 operations <var>X</var> and <var>Y</var>, both operating on some atomic object
4663 <var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
4664 <var>X</var> modifies <var>M</var> (either directly or through some side effect
4665 of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
4666 <var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
4667 <i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
4668 than an explicit <code>fence</code>, one (but not both) of the atomic operations
4669 <var>X</var> or <var>Y</var> might provide a <code>release</code> or
4670 <code>acquire</code> (resp.) ordering constraint and still
4671 <i>synchronize-with</i> the explicit <code>fence</code> and establish the
4672 <i>happens-before</i> edge.</p>
4674 <p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
4675 having both <code>acquire</code> and <code>release</code> semantics specified
4676 above, participates in the global program order of other <code>seq_cst</code>
4677 operations and/or fences.</p>
4679 <p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
4680 specifies that the fence only synchronizes with other fences in the same
4681 thread. (This is useful for interacting with signal handlers.)</p>
4683 <p>FIXME: This instruction is a work in progress; until it is finished, use
4684 llvm.memory.barrier.
4688 fence acquire <i>; yields {void}</i>
4689 fence singlethread seq_cst <i>; yields {void}</i>
4694 <!-- _______________________________________________________________________ -->
4695 <div class="doc_subsubsection"> <a name="i_cmpxchg">'<tt>cmpxchg</tt>'
4696 Instruction</a> </div>
4698 <div class="doc_text">
4702 [volatile] cmpxchg <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <ordering> <i>; yields {ty}</i>
4706 <p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
4707 It loads a value in memory and compares it to a given value. If they are
4708 equal, it stores a new value into the memory.</p>
4711 <p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
4712 address to operate on, a value to compare to the value currently be at that
4713 address, and a new value to place at that address if the compared values are
4714 equal. The type of '<var><cmp></var>' must be an integer type whose
4715 bit width is a power of two greater than or equal to eight and less than
4716 or equal to a target-specific size limit. '<var><cmp></var>' and
4717 '<var><new></var>' must have the same type, and the type of
4718 '<var><pointer></var>' must be a pointer to that type. If the
4719 <code>cmpxchg</code> is marked as <code>volatile</code>, then the
4720 optimizer is not allowed to modify the number or order of execution
4721 of this <code>cmpxchg</code> with other <a href="#volatile">volatile
4724 <!-- FIXME: Extend allowed types. -->
4726 <p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
4727 <code>cmpxchg</code> synchronizes with other atomic operations.</p>
4729 <p>The optional "<code>singlethread</code>" argument declares that the
4730 <code>cmpxchg</code> is only atomic with respect to code (usually signal
4731 handlers) running in the same thread as the <code>cmpxchg</code>. Otherwise the
4732 cmpxchg is atomic with respect to all other code in the system.</p>
4734 <p>The pointer passed into cmpxchg must have alignment greater than or equal to
4735 the size in memory of the operand.
4738 <p>The contents of memory at the location specified by the
4739 '<tt><pointer></tt>' operand is read and compared to
4740 '<tt><cmp></tt>'; if the read value is the equal,
4741 '<tt><new></tt>' is written. The original value at the location
4744 <p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
4745 purpose of identifying <a href="#release_sequence">release sequences</a>. A
4746 failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
4747 parameter determined by dropping any <code>release</code> part of the
4748 <code>cmpxchg</code>'s ordering.</p>
4751 FIXME: Is compare_exchange_weak() necessary? (Consider after we've done
4752 optimization work on ARM.)
4754 FIXME: Is a weaker ordering constraint on failure helpful in practice?
4760 %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered <i>; yields {i32}</i>
4761 <a href="#i_br">br</a> label %loop
4764 %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
4765 %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
4766 %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared <i>; yields {i32}</i>
4767 %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
4768 <a href="#i_br">br</a> i1 %success, label %done, label %loop
4776 <!-- _______________________________________________________________________ -->
4777 <div class="doc_subsubsection"> <a name="i_atomicrmw">'<tt>atomicrmw</tt>'
4778 Instruction</a> </div>
4780 <div class="doc_text">
4784 [volatile] atomicrmw <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> <i>; yields {ty}</i>
4788 <p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
4791 <p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
4792 operation to apply, an address whose value to modify, an argument to the
4793 operation. The operation must be one of the following keywords:</p>
4808 <p>The type of '<var><value></var>' must be an integer type whose
4809 bit width is a power of two greater than or equal to eight and less than
4810 or equal to a target-specific size limit. The type of the
4811 '<code><pointer></code>' operand must be a pointer to that type.
4812 If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
4813 optimizer is not allowed to modify the number or order of execution of this
4814 <code>atomicrmw</code> with other <a href="#volatile">volatile
4817 <!-- FIXME: Extend allowed types. -->
4820 <p>The contents of memory at the location specified by the
4821 '<tt><pointer></tt>' operand are atomically read, modified, and written
4822 back. The original value at the location is returned. The modification is
4823 specified by the <var>operation</var> argument:</p>
4826 <li>xchg: <code>*ptr = val</code></li>
4827 <li>add: <code>*ptr = *ptr + val</code></li>
4828 <li>sub: <code>*ptr = *ptr - val</code></li>
4829 <li>and: <code>*ptr = *ptr & val</code></li>
4830 <li>nand: <code>*ptr = ~(*ptr & val)</code></li>
4831 <li>or: <code>*ptr = *ptr | val</code></li>
4832 <li>xor: <code>*ptr = *ptr ^ val</code></li>
4833 <li>max: <code>*ptr = *ptr > val ? *ptr : val</code> (using a signed comparison)</li>
4834 <li>min: <code>*ptr = *ptr < val ? *ptr : val</code> (using a signed comparison)</li>
4835 <li>umax: <code>*ptr = *ptr > val ? *ptr : val</code> (using an unsigned comparison)</li>
4836 <li>umin: <code>*ptr = *ptr < val ? *ptr : val</code> (using an unsigned comparison)</li>
4841 %old = atomicrmw add i32* %ptr, i32 1 acquire <i>; yields {i32}</i>
4846 <!-- _______________________________________________________________________ -->
4848 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
4855 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4856 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4860 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
4861 subelement of an <a href="#t_aggregate">aggregate</a> data structure.
4862 It performs address calculation only and does not access memory.</p>
4865 <p>The first argument is always a pointer, and forms the basis of the
4866 calculation. The remaining arguments are indices that indicate which of the
4867 elements of the aggregate object are indexed. The interpretation of each
4868 index is dependent on the type being indexed into. The first index always
4869 indexes the pointer value given as the first argument, the second index
4870 indexes a value of the type pointed to (not necessarily the value directly
4871 pointed to, since the first index can be non-zero), etc. The first type
4872 indexed into must be a pointer value, subsequent types can be arrays,
4873 vectors, and structs. Note that subsequent types being indexed into
4874 can never be pointers, since that would require loading the pointer before
4875 continuing calculation.</p>
4877 <p>The type of each index argument depends on the type it is indexing into.
4878 When indexing into a (optionally packed) structure, only <tt>i32</tt>
4879 integer <b>constants</b> are allowed. When indexing into an array, pointer
4880 or vector, integers of any width are allowed, and they are not required to be
4883 <p>For example, let's consider a C code fragment and how it gets compiled to
4886 <pre class="doc_code">
4898 int *foo(struct ST *s) {
4899 return &s[1].Z.B[5][13];
4903 <p>The LLVM code generated by the GCC frontend is:</p>
4905 <pre class="doc_code">
4906 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
4907 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
4909 define i32* @foo(%ST* %s) {
4911 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
4917 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
4918 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
4919 }</tt>' type, a structure. The second index indexes into the third element
4920 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
4921 i8 }</tt>' type, another structure. The third index indexes into the second
4922 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
4923 array. The two dimensions of the array are subscripted into, yielding an
4924 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
4925 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
4927 <p>Note that it is perfectly legal to index partially through a structure,
4928 returning a pointer to an inner element. Because of this, the LLVM code for
4929 the given testcase is equivalent to:</p>
4932 define i32* @foo(%ST* %s) {
4933 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
4934 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
4935 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
4936 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
4937 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
4942 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
4943 <tt>getelementptr</tt> is a <a href="#trapvalues">trap value</a> if the
4944 base pointer is not an <i>in bounds</i> address of an allocated object,
4945 or if any of the addresses that would be formed by successive addition of
4946 the offsets implied by the indices to the base address with infinitely
4947 precise arithmetic are not an <i>in bounds</i> address of that allocated
4948 object. The <i>in bounds</i> addresses for an allocated object are all
4949 the addresses that point into the object, plus the address one byte past
4952 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
4953 the base address with silently-wrapping two's complement arithmetic, and
4954 the result value of the <tt>getelementptr</tt> may be outside the object
4955 pointed to by the base pointer. The result value may not necessarily be
4956 used to access memory though, even if it happens to point into allocated
4957 storage. See the <a href="#pointeraliasing">Pointer Aliasing Rules</a>
4958 section for more information.</p>
4960 <p>The getelementptr instruction is often confusing. For some more insight into
4961 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
4965 <i>; yields [12 x i8]*:aptr</i>
4966 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4967 <i>; yields i8*:vptr</i>
4968 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4969 <i>; yields i8*:eptr</i>
4970 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4971 <i>; yields i32*:iptr</i>
4972 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4979 <!-- ======================================================================= -->
4981 <a name="convertops">Conversion Operations</a>
4986 <p>The instructions in this category are the conversion instructions (casting)
4987 which all take a single operand and a type. They perform various bit
4988 conversions on the operand.</p>
4990 <!-- _______________________________________________________________________ -->
4992 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
4999 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
5003 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
5004 type <tt>ty2</tt>.</p>
5007 <p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5008 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5009 of the same number of integers.
5010 The bit size of the <tt>value</tt> must be larger than
5011 the bit size of the destination type, <tt>ty2</tt>.
5012 Equal sized types are not allowed.</p>
5015 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
5016 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5017 source size must be larger than the destination size, <tt>trunc</tt> cannot
5018 be a <i>no-op cast</i>. It will always truncate bits.</p>
5022 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
5023 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
5024 %Z = trunc i32 122 to i1 <i>; yields i1:false</i>
5025 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> <i>; yields <i8 8, i8 7></i>
5030 <!-- _______________________________________________________________________ -->
5032 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5039 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
5043 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
5048 <p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5049 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5050 of the same number of integers.
5051 The bit size of the <tt>value</tt> must be smaller than
5052 the bit size of the destination type,
5056 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5057 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5059 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
5063 %X = zext i32 257 to i64 <i>; yields i64:257</i>
5064 %Y = zext i1 true to i32 <i>; yields i32:1</i>
5065 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5070 <!-- _______________________________________________________________________ -->
5072 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5079 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
5083 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5086 <p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5087 Both types must be of <a href="#t_integer">integer</a> types, or vectors
5088 of the same number of integers.
5089 The bit size of the <tt>value</tt> must be smaller than
5090 the bit size of the destination type,
5094 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5095 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5096 of the type <tt>ty2</tt>.</p>
5098 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
5102 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
5103 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
5104 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> <i>; yields <i32 8, i32 7></i>
5109 <!-- _______________________________________________________________________ -->
5111 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5118 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
5122 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5126 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5127 point</a> value to cast and a <a href="#t_floating">floating point</a> type
5128 to cast it to. The size of <tt>value</tt> must be larger than the size of
5129 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5130 <i>no-op cast</i>.</p>
5133 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5134 <a href="#t_floating">floating point</a> type to a smaller
5135 <a href="#t_floating">floating point</a> type. If the value cannot fit
5136 within the destination type, <tt>ty2</tt>, then the results are
5141 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
5142 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
5147 <!-- _______________________________________________________________________ -->
5149 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5156 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
5160 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5161 floating point value.</p>
5164 <p>The '<tt>fpext</tt>' instruction takes a
5165 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5166 a <a href="#t_floating">floating point</a> type to cast it to. The source
5167 type must be smaller than the destination type.</p>
5170 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5171 <a href="#t_floating">floating point</a> type to a larger
5172 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5173 used to make a <i>no-op cast</i> because it always changes bits. Use
5174 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5178 %X = fpext float 3.125 to double <i>; yields double:3.125000e+00</i>
5179 %Y = fpext double %X to fp128 <i>; yields fp128:0xL00000000000000004000900000000000</i>
5184 <!-- _______________________________________________________________________ -->
5186 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5193 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
5197 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5198 unsigned integer equivalent of type <tt>ty2</tt>.</p>
5201 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5202 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5203 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5204 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5205 vector integer type with the same number of elements as <tt>ty</tt></p>
5208 <p>The '<tt>fptoui</tt>' instruction converts its
5209 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5210 towards zero) unsigned integer value. If the value cannot fit
5211 in <tt>ty2</tt>, the results are undefined.</p>
5215 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
5216 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
5217 %Z = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
5222 <!-- _______________________________________________________________________ -->
5224 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5231 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
5235 <p>The '<tt>fptosi</tt>' instruction converts
5236 <a href="#t_floating">floating point</a> <tt>value</tt> to
5237 type <tt>ty2</tt>.</p>
5240 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5241 scalar or vector <a href="#t_floating">floating point</a> value, and a type
5242 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5243 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5244 vector integer type with the same number of elements as <tt>ty</tt></p>
5247 <p>The '<tt>fptosi</tt>' instruction converts its
5248 <a href="#t_floating">floating point</a> operand into the nearest (rounding
5249 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5250 the results are undefined.</p>
5254 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
5255 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
5256 %Z = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
5261 <!-- _______________________________________________________________________ -->
5263 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5270 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5274 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5275 integer and converts that value to the <tt>ty2</tt> type.</p>
5278 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5279 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5280 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5281 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5282 floating point type with the same number of elements as <tt>ty</tt></p>
5285 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5286 integer quantity and converts it to the corresponding floating point
5287 value. If the value cannot fit in the floating point value, the results are
5292 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
5293 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
5298 <!-- _______________________________________________________________________ -->
5300 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5307 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
5311 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5312 and converts that value to the <tt>ty2</tt> type.</p>
5315 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5316 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5317 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5318 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5319 floating point type with the same number of elements as <tt>ty</tt></p>
5322 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5323 quantity and converts it to the corresponding floating point value. If the
5324 value cannot fit in the floating point value, the results are undefined.</p>
5328 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
5329 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
5334 <!-- _______________________________________________________________________ -->
5336 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5343 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
5347 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
5348 the integer type <tt>ty2</tt>.</p>
5351 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5352 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
5353 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
5356 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5357 <tt>ty2</tt> by interpreting the pointer value as an integer and either
5358 truncating or zero extending that value to the size of the integer type. If
5359 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5360 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5361 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5366 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
5367 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
5372 <!-- _______________________________________________________________________ -->
5374 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5381 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
5385 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5386 pointer type, <tt>ty2</tt>.</p>
5389 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5390 value to cast, and a type to cast it to, which must be a
5391 <a href="#t_pointer">pointer</a> type.</p>
5394 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5395 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5396 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5397 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5398 than the size of a pointer then a zero extension is done. If they are the
5399 same size, nothing is done (<i>no-op cast</i>).</p>
5403 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
5404 %Y = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
5405 %Z = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
5410 <!-- _______________________________________________________________________ -->
5412 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5419 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
5423 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5424 <tt>ty2</tt> without changing any bits.</p>
5427 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5428 non-aggregate first class value, and a type to cast it to, which must also be
5429 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5430 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5431 identical. If the source type is a pointer, the destination type must also be
5432 a pointer. This instruction supports bitwise conversion of vectors to
5433 integers and to vectors of other types (as long as they have the same
5437 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5438 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5439 this conversion. The conversion is done as if the <tt>value</tt> had been
5440 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
5441 be converted to other pointer types with this instruction. To convert
5442 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5443 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5447 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
5448 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
5449 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
5456 <!-- ======================================================================= -->
5458 <a name="otherops">Other Operations</a>
5463 <p>The instructions in this category are the "miscellaneous" instructions, which
5464 defy better classification.</p>
5466 <!-- _______________________________________________________________________ -->
5468 <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5475 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5479 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5480 boolean values based on comparison of its two integer, integer vector, or
5481 pointer operands.</p>
5484 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
5485 the condition code indicating the kind of comparison to perform. It is not a
5486 value, just a keyword. The possible condition code are:</p>
5489 <li><tt>eq</tt>: equal</li>
5490 <li><tt>ne</tt>: not equal </li>
5491 <li><tt>ugt</tt>: unsigned greater than</li>
5492 <li><tt>uge</tt>: unsigned greater or equal</li>
5493 <li><tt>ult</tt>: unsigned less than</li>
5494 <li><tt>ule</tt>: unsigned less or equal</li>
5495 <li><tt>sgt</tt>: signed greater than</li>
5496 <li><tt>sge</tt>: signed greater or equal</li>
5497 <li><tt>slt</tt>: signed less than</li>
5498 <li><tt>sle</tt>: signed less or equal</li>
5501 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
5502 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
5503 typed. They must also be identical types.</p>
5506 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
5507 condition code given as <tt>cond</tt>. The comparison performed always yields
5508 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
5509 result, as follows:</p>
5512 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
5513 <tt>false</tt> otherwise. No sign interpretation is necessary or
5516 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
5517 <tt>false</tt> otherwise. No sign interpretation is necessary or
5520 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
5521 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5523 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
5524 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5525 to <tt>op2</tt>.</li>
5527 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
5528 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5530 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
5531 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5533 <li><tt>sgt</tt>: interprets the operands as signed values and yields
5534 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5536 <li><tt>sge</tt>: interprets the operands as signed values and yields
5537 <tt>true</tt> if <tt>op1</tt> is greater than or equal
5538 to <tt>op2</tt>.</li>
5540 <li><tt>slt</tt>: interprets the operands as signed values and yields
5541 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
5543 <li><tt>sle</tt>: interprets the operands as signed values and yields
5544 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5547 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
5548 values are compared as if they were integers.</p>
5550 <p>If the operands are integer vectors, then they are compared element by
5551 element. The result is an <tt>i1</tt> vector with the same number of elements
5552 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
5556 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
5557 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
5558 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
5559 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
5560 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
5561 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
5564 <p>Note that the code generator does not yet support vector types with
5565 the <tt>icmp</tt> instruction.</p>
5569 <!-- _______________________________________________________________________ -->
5571 <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
5578 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
5582 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
5583 values based on comparison of its operands.</p>
5585 <p>If the operands are floating point scalars, then the result type is a boolean
5586 (<a href="#t_integer"><tt>i1</tt></a>).</p>
5588 <p>If the operands are floating point vectors, then the result type is a vector
5589 of boolean with the same number of elements as the operands being
5593 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
5594 the condition code indicating the kind of comparison to perform. It is not a
5595 value, just a keyword. The possible condition code are:</p>
5598 <li><tt>false</tt>: no comparison, always returns false</li>
5599 <li><tt>oeq</tt>: ordered and equal</li>
5600 <li><tt>ogt</tt>: ordered and greater than </li>
5601 <li><tt>oge</tt>: ordered and greater than or equal</li>
5602 <li><tt>olt</tt>: ordered and less than </li>
5603 <li><tt>ole</tt>: ordered and less than or equal</li>
5604 <li><tt>one</tt>: ordered and not equal</li>
5605 <li><tt>ord</tt>: ordered (no nans)</li>
5606 <li><tt>ueq</tt>: unordered or equal</li>
5607 <li><tt>ugt</tt>: unordered or greater than </li>
5608 <li><tt>uge</tt>: unordered or greater than or equal</li>
5609 <li><tt>ult</tt>: unordered or less than </li>
5610 <li><tt>ule</tt>: unordered or less than or equal</li>
5611 <li><tt>une</tt>: unordered or not equal</li>
5612 <li><tt>uno</tt>: unordered (either nans)</li>
5613 <li><tt>true</tt>: no comparison, always returns true</li>
5616 <p><i>Ordered</i> means that neither operand is a QNAN while
5617 <i>unordered</i> means that either operand may be a QNAN.</p>
5619 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
5620 a <a href="#t_floating">floating point</a> type or
5621 a <a href="#t_vector">vector</a> of floating point type. They must have
5622 identical types.</p>
5625 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
5626 according to the condition code given as <tt>cond</tt>. If the operands are
5627 vectors, then the vectors are compared element by element. Each comparison
5628 performed always yields an <a href="#t_integer">i1</a> result, as
5632 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
5634 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5635 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5637 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5638 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5640 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5641 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5643 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5644 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5646 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5647 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5649 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
5650 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5652 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
5654 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
5655 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
5657 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
5658 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
5660 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
5661 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
5663 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
5664 <tt>op1</tt> is less than <tt>op2</tt>.</li>
5666 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
5667 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
5669 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
5670 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
5672 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
5674 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
5679 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
5680 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
5681 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
5682 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
5685 <p>Note that the code generator does not yet support vector types with
5686 the <tt>fcmp</tt> instruction.</p>
5690 <!-- _______________________________________________________________________ -->
5692 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
5699 <result> = phi <ty> [ <val0>, <label0>], ...
5703 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
5704 SSA graph representing the function.</p>
5707 <p>The type of the incoming values is specified with the first type field. After
5708 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
5709 one pair for each predecessor basic block of the current block. Only values
5710 of <a href="#t_firstclass">first class</a> type may be used as the value
5711 arguments to the PHI node. Only labels may be used as the label
5714 <p>There must be no non-phi instructions between the start of a basic block and
5715 the PHI instructions: i.e. PHI instructions must be first in a basic
5718 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
5719 occur on the edge from the corresponding predecessor block to the current
5720 block (but after any definition of an '<tt>invoke</tt>' instruction's return
5721 value on the same edge).</p>
5724 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
5725 specified by the pair corresponding to the predecessor basic block that
5726 executed just prior to the current block.</p>
5730 Loop: ; Infinite loop that counts from 0 on up...
5731 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
5732 %nextindvar = add i32 %indvar, 1
5738 <!-- _______________________________________________________________________ -->
5740 <a name="i_select">'<tt>select</tt>' Instruction</a>
5747 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
5749 <i>selty</i> is either i1 or {<N x i1>}
5753 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
5754 condition, without branching.</p>
5758 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
5759 values indicating the condition, and two values of the
5760 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
5761 vectors and the condition is a scalar, then entire vectors are selected, not
5762 individual elements.</p>
5765 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
5766 first value argument; otherwise, it returns the second value argument.</p>
5768 <p>If the condition is a vector of i1, then the value arguments must be vectors
5769 of the same size, and the selection is done element by element.</p>
5773 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
5776 <p>Note that the code generator does not yet support conditions
5777 with vector type.</p>
5781 <!-- _______________________________________________________________________ -->
5783 <a name="i_call">'<tt>call</tt>' Instruction</a>
5790 <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>]
5794 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
5797 <p>This instruction requires several arguments:</p>
5800 <li>The optional "tail" marker indicates that the callee function does not
5801 access any allocas or varargs in the caller. Note that calls may be
5802 marked "tail" even if they do not occur before
5803 a <a href="#i_ret"><tt>ret</tt></a> instruction. If the "tail" marker is
5804 present, the function call is eligible for tail call optimization,
5805 but <a href="CodeGenerator.html#tailcallopt">might not in fact be
5806 optimized into a jump</a>. The code generator may optimize calls marked
5807 "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
5808 sibling call optimization</a> when the caller and callee have
5809 matching signatures, or 2) forced tail call optimization when the
5810 following extra requirements are met:
5812 <li>Caller and callee both have the calling
5813 convention <tt>fastcc</tt>.</li>
5814 <li>The call is in tail position (ret immediately follows call and ret
5815 uses value of call or is void).</li>
5816 <li>Option <tt>-tailcallopt</tt> is enabled,
5817 or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
5818 <li><a href="CodeGenerator.html#tailcallopt">Platform specific
5819 constraints are met.</a></li>
5823 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
5824 convention</a> the call should use. If none is specified, the call
5825 defaults to using C calling conventions. The calling convention of the
5826 call must match the calling convention of the target function, or else the
5827 behavior is undefined.</li>
5829 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
5830 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
5831 '<tt>inreg</tt>' attributes are valid here.</li>
5833 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
5834 type of the return value. Functions that return no value are marked
5835 <tt><a href="#t_void">void</a></tt>.</li>
5837 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5838 being invoked. The argument types must match the types implied by this
5839 signature. This type can be omitted if the function is not varargs and if
5840 the function type does not return a pointer to a function.</li>
5842 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5843 be invoked. In most cases, this is a direct function invocation, but
5844 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5845 to function value.</li>
5847 <li>'<tt>function args</tt>': argument list whose types match the function
5848 signature argument types and parameter attributes. All arguments must be
5849 of <a href="#t_firstclass">first class</a> type. If the function
5850 signature indicates the function accepts a variable number of arguments,
5851 the extra arguments can be specified.</li>
5853 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
5854 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
5855 '<tt>readnone</tt>' attributes are valid here.</li>
5859 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
5860 a specified function, with its incoming arguments bound to the specified
5861 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
5862 function, control flow continues with the instruction after the function
5863 call, and the return value of the function is bound to the result
5868 %retval = call i32 @test(i32 %argc)
5869 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) <i>; yields i32</i>
5870 %X = tail call i32 @foo() <i>; yields i32</i>
5871 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
5872 call void %foo(i8 97 signext)
5874 %struct.A = type { i32, i8 }
5875 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
5876 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
5877 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
5878 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
5879 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
5882 <p>llvm treats calls to some functions with names and arguments that match the
5883 standard C99 library as being the C99 library functions, and may perform
5884 optimizations or generate code for them under that assumption. This is
5885 something we'd like to change in the future to provide better support for
5886 freestanding environments and non-C-based languages.</p>
5890 <!-- _______________________________________________________________________ -->
5892 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
5899 <resultval> = va_arg <va_list*> <arglist>, <argty>
5903 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
5904 the "variable argument" area of a function call. It is used to implement the
5905 <tt>va_arg</tt> macro in C.</p>
5908 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
5909 argument. It returns a value of the specified argument type and increments
5910 the <tt>va_list</tt> to point to the next argument. The actual type
5911 of <tt>va_list</tt> is target specific.</p>
5914 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
5915 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
5916 to the next argument. For more information, see the variable argument
5917 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
5919 <p>It is legal for this instruction to be called in a function which does not
5920 take a variable number of arguments, for example, the <tt>vfprintf</tt>
5923 <p><tt>va_arg</tt> is an LLVM instruction instead of
5924 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
5928 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
5930 <p>Note that the code generator does not yet fully support va_arg on many
5931 targets. Also, it does not currently support va_arg with aggregate types on
5940 <!-- *********************************************************************** -->
5941 <h2><a name="intrinsics">Intrinsic Functions</a></h2>
5942 <!-- *********************************************************************** -->
5946 <p>LLVM supports the notion of an "intrinsic function". These functions have
5947 well known names and semantics and are required to follow certain
5948 restrictions. Overall, these intrinsics represent an extension mechanism for
5949 the LLVM language that does not require changing all of the transformations
5950 in LLVM when adding to the language (or the bitcode reader/writer, the
5951 parser, etc...).</p>
5953 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
5954 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5955 begin with this prefix. Intrinsic functions must always be external
5956 functions: you cannot define the body of intrinsic functions. Intrinsic
5957 functions may only be used in call or invoke instructions: it is illegal to
5958 take the address of an intrinsic function. Additionally, because intrinsic
5959 functions are part of the LLVM language, it is required if any are added that
5960 they be documented here.</p>
5962 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5963 family of functions that perform the same operation but on different data
5964 types. Because LLVM can represent over 8 million different integer types,
5965 overloading is used commonly to allow an intrinsic function to operate on any
5966 integer type. One or more of the argument types or the result type can be
5967 overloaded to accept any integer type. Argument types may also be defined as
5968 exactly matching a previous argument's type or the result type. This allows
5969 an intrinsic function which accepts multiple arguments, but needs all of them
5970 to be of the same type, to only be overloaded with respect to a single
5971 argument or the result.</p>
5973 <p>Overloaded intrinsics will have the names of its overloaded argument types
5974 encoded into its function name, each preceded by a period. Only those types
5975 which are overloaded result in a name suffix. Arguments whose type is matched
5976 against another type do not. For example, the <tt>llvm.ctpop</tt> function
5977 can take an integer of any width and returns an integer of exactly the same
5978 integer width. This leads to a family of functions such as
5979 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
5980 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5981 suffix is required. Because the argument's type is matched against the return
5982 type, it does not require its own name suffix.</p>
5984 <p>To learn how to add an intrinsic function, please see the
5985 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
5987 <!-- ======================================================================= -->
5989 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
5994 <p>Variable argument support is defined in LLVM with
5995 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
5996 intrinsic functions. These functions are related to the similarly named
5997 macros defined in the <tt><stdarg.h></tt> header file.</p>
5999 <p>All of these functions operate on arguments that use a target-specific value
6000 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
6001 not define what this type is, so all transformations should be prepared to
6002 handle these functions regardless of the type used.</p>
6004 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6005 instruction and the variable argument handling intrinsic functions are
6008 <pre class="doc_code">
6009 define i32 @test(i32 %X, ...) {
6010 ; Initialize variable argument processing
6012 %ap2 = bitcast i8** %ap to i8*
6013 call void @llvm.va_start(i8* %ap2)
6015 ; Read a single integer argument
6016 %tmp = va_arg i8** %ap, i32
6018 ; Demonstrate usage of llvm.va_copy and llvm.va_end
6020 %aq2 = bitcast i8** %aq to i8*
6021 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6022 call void @llvm.va_end(i8* %aq2)
6024 ; Stop processing of arguments.
6025 call void @llvm.va_end(i8* %ap2)
6029 declare void @llvm.va_start(i8*)
6030 declare void @llvm.va_copy(i8*, i8*)
6031 declare void @llvm.va_end(i8*)
6034 <!-- _______________________________________________________________________ -->
6036 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6044 declare void %llvm.va_start(i8* <arglist>)
6048 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
6049 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6052 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6055 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6056 macro available in C. In a target-dependent way, it initializes
6057 the <tt>va_list</tt> element to which the argument points, so that the next
6058 call to <tt>va_arg</tt> will produce the first variable argument passed to
6059 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6060 need to know the last argument of the function as the compiler can figure
6065 <!-- _______________________________________________________________________ -->
6067 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6074 declare void @llvm.va_end(i8* <arglist>)
6078 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
6079 which has been initialized previously
6080 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6081 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6084 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6087 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6088 macro available in C. In a target-dependent way, it destroys
6089 the <tt>va_list</tt> element to which the argument points. Calls
6090 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6091 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6092 with calls to <tt>llvm.va_end</tt>.</p>
6096 <!-- _______________________________________________________________________ -->
6098 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6105 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
6109 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6110 from the source argument list to the destination argument list.</p>
6113 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6114 The second argument is a pointer to a <tt>va_list</tt> element to copy
6118 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6119 macro available in C. In a target-dependent way, it copies the
6120 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6121 element. This intrinsic is necessary because
6122 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6123 arbitrarily complex and require, for example, memory allocation.</p>
6129 <!-- ======================================================================= -->
6131 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6136 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6137 Collection</a> (GC) requires the implementation and generation of these
6138 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6139 roots on the stack</a>, as well as garbage collector implementations that
6140 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6141 barriers. Front-ends for type-safe garbage collected languages should generate
6142 these intrinsics to make use of the LLVM garbage collectors. For more details,
6143 see <a href="GarbageCollection.html">Accurate Garbage Collection with
6146 <p>The garbage collection intrinsics only operate on objects in the generic
6147 address space (address space zero).</p>
6149 <!-- _______________________________________________________________________ -->
6151 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6158 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6162 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6163 the code generator, and allows some metadata to be associated with it.</p>
6166 <p>The first argument specifies the address of a stack object that contains the
6167 root pointer. The second pointer (which must be either a constant or a
6168 global value address) contains the meta-data to be associated with the
6172 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6173 location. At compile-time, the code generator generates information to allow
6174 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6175 intrinsic may only be used in a function which <a href="#gc">specifies a GC
6180 <!-- _______________________________________________________________________ -->
6182 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6189 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6193 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6194 locations, allowing garbage collector implementations that require read
6198 <p>The second argument is the address to read from, which should be an address
6199 allocated from the garbage collector. The first object is a pointer to the
6200 start of the referenced object, if needed by the language runtime (otherwise
6204 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6205 instruction, but may be replaced with substantially more complex code by the
6206 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6207 may only be used in a function which <a href="#gc">specifies a GC
6212 <!-- _______________________________________________________________________ -->
6214 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6221 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6225 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6226 locations, allowing garbage collector implementations that require write
6227 barriers (such as generational or reference counting collectors).</p>
6230 <p>The first argument is the reference to store, the second is the start of the
6231 object to store it to, and the third is the address of the field of Obj to
6232 store to. If the runtime does not require a pointer to the object, Obj may
6236 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6237 instruction, but may be replaced with substantially more complex code by the
6238 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6239 may only be used in a function which <a href="#gc">specifies a GC
6246 <!-- ======================================================================= -->
6248 <a name="int_codegen">Code Generator Intrinsics</a>
6253 <p>These intrinsics are provided by LLVM to expose special features that may
6254 only be implemented with code generator support.</p>
6256 <!-- _______________________________________________________________________ -->
6258 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6265 declare i8 *@llvm.returnaddress(i32 <level>)
6269 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6270 target-specific value indicating the return address of the current function
6271 or one of its callers.</p>
6274 <p>The argument to this intrinsic indicates which function to return the address
6275 for. Zero indicates the calling function, one indicates its caller, etc.
6276 The argument is <b>required</b> to be a constant integer value.</p>
6279 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6280 indicating the return address of the specified call frame, or zero if it
6281 cannot be identified. The value returned by this intrinsic is likely to be
6282 incorrect or 0 for arguments other than zero, so it should only be used for
6283 debugging purposes.</p>
6285 <p>Note that calling this intrinsic does not prevent function inlining or other
6286 aggressive transformations, so the value returned may not be that of the
6287 obvious source-language caller.</p>
6291 <!-- _______________________________________________________________________ -->
6293 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6300 declare i8* @llvm.frameaddress(i32 <level>)
6304 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6305 target-specific frame pointer value for the specified stack frame.</p>
6308 <p>The argument to this intrinsic indicates which function to return the frame
6309 pointer for. Zero indicates the calling function, one indicates its caller,
6310 etc. The argument is <b>required</b> to be a constant integer value.</p>
6313 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6314 indicating the frame address of the specified call frame, or zero if it
6315 cannot be identified. The value returned by this intrinsic is likely to be
6316 incorrect or 0 for arguments other than zero, so it should only be used for
6317 debugging purposes.</p>
6319 <p>Note that calling this intrinsic does not prevent function inlining or other
6320 aggressive transformations, so the value returned may not be that of the
6321 obvious source-language caller.</p>
6325 <!-- _______________________________________________________________________ -->
6327 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6334 declare i8* @llvm.stacksave()
6338 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6339 of the function stack, for use
6340 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
6341 useful for implementing language features like scoped automatic variable
6342 sized arrays in C99.</p>
6345 <p>This intrinsic returns a opaque pointer value that can be passed
6346 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
6347 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6348 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6349 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6350 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6351 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6355 <!-- _______________________________________________________________________ -->
6357 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6364 declare void @llvm.stackrestore(i8* %ptr)
6368 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6369 the function stack to the state it was in when the
6370 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6371 executed. This is useful for implementing language features like scoped
6372 automatic variable sized arrays in C99.</p>
6375 <p>See the description
6376 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6380 <!-- _______________________________________________________________________ -->
6382 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6389 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
6393 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6394 insert a prefetch instruction if supported; otherwise, it is a noop.
6395 Prefetches have no effect on the behavior of the program but can change its
6396 performance characteristics.</p>
6399 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6400 specifier determining if the fetch should be for a read (0) or write (1),
6401 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
6402 locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
6403 specifies whether the prefetch is performed on the data (1) or instruction (0)
6404 cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
6405 must be constant integers.</p>
6408 <p>This intrinsic does not modify the behavior of the program. In particular,
6409 prefetches cannot trap and do not produce a value. On targets that support
6410 this intrinsic, the prefetch can provide hints to the processor cache for
6411 better performance.</p>
6415 <!-- _______________________________________________________________________ -->
6417 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
6424 declare void @llvm.pcmarker(i32 <id>)
6428 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
6429 Counter (PC) in a region of code to simulators and other tools. The method
6430 is target specific, but it is expected that the marker will use exported
6431 symbols to transmit the PC of the marker. The marker makes no guarantees
6432 that it will remain with any specific instruction after optimizations. It is
6433 possible that the presence of a marker will inhibit optimizations. The
6434 intended use is to be inserted after optimizations to allow correlations of
6435 simulation runs.</p>
6438 <p><tt>id</tt> is a numerical id identifying the marker.</p>
6441 <p>This intrinsic does not modify the behavior of the program. Backends that do
6442 not support this intrinsic may ignore it.</p>
6446 <!-- _______________________________________________________________________ -->
6448 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
6455 declare i64 @llvm.readcyclecounter()
6459 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
6460 counter register (or similar low latency, high accuracy clocks) on those
6461 targets that support it. On X86, it should map to RDTSC. On Alpha, it
6462 should map to RPCC. As the backing counters overflow quickly (on the order
6463 of 9 seconds on alpha), this should only be used for small timings.</p>
6466 <p>When directly supported, reading the cycle counter should not modify any
6467 memory. Implementations are allowed to either return a application specific
6468 value or a system wide value. On backends without support, this is lowered
6469 to a constant 0.</p>
6475 <!-- ======================================================================= -->
6477 <a name="int_libc">Standard C Library Intrinsics</a>
6482 <p>LLVM provides intrinsics for a few important standard C library functions.
6483 These intrinsics allow source-language front-ends to pass information about
6484 the alignment of the pointer arguments to the code generator, providing
6485 opportunity for more efficient code generation.</p>
6487 <!-- _______________________________________________________________________ -->
6489 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
6495 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
6496 integer bit width and for different address spaces. Not all targets support
6497 all bit widths however.</p>
6500 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6501 i32 <len>, i32 <align>, i1 <isvolatile>)
6502 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6503 i64 <len>, i32 <align>, i1 <isvolatile>)
6507 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6508 source location to the destination location.</p>
6510 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
6511 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6512 and the pointers can be in specified address spaces.</p>
6516 <p>The first argument is a pointer to the destination, the second is a pointer
6517 to the source. The third argument is an integer argument specifying the
6518 number of bytes to copy, the fourth argument is the alignment of the
6519 source and destination locations, and the fifth is a boolean indicating a
6520 volatile access.</p>
6522 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6523 then the caller guarantees that both the source and destination pointers are
6524 aligned to that boundary.</p>
6526 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6527 <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
6528 The detailed access behavior is not very cleanly specified and it is unwise
6529 to depend on it.</p>
6533 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
6534 source location to the destination location, which are not allowed to
6535 overlap. It copies "len" bytes of memory over. If the argument is known to
6536 be aligned to some boundary, this can be specified as the fourth argument,
6537 otherwise it should be set to 0 or 1.</p>
6541 <!-- _______________________________________________________________________ -->
6543 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
6549 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
6550 width and for different address space. Not all targets support all bit
6554 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
6555 i32 <len>, i32 <align>, i1 <isvolatile>)
6556 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
6557 i64 <len>, i32 <align>, i1 <isvolatile>)
6561 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
6562 source location to the destination location. It is similar to the
6563 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
6566 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
6567 intrinsics do not return a value, takes extra alignment/isvolatile arguments
6568 and the pointers can be in specified address spaces.</p>
6572 <p>The first argument is a pointer to the destination, the second is a pointer
6573 to the source. The third argument is an integer argument specifying the
6574 number of bytes to copy, the fourth argument is the alignment of the
6575 source and destination locations, and the fifth is a boolean indicating a
6576 volatile access.</p>
6578 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6579 then the caller guarantees that the source and destination pointers are
6580 aligned to that boundary.</p>
6582 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6583 <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
6584 The detailed access behavior is not very cleanly specified and it is unwise
6585 to depend on it.</p>
6589 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
6590 source location to the destination location, which may overlap. It copies
6591 "len" bytes of memory over. If the argument is known to be aligned to some
6592 boundary, this can be specified as the fourth argument, otherwise it should
6593 be set to 0 or 1.</p>
6597 <!-- _______________________________________________________________________ -->
6599 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
6605 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
6606 width and for different address spaces. However, not all targets support all
6610 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
6611 i32 <len>, i32 <align>, i1 <isvolatile>)
6612 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
6613 i64 <len>, i32 <align>, i1 <isvolatile>)
6617 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
6618 particular byte value.</p>
6620 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
6621 intrinsic does not return a value and takes extra alignment/volatile
6622 arguments. Also, the destination can be in an arbitrary address space.</p>
6625 <p>The first argument is a pointer to the destination to fill, the second is the
6626 byte value with which to fill it, the third argument is an integer argument
6627 specifying the number of bytes to fill, and the fourth argument is the known
6628 alignment of the destination location.</p>
6630 <p>If the call to this intrinsic has an alignment value that is not 0 or 1,
6631 then the caller guarantees that the destination pointer is aligned to that
6634 <p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
6635 <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
6636 The detailed access behavior is not very cleanly specified and it is unwise
6637 to depend on it.</p>
6640 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
6641 at the destination location. If the argument is known to be aligned to some
6642 boundary, this can be specified as the fourth argument, otherwise it should
6643 be set to 0 or 1.</p>
6647 <!-- _______________________________________________________________________ -->
6649 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
6655 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
6656 floating point or vector of floating point type. Not all targets support all
6660 declare float @llvm.sqrt.f32(float %Val)
6661 declare double @llvm.sqrt.f64(double %Val)
6662 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
6663 declare fp128 @llvm.sqrt.f128(fp128 %Val)
6664 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
6668 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
6669 returning the same value as the libm '<tt>sqrt</tt>' functions would.
6670 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
6671 behavior for negative numbers other than -0.0 (which allows for better
6672 optimization, because there is no need to worry about errno being
6673 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
6676 <p>The argument and return value are floating point numbers of the same
6680 <p>This function returns the sqrt of the specified operand if it is a
6681 nonnegative floating point number.</p>
6685 <!-- _______________________________________________________________________ -->
6687 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
6693 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
6694 floating point or vector of floating point type. Not all targets support all
6698 declare float @llvm.powi.f32(float %Val, i32 %power)
6699 declare double @llvm.powi.f64(double %Val, i32 %power)
6700 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
6701 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
6702 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
6706 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
6707 specified (positive or negative) power. The order of evaluation of
6708 multiplications is not defined. When a vector of floating point type is
6709 used, the second argument remains a scalar integer value.</p>
6712 <p>The second argument is an integer power, and the first is a value to raise to
6716 <p>This function returns the first value raised to the second power with an
6717 unspecified sequence of rounding operations.</p>
6721 <!-- _______________________________________________________________________ -->
6723 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
6729 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
6730 floating point or vector of floating point type. Not all targets support all
6734 declare float @llvm.sin.f32(float %Val)
6735 declare double @llvm.sin.f64(double %Val)
6736 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
6737 declare fp128 @llvm.sin.f128(fp128 %Val)
6738 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
6742 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
6745 <p>The argument and return value are floating point numbers of the same
6749 <p>This function returns the sine of the specified operand, returning the same
6750 values as the libm <tt>sin</tt> functions would, and handles error conditions
6751 in the same way.</p>
6755 <!-- _______________________________________________________________________ -->
6757 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
6763 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
6764 floating point or vector of floating point type. Not all targets support all
6768 declare float @llvm.cos.f32(float %Val)
6769 declare double @llvm.cos.f64(double %Val)
6770 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
6771 declare fp128 @llvm.cos.f128(fp128 %Val)
6772 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
6776 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
6779 <p>The argument and return value are floating point numbers of the same
6783 <p>This function returns the cosine of the specified operand, returning the same
6784 values as the libm <tt>cos</tt> functions would, and handles error conditions
6785 in the same way.</p>
6789 <!-- _______________________________________________________________________ -->
6791 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
6797 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
6798 floating point or vector of floating point type. Not all targets support all
6802 declare float @llvm.pow.f32(float %Val, float %Power)
6803 declare double @llvm.pow.f64(double %Val, double %Power)
6804 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
6805 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
6806 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
6810 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
6811 specified (positive or negative) power.</p>
6814 <p>The second argument is a floating point power, and the first is a value to
6815 raise to that power.</p>
6818 <p>This function returns the first value raised to the second power, returning
6819 the same values as the libm <tt>pow</tt> functions would, and handles error
6820 conditions in the same way.</p>
6826 <!-- _______________________________________________________________________ -->
6828 <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
6834 <p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
6835 floating point or vector of floating point type. Not all targets support all
6839 declare float @llvm.exp.f32(float %Val)
6840 declare double @llvm.exp.f64(double %Val)
6841 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
6842 declare fp128 @llvm.exp.f128(fp128 %Val)
6843 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
6847 <p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
6850 <p>The argument and return value are floating point numbers of the same
6854 <p>This function returns the same values as the libm <tt>exp</tt> functions
6855 would, and handles error conditions in the same way.</p>
6859 <!-- _______________________________________________________________________ -->
6861 <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
6867 <p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
6868 floating point or vector of floating point type. Not all targets support all
6872 declare float @llvm.log.f32(float %Val)
6873 declare double @llvm.log.f64(double %Val)
6874 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
6875 declare fp128 @llvm.log.f128(fp128 %Val)
6876 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
6880 <p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
6883 <p>The argument and return value are floating point numbers of the same
6887 <p>This function returns the same values as the libm <tt>log</tt> functions
6888 would, and handles error conditions in the same way.</p>
6891 <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
6897 <p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
6898 floating point or vector of floating point type. Not all targets support all
6902 declare float @llvm.fma.f32(float %a, float %b, float %c)
6903 declare double @llvm.fma.f64(double %a, double %b, double %c)
6904 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
6905 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
6906 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
6910 <p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
6914 <p>The argument and return value are floating point numbers of the same
6918 <p>This function returns the same values as the libm <tt>fma</tt> functions
6923 <!-- ======================================================================= -->
6925 <a name="int_manip">Bit Manipulation Intrinsics</a>
6930 <p>LLVM provides intrinsics for a few important bit manipulation operations.
6931 These allow efficient code generation for some algorithms.</p>
6933 <!-- _______________________________________________________________________ -->
6935 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
6941 <p>This is an overloaded intrinsic function. You can use bswap on any integer
6942 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
6945 declare i16 @llvm.bswap.i16(i16 <id>)
6946 declare i32 @llvm.bswap.i32(i32 <id>)
6947 declare i64 @llvm.bswap.i64(i64 <id>)
6951 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
6952 values with an even number of bytes (positive multiple of 16 bits). These
6953 are useful for performing operations on data that is not in the target's
6954 native byte order.</p>
6957 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
6958 and low byte of the input i16 swapped. Similarly,
6959 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
6960 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
6961 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
6962 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
6963 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
6964 more, respectively).</p>
6968 <!-- _______________________________________________________________________ -->
6970 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
6976 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
6977 width, or on any vector with integer elements. Not all targets support all
6978 bit widths or vector types, however.</p>
6981 declare i8 @llvm.ctpop.i8(i8 <src>)
6982 declare i16 @llvm.ctpop.i16(i16 <src>)
6983 declare i32 @llvm.ctpop.i32(i32 <src>)
6984 declare i64 @llvm.ctpop.i64(i64 <src>)
6985 declare i256 @llvm.ctpop.i256(i256 <src>)
6986 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
6990 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
6994 <p>The only argument is the value to be counted. The argument may be of any
6995 integer type, or a vector with integer elements.
6996 The return type must match the argument type.</p>
6999 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7000 element of a vector.</p>
7004 <!-- _______________________________________________________________________ -->
7006 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7012 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7013 integer bit width, or any vector whose elements are integers. Not all
7014 targets support all bit widths or vector types, however.</p>
7017 declare i8 @llvm.ctlz.i8 (i8 <src>)
7018 declare i16 @llvm.ctlz.i16(i16 <src>)
7019 declare i32 @llvm.ctlz.i32(i32 <src>)
7020 declare i64 @llvm.ctlz.i64(i64 <src>)
7021 declare i256 @llvm.ctlz.i256(i256 <src>)
7022 declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src;gt)
7026 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7027 leading zeros in a variable.</p>
7030 <p>The only argument is the value to be counted. The argument may be of any
7031 integer type, or any vector type with integer element type.
7032 The return type must match the argument type.</p>
7035 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7036 zeros in a variable, or within each element of the vector if the operation
7037 is of vector type. If the src == 0 then the result is the size in bits of
7038 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7042 <!-- _______________________________________________________________________ -->
7044 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7050 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7051 integer bit width, or any vector of integer elements. Not all targets
7052 support all bit widths or vector types, however.</p>
7055 declare i8 @llvm.cttz.i8 (i8 <src>)
7056 declare i16 @llvm.cttz.i16(i16 <src>)
7057 declare i32 @llvm.cttz.i32(i32 <src>)
7058 declare i64 @llvm.cttz.i64(i64 <src>)
7059 declare i256 @llvm.cttz.i256(i256 <src>)
7060 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>)
7064 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7068 <p>The only argument is the value to be counted. The argument may be of any
7069 integer type, or a vectory with integer element type.. The return type
7070 must match the argument type.</p>
7073 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7074 zeros in a variable, or within each element of a vector.
7075 If the src == 0 then the result is the size in bits of
7076 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7082 <!-- ======================================================================= -->
7084 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7089 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7091 <!-- _______________________________________________________________________ -->
7093 <a name="int_sadd_overflow">
7094 '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7101 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7102 on any integer bit width.</p>
7105 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7106 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7107 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7111 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7112 a signed addition of the two arguments, and indicate whether an overflow
7113 occurred during the signed summation.</p>
7116 <p>The arguments (%a and %b) and the first element of the result structure may
7117 be of integer types of any bit width, but they must have the same bit
7118 width. The second element of the result structure must be of
7119 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7120 undergo signed addition.</p>
7123 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7124 a signed addition of the two variables. They return a structure — the
7125 first element of which is the signed summation, and the second element of
7126 which is a bit specifying if the signed summation resulted in an
7131 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7132 %sum = extractvalue {i32, i1} %res, 0
7133 %obit = extractvalue {i32, i1} %res, 1
7134 br i1 %obit, label %overflow, label %normal
7139 <!-- _______________________________________________________________________ -->
7141 <a name="int_uadd_overflow">
7142 '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7149 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7150 on any integer bit width.</p>
7153 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7154 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7155 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7159 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7160 an unsigned addition of the two arguments, and indicate whether a carry
7161 occurred during the unsigned summation.</p>
7164 <p>The arguments (%a and %b) and the first element of the result structure may
7165 be of integer types of any bit width, but they must have the same bit
7166 width. The second element of the result structure must be of
7167 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7168 undergo unsigned addition.</p>
7171 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7172 an unsigned addition of the two arguments. They return a structure —
7173 the first element of which is the sum, and the second element of which is a
7174 bit specifying if the unsigned summation resulted in a carry.</p>
7178 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7179 %sum = extractvalue {i32, i1} %res, 0
7180 %obit = extractvalue {i32, i1} %res, 1
7181 br i1 %obit, label %carry, label %normal
7186 <!-- _______________________________________________________________________ -->
7188 <a name="int_ssub_overflow">
7189 '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7196 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7197 on any integer bit width.</p>
7200 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7201 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7202 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7206 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7207 a signed subtraction of the two arguments, and indicate whether an overflow
7208 occurred during the signed subtraction.</p>
7211 <p>The arguments (%a and %b) and the first element of the result structure may
7212 be of integer types of any bit width, but they must have the same bit
7213 width. The second element of the result structure must be of
7214 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7215 undergo signed subtraction.</p>
7218 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7219 a signed subtraction of the two arguments. They return a structure —
7220 the first element of which is the subtraction, and the second element of
7221 which is a bit specifying if the signed subtraction resulted in an
7226 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7227 %sum = extractvalue {i32, i1} %res, 0
7228 %obit = extractvalue {i32, i1} %res, 1
7229 br i1 %obit, label %overflow, label %normal
7234 <!-- _______________________________________________________________________ -->
7236 <a name="int_usub_overflow">
7237 '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7244 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7245 on any integer bit width.</p>
7248 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7249 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7250 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7254 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7255 an unsigned subtraction of the two arguments, and indicate whether an
7256 overflow occurred during the unsigned subtraction.</p>
7259 <p>The arguments (%a and %b) and the first element of the result structure may
7260 be of integer types of any bit width, but they must have the same bit
7261 width. The second element of the result structure must be of
7262 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7263 undergo unsigned subtraction.</p>
7266 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7267 an unsigned subtraction of the two arguments. They return a structure —
7268 the first element of which is the subtraction, and the second element of
7269 which is a bit specifying if the unsigned subtraction resulted in an
7274 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7275 %sum = extractvalue {i32, i1} %res, 0
7276 %obit = extractvalue {i32, i1} %res, 1
7277 br i1 %obit, label %overflow, label %normal
7282 <!-- _______________________________________________________________________ -->
7284 <a name="int_smul_overflow">
7285 '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7292 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7293 on any integer bit width.</p>
7296 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7297 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7298 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7303 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7304 a signed multiplication of the two arguments, and indicate whether an
7305 overflow occurred during the signed multiplication.</p>
7308 <p>The arguments (%a and %b) and the first element of the result structure may
7309 be of integer types of any bit width, but they must have the same bit
7310 width. The second element of the result structure must be of
7311 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7312 undergo signed multiplication.</p>
7315 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7316 a signed multiplication of the two arguments. They return a structure —
7317 the first element of which is the multiplication, and the second element of
7318 which is a bit specifying if the signed multiplication resulted in an
7323 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7324 %sum = extractvalue {i32, i1} %res, 0
7325 %obit = extractvalue {i32, i1} %res, 1
7326 br i1 %obit, label %overflow, label %normal
7331 <!-- _______________________________________________________________________ -->
7333 <a name="int_umul_overflow">
7334 '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
7341 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
7342 on any integer bit width.</p>
7345 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
7346 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7347 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
7351 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7352 a unsigned multiplication of the two arguments, and indicate whether an
7353 overflow occurred during the unsigned multiplication.</p>
7356 <p>The arguments (%a and %b) and the first element of the result structure may
7357 be of integer types of any bit width, but they must have the same bit
7358 width. The second element of the result structure must be of
7359 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7360 undergo unsigned multiplication.</p>
7363 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
7364 an unsigned multiplication of the two arguments. They return a structure
7365 — the first element of which is the multiplication, and the second
7366 element of which is a bit specifying if the unsigned multiplication resulted
7371 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
7372 %sum = extractvalue {i32, i1} %res, 0
7373 %obit = extractvalue {i32, i1} %res, 1
7374 br i1 %obit, label %overflow, label %normal
7381 <!-- ======================================================================= -->
7383 <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
7388 <p>Half precision floating point is a storage-only format. This means that it is
7389 a dense encoding (in memory) but does not support computation in the
7392 <p>This means that code must first load the half-precision floating point
7393 value as an i16, then convert it to float with <a
7394 href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
7395 Computation can then be performed on the float value (including extending to
7396 double etc). To store the value back to memory, it is first converted to
7397 float if needed, then converted to i16 with
7398 <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
7399 storing as an i16 value.</p>
7401 <!-- _______________________________________________________________________ -->
7403 <a name="int_convert_to_fp16">
7404 '<tt>llvm.convert.to.fp16</tt>' Intrinsic
7412 declare i16 @llvm.convert.to.fp16(f32 %a)
7416 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7417 a conversion from single precision floating point format to half precision
7418 floating point format.</p>
7421 <p>The intrinsic function contains single argument - the value to be
7425 <p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
7426 a conversion from single precision floating point format to half precision
7427 floating point format. The return value is an <tt>i16</tt> which
7428 contains the converted number.</p>
7432 %res = call i16 @llvm.convert.to.fp16(f32 %a)
7433 store i16 %res, i16* @x, align 2
7438 <!-- _______________________________________________________________________ -->
7440 <a name="int_convert_from_fp16">
7441 '<tt>llvm.convert.from.fp16</tt>' Intrinsic
7449 declare f32 @llvm.convert.from.fp16(i16 %a)
7453 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
7454 a conversion from half precision floating point format to single precision
7455 floating point format.</p>
7458 <p>The intrinsic function contains single argument - the value to be
7462 <p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
7463 conversion from half single precision floating point format to single
7464 precision floating point format. The input half-float value is represented by
7465 an <tt>i16</tt> value.</p>
7469 %a = load i16* @x, align 2
7470 %res = call f32 @llvm.convert.from.fp16(i16 %a)
7477 <!-- ======================================================================= -->
7479 <a name="int_debugger">Debugger Intrinsics</a>
7484 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
7485 prefix), are described in
7486 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
7487 Level Debugging</a> document.</p>
7491 <!-- ======================================================================= -->
7493 <a name="int_eh">Exception Handling Intrinsics</a>
7498 <p>The LLVM exception handling intrinsics (which all start with
7499 <tt>llvm.eh.</tt> prefix), are described in
7500 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
7501 Handling</a> document.</p>
7505 <!-- ======================================================================= -->
7507 <a name="int_trampoline">Trampoline Intrinsic</a>
7512 <p>This intrinsic makes it possible to excise one parameter, marked with
7513 the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
7514 The result is a callable
7515 function pointer lacking the nest parameter - the caller does not need to
7516 provide a value for it. Instead, the value to use is stored in advance in a
7517 "trampoline", a block of memory usually allocated on the stack, which also
7518 contains code to splice the nest value into the argument list. This is used
7519 to implement the GCC nested function address extension.</p>
7521 <p>For example, if the function is
7522 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
7523 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
7526 <pre class="doc_code">
7527 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
7528 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
7529 %p = call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval)
7530 %fp = bitcast i8* %p to i32 (i32, i32)*
7533 <p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
7534 to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
7536 <!-- _______________________________________________________________________ -->
7539 '<tt>llvm.init.trampoline</tt>' Intrinsic
7547 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
7551 <p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
7552 function pointer suitable for executing it.</p>
7555 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
7556 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
7557 sufficiently aligned block of memory; this memory is written to by the
7558 intrinsic. Note that the size and the alignment are target-specific - LLVM
7559 currently provides no portable way of determining them, so a front-end that
7560 generates this intrinsic needs to have some target-specific knowledge.
7561 The <tt>func</tt> argument must hold a function bitcast to
7562 an <tt>i8*</tt>.</p>
7565 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
7566 dependent code, turning it into a function. A pointer to this function is
7567 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
7568 function pointer type</a> before being called. The new function's signature
7569 is the same as that of <tt>func</tt> with any arguments marked with
7570 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
7571 is allowed, and it must be of pointer type. Calling the new function is
7572 equivalent to calling <tt>func</tt> with the same argument list, but
7573 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
7574 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
7575 by <tt>tramp</tt> is modified, then the effect of any later call to the
7576 returned function pointer is undefined.</p>
7582 <!-- ======================================================================= -->
7584 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
7589 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
7590 hardware constructs for atomic operations and memory synchronization. This
7591 provides an interface to the hardware, not an interface to the programmer. It
7592 is aimed at a low enough level to allow any programming models or APIs
7593 (Application Programming Interfaces) which need atomic behaviors to map
7594 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
7595 hardware provides a "universal IR" for source languages, it also provides a
7596 starting point for developing a "universal" atomic operation and
7597 synchronization IR.</p>
7599 <p>These do <em>not</em> form an API such as high-level threading libraries,
7600 software transaction memory systems, atomic primitives, and intrinsic
7601 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
7602 application libraries. The hardware interface provided by LLVM should allow
7603 a clean implementation of all of these APIs and parallel programming models.
7604 No one model or paradigm should be selected above others unless the hardware
7605 itself ubiquitously does so.</p>
7607 <!-- _______________________________________________________________________ -->
7609 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
7615 declare void @llvm.memory.barrier(i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device>)
7619 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
7620 specific pairs of memory access types.</p>
7623 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
7624 The first four arguments enables a specific barrier as listed below. The
7625 fifth argument specifies that the barrier applies to io or device or uncached
7629 <li><tt>ll</tt>: load-load barrier</li>
7630 <li><tt>ls</tt>: load-store barrier</li>
7631 <li><tt>sl</tt>: store-load barrier</li>
7632 <li><tt>ss</tt>: store-store barrier</li>
7633 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
7637 <p>This intrinsic causes the system to enforce some ordering constraints upon
7638 the loads and stores of the program. This barrier does not
7639 indicate <em>when</em> any events will occur, it only enforces
7640 an <em>order</em> in which they occur. For any of the specified pairs of load
7641 and store operations (f.ex. load-load, or store-load), all of the first
7642 operations preceding the barrier will complete before any of the second
7643 operations succeeding the barrier begin. Specifically the semantics for each
7644 pairing is as follows:</p>
7647 <li><tt>ll</tt>: All loads before the barrier must complete before any load
7648 after the barrier begins.</li>
7649 <li><tt>ls</tt>: All loads before the barrier must complete before any
7650 store after the barrier begins.</li>
7651 <li><tt>ss</tt>: All stores before the barrier must complete before any
7652 store after the barrier begins.</li>
7653 <li><tt>sl</tt>: All stores before the barrier must complete before any
7654 load after the barrier begins.</li>
7657 <p>These semantics are applied with a logical "and" behavior when more than one
7658 is enabled in a single memory barrier intrinsic.</p>
7660 <p>Backends may implement stronger barriers than those requested when they do
7661 not support as fine grained a barrier as requested. Some architectures do
7662 not need all types of barriers and on such architectures, these become
7667 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7668 %ptr = bitcast i8* %mallocP to i32*
7671 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
7672 call void @llvm.memory.barrier(i1 false, i1 true, i1 false, i1 false, i1 true)
7673 <i>; guarantee the above finishes</i>
7674 store i32 8, %ptr <i>; before this begins</i>
7679 <!-- _______________________________________________________________________ -->
7681 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
7687 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
7688 any integer bit width and for different address spaces. Not all targets
7689 support all bit widths however.</p>
7692 declare i8 @llvm.atomic.cmp.swap.i8.p0i8(i8* <ptr>, i8 <cmp>, i8 <val>)
7693 declare i16 @llvm.atomic.cmp.swap.i16.p0i16(i16* <ptr>, i16 <cmp>, i16 <val>)
7694 declare i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* <ptr>, i32 <cmp>, i32 <val>)
7695 declare i64 @llvm.atomic.cmp.swap.i64.p0i64(i64* <ptr>, i64 <cmp>, i64 <val>)
7699 <p>This loads a value in memory and compares it to a given value. If they are
7700 equal, it stores a new value into the memory.</p>
7703 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
7704 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
7705 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
7706 this integer type. While any bit width integer may be used, targets may only
7707 lower representations they support in hardware.</p>
7710 <p>This entire intrinsic must be executed atomically. It first loads the value
7711 in memory pointed to by <tt>ptr</tt> and compares it with the
7712 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
7713 memory. The loaded value is yielded in all cases. This provides the
7714 equivalent of an atomic compare-and-swap operation within the SSA
7719 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7720 %ptr = bitcast i8* %mallocP to i32*
7723 %val1 = add i32 4, 4
7724 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 4, %val1)
7725 <i>; yields {i32}:result1 = 4</i>
7726 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7727 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7729 %val2 = add i32 1, 1
7730 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32(i32* %ptr, i32 5, %val2)
7731 <i>; yields {i32}:result2 = 8</i>
7732 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
7734 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
7739 <!-- _______________________________________________________________________ -->
7741 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
7747 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
7748 integer bit width. Not all targets support all bit widths however.</p>
7751 declare i8 @llvm.atomic.swap.i8.p0i8(i8* <ptr>, i8 <val>)
7752 declare i16 @llvm.atomic.swap.i16.p0i16(i16* <ptr>, i16 <val>)
7753 declare i32 @llvm.atomic.swap.i32.p0i32(i32* <ptr>, i32 <val>)
7754 declare i64 @llvm.atomic.swap.i64.p0i64(i64* <ptr>, i64 <val>)
7758 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
7759 the value from memory. It then stores the value in <tt>val</tt> in the memory
7760 at <tt>ptr</tt>.</p>
7763 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
7764 the <tt>val</tt> argument and the result must be integers of the same bit
7765 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
7766 integer type. The targets may only lower integer representations they
7770 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
7771 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
7772 equivalent of an atomic swap operation within the SSA framework.</p>
7776 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7777 %ptr = bitcast i8* %mallocP to i32*
7780 %val1 = add i32 4, 4
7781 %result1 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val1)
7782 <i>; yields {i32}:result1 = 4</i>
7783 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
7784 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
7786 %val2 = add i32 1, 1
7787 %result2 = call i32 @llvm.atomic.swap.i32.p0i32(i32* %ptr, i32 %val2)
7788 <i>; yields {i32}:result2 = 8</i>
7790 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
7791 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
7796 <!-- _______________________________________________________________________ -->
7798 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
7804 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
7805 any integer bit width. Not all targets support all bit widths however.</p>
7808 declare i8 @llvm.atomic.load.add.i8.p0i8(i8* <ptr>, i8 <delta>)
7809 declare i16 @llvm.atomic.load.add.i16.p0i16(i16* <ptr>, i16 <delta>)
7810 declare i32 @llvm.atomic.load.add.i32.p0i32(i32* <ptr>, i32 <delta>)
7811 declare i64 @llvm.atomic.load.add.i64.p0i64(i64* <ptr>, i64 <delta>)
7815 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
7816 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
7819 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7820 and the second an integer value. The result is also an integer value. These
7821 integer types can have any bit width, but they must all have the same bit
7822 width. The targets may only lower integer representations they support.</p>
7825 <p>This intrinsic does a series of operations atomically. It first loads the
7826 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
7827 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
7831 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7832 %ptr = bitcast i8* %mallocP to i32*
7834 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 4)
7835 <i>; yields {i32}:result1 = 4</i>
7836 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 2)
7837 <i>; yields {i32}:result2 = 8</i>
7838 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32(i32* %ptr, i32 5)
7839 <i>; yields {i32}:result3 = 10</i>
7840 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
7845 <!-- _______________________________________________________________________ -->
7847 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
7853 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
7854 any integer bit width and for different address spaces. Not all targets
7855 support all bit widths however.</p>
7858 declare i8 @llvm.atomic.load.sub.i8.p0i32(i8* <ptr>, i8 <delta>)
7859 declare i16 @llvm.atomic.load.sub.i16.p0i32(i16* <ptr>, i16 <delta>)
7860 declare i32 @llvm.atomic.load.sub.i32.p0i32(i32* <ptr>, i32 <delta>)
7861 declare i64 @llvm.atomic.load.sub.i64.p0i32(i64* <ptr>, i64 <delta>)
7865 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
7866 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
7869 <p>The intrinsic takes two arguments, the first a pointer to an integer value
7870 and the second an integer value. The result is also an integer value. These
7871 integer types can have any bit width, but they must all have the same bit
7872 width. The targets may only lower integer representations they support.</p>
7875 <p>This intrinsic does a series of operations atomically. It first loads the
7876 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
7877 result to <tt>ptr</tt>. It yields the original value stored
7878 at <tt>ptr</tt>.</p>
7882 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7883 %ptr = bitcast i8* %mallocP to i32*
7885 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 4)
7886 <i>; yields {i32}:result1 = 8</i>
7887 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 2)
7888 <i>; yields {i32}:result2 = 4</i>
7889 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32(i32* %ptr, i32 5)
7890 <i>; yields {i32}:result3 = 2</i>
7891 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
7896 <!-- _______________________________________________________________________ -->
7898 <a name="int_atomic_load_and">
7899 '<tt>llvm.atomic.load.and.*</tt>' Intrinsic
7902 <a name="int_atomic_load_nand">
7903 '<tt>llvm.atomic.load.nand.*</tt>' Intrinsic
7906 <a name="int_atomic_load_or">
7907 '<tt>llvm.atomic.load.or.*</tt>' Intrinsic
7910 <a name="int_atomic_load_xor">
7911 '<tt>llvm.atomic.load.xor.*</tt>' Intrinsic
7918 <p>These are overloaded intrinsics. You can
7919 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
7920 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
7921 bit width and for different address spaces. Not all targets support all bit
7925 declare i8 @llvm.atomic.load.and.i8.p0i8(i8* <ptr>, i8 <delta>)
7926 declare i16 @llvm.atomic.load.and.i16.p0i16(i16* <ptr>, i16 <delta>)
7927 declare i32 @llvm.atomic.load.and.i32.p0i32(i32* <ptr>, i32 <delta>)
7928 declare i64 @llvm.atomic.load.and.i64.p0i64(i64* <ptr>, i64 <delta>)
7932 declare i8 @llvm.atomic.load.or.i8.p0i8(i8* <ptr>, i8 <delta>)
7933 declare i16 @llvm.atomic.load.or.i16.p0i16(i16* <ptr>, i16 <delta>)
7934 declare i32 @llvm.atomic.load.or.i32.p0i32(i32* <ptr>, i32 <delta>)
7935 declare i64 @llvm.atomic.load.or.i64.p0i64(i64* <ptr>, i64 <delta>)
7939 declare i8 @llvm.atomic.load.nand.i8.p0i32(i8* <ptr>, i8 <delta>)
7940 declare i16 @llvm.atomic.load.nand.i16.p0i32(i16* <ptr>, i16 <delta>)
7941 declare i32 @llvm.atomic.load.nand.i32.p0i32(i32* <ptr>, i32 <delta>)
7942 declare i64 @llvm.atomic.load.nand.i64.p0i32(i64* <ptr>, i64 <delta>)
7946 declare i8 @llvm.atomic.load.xor.i8.p0i32(i8* <ptr>, i8 <delta>)
7947 declare i16 @llvm.atomic.load.xor.i16.p0i32(i16* <ptr>, i16 <delta>)
7948 declare i32 @llvm.atomic.load.xor.i32.p0i32(i32* <ptr>, i32 <delta>)
7949 declare i64 @llvm.atomic.load.xor.i64.p0i32(i64* <ptr>, i64 <delta>)
7953 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
7954 the value stored in memory at <tt>ptr</tt>. It yields the original value
7955 at <tt>ptr</tt>.</p>
7958 <p>These intrinsics take two arguments, the first a pointer to an integer value
7959 and the second an integer value. The result is also an integer value. These
7960 integer types can have any bit width, but they must all have the same bit
7961 width. The targets may only lower integer representations they support.</p>
7964 <p>These intrinsics does a series of operations atomically. They first load the
7965 value stored at <tt>ptr</tt>. They then do the bitwise
7966 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
7967 original value stored at <tt>ptr</tt>.</p>
7971 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
7972 %ptr = bitcast i8* %mallocP to i32*
7973 store i32 0x0F0F, %ptr
7974 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32(i32* %ptr, i32 0xFF)
7975 <i>; yields {i32}:result0 = 0x0F0F</i>
7976 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32(i32* %ptr, i32 0xFF)
7977 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
7978 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32(i32* %ptr, i32 0F)
7979 <i>; yields {i32}:result2 = 0xF0</i>
7980 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32(i32* %ptr, i32 0F)
7981 <i>; yields {i32}:result3 = FF</i>
7982 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
7987 <!-- _______________________________________________________________________ -->
7989 <a name="int_atomic_load_max">
7990 '<tt>llvm.atomic.load.max.*</tt>' Intrinsic
7993 <a name="int_atomic_load_min">
7994 '<tt>llvm.atomic.load.min.*</tt>' Intrinsic
7997 <a name="int_atomic_load_umax">
7998 '<tt>llvm.atomic.load.umax.*</tt>' Intrinsic
8001 <a name="int_atomic_load_umin">
8002 '<tt>llvm.atomic.load.umin.*</tt>' Intrinsic
8009 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
8010 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
8011 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
8012 address spaces. Not all targets support all bit widths however.</p>
8015 declare i8 @llvm.atomic.load.max.i8.p0i8(i8* <ptr>, i8 <delta>)
8016 declare i16 @llvm.atomic.load.max.i16.p0i16(i16* <ptr>, i16 <delta>)
8017 declare i32 @llvm.atomic.load.max.i32.p0i32(i32* <ptr>, i32 <delta>)
8018 declare i64 @llvm.atomic.load.max.i64.p0i64(i64* <ptr>, i64 <delta>)
8022 declare i8 @llvm.atomic.load.min.i8.p0i8(i8* <ptr>, i8 <delta>)
8023 declare i16 @llvm.atomic.load.min.i16.p0i16(i16* <ptr>, i16 <delta>)
8024 declare i32 @llvm.atomic.load.min.i32.p0i32(i32* <ptr>, i32 <delta>)
8025 declare i64 @llvm.atomic.load.min.i64.p0i64(i64* <ptr>, i64 <delta>)
8029 declare i8 @llvm.atomic.load.umax.i8.p0i8(i8* <ptr>, i8 <delta>)
8030 declare i16 @llvm.atomic.load.umax.i16.p0i16(i16* <ptr>, i16 <delta>)
8031 declare i32 @llvm.atomic.load.umax.i32.p0i32(i32* <ptr>, i32 <delta>)
8032 declare i64 @llvm.atomic.load.umax.i64.p0i64(i64* <ptr>, i64 <delta>)
8036 declare i8 @llvm.atomic.load.umin.i8.p0i8(i8* <ptr>, i8 <delta>)
8037 declare i16 @llvm.atomic.load.umin.i16.p0i16(i16* <ptr>, i16 <delta>)
8038 declare i32 @llvm.atomic.load.umin.i32.p0i32(i32* <ptr>, i32 <delta>)
8039 declare i64 @llvm.atomic.load.umin.i64.p0i64(i64* <ptr>, i64 <delta>)
8043 <p>These intrinsics takes the signed or unsigned minimum or maximum of
8044 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
8045 original value at <tt>ptr</tt>.</p>
8048 <p>These intrinsics take two arguments, the first a pointer to an integer value
8049 and the second an integer value. The result is also an integer value. These
8050 integer types can have any bit width, but they must all have the same bit
8051 width. The targets may only lower integer representations they support.</p>
8054 <p>These intrinsics does a series of operations atomically. They first load the
8055 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
8056 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
8057 yield the original value stored at <tt>ptr</tt>.</p>
8061 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
8062 %ptr = bitcast i8* %mallocP to i32*
8064 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32(i32* %ptr, i32 -2)
8065 <i>; yields {i32}:result0 = 7</i>
8066 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32(i32* %ptr, i32 8)
8067 <i>; yields {i32}:result1 = -2</i>
8068 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32(i32* %ptr, i32 10)
8069 <i>; yields {i32}:result2 = 8</i>
8070 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32(i32* %ptr, i32 30)
8071 <i>; yields {i32}:result3 = 8</i>
8072 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
8079 <!-- ======================================================================= -->
8081 <a name="int_memorymarkers">Memory Use Markers</a>
8086 <p>This class of intrinsics exists to information about the lifetime of memory
8087 objects and ranges where variables are immutable.</p>
8089 <!-- _______________________________________________________________________ -->
8091 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8098 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
8102 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8103 object's lifetime.</p>
8106 <p>The first argument is a constant integer representing the size of the
8107 object, or -1 if it is variable sized. The second argument is a pointer to
8111 <p>This intrinsic indicates that before this point in the code, the value of the
8112 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8113 never be used and has an undefined value. A load from the pointer that
8114 precedes this intrinsic can be replaced with
8115 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8119 <!-- _______________________________________________________________________ -->
8121 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8128 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
8132 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8133 object's lifetime.</p>
8136 <p>The first argument is a constant integer representing the size of the
8137 object, or -1 if it is variable sized. The second argument is a pointer to
8141 <p>This intrinsic indicates that after this point in the code, the value of the
8142 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
8143 never be used and has an undefined value. Any stores into the memory object
8144 following this intrinsic may be removed as dead.
8148 <!-- _______________________________________________________________________ -->
8150 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8157 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
8161 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8162 a memory object will not change.</p>
8165 <p>The first argument is a constant integer representing the size of the
8166 object, or -1 if it is variable sized. The second argument is a pointer to
8170 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8171 the return value, the referenced memory location is constant and
8176 <!-- _______________________________________________________________________ -->
8178 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8185 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
8189 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8190 a memory object are mutable.</p>
8193 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8194 The second argument is a constant integer representing the size of the
8195 object, or -1 if it is variable sized and the third argument is a pointer
8199 <p>This intrinsic indicates that the memory is mutable again.</p>
8205 <!-- ======================================================================= -->
8207 <a name="int_general">General Intrinsics</a>
8212 <p>This class of intrinsics is designed to be generic and has no specific
8215 <!-- _______________________________________________________________________ -->
8217 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8224 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
8228 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8231 <p>The first argument is a pointer to a value, the second is a pointer to a
8232 global string, the third is a pointer to a global string which is the source
8233 file name, and the last argument is the line number.</p>
8236 <p>This intrinsic allows annotation of local variables with arbitrary strings.
8237 This can be useful for special purpose optimizations that want to look for
8238 these annotations. These have no other defined use, they are ignored by code
8239 generation and optimization.</p>
8243 <!-- _______________________________________________________________________ -->
8245 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8251 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8252 any integer bit width.</p>
8255 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
8256 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
8257 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
8258 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
8259 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
8263 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8266 <p>The first argument is an integer value (result of some expression), the
8267 second is a pointer to a global string, the third is a pointer to a global
8268 string which is the source file name, and the last argument is the line
8269 number. It returns the value of the first argument.</p>
8272 <p>This intrinsic allows annotations to be put on arbitrary expressions with
8273 arbitrary strings. This can be useful for special purpose optimizations that
8274 want to look for these annotations. These have no other defined use, they
8275 are ignored by code generation and optimization.</p>
8279 <!-- _______________________________________________________________________ -->
8281 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8288 declare void @llvm.trap()
8292 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8298 <p>This intrinsics is lowered to the target dependent trap instruction. If the
8299 target does not have a trap instruction, this intrinsic will be lowered to
8300 the call of the <tt>abort()</tt> function.</p>
8304 <!-- _______________________________________________________________________ -->
8306 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8313 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
8317 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8318 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8319 ensure that it is placed on the stack before local variables.</p>
8322 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8323 arguments. The first argument is the value loaded from the stack
8324 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8325 that has enough space to hold the value of the guard.</p>
8328 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
8329 the <tt>AllocaInst</tt> stack slot to be before local variables on the
8330 stack. This is to ensure that if a local variable on the stack is
8331 overwritten, it will destroy the value of the guard. When the function exits,
8332 the guard on the stack is checked against the original guard. If they are
8333 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8338 <!-- _______________________________________________________________________ -->
8340 <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8347 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <type>)
8348 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <type>)
8352 <p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8353 the optimizers to determine at compile time whether a) an operation (like
8354 memcpy) will overflow a buffer that corresponds to an object, or b) that a
8355 runtime check for overflow isn't necessary. An object in this context means
8356 an allocation of a specific class, structure, array, or other object.</p>
8359 <p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8360 argument is a pointer to or into the <tt>object</tt>. The second argument
8361 is a boolean 0 or 1. This argument determines whether you want the
8362 maximum (0) or minimum (1) bytes remaining. This needs to be a literal 0 or
8363 1, variables are not allowed.</p>
8366 <p>The <tt>llvm.objectsize</tt> intrinsic is lowered to either a constant
8367 representing the size of the object concerned, or <tt>i32/i64 -1 or 0</tt>,
8368 depending on the <tt>type</tt> argument, if the size cannot be determined at
8377 <!-- *********************************************************************** -->
8380 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8381 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8382 <a href="http://validator.w3.org/check/referer"><img
8383 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8385 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8386 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8387 Last modified: $Date$